Professional Documents
Culture Documents
ECOLOGICAL MODELING
No part of this digital document may be reproduced, stored in a retrieval system or transmitted in any form or
by any means. The publisher has taken reasonable care in the preparation of this digital document, but makes no
expressed or implied warranty of any kind and assumes no responsibility for any errors or omissions. No
liability is assumed for incidental or consequential damages in connection with or arising out of information
contained herein. This digital document is sold with the clear understanding that the publisher is not engaged in
rendering legal, medical or any other professional services.
ENVIRONMENTAL SCIENCE,
ENGINEERING AND TECHNOLOGY
ECOLOGICAL MODELING
WENJUN ZHANG
EDITOR
All rights reserved. No part of this book may be reproduced, stored in a retrieval system or
transmitted in any form or by any means: electronic, electrostatic, magnetic, tape, mechanical
photocopying, recording or otherwise without the written permission of the Publisher.
For permission to use material from this book please contact us:
Telephone 631-231-7269; Fax 631-231-8175
Web Site: http://www.novapublishers.com
Independent verification should be sought for any data, advice or recommendations contained in
this book. In addition, no responsibility is assumed by the publisher for any injury and/or damage
to persons or property arising from any methods, products, instructions, ideas or otherwise
contained in this publication.
This publication is designed to provide accurate and authoritative information with regard to the
subject matter covered herein. It is sold with the clear understanding that the Publisher is not
engaged in rendering legal or any other professional services. If legal or any other expert
assistance is required, the services of a competent person should be sought. FROM A
DECLARATION OF PARTICIPANTS JOINTLY ADOPTED BY A COMMITTEE OF THE
AMERICAN BAR ASSOCIATION AND A COMMITTEE OF PUBLISHERS.
Additional color graphics may be available in the e-book version of this book.
Preface vii
Chapter 1 Artificial Neural Network Simulation of Spatial Distribution of
Arthropods: A Multi-Model Comparison 1
WenJun Zhang and GuangHua Liu
Chapter 2 Multispectral Vegetation Indices in Remote Sensing:
An Overview 15
George P. Petropoulos and
Chariton Kalaitzidis
Chapter 3 Development of a Decision Support System for the Estimation of
Surface Water Pollution Risk From Olive Mill Waste Discharges 41
Anas Altartouri, Kalliope Pediaditi, George P. Petropoulos,
Dimitris Zianis and Nikos Boretos
Chapter 4 Analysis of Green Oak Leaf Roller Population Dynamics
in Various Locations 65
L. V. Nedorezov
Chapter 5 Individual Based Modelling of Planktonic Organisms 83
Daniela Cianelli, Marco Uttieri
and Enrico Zambianchi
Chapter 6 The Effectiveness of Artificial Neural Networks
in Modelling the Nutritional Ecology of a Blowfly Species 97
Michael J. Watts, Andre Bianconi,
Adriane Beatriz S. Serapiao, Jose S. Govone
and Claudio J. Von Zuben
Chapter 7 Development and Utility of an Ecological-based
Decision-Support System for Managing Mixed Coniferous
Forest Stands for Multiple Objectives 115
Peter F. Newton
vi Contents
Ecological modeling is a fast growing science. More and more innovative methodologies
and theories on ecological modeling are emerging around the world. This book presents
models, methods and theories on ecological modeling. Topics discussed include artificial
neural networks; individual-based modeling; ecological nich models; landscape and GIS
modeling; population dynamics; nutritional ecology; remote sensing and decision support
systems. This book reflects recent achievements of scientists in the study of ecological
modeling.
Chapter 1 - Probability distribution functions have been widely used to model the spatial
distribution of arthropods. Aggregation types (i.e., randomly distributed, uniformly
distributed, aggregately distributed, etc.) of arthropods can be detected based on probability
distribution functions, but the abundance at given location is not able to be predicted by them.
This study aimed to present an artificial neural network to simulate spatial distribution of
arthropods. Response surface model and spline function were compared and evaluated against
the neural network model for their simulation performance.
The results showed that the artificial neural network exhibited good simulation
performance. Simulated spatial distribution was highly in accordant with the observed one.
Overall the neural network performed better in the case of lower total abundance of
arthropods. Response surface model could fit the spatial distribution of arthropods but the
simulation performance was worse than neural network. Cross validation revealed that neural
network performed better than response surface model and spline function in predicting
spatial distribution of arthropods. Confidence interval of predicted abundance could be
obtained using randomized submission of quadrate sequences in the neural network
simulation. It is concluded that artificial neural network is a valuable model to simulate the
spatial distribution of arthropods.
Chapter 2 – Remote sensing has generally demonstrated a great potential in mapping
spatial patterns of vegetation. By employing the amount of reflected radiation at particular
regions of the electromagnetic spectrum, it is possible to make estimates on certain
characteristic of vegetation. The use of radiometric vegetation indices is a fast and efficient
method for vegetation monitoring, exploiting information acquired from remote sensing data.
These indices are dimensionless radiometric measures that generally function as indicators of
relative abundance and activity of green vegetation.
Throughout the years, a large number of multispectral vegetation indices have been
formulated. Each has variable degree of efficiency in estimating one or more vegetation
viii WenJun Zhang
parameters such as, health status, nutrient or water deficiency, crop yield, vegetation cover
fraction, leaf area index, absorbed photosynthetically active radiation, net primary production
and above-ground biomass. Additionally some of them also consider atmospheric effects and/
or the soil background for an enhanced retrieval. The present chapter aims in providing an
overview on the use of radiometric vegetation indices developed over the last few decades,
utilizing spectral information acquired from multispectral optical remote sensing sensors.
This overview is preceded an introduction to some important principles of remote sensing
relevant to the vegetation spectral response is made available, as this was considered
necessary to better understand the context of the present overview.
Chapter 3 – According to the Water Framework Directive (WFD, 2000/60/EC),
Integrated River Basin Management Plans (RBMP) are required at different scales, in order to
prevent amongst other things, water resource deterioration and ensure water pollution
reduction. An integrated river basin management approach underpins a risk-based land
management framework for all activities within a spatial land-use planning framework. To
this end, a risk assessment methodology is required to identify water pollution hazards in
order to set appropriate environmental objectives and in turn design suitable mitigation
measures. Surface water pollution as a result of Olive Mill Waste (OMW) discharge is a
serious hazard in the olive oil producing regions of the Mediterranean. However, there is no
standardised method to assess the risk of water pollution from olive mill waste for any given
river basin. The present chapter shows the results from a study conducted addressing the
above issue by designing a detailed risk assessment methodology, which utilises GIS
modelling to classify within a watershed individual sub-catchment risk of water pollution
occurring from olive mill waste discharges. The chapter presents the proposed criteria and
calculations required to estimate sub-catchment risk significance and comments on the
methods potential for wider application. It combines elements from risk assessment
frameworks, Multi Criteria Analysis (MCA), and Geographic Information Systems (GIS).
MCA is used to aggregate different aspects and elements associated with this environmental
problem, while GIS modeling tools helped in obtaining many criterion values and providing
insight into how different objects interact in nature and how these interactions influence risk
at the watershed level. The proposed method was trialed in the Keritis watershed in Crete,
Greece and the results indicated that this method has the potential to be a useful guide to
prioritise risk management actions and mitigation measures which can subsequently be
incorporated in river basin management plans.
Chapter 4 - Publication is devoted to the problem of population time series analysis with
various discrete time models of population dynamics. Applications of various statistical
criterions, which are normally used for determination of mathematical model parameters, are
under the discussion. With a particular example on green oak leaf roller (Tortrix viridana L.)
population fluctuations, which had been presented in publications by Rubtsov (1992), and
Korzukhin and Semevskiy (1992) for three different locations in Europe, the possibilities of
considering approach to the analysis of population dynamics are demonstrated. For
approximations of empirical datasets the well-known models of population dynamics with a
discrete time (Kostitzin model, Skellam model, Moran – Ricker model, Morris – Varley –
Gradwell model, and discrete logistic model) were applied. For every model the final decision
about the possibility to use the concrete model for approximation of datasets are based on
analyses of deviations between theoretical (model) and empirical trajectories: the
correspondence of distribution of deviations to Normal distribution with zero average was
Preface ix
checked with Kolmogorov – Smirnov and Shapiro – Wilk tests, and existence/absence of
serial correlation was determined with Durbin – Watson criteria. It was shown that for two
experimental trajectories Kostitzin model and discrete logistic model give good
approximations; it means that population dynamics can be explained as a result of influence
of intra-population self-regulative mechanisms only. The third considering empirical
trajectory needs in use more complicated mathematical models for fitting.
Chapter 5 - In the last decades, numerical modelling has gained increasing consensus in
the scientific world, and particularly in the framework of behavioural and population ecology.
Through numerical models it is possible to reconstruct what is observed in the environment or
in the laboratory and to get a more in-depth comprehension of the factors regulating the
phenomena under examination.
Numerous approaches have been developed in this framework, but probably one of the
most promising is the individual-based modelling. With this type of approach it is relatively
straightforward to investigate aspects related to the ecology of a population starting from the
characterisation of processes taking place at the scale of the individual organism.
This contribution is intended to provide a general view of the main features of the
individual-based models and of their peculiarities in comparison to other modelling strategies.
Special emphasis will be given to applications in the field of phyto- and zooplankton ecology
and behaviour, and results from the available literature on this topic will be used as examples.
Chapter 6 - The larval phase of most blowfly species is considered a critical
developmental period in which intense limitation of feeding resources frequently occurs.
Furthermore, such a period is characterised by complex ecological processes occurring at
both individual and population levels. These processes have been analysed by means of
traditional statistical techniques such as simple and multiple linear regression models.
Nonetheless, it has been suggested that some important explanatory variables could well
introduce non-linearity into the modelling of the nutritional ecology of blowflies. In this
context, dynamic aspects of the life history of blowflies could be clarified and detailed by the
deployment of machine learning approaches such as artificial neural networks (ANNs), which
are mathematical tools widely applied to the resolution of complex problems. A
distinguishing feature of neural network models is that their effective implementation is not
precluded by the theoretical distribution of the data used. Therefore, the principal aim of this
investigation was to use neural network models (namely multi-layer perceptrons and fuzzy
neural networks) in order to ascertain whether these tools would be able to outperform a
general quadratic model (that is, a second-order regression model with three predictor
variables) in predicting pupal weight values (outputs) of experimental populations of
Chrysomya megacephala (F.) (Diptera: Calliphoridae), using initial larval density (number of
larvae), amount of available food, and pupal size as input variables. These input variables
may have generated non-linear variation in the output values, and fuzzy neural networks
provided more accurate outcomes than the general quadratic model (i.e. the statistical model).
The superiority of fuzzy neural networks over a regression-based statistical method does
represent an important fact, because more accurate models may well clarify several intricate
aspects regarding the nutritional ecology of blowflies. Additionally, the extraction of fuzzy
rules from the fuzzy neural networks provided an easily comprehensible way of describing
what the networks had learnt.
Chapter 7 - An ecological-based decision-support system and corresponding algorithmic
analogue for managing natural black spruce (Picea mariana (Mill) BSP.) and jack pine (Pinus
x WenJun Zhang
banksiana Lamb.) mixed stands was developed. The integrated hierarchical system consisted
of six sequentially-linked estimation modules. The first module consisted of a key set of
empirical yield-density relationships and theoretically-based functions derived from allometry
and self-thinning theory that were used to describe overall stand dynamics including temporal
size-density interrelationships and expected stand development trajectories. The second
module was comprised of a Weibull-based parameter prediction equation system and an
accompanying composite height-diameter function that were used to recover diameter and
height distributions. The third module included a set of species-specific composite taper
equations that were used to derive log product distributions and volumetric yields. The fourth
module was composed of a set of species-specific allometric-based composite biomass
equations that were used to estimate mass distributions and associated carbon-based
equivalents for each above-ground component (bark, stem, branch and foliage). The fifth
module incorporated a set of species-specific end-product and value equations that were used
to predict chip and lumber volumes and associated monetary equivalents by sawmill type
(stud and randomized length mill configurations). The sixth module encompassed a set of
species-specific composite equations that were used to derive wood and log quality metrics
(specific gravity and mean maximum branch diameter, respectively). The stand dynamic and
structural recovery modules were developed employing 382 stand-level measurements
derived from 155 permanent and temporary sample plots situated throughout the central
portion of the Canadian Boreal Forest Region, the taper and end-product modules were
developed employing published results from taper and sawmill simulation studies, and the
biomass and fibre attribute modules were developed using data from density control
experiments.
The potential of the system in facilitating the transformative change towards the
production of higher value end-products and a broader array of ecosystem services was
exemplified by simultaneously contrasting the consequences of density management regimes
involving commercial thinning treatments in terms of overall productivity, end-product
yields, economic efficiency, and ecological impact. This integration of quantitative
relationships derived from applied ecology, plant population biology and forest science into a
common analytical platform, illustrates the synergy that can be realized through a multi-
disciplinary approach to forest modeling.
Chapter 8 – The authors present a review of the concepts and methods associated to
ecological niche modeling illustrated with the published works on amphibians and reptiles of
the Mediterranean Basin, one of the world's biodiversity hotspots for conservation priorities.
They start by introducing ecological niche models, analyzing the various concepts of niche
and the modeling methods associated to each of them. The authors list some conceptual and
practical steps that should be followed when modeling, and highlight the pitfalls that should
be avoided. The authors then outline the history of ecological modeling of Mediterranean
amphibians and reptiles, including a variety of aspects: identification of the ecological niche;
detection of common distribution areas (chorotypes) and other biogeographical patterns;
analysis and prediction of species richness patterns; analysis of the expansion of native and
invasive species; integration of molecular data with spatial modeling; identification of contact
zones between related taxa; assessment of species' conservation status; and prediction of
future conservation problems, including the effects of global change. They conclude this
review with a discussion of the research that still needs to be developed in this area.
Preface xi
landscape pattern indices are numerous, and the need for a synthetic representation is more
and more impelling.
I provide here the description and application of a novel approach to landscape structural
modelling based on the combined use of GIS (Geographical Information Systems) and
multivariate statistics. First, landscape structure of the study area (Ceno valley, Italy) is
analyzed through 5 patch-based, non-redundant indicators (area, isolation, compactness,
shape complexity, interspersion) with indirect link to functional aspects. Second, PCA
(principal component analysis) is used in order to synthesize structural indicators, and
cartographic output is given. Third, KCA (k-means cluster analysis) is applied in order to
group landscape patches into homogeneous clusters, and again GIS output is supplied. Last,
LDA (linear discriminant analysis) is employed to provide evidence for the differences
among clusters.
This modelling approach provides the chance for a deep and cost-effective exegesis of
landscape structure, with promising consequences on conjecture formulation about functional
aspects as well.
Chapter 14 - Our days, the climatic change, manifested by strong and brutal precipitation,
violent wind and long drought, has as direct consequence to damage the plant canopies
(forests, sylviculture, oasis, pastoral lands and agricultural fields) so menacing the human
feeding either from plants or animals (caprine, ovine, bovine, cameline..), exhausting the
water resources, increasing the need for energy in buildings used for all activities (industrial,
agricultural and services). Which solution the ecological modelling is capable to participate
with, at short and long dated, in order to buffer the climatic change effect and to assume the
need of food and clean energy for human? In this chapter we will present the basic concepts
to model the plant architecture (species, densities, positions and orientation) the most
adaptable to the sudden calamity, the energy use efficiency in building (material of
construction, isolation system, organisation of accessories and apparatus), and the produce of
clean energy from the wind velocity (founding wind sources and evaluating regional wind
potential offshore and inshore, conceptualising wind turbine and testing their efficiencies)
In: Ecological Modeling ISBN: 978-1-61324-567-5
Editor: WenJun Zhang, pp. 1-14 © 2012 Nova Science Publishers, Inc.
Chapter 1
ABSTRACT
Probability distribution functions have been widely used to model the spatial
distribution of arthropods. Aggregation types (i.e., randomly distributed, uniformly
distributed, aggregately distributed, etc.) of arthropods can be detected based on
probability distribution functions, but the abundance at given location is not able to be
predicted by them. This study aimed to present an artificial neural network to simulate
spatial distribution of arthropods. Response surface model and spline function were
compared and evaluated against the neural network model for their simulation
performance.
The results showed that the artificial neural network exhibited good simulation
performance. Simulated spatial distribution was highly in accordant with the observed
one. Overall the neural network performed better in the case of lower total abundance of
arthropods. Response surface model could fit the spatial distribution of arthropods but the
simulation performance was worse than neural network. Cross validation revealed that
neural network performed better than response surface model and spline function in
predicting spatial distribution of arthropods. Confidence interval of predicted abundance
could be obtained using randomized submission of quadrate sequences in the neural
network simulation. It is concluded that artificial neural network is a valuable model to
simulate the spatial distribution of arthropods.
*
Correspondence: zhwj@mail.sysu.edu.cn.
+
Correspondence: ghliu@gdaib.edu.cn.
2 WenJun Zhang and GuangHua Liu
Keywords: Artificial neural network; response surface model; spline function; arthropods;
spatial distribution; simulation.
1. INTRODUCTION
Arthropods account for 90% of global species. Biomass of arthropods reaches 1,000kg
/ha in a temperate grassland, which is lower than plant (20,000kg/ha) and microorganisms
(7,000kg /ha) but higher than mammals (1.2kg/ha) and birds (0.3kg/ha) (Pimental et al.,
1992). They control the structures and functions of ecosystems (Wilson, 1987).Arthropods
have been used as the sensitive indicators of environment health (Brown, 1991; Kremen et al.,
1993).
In ecological research the spatial distribution means the two-dimensional distribution of
animal or plant individuals in the field. Many methods such as GIS (Dantas et al., 2009) and
probability theory are used in the spatial pattern analysis. In the arthropod researches, a lot of
probability distribution functions have been developed and used to describe spatial
distribution of arthropod individuals (Krebs, 1989; Zhang, 2007b). In such methods the
number of individuals found in a sample (plot, quadrat, etc.) is supposed to be a random
variable and the random variable follows some probability distribution, e.g., binomial
distribution, Poisson distribution, negative binomial distribution, Neyman‘s distribution, etc.
Because lack of spatial variables in the function, it can not be used to predict the abundance at
given location (Krebs, 1989; Zhang, 2007b). This is a substantial problem arisen from the use
of probability distribution functions in spatial distribution researches. However, the spatial
information is not available from those models. Due to the lack of theoretical background it is
also hard to construct a mechanistic model that calculates individual distribution from spatial
information. Questions on spatial distribution are therefore data-driven (Schultz and Wieland,
1997). The relationship between individual distribution and spatial information is usually a
nonlinear relationship (Pastor-Barcenas et al., 2005).
Artificial neural networks are known to be flexible and adaptable function approximators
for nonlinear relationships (Bianconi, 2010; Cereghino et al., 2001; Marchant and Onyango,
2003; Acharya et al., 2006; Filippi and Jensen, 2006; Nour et al., 2006; Zhang, 2007a,b;
Zhang and Barrion, 2006; Zhang et al., 2007; Zhang et al., 2008). They can offer the
advantages of simplified and more automated model synthesis and analytical input-output
models (Abdel-Aal, 2004; Tan et al., 2006). A large number of studies were reported
concerning applications of neural networks. For examples, they are considered to be more
effective in time series prediction than previous procedures based on dynamical system theory
(Ballester et al., 2002). They are used in the forecast of short and middle long-term
concentration levels (Viotti et al., 2002), subsurface modeling (Almasri and Kaluarachchi,
2005), modeling hourly temperature with the alternative abductive networks (Abdel-Aal,
2004), modeling sediment transfer (Abrahart and White, 2001), rule extraction (Drumm et al.,
1999) and subsurface drain outflow and nitrate-nitrogen concentration in tile effluent and
surface ozone (Sharma et al., 2003; Pastor-Barcenas, et al., 2005), and estimation of
endoparasitic load using morphological descriptors (Loot et al., 2002), the uses of BP to
describe nitrogen dioxide dispersion (Nagendra and Khare, 2006), and for reservoir
eutrophication prediction (Kuo et al., 2007). Empirical models regained popularity in recent
Artificial Neural Network Simulation of Spatial Distribution of Arthropods 3
years due to the complexity and nonlinearity of ecosystems (Tan et al., 2006). Various
conventional models, including empirical models, were thus used to compare simulation
performances between neural networks and these models. For instance, it demonstrated that
neural network was superior to linear models, generalized additive models, and classification
and regression trees (Moisen and Frescino, 2002). Neural network was proved to outperform
other models like multiple regression, logistic regression, and multiple discriminant model in
predicting the number of salmonids and community composition (McKenna, 2005; Olden et
al., 2006). In the research areas of arthropods or related taxonomic groups, neural networks
have been used to make simulation and prediction. A stream classification based on
characteristic invertebrate species assemblages was also satisfactorily conducted using self-
organizing map neural network (Cereghino et al., 2001). They were used to explain the
observed structure of functional feeding groups of aquatic macro-invertebrates (Jorgensen et
al., 2002); Self-Organizing Map (SOM) neural network was used to determine pest species
assemblages for global regions (Worner and Gevrey, 2006); BP and RBF (radial basis
function) neural networks were used to simulate and predict species richness of rice
arthropods (Zhang and Barrion, 2006), reconstruct spatial pattern of insects (Zhang et al.,
2007), and simulate survival dynamics of insects (Zhang and Zhang, 2008).
This study aimed to present several models, and evaluate their effectiveness in the
simulation of spatial distribution of arthropods. Arthropods were investigated on the
grassland. A neural network and a partial differential equation were developed to model
above-ground distribution of arthropods (Zhang, 2010). Models were validated and compared
for their power in predictability. Details for developing and using neural network were
discussed.
Investigation was conducted on the grassland with 8×8 quadrates. Each quadrate has an
area of 1×1 m2. Arthropods were collected, identified, and counted for every quadrate. Insects
were sorted and identified to order level and the other arthropods were identified to classis
level.
tansig(x) = 2/(1+exp(-2*x))-1,
logsig(x) =1/(1+exp(-x)),
purelin(x) =x,
respectively. Initialization of network, and weights and bias for each layer, was performed by
a function that initializes each layer i (i=1,2,3) according to its own initialization function
(Hagan et al., 1996; Mathworks, 2002; Fecit, 2003). Network was trained using Levenberg-
Marquardt backpropagation algorithm. Desired performance function was mean squared error
performance function (mse). The first and second layers received inputs from input space and
produced outputs for the third layer. There was a closed loop for the third layer. For each
layer, the net input functions calculated the layer‘s net input by combining its weighted inputs
and biases.
where
In eq. (1), x=(x, y) T is the input, u=u(x) is the output; ωi, i=1,2,3; ωij, i,j=1,2; ωi3,
i=1,2,3…, and bi, i=1,2,3, are the parameters.
The artificial neural network was developed using Matlab (Mathworks, 2002). Simulation
performance of the neural network was expressed as mse, Pearson correlation coefficient, and
significance level for the linear regression between the simulated and observed.
The response surface model (He, 2001; Mathworks, 2002), i.e., the trend surface model
(Zhang and Fang, 1982), was also used in present simulation:
where u(x): arthropod abundance (individuals per quadrate); x=(x, y) T: the coordinate of
quadrate; b=(b1,b2)T, c=(c1,c2)T: parametric vectors; a: constant.
Spline interpolation is one of the most efficient interpolation models, among which the
cubic spline interpolation is widely used (Burden and Faires, 2001). The cubic spline function
used in present study was:
where, xi=i+1, Mi=S’’(xi), i=0,1,…,n; li =xi+1-xi, i=0,1,…,n-1. Mi, i=0,1,…,n, were obtained
from three-bending moment equation (Zhang, 2007b).
two-dimensional space (coordinates of quadrate, e.g., (1,2), (5,7), etc.), and the
output space was a one-dimensional space (arthropod abundance).
2) Cross validation. There are several cross validation methods. I used a widely
applicable method (Olden et al., 2006). Using this method, each quadrate was
separately removed from the input set of 64 quadrates, and the remaining quadrates
were used to train model and to predict the removed quadrates using the trained
model. As a consequence, the cross validation may be conducted within the data set
in the same study. Comparisons between the predicted and observed arthropod
abundances were made and Pearson correlation coefficient (r) and statistic
significance were calculated to validate models.
3) Quadrates were submitted to neural network in two ways, i.e., their fixed sequences,
and randomized sequences of quadrates.
3. RESULTS
Most arthropods found on the grassland were insects, which belong to the orders
Homoptera (523 individuals), Orthoptera (230 individuals), Hymenoptera (110 individuals),
Coleoptera (55 individuals), and Diptera (40 individuals), etc. Other arthropods were sparsely
distributed on the grassland. The spatial distribution of arthropods exhibited a saddle-like
shape, which was similar to the most abundant Homoptera (Figure 2).
Using the artificial neural network developed above (Figure1, eq.(1)) to simulate spatial
distributions of arthropods and the most abundant orders Orthoptera, Hymenoptera, and
Homoptera. Neural network was trained by 10000 epochs and the desired accuracy (mse) was
0.00001. The results revealed that neural network exhibited excellent simulation performance.
The simulated spatial distribution perfectly coincided with the observed (intercept≈0,
slope≈1, r≈1, p<0.0001), as illustrated by Figure 3.
Using a deviation function, stmse=mse/u2, where u is the averaged individuals per
quadrate, together with Figure 4, it was found that the lower abundance would overall lead
neural network to yield the better simulation performance (Arthropods (stmse=6.39*10-3),
Homoptera (stmse=2.16*10-2), Orthoptera (stmse=5.88*10-7), Hymenoptera
-6
(stmse=1.46*10 )).
Figure 3. Simulating spatial distribution of arthropods using neural network. Quadrates were submitted
to neural network in their fixed sequences.
Response surface model could better fit the spatial distribution of arthropods. However,
its simulation performance was lower than neural network (Figure 4).
8 WenJun Zhang and GuangHua Liu
The cross validation, in which quadrates were submitted to neural network in their fixed
sequences, demonstrated that neural network exhibited much better performance than
response surface model in predicting unknown quadrates (Figure 5). In most cases, response
surface model yielded the negative correlation between the predicted and observed
abundances. As a result, it is not suggested using response surface model to predict spatial
distribution of arthropods.
Different from the situation above, neural network exhibited better prediction
performance in the case of larger abundance than lower abundance on the grassland (Figure
5). This means that compared to simulation the neural network needs more information to
train itself for producing a reasonable extrapolation.
In an additional cross validation of neural network to predict the spatial distribution of
arthropods, quadrates were submitted in randomized sequences and five randomizations were
used. The results showed that neural network yielded the better performance (r=0.5323,
p<0.0001). More than 60% of quadrates were correctly predicted and they fell inside 95%
confidence intervals of the predicted (Figure 6).
Artificial Neural Network Simulation of Spatial Distribution of Arthropods 9
Figure 5. Cross validation of neural network and response surface model for predicting spatial
distribution of arthropods. Quadrates were submitted to neural network in their fixed sequences.
Figure 6. Cross validation of neural network for predicting distribution of arthropods. Quadrates were
submitted to neural network in randomized sequences. Five randomizations were conducted.
10 WenJun Zhang and GuangHua Liu
The cross validation indicated that spline function performed badly than both neural
network and response surface model (Table 1).
Arthropods Orthoptera
Observed=15.6520-0.0041*Simulated Observed=3.8052-0.0545
Hymenoptera Homoptera
Observed=1.7092+0.0067*Simulated Observed=8.3154-0.0148*Simulated
(layers, neurons, etc.), by training neural network with noise, and by using techniques as
weight decay, and so on (Ozesmi et al., 2006).
Both spatial information (Euclidean coordinates) and environmental factors like plant
composition, climate conditions, etc., can be considered to be input components. By doing
this, we obtain a more interpretative neural network. This model may be used to interpret
data. Some methods on data interpretation of explore neural network, like sensitivity analysis,
inference rule extraction, randomization approach, neural interpretation diagram, etc., are
available now for this purpose (Bradshaw et al., 2002; Olden and Jackson, 2002; Gevrey et
al., 2006).
Similar to the conclusion from present study, most previous studies showed that neural
networks outperfomed conventional models (Lek et al., 1997; Paruelo and Tomasel, 1997;
Brosse et al., 1999; Abrahart and White, 2001; Yu et al., 2006; Zhang, 2007a,b), although
varied results were also produced in using neural networks (Marchant and Onyango, 2003;
Filippi and Jensen, 2006). This study demonstrates that artificial neural network is more
robust than the conventional model in the modelling of spatial distribution. It can provide a
feasible alternative to more classical spatial statistical techniques (Pearson et al., 2002).
However, researches towards complicate spatial distribution are further desired in the future.
This study used the data set of 64 quadrates to model spatial distribution of arthropods
and a better performance was achieved. As fitting models, their simulation performance is
expected to be improved with the increase of the size of data set. However, more studies with
larger data sets are still needed in the future to further validate these models.
ACKNOWLEDGMENTS
This project was supported by ―N ational Basic Research Program of China‖(973
Program)(No. 2006CB102005). We thank all participants of arthropod investigation, Mr. WG
Zhou, HQ Dai, and undergraduates of ecology 2004, Sun Yat-sen University, China.
REFERENCES
Abdel-Aal RE. Hourly temperature forecasting using abductive networks. Engineering
Applications of Artificial Intelligence, 17, 543-556, 2004.
Abrahart RJ, White SM. Modelling Sediment Transfer in Malawi: Comparing
Backpropagation Neural Network Solutions Against a Multiple Linear Regression
Benchmark Using Small Data Set. Phys. Chem. Earth (B), 26(1), 19-24, 2001.
Acharya C, Mohanty S, Sukla LB, et al. Prediction of sulphur removal with Acidithiobacillus
sp. using artificial neural networks. Ecological Modelling, 190(1-2), 223-230, 2006.
Almasri MN, Kaluarachchi JJ. Modular neural networks to predict the nitrate distribution in
ground water using the on-ground nitrogen loading and recharge data. Environmental
Modelling and Software, 20, 851-871, 2005.
Ballester EB, Valls GCI, Carrasco-Rodriguez JL, et al. Effective 1-day ahead prediction of
hourly surface ozone concentrations in eastern Spain using linear models and neural
networks. Ecological Modelling, 156 (1), 27-41, 2002.
12 WenJun Zhang and GuangHua Liu
Bianconi A, Zuben CJV, de Souza Serapião AB, et al. 2010. The use of artificial neural
networks in analysing the nutritional ecology of Chrysomya megacephala (F.) (Diptera:
Calliphoridae), compared with a statistical model. Australian Journal of Entomology, 49:
201–212.
Bradshaw CJA, Davis LS, Purvis M, et al. Using artificial neural networks to model the
suitability of coastline for breeding by New Zealand fur seals (Arctocephalus forsteri).
Ecological Modelling, 148 (2), 111-131, 2002.
Brown KS Jr. Conservation of Neotropical insects: Insects as indicators. In: Collins, NM,
Thomas, JA (eds.), The Conservation of Insects and Their Habitats. Academic Press,
London, 349-404, 1991.
Burden R, Faires JD. Numerical Analysis (Seventh Edition). Thomson Learning, Inc., 2001.
Cereghino R, Giraudel JL, Compin A. Spatial analysis of stream invertebrates distribution in
the Adour-Garonne drainage basin (France), using Kohonen self organizing maps.
Ecological Modelling,146,167–180, 2001.
Dantas A, Yamamoto K, Lamar MV, Yamashita Y. 2009. Neural geo-spatial model: a
strategic planning tool for urban transportation. Transactions of the Wessex Institute,
DOI: 10.2495/UT000251.
Drumm D, Purvis M, Zhou QQ. Spatial ecology and artificial neural networks: modeling the
habitat preference of the sea cucumber (Holothuria leucospilota) on Rarotonga, Cook
Islands. SIRC 99 – The 11th Annual Colloquium of the Spatial Information Research
Centre, University of Otago, Dunedin, New Zealand December 13-15th, 1999.
Fecit. Analysis and Design of Neural Networks in MATLAB 6.5, Electronics Industry Press,
Beijing, China, 2003.
Filippi AM, Jensen JR. Fuzzy learning vector quantization for hyperspectral coastal
vegetation classification. Remote Sensing of Environment, 100, 512-530, 2006.
Gevrey M, Dimopoulos I, Lek S. Two-way interaction of input variables in the sensitivity
analysis of neural network models. Ecological Modelling, 195 (1-2), 43-50, 2006.
Hagan MT, Demuth HB, Beale MH. Neural Network Design. PWS Publishing Company,
USA, 1996.
He RB. MATLAB 6: Engineering Computation and Applications. Chongqing University
Press, Chongqing, China, 2001.
JØrgensen SE, Verdonschot P, Lek S. Explanation of the observed structure of functional
feeding groups of aquatic macro-invertebrates by an ecological model and the maximum
exergy principle. Ecological Modelling, 158 (3),223-231, 2002.
Kilic H, Soyupak S, Tuzun I et al. An automata networks based preprocessing technique for
artificial neural network modelling of primary production levels in reservoirs. Ecological
Modelling, 201 (3-4), 359-368, 2007.
Krebs CJ. Ecological Methodology. HarperCollinsPublishers, New York, 1989.
Kremen C, Colwell RK, Erwin TL, et al. Invertebrate assemblges: their use as indicators in
conservation planning. Conservation Biology, 7, 796-808, 1993.
Kuo JT, Hsieh MH, Lung WS et al. Using artificial neural network for reservoir
eutrophication prediction. Ecological Modelling, 200(1-2), 171-177, 2007.
Lek S, Baran P. Estimations of trout density and biomass: a neural networks approach.
Nonlinear Analysis, Theory, Methods & Applications, 30(8), 4985-4990, 1997.
Artificial Neural Network Simulation of Spatial Distribution of Arthropods 13
Yu R, Leung PS, Bienfang P. Predicting shrimp growth: Artificial neural network versus
nonlinear regression models. Aquacultural Engineering, 34, 26–32, 2006.
Zhang WJ. Computational Ecology: Artificial Neural Networks and Their Applications.
World Scientific, Singapore, 2010.
Zhang WJ. Supervised neural network recognition of habitat zones of rice invertebrates.
Stochastic Environmental Research and Risk Assessment, 21, 729-735, 2007a.
Zhang WJ. Methodology on Ecology Research. Sun Yat-sen University Press, Guangzhou,
China, 2007b.
Zhang WJ, Barrion AT. Function approximation and documentation of sampling data using
artificial neural networks. Environmental Monitoring and Assessment, 122, 185-201,
2006.
Zhang WJ, Bai CJ, Liu GD. Neural network modeling of ecosystems: a case study .on
cabbage growth system. Ecological Modelling, 201,317-325, 2007.
Zhang WJ, Liu GH, Dai HQ. Simulation of food intake dynamics of holometabolous insect
using functional link artificial neural network. Stochastic Environmental Research and
Risk Assessment, 22, 123-133, 2008.
Zhang WJ, Zhang XY. Neural network modeling of survival dynamics of holometabolous
insects: a case study. Ecological Modelling, 211, 433-443, 2008.
Zhang WJ, Zhong XQ, Liu GH. Recognizing spatial distribution patterns of grassland insects:
neural network approaches. Stochastic Environmental Research and Risk Assessment, 22,
207-216 , 2008.
Zhang YT, Fang KT. Introduction to Multivariate Statistics. Science Press, Beijing, China,
1982.
In: Ecological Modeling ISBN: 978-1-61324-567-5
Editor: WenJun Zhang, pp. 15-39 © 2012 Nova Science Publishers, Inc.
Chapter 2
ABSTRACT
Remote sensing has generally demonstrated a great potential in mapping spatial
patterns of vegetation. By employing the amount of reflected radiation at particular
regions of the electromagnetic spectrum, it is possible to make estimates on certain
characteristic of vegetation. The use of radiometric vegetation indices is a fast and
efficient method for vegetation monitoring, exploiting information acquired from remote
sensing data. These indices are dimensionless radiometric measures that generally
function as indicators of relative abundance and activity of green vegetation.
Throughout the years, a large number of multispectral vegetation indices have been
formulated. Each has variable degree of efficiency in estimating one or more vegetation
parameters such as, health status, nutrient or water deficiency, crop yield, vegetation
cover fraction, leaf area index, absorbed photosynthetically active radiation, net primary
production and above-ground biomass. Additionally some of them also consider
atmospheric effects and/ or the soil background for an enhanced retrieval. The present
chapter aims in providing an overview on the use of radiometric vegetation indices
developed over the last few decades, utilizing spectral information acquired from
multispectral optical remote sensing sensors. This overview is preceded an introduction
to some important principles of remote sensing relevant to the vegetation spectral
response is made available, as this was considered necessary to better understand the
context of the present overview.
1. INTRODUCTION
Remote sensing can be generally defined as the technique of gathering information about
an object (or target) without making actual contact with this object. Extraction of information
is done by analyzing the reflected or emitted electromagnetic radiation (EMR) energy of the
object recorded by the sensor of a remote sensing system. The first instances of remote
sensing occured in 1858 with the first aerial photographs, taken from balloons, with the
purpose of being used for mapping and military reconnaissance. Satellite remote sensing
began around 1960‘s with the launch of the first satellite, TIROS-1, which was primarily used
for meteorological purposes. Nowadays, remote sensing data are collected by both
multispectral and hyperspectral sensors operating in both airborne and satellite platforms,
with a very large number of satellite systems operating in orbit over the Earth‘s surface.
The advent of satellite-based remote sensing over the last few decades has lead to a
considerable amount of work being done in determining their potential usefulness in many
disciplines and applications. Among the main advantages of remote sensing include its ability
to provide synoptic views of large areas in a spatially contiguous fashion and in a repetitive
manner, without a disturbing influence on the area to be surveyed and without accessibility
issues to the site to be studied. Remote sensing observations from microwave sensors offer all
weather capability and daytime and nighttime observations, which, combined with their
strong dependence on the dielectric properties of the target, make them potentially very
powerful for the estimation of various parameters, such as of soil moisture content.
Specifically the potential use of remote sensing for mapping vegetation condition and
health has been explored by many scientists over the last decades. A common approach
employed for this purpose involves the use of radiometric vegetation indices computed from
multispectral remote sensing observations. The present chapter aims in providing an overview
of the approaches exploited for the estimation of vegetation health and vigor from the
computation of radiometric vegetation indices from multispectral optical remote sensing
observations. Nevertheless, before that, a discussion to the principles of remote sensing in the
reflective part of the EMR, namely the visible near-infrared (VNIR) and shortwave infrared
(SWIR) is made available. This was deemed necessary, as understanding the characteristics of
EMR and how the latter interacts with land surface targets is crucial in later understanding the
information that is extracted from remote sensing data and subsequently used in the
estimation of vegetation condition by employing these radiometric indices.
The satellite sensor measures the intensity of the electromagnetic waves reflected by the
surface in different parts of the spectrum. Comparison of the radiometric and spectral
characteristics of the reflected energy to the characteristics of the incident energy derives the
surface reflectivity, which is analyzed to determine the physical and chemical properties of
the surface. However, as the solar waves propagate through a planet‘s atmosphere, they
interact with atmospheric constituents, leading to significant effects on the intensity and
spectral composition of the energy recorded by a remote sensor. These effects are caused
principally through the mechanisms of atmospheric scattering and absorption (e.g. Cambell,
1981). Both these effects are directly related to the atmospheric path length and the
wavelengths involved. Scattering arises when particles of various sizes present in the
atmosphere or large gas molecules interact with the EMR resulting to a redirection of the
radiation from its original path. The magnitude of scattering depends on various factors, most
importantly the wavelength of the radiation, the abundance of particles or gases, and the
distance the radiation travels through the atmosphere (e.g. Verbyla, 1995). The atmospheric
absorption normally involves absorption of EMR energy at particular wavelengths by certain
gases present in the atmosphere. The dominant gases responsible for most of this absorption
are water vapor (H2O), carbon dioxide (CO2) and ozone (O3), and depending on atmospheric
conditions and the amount of radiation emitted from the surface one of these effects will be
dominant. Nevertheless, there are certain spectral domains within the EMR spectrum that are
relatively free from the effects of scattering and absorption, known as atmospheric windows.
Figure 1 illustrates these ―absorption-free‖ regions within the EMR, which are very important
in remote sensing, as in these regions atmospheric effects on radiation minimal in comparison
with other wavelengths.
All wavelengths shorter than 0.30 µm are unavailable for remote sensing and strong
absorption bands also exist in the near-infrared, particularly around 1.9, 1.4, 1.12, 0.95 and
0.76 µm. The first significant atmospheric window begins at 0.3 μm, providing good
transparency in the visible spectrum, between 0.30-0.75 μm. The atmospheric window
continues but with interruptions in the NIR region (i.e. 0.77-0.91 µm). In the near-infrared
(NIR) part of the EMR spectrum there are several atmospheric windows in narrow wavebands
between 1.0-1.12, 1.19-1.34, 1.55-1.75 and 2.05-2.4 μm. Consequently, it is natural to expect
that correction of these atmospheric effects can be particularly useful for improving the
quality of the remotely sensed data. Description of such atmospheric correction approaches
can be found elsewhere, including Slater (1980) and Rees (2001).
As the solar waves propagate through Earth‘s atmosphere from these atmospheric
windows, radiation that is not absorbed or scattered in the atmosphere can reach and interact
with the Earth's surface. EMR reaching the Earth‘s surface interacts with the Earth‘s surface
objects by the mechanisms of reflection, transmission or absorption. The interrelationship of
these three parameters is expressed by a direct application of the conservation of energy law,
from the equation below:
E I ( ) E R ( ) EA ( ) ET ( ) , (1)
where EI(λ) denotes the incident energy, ER(λ) is the reflected energy, EA(λ) the absorbed
energy and ET(λ) is the transmitted energy.
There are two very important points that should be considered, regarding the
aforementioned equation. The first one is that that the proportions of energy, which is
reflected, absorbed and transmitted, will vary for different Earth surface targets, depending on
their material type and condition. These differences allow the distinguishing of the different
features recorded on a satellite image. In addition, even for a given feature type, the
proportion of the energy reflected, absorbed and transmitted will vary at different
wavelengths (Elachi, 1987). Thus, two features may be possible to be discriminated in one
spectral region but be very different in another wavelength band. Furthermore, another
important consideration accounted in remote sensing is the geometric manner in which an
object is reflecting energy, a function of the surface roughness of the object. Specular
reflectors are flat surfaces that manifest mirror-like reflections, where the angle of reflection
equals the angle of incidence. Diffuse (or Lambertian) reflectors are rough surfaces that
reflect uniformly in all directions.
The satellite sensor records a digital number (DN) for each pixel and spectral band.
However, for the majority of practical applications in remote sensing, the DN values must be
converted in measurements of the amount of energy reaching the sensor in each band, using
an appropriate sensor calibration equation. This energy is expressed by the radiance (L).
Radiance (L) is considered to be the measure of the radiant flux per unit of solid angle leaving
an extended area source in a given direction per unit projected source area in that direction,
and is measured in Wm-2sr-1. This measure of radiance is obviously made in a particular
viewing direction where the perceived surface brightness is considered constant for the entire
hemisphere above the surface. The measure of the amount of energy reflected from the
surface can be related to the amount of energy arriving to the surface, from the Sun, by
Multispectral Vegetation Indices in Remote Sensing 19
introducing the reflectance of the surface term, which is mathematically defined as (e.g.
Lillesand & Kieffer, 1994):
where reflectance ρλ is expressed as a percentage whereas all the other parameters defined as
in equation 2, above.
Conventional research in remote sensing has concentrated on the use of spectral response
based on nadir or near-nadir reflectance measurements. However, for remote sensing
applications the spectral response of a target will also depend upon factors such as the
orientation of the Sun (solar azimuth), the height of the Sun in the sky (solar elevation angle),
the direction in which the sensor is pointing relative to nadir (the look angle) and the
wavelength used (Sabins, 1997). All these factors are combined in the bidirectional
reflectance distribution function (BRDF), which is a theoretical concept that describes
directional reflectance phenomena by relating the incident irradiance from one given direction
to its contribution to the reflected radiance in another specific direction (Nicodemus et al.,
1977). The bidirectional reflectance distribution function is normally defined as the ratio of
reflected radiance to incident irradiance at a particular wavelength:
where the subscripts i and r denote incident and reflected respectively, is the direction of
light propagation, λ is the wavelength of light, L is radiance, and E is irradiance.
Field devices, called ―f ield goniometers‖, which are essentially goniometric radiometric
instruments have been used for a number of years to practically assess the BRDF of natural
and man-made surfaces under natural illumination conditions (e.g Jackson et al., 1990;
Hosgood et al., 1999). Several bidirectional reflectance distribution function (BRDF) models
have been developed to predict the bidirectional reflectance properties (e.g. Verstraete et al.,
1990; Strahler and Jupp, 1991; Qin, 1993; Albuelgasim and Strahler, 1994). Although
detailed reference to these models is beyond the purposes of this discussion it should be
mentioned that essentially the existing approaches to the analytical modelling of the BRDF
are mainly distinguished in two groups; the physical and the statistical models. Physical
models relate the BRDF to various internal properties of the surface relying on physical
parameters (e.g. Myneni et al., 1990; Goel, 1988). On the other hand, statistical models of the
BRDF of the surface characterize the shape of the BRDF function using statistical parameter
(e.g. Kieffer et al., 1977; Pinty & Ramond, 1986).
Studies have addressed the importance of bidirectional effects in remote sensing (e.g.
Holben and Kimes, 1986; Schaaf and Strahler, 1994). Such studies, among others, showed
that correction / normalization of the sun / view angle effects is very important for
quantitative analysis of remotely sensed data. However, it should be taken into consideration
that, in general, most remote sensing applications assume Lambertian reflectance, which is
usually acceptable. Besides, the influence of the atmospheric conditions must also be
20 George P. Petropoulos and Chariton Kalaitzidis
considered for accurate BRDF determination, which practically is constantly changing in the
different viewing angles (Deering and Eck, 1987).
In remote sensing, by measuring the energy that is reflected by targets on the Earth's
surface over a variety of different wavelengths, it is possible to compile a spectral response
for that object. A spectral reflectance curve describes the spectral response of a target as a
function of wavelength, covering the visible to near-infrared region of the electromagnetic
spectrum. The configuration of spectral reflectance curves is particular important in remote
sensing, as it offers an insight into the spectral characteristics of an object and has a strong
influence on the choice of wavelength regions in which remotely sensed data should be
acquired. This important property makes it possible to identify the different substances or
classes and separate them by their spectral signatures denoted by their spectral curves.
The spectral reflectance curves of some typical surface materials are depicted in Figure 2
and their reflectance properties are briefly discussed below. Figure 2 shows reflectance
spectra for three different surface types of terrain cover, i.e. vegetation, soil and water. The
horizontal axis shows the wavelength of the incident energy, whereas the vertical axis shows
the percentage of incident energy reflected at the different wavelengths. Although these lines
for each cover type in Figure 2 represent average reflectance curves, it is prominent how
distinctive the curves are for each surface feature.
Figure 2. Spectral signatures of some terrestrial materials in the reflective part of the EMR (adopted
from http://www.satreponline.org/landsaf/print.htm#page_3.1.0).
Multispectral Vegetation Indices in Remote Sensing 21
increases. One of the most important parameters controlling reflectance from soil surfaces is
soil moisture content, which inversely related to the reflectance, especially in the mid-infrared
region. Furthermore, dry fine-textured soils (such as clay) will usually have higher reflectance
than dry coarse-textured soils (such as sand) (Verbyla, 2001). In addition, organic matter has
been found to have a strong influence in soil reflectance, where spectral reflectance generally
decreases over the entire shortwave region as organic matter content increases (Stoner and
Baumgardne, 1980). The presence of iron oxide in the soil will also significantly decrease
reflectance, at least in the visible wavelengths (Asrar, 1989). Last but not least, soil particle
size has been found to affect reflectance of particular soil minerals, which is generally
increased and the contrast of the absorption features decrease as the particle size decreases.
Huete (1989) gives a summary of the influences of soil background on measurements of
vegetation spectra.
Finally, the spectral reflectance curve of water shows a general reduction in reflectance
with increasing wavelength, so that in the NIR the reflectance of deep, clear water is
effectively almost zero. Clear water reflects very little in most spectral regions. However,
turbid water reflects significant amounts of radiation, especially in the red and NIR spectral
regions. There is a shift in the spectral reflectance regions, as water increases in turbidity
(Verbyla, 2001). However, the spectral reflectance of water is affected by the presence and
concentration of dissolved and suspended organic and inorganic material, and by the depth of
the water body (Jensen, 2000). Thus, the intensity and distribution of the radiance upwelling
from a water body are indicative of the nature of the dissolved and suspended mater in the
water and of the water depth. The peak of the reflectance curve moves to progressively longer
wavelengths as concentration of these materials increases.
Since 1960‘s with the initial attempt by Jordan (1969) scientists have extracted and
modelled various vegetation biophysical variables from remotely sensed data by exploiting
mathematical formulae, referred to as vegetation indices, defined as dimensionless
radiometric measures that function as indicators of relative abundance and activity of green
vegetation (Jensen, 2000). A vegetation index is effectively a numerical value without units,
resulting from the mathematical combination of radiance values at particular wavebands. The
measurements of those radiance values are almost always collected concurrently, and
represent the state of vegetation at that particular moment in time, under the specific
conditions that were in place during the time the data were acquired. Also, comparison of
vegetation indices between different vegetation targets, or the same vegetation target at
different time and conditions can potentially provide information on the differences between
the targets or the effects of the variable conditions to the vegetation.
The remaining part of this chapter provides an overview on the development of
radiometric vegetation indices developed over the past decades which are based on
multispectral remote sensing observations acquired in the reflective part of the EMR. A
summary of the indices reviewed herein is provided in Table 2.
Multispectral Vegetation Indices in Remote Sensing 23
The first use of a ratio between NIR and red radiance was reported by Jordan (1969),
when he measured transmittance value at the floor of a tropical forest, to estimate the Leaf
Area Index (LAI). As explained by Knipling (1970), increased Leaf Area Index (LAI) values,
result in lower red and higher NIR reflectance. The cause for the first is the increased amount
of chlorophyll in the sensor‘s field of view, leading to increased absorption and reduced
reflectance. The increase in NIR reflectance is attributed to the increased number of cells
present in the field of view, resulting in an increased amount of scattered NIR radiation
reaching the sensor. The ratio was coined as a vegetation index by Pearson and Miller (1972),
when they introduced the Ratio Vegetation Index (RVI), which is the ratio of the NIR over
the red signal (Eq. 4), in order to estimate grass canopy biomass.
NIR
RVI (4)
RED
The simple ratio of NIR reflectance over red reflectance has been shown on numerous
occasions to be related to crop yield and dry matter accumulation (Markham et. al., 1981;
Tucker et. al., 1981).
24 George P. Petropoulos and Chariton Kalaitzidis
An evolved version of the RVI is the Normalised Difference Vegetation Index (NDVI),
introduced for the first time by Deering et et. al. (1975). This index is a ratio between the
difference and the sum of NIR and red radiation (Eq. 5).
NIR red
NDVI (5)
NIR red
Typically NDVI values can scale between -1 to +1 with water surfaces typically having
an NDVI value less than 0, bare soils between 0 and 0.1, clouds about 0.23, snow and ice
about 0.38 and vegetation over 0.1 (Jensen, 2000). NDVI is considered to be superior of the
RVI. The effectiveness especially of NDVI is because chlorophyll absorbs light in the visible
(0.58-0.68 µm) and foliage reflects light in the NIR part of the EMR (0.72-1.10 µm). The
combination of NIR and red reflectance in one index, succeeded in combining information
regarding the chlorophyll content of the vegetation, as well as information about the leaf
anatomy. Therefore, higher photosynthetic activity would result in lower reflectance in the
red channel and higher reflectance in the NIR channel. Also, the normalisation of the
difference makes the index more robust and less affected by variations in the illumination
intensity, allowing for comparisons of target located at different locations, with data acquired
at different times.
NDVI became very popular and was probably the most broadly used index in many
applications related with vegetation. NDVI has been linked in many studies in a positive
correlation with the amount of green biomass, leaf area index, vegetation percentage cover,
plant vigour and health, plant stress, photosynthetic activity and agricultural crop yield (Asrar
et. al., 1984; Sellers, 1987). However, it should be mentioned here that NDVI values can vary
significantly as a function of sensor calibration (Goward et al., 1991), atmospheric conditions
(Myneni and Asrar, 1994), directional surface reflectance effects (Holben et al., 1986), terrain
relief and soil background effects (Major et al., 1990). What is more, when the vegetation
cover is complete and chlorophyll content reaches a certain level, the relationship between
NDVI and any characteristic that is derived from the amount of chlorophyll is saturated
(Huete et. al., 1997), and further increases in chlorophyll content, do not result in proportional
increases of the NDVI. Last but not least, another important limitation of the NDVI, is its
sensitivity to the contribution of the background beneath the vegetation, when vegetation
cover is not complete. The contribution of soil reflectance to total canopy reflectance in the
NIR is significant (Allen and Richardson, 1968). For a given amount of vegetation and
vegetation cover, darker soils in the background result in higher values of vegetation indices
(Elvidge and Lyon, 1985; Huete et. al., 1985).
The Transformed Vegetation Index proposed by Rouse et. al. (1974), was effectively the
NDVI with the addition of a 0.5 constant and the square-rooting of the sum (Eq. 6). The TVI
was produced in order to avoid the negative values of NDVI, and also to avoid the possibility
that the variances of the ratio would be proportional to the mean values.
When dealing with vegetation canopies, the contribution of the soil reflectance to the total
canopy reflectance is significant, particularly in the NIR region, where there is no pigment
Multispectral Vegetation Indices in Remote Sensing 25
absorption (Allen and Richardson, 1968). In order to discriminate between vegetation and soil
signal, Richardson and Wiegand (1977) proposed the Perpendicular Vegetation Index (PVI;
Eq. 7). The index calculates the difference between the soil and vegetation signals in both the
red (Redsoil, Redveg.) and NIR region (NIRsoil, NIRveg.), and employs those differences.
PVI was one of the first attempts to discriminate vegetation and soil background
reflectance, however it proved to be very sensitive to variable soil brightness and the index
value increased for brighter soils, when vegetation cover remained constant (Elvidge and
Lyon, 1985; Huete, 1988; Baret and Guyot, 1991). Even though the soil line causes the
NIR/red ratio to remain constant, the actual slope and intercept parameters of the soil line
vary depending on the soil properties (Huete et. al., 1984). In addition to soil brightness, other
properties, depending on the soil type, could affect the greenness assessment, even when the
percentage of vegetation cover was as high as 75% (Huete et. al., 1985).
The Soil-Adjusted Vegetation Index (SAVI; Eq. 8) was later introduced by Huete (1988)
as an alternative index dealing with the background signal. This index was employing an L
factor, representing the amount of vegetation present and the extent of vegetation cover. For
total vegetation cover the L receives a value of 0 and the index effectively becomes NDVI,
while for very low vegetation cover, the L acquires a value near 0. When the extent of
vegetation cover is unknown, the author suggested a value of 0.5 as optimal. For this value
and for intermediate vegetation cover, SAVI was found to be superior to both the NDVI and
PVI (Huete, 1988).
NIR red
SAVI (1 L) (8)
NIR red L
The index performs best when the slope of the soil line (a) is equal to 1 and the intercept
(b) equals zero. Deviations from these values tend to reduce the accuracy of SAVI (Baret et.
al., 1989). However, Rondeaux et. al. (1996) suggested that the optimal value of the L factor
for SAVI was 0.16, thus coining the OSAVI index. To address the issue of fixed L values,
another version of the SAVI index was introduced by Baret et. al. (1989). Instead of simply
adding a subjective factor, the NDVI was equipped with information of the soil line, namely
its slope (a) and intercept (b), producing the Transformed Soil-Adjusted Vegetation Index
(TSAVI; Eq. 9). The index ranges from zero for bare soil to a maximum of around 0.7 for
very dense vegetation cover. The index was shown to be able to compensate for changes in
solar elevation and canopy structure (Baret et. al., 1989).
aNIR - a red - b
TSAVI (9)
red a NIR - ab
Another incarnation of the SAVI index was proposed by Major et. al. (1990), employing
once again the soil line slope and intercept. In this instance, instead of using the NDVI as a
basis, the information was added at the denominator of the RVI, producing SAVI2 (Eq. 10).
26 George P. Petropoulos and Chariton Kalaitzidis
NIR
SAVI 2 (10)
red b a
SAVI was derived specifically to reduce the effects due to ground reflectance and
implicitly assumes a linear relationship between red and near-infrared ground reflectances. Qi
et al (1994) suggested that SAVI was significantly less susceptible to changes associated with
soil variations compared to SR an NDVI. On the other hand, Rondeaux et al (1996) compared
NDVI, SAVI and TSAVI and determined that TSAVI was least prone to perturbations
associated with soil changes.
At the time when Huete (1988) was incorporating the soil line to NDVI to produce SAVI,
Clevers (1989) proposed the Weighted Difference Vegetation index (WDVI) stating that this
index could reportedly estimate LAI, with the assumption that red and NIR reflectance was
independent of soil moisture content. However, further studies by Baret and Guyot (1991)
found that the index had no particular advantage over the PVI and shared the same
weaknesses.
Qi et. al. (1994) suggested an iterative process for the determination of the L factor,
through which the initial L value is combining the NDVI and WDVI and is also employing a
primary soil-line parameter γ (a value of 1.06 was used in that study; Eq.11). Each subsequent
L value is calculated by the difference between the resulting MSAVI (Eq. 12) and 1 (Eq. 13).
NIR red
MSAVI (1 L0 ) (12)
0 NIR red L
0
A further attempt to create a more accurate index in the SAVI family, was made by
Gilabert et. al. (2002). They proposed a generalised soil-adjusted vegetation index
(GESAVI), which utilises the red and NIR reflectance, along with the soil parameters a and b,
plus a soil adjustment coefficient (Eq. 14). The difference between this index and the previous
SAVI indices, is the fact that the vegetation isolines in the NIR-red plane, are neither parallel
to the soil line (as required by PVI), nor converging from the same point (as required by the
NDVI), but somewhere in between.
NIR - b Red - a
GESAVI (14)
Red z
The z factor is related to the red reflectance, at the point where the soil line and the
vegetation isolines converge. The authors have suggested the use of a value of 0.35, in case
additional data to derive the actual z value is not available.
A different approach in dealing with partial vegetation cover was followed by Nemani et.
al. (1993), when they introduced a middle infrared band to the traditional NDVI, producing
Multispectral Vegetation Indices in Remote Sensing 27
the corrected NDVI (NDVIc; Eq 15). The middle infrared helped account for understory
effects, when the vegetation cover was partial and the understory vegetation had a
significantly different spectral signature than the tree canopies. As a result the index was
more closely related to the LAI of conifer forests.
In the above equation (Eq. 15), MIR is the middle infrared reflectance of (band 5 of
LANDSAT TM), and MIRmin and MIRmax are the middle infrared reflectance signals from a
completely open and completely closed canopy respectively, from the general study area.
However, atmosphere can affect considerably vegetation indices (Kaufman and Sendra,
1988). As it was discussed earlier, the interference of the atmosphere with the incident solar
radiation affects its intensity and makes it more diffuse, through scattering. The NDVI, like
most vegetation indices, is suffering from those effects. Subsequently, in addition to
minimizing the effect of background, spectral radiance values must be corrected for
atmospheric effects to recover the vegetation signal.
In an attempt to correct these atmospheric effects, Pinty and Verstraete (1992) proposed
the Global Environmental Monitoring Index (GEMI), a complex non-linear polynomial
equation, which combined the red and near-infrared reflectance. The GEMI was shown to be
more useful in comparing observations under variable atmospheric conditions and also more
representative of surface conditions, compared to the simple ratio (SR) and the NDVI. Also
aiming to create an index resistant to atmospheric influence, Kaufman and Tanre (1992) have
developed the Atmospherically Resistant Vegetation Index (ARVI) for the MODIS sensor.
This index is using the blue and red channel to isolate the atmospheric effects and
subsequently apply the corrections on the red and NIR reflectance. Kaufman and Tanre
(1992) showed that ARVI is four times less sensitive to atmospheric changes than NDVI.
The same principle which was applied to SAVI, correcting the red band with the
information contained in the blue band, producing the Soil-adjusted Atmospherically
Resistant Vegetation Index (SARVI; Huete et. al., 1994). Another index which accounts for
residual atmospheric contamination (e.g., aerosols) and variable soil background reflectance,
is the Enhanced Vegetation Index (EVI) developed by Huete et al. (1997; Eq. 16), which
normalizes the reflectance in the red band as a function of the reflectance in the blue band.
Evaluation of radiometric and biophysical performance of EVI implemented from the
Moderate Resolution Imaging Spectroradiometer (MODIS) radiometer revealed that EVI
remained susceptible to canopy variations (Huete et al., 2002).
NIR red
EVI G * (16)
NIR C1red - C 2 blue L
In equation 16, G is a gain factor, C1 and C2 are coefficients to correct aerosol effects and
L is a coefficient do account for canopy background effects. The index was produced for use
with the MODIS sensor.
28 George P. Petropoulos and Chariton Kalaitzidis
Gitelson et. al. (1996) have found the spectral region between 520 and 630 nm (primarily
in the green region of the spectrum) to be more sensitive to chlorophyll fluctuations, even at
very high concentrations. The difference between the red and green bands is that the former is
not affected by the presence of carotenoids. However, the authors have found that the 530 –
570 nm region is not affected by carotenoids absorption either. Instead of using the red band,
they adapted the NDVI and ARVI indices to employ the narrower green band (530 – 570
nm), creating the two ―G reen‖ indices, the Green NDVI (GNDVI) and the Green
Atmospherically Resistant Vegetation Index (GARI).
With the aim of creating an index with the ability to ignore atmospheric effects, while
being sensitive to the fraction of absorbed photosynthetically active radiation (fAPAR) by
vegetation, Gobron et. al. (1999) have formulated a vegetation index to be used the MERIS
sensor data. The MERIS Global Vegetation Index (MGVI) is using a red and NIR band at 681
and 865 nm respectively, in order to estimate the fAPAR. Before the index is calibrated, the
information in the blue band (442 nm) is used, in order to remove the atmospheric effects
present in the red and NIR bands. Comparison of the index with the traditional NDVI has
shown that the MGVI is equally efficient and additionally has a global application. Similarly
to the MGVI, a global vegetation index was designed to be used with SeaWiFS data (Gobron
et. al., 2001). This index also used a red and NIR band combination and was corrected for
atmospheric and geometric effects, before being adjusted to maximise its sensitivity to fAPAR.
Due to the limitations of the use of NIR reflectance, when vegetation canopy cover is not
complete (Colwell, 1974), Gitelson et. al. (2002) have formulated an index that only employs
visible reflectance data. The Visible Atmospherically Resistant Index (VARI) is using the
green and red bands for the estimation of vegetation fraction (VF), as well as the blue band
for the compensation of atmospheric effects in the other two bands (Eq. 17).
R green R red
VARI (17)
R green R red R blue
Despite the fact that the contrast between the bands in the visible region is not as high as
that between red and NIR, the index was found to be more sensitive than NDVI at high VF,
and the error in estimating VF did not exceed 10% (Gitelson et. al., 2002).
The well-known issue of early saturation of many vegetation indices, has also been the
focus of research recently. In their study, Unsalan and Boyer (2004) have analysed the
statistical framework for the NDVI. Subsequently, they represented the index as a slope,
converting its relationship with LAI into a linear form, reducing, in effect, the extend of
saturation. An alternative path was followed by Gitelson (2004), who observed that, while red
reflectance exhibits a flat response once LAI exceeds the value of 2, NIR reflectance
remained sensitive, to LAI values between 2 and 6. On the other hand, the sensitivity of NIR
reflectance was reduced, when its value exceeded 30%. The Wide Dynamic Range
Vegetation Index (WDRVI) is effectively the NDVI, with a factor between 0.1 and 0.2
applied to the NIR reflectance (Eq. 18):
Multispectral Vegetation Indices in Remote Sensing 29
a R NIR R red
WDRVI (18)
a R NIR R red
The value of the a factor is dependent on the vegetation fraction (VF). The new index
was found to be up to three times more sensitive to moderate-to-high LAI values (between 2
and 6). The cause for the increased sensitivity is the effective linearisation of the relationship
with vegetation fraction.
However, in contrast with the numerous studies that use red and NIR spectral bands
discussed so far, a limited number of studies have explored the SWIR spectral bands (e.g., 1.6
and 2.1 µm) for vegetation study. A number of studies have suggested that a combination of
NIR and SWIR bands have the potential for retrieving leaf and canopy water content (e.g.
Hunt & Rock, 1989; Ceccato et al., 2001). One such index is the Moisture Stress Index
(MSI), proposed by Hunt and Rock (1989), which is calculated as a simple ratio between
SWIR (1.6 µm) and NIR (0.82 µm) spectral bands, was proposed to estimate leaf relative
water content (%) and equivalent water thickness (gr cm-2) of different plant species (Eq 19).
SWIR
MSI (19)
NIR
In analyses of the 10-day composite of VGT data another water index was calculated as
the normalized difference between the NIR (0.78–0.89 µm) and SWIR (1.58–1.75 µm)
spectral bands (Xiao et al., 2002), here it is called Land Surface Water Index (LSWI, Eq 20):
NIR SWIR
LSWI (20)
NIR SWIR
More recently, Ceccato et al. (2002 a, b) proposed the Global Vegetation Moisture Index
(GVMI) to retrieve equivalent water thickness (grcm-2) at canopy level, using images from
the SPOT-VGT sensor. This index uses the reflectance values of the rectified NIR band,
which are derived from a complex procedure that involves blue spectral band and uses the
apparent reflectance as seen at the top-of-atmosphere (Gobron et al., 1999), and shown in Eq
21 below:
* *
( pnir 0.1) ( pswir 0.02 )
GVMI (21)
* *
( pnir 0.1) ( pswir 0.02 )
A comparison between GVMI and NDVI showed that the former provided information
related to canopy water content (EWT), while NDVI supplied the information related to
vegetation greenness (Ceccato et al., 2002a).
30 George P. Petropoulos and Chariton Kalaitzidis
The Leaff Area Index (LAI) is a very important plant parameter because its magnitude
affects the amount of radiation that can be absorbed by the canopy. As a result, the LAI has
been related with leaf mass and overall biomass (Wiegand et. al., 1990). Vegetation indices
have been evaluated for their relationship with LAI in many studies, for both forest species
and agricultural crops. Estimation of LAI was found to be possible by the simple ratio of
NIR/red (Asrar et. al., 1985a; Maas, 1993) and the NDVI (Asrar et. al., 1985a) in wheat.
Running et. al. (1986) reached the same conclusion for coniferous forests (r2 = 0.76 for the
simple ratio and r2 = 0.55 for the NDVI). Studying the same area, Peterson et. al. (1987)
improved the relationship of the simple ratio with LAI from r2 of 0.83 to 0.91, by using a log-
linear transformation.
The Normalised Difference Vegetation Index (NDVI) has been the most commonly used
index in studies of estimating LAI and the accuracy of those estimations through this index
have been met with variable results. The highest coefficient of determination between NDVI
and LAI was r2 = 0.95, in a study with corn (Gilabert et. al., 1996). However, the limitations
of using the NDVI for LAI estimations are quite severe. The index has been proven to be very
sensitive to the contribution of soil or background vegetation at low LAI values (Baret and
Guyot, 1991). On the other hand, the NDVI-LAI relationship seems to saturate at high LAI
values, because of the reduced contribution of the lower canopy leaves to the overall canopy
reflectance (Asrar et. al., 1984; Turner et. al., 1999). A corrected version of the NDVI, using
middle-infrared information was shown to improve the estimation of LAI (Nemani et. al.,
1993) and the short-wave infrared reflectance signal has also been suggested as being able to
estimate LAI, taking advantage of the intense water absorption features in that region (Gong
et. al., 2003). However, in addition to the soil sensitivity and saturation issues, the NDVI-LAI
relationship also appears to be affected by leaf orientation (Baret et. al., 1989) and growth
stage (Hatfield et. al., 1984). In particular, Curran et. al. (1992) have shown that the
coefficient of determination between the NDVI and LAI for slash pine is a low r2 = 0.35
during February, increasing to r2 = 0.86 in March and falling to r2 = 0.75 for September data.
Alternative indices have also been evaluated, in an attempt to deal with the issues faced
with the NDVI. The Green Vegetation Index (Jackson, 1983), was evaluated on various fields
of corn and was found to provide accurate estimates of LAI with r2 = 0.78 – 0.93 (Wiegand
et. al., 1990). In order to account for background soil contribution, the PVI was used to
estimate green LAI (Maas, 1988). Wiegand et. al. (1990) evaluated the GVI and PVI on corn
and found that a function of those two indices provided coefficients of determination of r2 =
0.937, higher than those produced by the simple ratio and the NDVI. The introduction of
vegetation indices such as SAVI (Huete, 1988) and TSAVI (Baret et. al., 1989), which tend to
deal better with soil contribution, prompted the comparison with the more traditional PVI and
NDVI indices. The first two indices appeared to be superior to the latter two, due to their
ability to account for soil contribution (Baret and Guyot, 1991). The introduction of
atmospheric resistance employed in the ARVI index (Kaufman and Tanre, 1992) in SAVI,
created a stronger version of the index (SARVI), which appeared to reduce the accuracy error
of estimating LAI in half, in comparison with NDVI (Huete et. al., 1994).
All vegetation indices have an asymptotic relationship with LAI that saturates at high
LAI values (Spanner et. al., 1990; Baret and Guyot, 1991; Turner et. al., 1999). Fassnacht et.
al. (1997) found that a linear relationship was sufficient to connect LAI and a VI, but a log-
Multispectral Vegetation Indices in Remote Sensing 31
The Photosynthetically Active Radiation (PAR) is the spectral range of light that can be
used by vegetation for the process of photosynthesis. This region is between 400 and 700 nm
(the ―v isible‖ region), because in this region the vegetation pigments absorb energy. The
amount of energy that is actually absorbed is known as Absorbed PAR (APAR) and when it
is expressed as a fraction of the total incident radiation it is referred to as the Fraction of
APAR (fAPAR). This energy is very closely related to the primary productivity of plants and
the production of biomass. Dry matter production and the accumulated intercepted
photosynthetically active radiation in the 400-700 nm region of the electromagnetic spectrum
were also shown to be closely related (Biscoe et. al., 1975; Monteith, 1977; Gallagher and
Biscoe, 1978). In addition, a quantitative relationship between dry matter production and
intercepted radiation can be established, as Hodges and Kanemasu (1977) showed for barley
canopies. Hence, it is possible to use remote sensing to estimate the solar radiation that is
intercepted by canopies, and then converted into dry matter (Daughtry et. al., 1983).
Early studies have shown that both the simple ratio (NIR/red) and the NDVI were closely
related to dry matter accumulation (Tucker et. al., 1981). The simple ratio (NIR/red) was used
to derive an empirical relationship with fAPAR on sugarbeet, but that relationship was not
transferable to different crops or different growth stages (Steven et. al., 1983). Christensen
and Gourdiaan (1993) also used the simple ratio to calculate the cumulative PAR and the
result was found to be closely related with the above-ground biomass. The NDVI was also
used in many cases for the estimation of PAR. Asrar et. al. (1985b) found a strong correlation
between the index and PAR, which was then used to calculate above-ground phytomass of
wheat. Steinmetz et. al. (1990) also found a good relationship between NDVI and PAR, but
highlighted the fact that the relationship is affected by nitrogen and water stress and also that
the rate of conversion from PAR to biomass was dependent on growth stage of wheat.
The relationship between the NDVI and fAPAR has a low signal-to-noise ratio (North,
2002) and is linear, but it is only valid during the growth stage of the crops (Ruimy et. al.,
1994). The possible reason is that the canopy continues absorbing radiation at later crop
stages but it contains less photosynthetic pigments, which leads to a decrease in the NDVI
(Hatfield et. al., 1984; Gallo et. al., 1985). On the other hand, when fAPAR is high, NDVI is
less sensitive to fAPAR changes. In those cases the WDRVI (Gitelson, 2004) appears to be
more sensitive, and in cases where hyperspectral data are available (e.g. ESA‘s MERIS or
NASA‘s Hyperion sensors), the red-edge NDVI is the most sensitive index (Vina and
Gitelson, 2005).
32 George P. Petropoulos and Chariton Kalaitzidis
Other indices were also evaluated for their ability to estimate PAR, fAPAR and indirectly,
biomass and in many cases they provided superior results to both the simple ratio and NDVI.
In a study comparing the GVI, PVI, NDVI and RVI, Wiegand et. al. (1990) found that a
combination of GVI and PVI gave the most accurate estimates of f APAR (r2 = 0.94). On the
other hand, a similar study showed that the NDVI and PVI were the most accurate indices in
the estimation of fAPAR in cotton (Wiegand et. al., 1991).Gobron et. al. (1999) evaluated the
MERIS Global Vegetation Index (MGVI) using MERIS data and found that the index
provided more information than the simple NDVI. The WDRVI was used on maize and
soybean plants, employing field spectra, and it was found to be more sensitive than the NDVI
at high fAPAR values (Vina and Gitelson, 2005).
CONCLUSION
In this chapter it was provided an overview of the different radiometric vegetation indices
usilising multispectral remote sensing observations acquired in the reflective part of the EMR
spectrum. In this framework, it was first provided an overview of the main properties of
remote sensing in this part of the EMR, as this was deemed necessary in order to cover the
theoretical background required in understanding the basic principles on which these
radiometric indices have their basis.
As indicated by the overview presented herein, a wide range of radiometric indices have
been developed in order to establish relationships between such data and biomass, or other
vegetation characteristics that can be indirectly linked to the amount of biomass present. As
was also made clear from the overview conducted herein, the factors that affect the efficiency
of vegetation indices are concerned with the characteristics of the recorded signal, which is
affected by bidirectional and atmospheric effects, canopy structure and background
vegetation or soil contribution, scattering, spatial heterogeneity, adjacency effects, non-linear
mixing and topographic effects. These factors are a major concern for the transferability of an
established methodology to a different plant, location, or time. What is more, on many
occasions the proven relationships between vegetation indices and vegetation properties were
empirical in nature, performing well on the particular study, but facing transferability issues
when the same indices were evaluated on a different vegetation type, different location or
even different time of the year.
As was also indicated in the present review, methods of using vegetation indices for the
estimation of the leaf area index (LAI) have had variable results. Canopy structure and
background contribution appear to play a crucial role in the determination of the canopy
reflectance signal, influencing the performance of any VI – LAI relationship. When
vegetation cover is low, the presence of understory vegetation (primarily in the case of
forests) and the spectral characteristics of the underlying soil, affect canopy reflectance and
give erroneous LAI estimates. Soil-adjusted indices that either use information on the soil line
(NIR/red ratio) underneath the canopy, or a factor estimating the vegetation cover, tend to
somewhat deal with the background contribution issue at low LAI values. However, those
indices still present weaknesses that render their operational ability questionable. On the other
hand, when the LAI is high, the multiple layers of leaves within the canopy have a different
contribution to the canopy reflectance signal, causing the VI – LAI relationship to saturate.
Multispectral Vegetation Indices in Remote Sensing 33
ACKNOWLEDGMENTS
Part of this work was undertaken within the frame of an FP7 funded programme with the
acronym CEUBIOM, and full title ―Cla ssification of European Biomass Potential for
Bioenergy Using Terrestrial and Earth Observations‖. Authors wish to thank the anonymous
reviewers for the valuable comments which resulted to the improvement of the originally
submitted chapter. Dr. Petropoulos is also grateful to INFOCOSMOS E.E.
(http://www.infocosmos.eu) for supporting his participation to the present work.
34 George P. Petropoulos and Chariton Kalaitzidis
REFERENCES
Albuelgasim, A. A., and Strahler, A. H., 1994. Modeling bidirectional radiance measurements
collected by the Advanced Solid-state Array Spectroradiometer over Oregon transect
conifer forests, Remote Sens. Environ. 47:261-275.
Allen, W.A. and Richardson, A.J. 1968. Interaction of light with plant canopy. Journal of the
Optical Society of America, 58, 1023-1028.
Asrar, G., Fuchs, M., Kanemasu, E. T. and Hatfield, J. L. 1984. Estimating absorbed
photosynthetic radiation and leaf area index from spectral reflectance in wheat.
Agronomy Journal, 76, 300-306.
Asrar, G., Kanemasu, E. T. and Yoshida, M. 1985a. Estimates of leaf area index from spectral
reflectance of wheat under different cultural practices and solar angle. Remote Sensing of
Environment, 17, 1-11.
Asrar, G., Kanemasu, E. T., Jackson, R. D. and Pinter, P. J. 1985b. Estimation of total above
ground phytomass production using remotely sensed data. Remote Sensing of
Environment, 17, 211-220.
Baret, F. and Guyot, G. 1991. Potential and limits of vegetation indices for LAI and APAR
assessment. Remote Sensing of Environment, 35, 161-173.
Baret, F., Guyot, G. and Major, D. J. 1989. TSAVI: A vegetation index which minimizes soil
brightness effects on LAI and APAR estimation. In Proceedings IGARRS ‘90 / 12th
Canadian Symposium on Remote Sensing, Vancouver, Canada, 10-14 July 1989, vol. 3,
pp. 1355-1358.
Biscoe, P. V., Gallagher, J. N., Littleton, E. J., Monteith, J. L. and Scott, R. K. 1975. Barley
and its environment, IV. Sources of assimilate for the grain. Journal of Applied Ecology,
12, 295-318.
Buschmann, C. and Nagel, E. 1991. Reflection spectra of terrestrial vegetation as influenced
by pigment-protein complexes and the internal optics of the leaf structure. In: Putkonen J,
editor. Proceedings of the International Geoscience and Remote Sensing Symposium
(IGARSS '91). Espoo, Finland, 3-7 June 1991: New York: IEEE. p 1909-1912.
Cambell, J. E., 1981.Introduction to remote Sensing, 2nd Ed., New York, Guilford Press.
Ceccato, P., Flasse, S., & Gregoire, J. M. 2002a. Designing a spectral index to estimate
vegetation water content from remote sensing data: Part 2. Validation and applications.
Remote Sensing of Environment, 82, 198– 207
Ceccato, P., Flasse, S., Tarantola, S., Jacquemoud, S., & Gregoire, J. M.2001. Detecting
vegetation leaf water content using reflectance in the optical domain. Remote Sensing of
Environment, 77, 22–33
Ceccato, P., Gobron, N., Flasse, S., Pinty, B., & Tarantola, S. 2002b. Designing a spectral
index to estimate vegetation water content from remote sensing data: Part 1. Theoretical
approach. Remote Sensing of Environment, 82, 188– 197
Christensen, S. and Goudriaan, J. 1993. Deriving light interception and biomass from spectral
reflectance ratio. Remote Sensing of Environment, 43, 87-95.
Clevers, J. G. P. W. 1988. The derivation of a simplified reflectance model for the estimation
of leaf area index. Remote Sensing of Environment, 25, 53-69.
Multispectral Vegetation Indices in Remote Sensing 35
Gitelson, A. A., Kaufman, Y. J. and Merzlyak, M. N. 1996. Use of a green channel in remote
sensing of global vegetation from EOS-MODIS. Remote Sensing of Environment, 58,
289-298.
Gitelson, A. A., Kaufman, Y. J., Stark, R. and Rundquist, D. 2002. Novel algorithms for
remote estimation of vegetation fraction. Remote Sensing of Environment, 80, 76-87.
Gobron, N., Melin, F., Pinty, B., Verstraete, M. M., Widlowski, J. L. and Bucini, G. 2001. A
global vegetation index for SeaWiFS: Design and applications. In: Beniston M, and
Verstraete, M. M., (eds.) Remote Sensing and Climate Modeling: Synergies and
Limitations, Kluwer Academic Publications. p 5-21.
Gobron, N., Pinty, B., Verstraete, M. and Govaerts, Y. 1999. The MERIS Global Vegetation
Index (MGVI): description and preliminary application. International Journal of Remote
Sensing, 20, 1917-1927.
Goel, N. S., 1988. Models of vegetation canopy reflectance and their use in estimation of
biophysical parameters from reflectance data, Remote Sens. Rev. 4:1-222.
Gong, P., Pu, R., Biging, G. S. and Larrieu, M. R. 2003. Estimation of forest leaf area index
using vegetation indices derived from Hyperion hyperspectral data. IEEE Transactions
on Geoscience and Remote Sensing, 40, 1355-1362.
Goward, S.N, Marklam, B.L., Dye, D.G., and Yang, J. 1991. Normalised difference
vegetation index measurements from the Advanced Very High Resolution Radiometer,
Rem. Sen. Envir., 35, pp; 257-277of Environment, ERIM, Ann Arbor, Michighan, August
23-25, pp: 1169-1178
Hansen, P.M., Schjoerring, J.K., 2003. Reflectance measurement of canopy biomass and
nitrogen status in wheat crops using normalized difference vegetation indices and partial
least squares regression. Remote Sens. Environ. 86, 542–553.
Hatfield, J. L., Asrar, G. and Kanemasu, E. T. 1984. Intercepted photosynthetically active
radiation estimated by spectral reflectance. Remote Sensing of Environment, 14, 65-75.
Hodges, T. and Kanemasu, E. T. 1977. Modeling daily dry matter production of winter wheat.
Agronomy Journal, 69, 974-978.
Holben, B., and Kimes, D., 1986. Directional reflectance response in AVHRR red and near-
IR bands for three cover types and varying atmospheric conditions, Remote Sens.
Environ. 19:213-236.
Hosgood, B., Sandmeier, S., Piironen, J., Andreoli, G., and Koechler, C., 1999. Goniometers.
ncyclopedia of Electrical and Electronics Engineering, Wiley, New York, pp. 424-433
Houborg, R., Soegaard, H. and Boegh, E. 2007. Combining vegetation index and model
inversion methods for the extraction of key vegetation biophysical parameters using
Terra and Aqua MODIS reflectance data. Remote Sensing of Environment, 106, 39-58.
Huete, A. R. 1988. A soil-adjusted vegetation index (SAVI). Remote Sensing of Environment,
25, 295-309.
Huete, A. R., Liu, H. Q., Batchily, K. and van Leeuwen, W. 1997. A comparison of
vegetation indices over a global set of TM images for EOS-MODIS. Remote Sensing of
Environment, 59, 440-451.
Huete, A., Didan, K., Miura, T., Rodriguez, E. P., Gao, X., & Ferreira, L. G. 2002. Overview
of the radiometric and biophysical performance of the MODIS vegetation indices. Remote
Sensing of Environment, 83, 195– 213
Huete, A., Justice, C. and Liu, H. 1994. Development of vegetation and soil indices for
MODIS-EOS. Remote Sensing of Environment, 49, 224-234.
Multispectral Vegetation Indices in Remote Sensing 37
Huete, A.R., Jackson, R.D. and Post, D.F. 1985. Spectral response of a plant canopy with
different soil backgrounds. Remote Sensing of Environment, 17, 37-53.
Huete, A.R., Post, D.F. and Jackson, R.D. 1984. Soil spectral effect on 4-space vegetation
discrimination. Remote Sensing of Environment, 15, 155-165.
Hunt, E. R., & Rock, B. N. 1989. Detection of changes in leaf water content using near-
infrared and middle-infrared reflectance. Remote Sensing of Environment, 30, 43– 54
Jackson, R. D. 1983. Spectral indices in n-Space. Remote Sensing of Environment, 13, 409-
421.
Jackson, R. D., Teillet, P. M., Slater, P. N., Fedosejevs, G., Jasinski, M. F., Aase, J. K., and
Moran, M. S., 1990. Bidirectional measurements of surface reflectance for view angle
corrections of oblique imagery. Remote Sens. Environ. 32: 189-202
Jensen, J. R., 2000. Remote Sensing of the Environment: An Earth Resource Perspective.
Prentice Hall: Saddle River, N.J.
Jordan, C. F. 1969. Derivation of leaf area index from quality of light on the forest floor.
Ecology, 50, 663-666.
Kaufman, Y. J. and Tanre, D. 1992. Atmospherically resistant vegetation index (ARVI) for
EOS-MODIS. IEEE Transactions on Geoscience and Remote Sensing, 30, 261-270.
Kaufman, Y.J. and Sendra, C. 1988. Algorithm for automatic atmospheric corrections to
visible and near-IR satellite imagery. Int. J. Rem. Sens., 9, pp:1357-1381
Kieffer, H. H., Martin, T. Z., Peterfreund, A. R., and Jakosky, B. M., 1977. Thermal and
albedo mapping of Mars during the Viking primary mission, J. Geophys. Res. 82:4249-
4291.
Knipling, E. B. 1970. Physical and physiological basis for the reflectance and near infrared
radiation from vegetation. Remote Sensing of Environment, 1, 155-159.
Lillesand, T. M. and Kiefer, 1994. Remote Sensing and Image Interpretation, New York, John
Wiley and Sons, 750 pp
Maas, S. J. 1988. Using satellite data to improve model estimates of crop yield. Agronomy
Journal, 80, 655-662.
Maas, S. J. 1993. Within-season calibration of modeled wheat growth using remote sensing
and field sampling. Agronomy Journal, 85, 669-672.
Major, D. J., Baret, F. and Guyot, G. 1990. A ratio vegetation index adjusted for soil
brightness. International Journal of Remote Sensing, 11, 727-740.
Markham, B. L., Kimes, D. S. and Tucker, C. J. 1981. Temporal spectral response of a corn
canopy. Photogrammetric Engineering and Remote Sensing, 48, 1599-1605.
Monteith, J. L. 1977. Climate and the efficiency of crop production in Britain. Philosophical
Transactions of the Royal Society in London Ser. B., 281, 277-294.
Myneni, R. B., Ross, J., and Asrar, G., 1990. A review on the theory of photon transport in
leaf canopies, Agric. For. Meteorol. 45:1-153.
Nemani, R. P. L., Running, S. W. and Band, L. 1993. Forest ecosystem processes at the
watershed scale: Sensitivity to remotely sensed leaf area index estimates. International
Journal of Remote Sensing, 14, 2519-2534.
Nicodemus, F. E., Richmond, J. C., Hsia, J. J., Ginsberg, I. W.,and Limperis, T., 1977.
Geometrical considerations and nomenclature for reflectance. Nat. Bur. Standards Mono.
160:52.
North, P. R. J. 2002. Estimation of fAPAR, LAI, and vegetation fractional cover from ATSR-
2 imagery. Remote Sensing of Environment, 80, 114-121.
38 George P. Petropoulos and Chariton Kalaitzidis
Steinmetz, S., Guerif, M., Delecolle, R. and Baret, F. 1990. Spectral estimates of the absorbed
phptpsynthetically active radiation and light-use efficiency of a winter wheat crop
subjected to nitrogen and water deficiencies. International Journal of Remote Sensing,
11, 1797-1808.
Steven, M. D., Biscoe, P. V. and Jaggard, K. W. 1983. Estimation of sugarbeet productivity
from reflection in the red and infrared spectral bands. International Journal of Remote
Sensing, 4, 325-334.
Strahler, A. H., and Jupp, D. L. B., 1991.Geometric-optical modeling of forests as remotely-
sensed scenes composed of three-dimensional, discrete objects, in Photon-Vegetation
Tucker, C. J., Holben, B. N., Elgin, J. H. and McMurtrey, J. E. 1981. Remote sensing of total
dry matter accumulation in winter wheat. Remote Sensing of Environment, 11, 171-189.
Turner, D. P., Cohen, W. B., Kennedy, R. E., Fassnacht, K. S. and Briggs, J. M. 1999.
Relationships between Leaf Area Index and Landsat TM spectral vegetation indices
across three temperate zone sites. Remote Sensing of Environment, 70, 52-68.
Unsalan, C. and Boyer, K. L. 2004. Linearized vegetation indices based on formal statistical
framework. IEEE Transactions on Geoscience and Remote Sensing, 42, 1575-1585.
Verbyla, D. L. 2001. Practical GIS Analysis. Taylor & Francis Press. 12 chapters.
Verbyla, D., 1995. Satellite Remote Sensing of Natural resources. Lewis publ., Florida.
Verstraete, M. M., Pinty, B., and Dickinson, R. E., 1990.A physical model of the bidirectional
reflectance of vegetation canopies, 1: Theory, J. Geophys. Res. 95:11,755-11,765.
Vina, A. and Gitelson, A. A. 2005. New developments in the remote estimation of the
fraction of absorbed photosynthetically active radiation in crops. Geophysical Research
Letters, 32, L17403.
Wiegand, C. L., Gerbermann, A. H., Gallo, K. P., Blad, B. L. and Dusek, D. 1990. Multisite
analyses of specrtal-biophysical data for corn. Remote Sensing of Environment, 33, 1-16.
Wiegand, C. L., Maas, S. J., Aase, J. K., Hatfield, J. L., Pinter Jr., P. L., Jackson, R. D.,
Kanemasu, E. T. and Lapitan, R. L. 1992. Multisite analyses of spectral-biophysical data
for wheat. Remote Sensing of Environment, 42, 1-21.
Wiegand, C. L., Richardson, A. J., Escobar, D. E. and Gergermann, A. H. 1991. Vegetation
indices in crop assessments. Remote Sensing of Environment, 35, 105-119.
Xiao, X., Boles, S., Liu, J. Y., Zhuang, D. F., & Liu, M. L. (2002). Characterization of forest
types in Northeastern China, using multi-temporal SPOT-4 VEGETATION sensor data.
Remote Sensing of Environment, 82, 335– 348
In: Ecological Modeling ISBN: 978-1-61324-567-5
Editor: WenJun Zhang, pp. 41-64 © 2012 Nova Science Publishers, Inc.
Chapter 3
ABSTRACT
According to the Water Framework Directive (WFD, 2000/60/EC), Integrated River
Basin Management Plans (RBMP) are required at different scales, in order to prevent
amongst other things, water resource deterioration and ensure water pollution reduction.
An integrated river basin management approach underpins a risk-based land management
framework for all activities within a spatial land-use planning framework. To this end, a
risk assessment methodology is required to identify water pollution hazards in order to
set appropriate environmental objectives and in turn design suitable mitigation measures.
42 Anas Altartouri, Kalliope Pediaditi, George P. Petropoulos et al.
Surface water pollution as a result of Olive Mill Waste (OMW) discharge is a serious
hazard in the olive oil producing regions of the Mediterranean. However, there is no
standardised method to assess the risk of water pollution from olive mill waste for any
given river basin. The present chapter shows the results from a study conducted
addressing the above issue by designing a detailed risk assessment methodology, which
utilises GIS modelling to classify within a watershed individual sub-catchment risk of
water pollution occurring from olive mill waste discharges. The chapter presents the
proposed criteria and calculations required to estimate sub-catchment risk significance
and comments on the methods potential for wider application. It combines elements from
risk assessment frameworks, Multi Criteria Analysis (MCA), and Geographic
Information Systems (GIS). MCA is used to aggregate different aspects and elements
associated with this environmental problem, while GIS modeling tools helped in
obtaining many criterion values and providing insight into how different objects interact
in nature and how these interactions influence risk at the watershed level. The proposed
method was trialed in the Keritis watershed in Crete, Greece and the results indicated that
this method has the potential to be a useful guide to prioritise risk management actions
and mitigation measures which can subsequently be incorporated in river basin
management plans.
Keywords: Decision Support Systems (DSS), Geographic Information System (GIS), Multi
Criteria Analysis (MCA), Olive Mill Wastewater (OMW), risk assessment, water
pollution.
1. INTRODUCTION
A major environmental issue in Mediterranean region is the pollution of aquatic
ecosystems through the discharge of industrial and domestic effluents in water bodies
(Karageorgis et al., 2003). One of the main polluting activities is the discharge of effluents
generated from olive mill agricultural industries. Olive mill wastewater (OMW) is the liquid
by-product generated during olive oil production. OMW contains pollutants and hazardous
materials in different concentrations which may cause negative impacts on the natural water
bodies and, consequently, human and environmental health. Indicatively, Paliatziki (2006)
states that 50 m3 of olive oil mill wastewater are equivalent to the waste produced by 30,000
citizens. The dispersed spatial location of a large number of small-sized olive oil mills
together with the concentration and seasonal production of OMW, as the Mediterranean
region accounts for 95% of the global OMW production, are the main reasons for the
environmental degradation caused by OMW (Kapellakis et al., 2002; Niaounakis and
Halvadakis, 2006).
According to Niaounakis and Halvadakis (2006), the OMW composition is: water (80-
83%); organic compounds (15-18%); and inorganic compounds (mainly potassium, salts and
phosphates, 2%). It contains phytotoxic and biotoxic substances and is non-biodegradable. It
has a high organic load and classified among the ‗strongest‘ industrial effluents, with
Chemical Oxygen Demand (COD) up to 220 g/l. The consequence of this is a high
consumption of oxygen dissolved in the water bodies which negatively affects the living
organisms and, thus, an imbalance of the whole ecosystem may be caused (Niaounakis and
Halvadakis, 2006). Similar effects can also result from high phosphorus content in OMW
which accelerate the growth of algae leading to eutrophication. Beside the high number of
Development of a Decision Support System … 43
bacteria and fungi in OMW, the presence of high concentration of nutrients may cause
infection of water bodies since it make perfect medium of pathogens to multiply in these
water bodies. Moreover, OMW has long chain fatty acids and phenolic compounds with high
percentage of dissolved mineral salts (ibid). All these contaminants and their impacts on
natural water bodies can result in significant consequences on the environment and people
who may be in contact with these water bodies (Morrison et al., 2001; Ogunfowokan et al.,
2005, Niaounakis and Halvadakis, 2006).
The EU Water Framework Directive (WFD, 2000/60/EC) aims at harmonizing existing
European water policies and to improve water quality in all aquatic environments within the
community area. It emphasizes the need of new Integrated River Basin Management Plans
(RBMP) at national and regional/local scale resulting in the protection and improvement of
the sustainable use of all waters (Heathwite et al., 2005; Rekolainen et al., 2003). The main
objectives of the RBMP include the prevention of further deterioration of water resources and
the promotion of sustainable water use that ensures the progressive reduction of pollution.
These elements in the EU legislation, policies, and programs underpin the need of including
risk-based land management framework to all activities within a spatial land-use planning
framework (Heathwite et al., 2005). However, there is no standardised method to assess the
risk of water pollution from OMW for any given river basin. Due to the small scale and
dispersed nature of OMW processing and disposal method, regulation using a risk-based
approach which uses planning, for example through RBMP as pollution preventing, rather
than relying on post development or pollution incident mitigation is required. To this end, risk
assessment methodology is needed for this point-source water pollution in order to set
appropriate environmental objectives and risk zones to integrate within RBMP as well as
design suitable mitigation measures.
This chapter presents a risk assessment method based on general frameworks to address
different aspects of the environmental problems associated with the OMW pollution. It
provides an analytical approach based on deep investigation of elements of OMW pollution
risk and linkages between them resulting in a conceptual model to understand and cope with
such problem. This model provides insight into how different objects interact in the nature
and how this interactions influences water resources in a watershed. As number of tools and
techniques are needed for risk assessment process to effectively support decision makers
(Allan et al., 2006), the proposed method combines field and desk-based techniques and
utilizes different tools, such as Geographic Information Systems (GIS) and Multi Criteria
Analysis (MCA). This integration between GIS and MCA in developed risk assessment
method takes advantage of the analytical capabilities of MCA on the one hand. On the other
hand, this method benefits from the information processing and display capabilities of GIS as
risk assessment should involve analysis of spatial variability which consider differences
between locations (Allan et al., 2006; Lapucci et al., 2005). The chapter presents the proposed
approach consisting of all criteria and calculation models required to quantitatively estimate
risk significance at each sub-catchment within the river basin under investigation and
comments on the methods potential for wider application.
44 Anas Altartouri, Kalliope Pediaditi, George P. Petropoulos et al.
As this developed surface water OMW pollution risk assessment method has been
designed to be used in RBMP, the method was based on the generic framework of DEFRA
Guidelines for Environmental Risk Assessment and Management (2002) which makes clear
links between risk assessment and risk management focusing on the practical implementation
of risk assessment result to generate risk management solution (Power & McCarty, 1998). In
addition, it provides clear breakdown of risk into basic components upon which several
criteria have been developed (see Box 1). For the determination of these components, the
olive mill waste process was studied from cradle to grave using field observations, expert
opinion, and literature review. Based on the above, the conceptual model, illustrated in Figure
1, was developed which, subsequently, served as the basis for the development of the
quantitative assessment of risk from OMW pollution.
As illustrated in Figure 1, it is essential for risk assessment process to define the three
elements of risk; sources, pathways, and receptors which, subsequently, help developing
quantitative criteria. A source of stressors can be defined as the place where the stressor
Development of a Decision Support System … 45
originates or is released (EPA, 1998), which in this case are the lagoons where OMW is
gathered. These lagoons are the first element of the exposure pathway and the entities where
hazardous events, such as heavy storms or spillage, may occur and, hence, the probability of
hazard occurring is strongly associated with them (Component 3, Box 1).
In the case of OMW pollution risk, the pathway consists of the mechanism by which
receptors are exposed to the water polluted by OMW and, therefore, it controls the probability
of receptors being exposed to hazard (Component 4, Box 1). Water bodies, ground and
surface, are considered as the potential pathways by which receptors may be exposed to the
hazard of OMW. However, it has been noticed that surface water bodies are more likely to
transmit the pollutants released from the lagoons. This is due to the fact that the most
probable hazardous event to cause the lagoon overflowing is the heavy storms during which a
large amount of water is running off to the drainage network, meaning that released OMW is
more likely to be washed away into surface water bodies rather than infiltrated into
groundwater bodies.
The last elements in this chain are the receptors which in this case are people or actual
environmental values to be protected. Studying the probability of adverse effects resulting
46 Anas Altartouri, Kalliope Pediaditi, George P. Petropoulos et al.
requires a clear definition of receptors to be addressed (Allan et al., 2006). For the purpose of
this study, humans and high-value ecological sites (such as NATURA 2000 sites) are
considered as the potential receptors that may be harmed from water pollution from OMW.
The choice of these two generic receptors is underpinned by the fact that they are the most
sensitive and protected according to the EU legislation. As specific dose-response exposure
data is limited in the literature, the use of highly valued and protected by legislation receptors
is the first step in the practical implementation of this method in the context of the WFD and
RBMP.
This criterion is proposed to calculate the magnitude of OMW pollution risk at the spatial
scale (Component 1, Box 1). As the main receptor groups of OMW pollution risk are humans
and protected areas, the calculation of this criterion is strongly associated with the spatial
distribution of these receptors and is divided into two directions based on the addressed
receptor group.
Regarding the human receptor group, the magnitude of consequences can be estimated by
the number of affected inhabitants. It can be stated that the larger the population is, the greater
the potential consequences of water pollution in the watershed. Thus, the distribution of
towns and population within sub-catchments in the river basin under investigation should be
drawn. However, the term ‗potentially exposed inhabitants‘ include also all people that may
be exposed to the polluted water in the area regardless their residence place. This may include
people whose drinking water source is located in this area. Also, it may include people who
may have come in contact with the surface water bodies such as tourists or farmers.
However, when addressing the other receptor group, the area of protected areas which
represent a high ecological value should be taken into account in order to calculate the spatial
scale of the magnitude of consequences (Component 1, Box 1). The larger the extension of
these protected areas in the addressed river basin, the greater number of impacted species.
Potentially exposed area within these sites can be delineated by creating a buffer around the
surface water bodies through which the pollutants may be transported. Historical observations
of the heavy storms and floods in the region should take part in determining the extension of
Development of a Decision Support System … 47
the buffer zone. For the EU territories, NATURA 2000 sites can represent these ecologically
valuable areas. NATURA 2000 is a European network of protected sites which represent
areas of the highest environmental value for natural habitats (for several plant and animal
species) which are rare, endangered or vulnerable in the European Community. However,
should designated environmental sites other than NATURA, they should also be included as
receptors1. Ideally, an ecological baseline assessment of areas under consideration should be
undertaken in order to identify areas of potential high ecological value which might not be
designated. However, following discussion with potential end users of this tool, the feasibility
of carrying this out was questioned therefore at a coarser level and, thus as a first step,
calculating the area of potentially exposed protected sites can be used although the first is
preferable.
Precipitation is one of the factors that may lead to hazard occurrence (Component 3, Box
1). It is important in terms of the potential overflow (hazardous event) of the lagoons where
the waste is deposited. Overflowing, however, causes the pollutants to reach the water bodies
and, therefore, hazard occurs. The greater the precipitation over the area it is, the higher the
probability of overflowing. Records of occasional, heavy storm events should be considered
in estimating the probability of a lagoon to overflow. Alternatively, the maximum monthly
precipitation may substitute in estimating the overflowing probability.
The ratio between generated waste volume and lagoon capacity is another factor which
affects the probability of hazard occurrence, i.e. the probability that OMW lagoons overflow
(Component 3, Box 1). If the amount of produced waste is larger than the lagoon capacity,
then the hazard is more likely to occur. On the contrary, if the lagoons are well designed to
include a buffer margin which can accommodate for excess precipitation, then the probability
1
This is particularly relevant for the application of this method in NON-EU countries which do not belong to
NATRA 2000 network.
48 Anas Altartouri, Kalliope Pediaditi, George P. Petropoulos et al.
of hazard occurrence goes to zero unless of other events like extreme rainstorm occur.
Therefore, the higher the ratio is, the higher the probability of hazard to occur.
The structural conditions of the lagoons where the waste is gathered have to be
considered. It is an essential factor which affects the probability of hazard occurring
(Component 3, Box 1). Infiltration of the pollutants through the basement of lagoons and
spillage through the walls, due to poor lagoon conditions e.g. lack of maintenance, structural
malfunction, etc., lead the hazard occur. Therefore, the better the structural conditions of
lagoons, the lower the probability of pollutants to reach water bodies. In order to calculate
this criterion, site visits to lagoon are required as well as an investigation of their specification
and permits.
3.6. Length of the Flow Path to Surface Water Bodies Criterion (Cr.6)
Water bodies close to lagoons have a higher probability of being contaminated should the
OMW discharge event occur. The longer the flow path is to a sub-catchment, the greater the
probability of pollutants to be diluted before reaching receptors. This criterion plays a main
role in the estimation of probability of stressor and receptor co-occurrence (Component 4,
Box 1). The actual length of the flow paths can be determined using the 3D tools in GIS
associated with the Digital Elevation Model (DEM). Several flow paths can be drawn for a
single lagoon. Each of these flow paths starts from the lagoon and ends at the point where this
flow path joins the water bodies in a receiving sub-catchment.
The probability of receptors to be harmed resulting from exposure to the hazard of water
pollution (Component 5, Box 1) depends on the concentrations of polluting substances in the
water bodies they be in touch with. Water quality parameters are the key factors to estimate
the impact of these pollutants on the receptors. If the concentrations of polluting substances
are within the allowable limits in legislation, then the probability of harm is near to zero. On
the contrary, the probability of harm is very high when the concentrations exceed the
contamination thresholds. For example, if a lagoon is located in a catchment, then iron (Fe)
could be one of the pollutants (hazard) that may be found in the water samples taken from the
water bodies of that catchment. According to the WFD, the guide level and the maximum
admissible concentration for the drinking water are 50µg/l and 200µg/l, respectively.
Therefore, a scale from zero to one is established to express the likelihood of harm resulting.
In this scale, the probability of 0 is assigned for samples with 50µg/l or less iron
concentration, and 1 for those with 200µg/l iron concentration. The intermediate
concentrations are assigned values between 0 and 1 according to the linear equation which
interpolating these two limit values (section 4.2). This should be applied on all the pollutants
associated with the OMW. Table 1 lists the chemicals associated with the OMW (Niaounakis
Development of a Decision Support System … 49
and Halvadakis, 2006) and their preferable and maximum admissible levels according to the
WFD (2000/60/EC).
However, as data about dose-response is available only for human in the WFD, this
method proposes the dilution degree as an alternative approach to calculate its value when
addressing risk of OMW pollution on the protected areas. The degree of pollutants‘
concentrations in the surface water bodies located in the protected areas can be indicated
taking in consideration the pathway by which these pollutants are transported. Pollutants
could be carried by direct surface runoff or, alternatively, by streams. By direct surface
runoff, the concentrations of pollutants remain almost the same as in the origin pollution
source or with a slight dilution degree in the rain water while pollutants carried by streams are
subjected to higher dilution degrees which decrease their concentration before reaching
ecological sites. However, the degree of dilution and its speed depend on stream velocity as
well as the quantity of stream water. These two parameters can be expressed in terms of
stream order. Therefore, the higher the stream order, the faster the dilution and, therefore, the
lower the probability of protected areas to be harmed.
model which consists of a set of formulae. These steps are illustrated in Figure 3, 4, and 5 and
discussed in the following subheadings.
Practically, there are some initiative steps to be performed in order to proceed with this
method. These steps mainly deal with data collection and preparation of appropriate datasets
which serves as an input into the calculation model in order to obtain quantitative values of
risk. This data consists of all the initial values of criteria which result from the application of
a variety of data processing procedures. Such procedures may include hydrological
processing, map derivation, tabular calculations, and other procedures depending on the
available data.
In order to define surface water bodies, namely streams, and determine some criterion
values (Cr.6, Box 1), a hydrological model is needed. As the function of the spatial hydrology
modeling tool is to simulate the water flow and transport on a specified area using GIS data,
a hydrological model is needed in this analysis in order to define the drainage network in the
target watershed, including streams and possible flow paths from the sources of hazard to
surface water bodies. In addition, the hydrological model is essential for dividing the
watershed under investigation into sub-catchments which represent, together with the
lagoons, the analysis units.
Additionally, data from the field is necessary to determine criterion values of water
quality parameters (Cr.7). Samples from surface water bodies in the target watershed should
be collected within a pre-planned sampling strategy and analyzed for the chemicals and
parameters associated with polluting source, namely OMW. The first consideration is the
timing of sampling process. Regarding OMW, water samples should be collected from the
surface water bodies (streams) in the target watershed during the period when lagoons contain
the maximum volume of wastewater and the water is still present in the drainage network.
The proposed sampling strategy consists of collecting samples from the points where possible
flow paths from lagoons join the receiving streams. This sampling strategy is designed in
order that the chemical tests to show the worst case where the pollutants are in the highest
possible concentrations and have not yet been affected by the dilution process. Such an
approach restricts biased sampling and minimizes the bias of one criterion‘s effect to the
other. The concentrations of the analyzed chemicals in each sample should be assigned to the
corresponding lagoon as shown in the analysis below.
Because of the different scales upon which criteria are measured, it is necessary to
standardize them before being combined. There are various standardization procedures,
typically using the minimum and maximum values as scaling points (Eastman, 2003). These
functions transform the values with dimensions to dimensionless values between 0 and 1,
which makes the criteria of different dimensions comparable (Mendoza et al., 1999). The
functions are implemented either as a benefit (the higher the criterion score, the higher the
likelihood or magnitude of risk) or as a cost (the lower the criterion score the higher the
Development of a Decision Support System … 51
R
ci
Rmax for benefit criteria, and (1)
Rmin
ci
R for cost criteria (2)
where:
ci = the standardized criterion value;R = the origin criterion value; Rmin = the minimum
value of the criterion of all units (lagoons or sub-catchments); and Rmax = the maximum value
of the criterion of all units (lagoons or sub-catchments).
The other standardization method is applied for water quality parameters (Cr.7, Box1).
As the probability of harm resulting is strongly correlated with the amounts of different
OMW hazardous chemical in the water, it can be given as a function of their concentrations.
These functions (see Figure 2) consist of linear multi-segments based on the following linear
scaling functions:
R Rmin
ci
Rmax Rmin for benefit criteria, and (3)
Rmax R
ci
Rmax Rmin for cost criteria (4)
where:
ci = the standardized criterion value; R = the origin criterion value; Rmin = the guide level
according to the WFD (see Table 1) and; and Rmax = the maximum admissible level according
to the WFD (see Table 1).
Finally, as some criteria are assigned dimensionless values, there is no need to
standardize them. These criteria are the ratio between waste volume and lagoon capacity and
the lagoon conditions (see Table 2). The ratio is result of dividing the amount of produced
waste on the volume of the lagoon where it is located which does not exceed the value of one
since the permission of olive mills requires a lagoon size larger than the expected produced
waste. From another side, the lagoon condition criterion is assigned a Boolean value of zero,
where the lagoon has no constructional problems, or one, where obvious construction
problems are detected.
52 Anas Altartouri, Kalliope Pediaditi, George P. Petropoulos et al.
Acknowledging the context specific nature of risk, this method enables through its design
the assignment of weights according to contextual criterion significance. Hence, criteria that
are deemed more significant indicators of risk in the river basin under consideration can be
assigned higher weights thereby giving them greater importance in the estimation of the risk
of OMW pollution (Mendoza et al., 2002). These weights are assigned separately within
every risk component (Box 1) to show the importance of every criterion in relation to that
component. In order to calculate the weights of multiple criteria, the Analytic Hierarchy
Process (AHP) can be used. AHP is a multi-criteria decision support method which uses
paired comparisons in order to calculate the weights of multiple criteria (Saaty, 1990).
Thereafter, the analysis is separated into two directions according to the addressed
receptors. The first direction addresses the hazard of OMW transported by surface water
bodies and affecting population, while the second direction addresses the hazard of OMW
transported by surface water bodies affecting protected areas. This implies producing two risk
Development of a Decision Support System … 53
maps (one for each direction) as a result of this analysis. This separation is due to the slight
differences in the criteria used for each receptor (see Table 3). Assigning the weight for
criteria is a subjective process which results from the pair wise comparisons. The
specifications of every site greatly influence this process. Hence, for more accurate and
unbiased weight assignment, expert opinion should be considered as well as conducting site
visits.
Unit to Direction
Criterion which it is Component (receptors
assigned addressed)
Cr.1 Number of potentially exposed inhabitants Sub-catch. Component 1 (Box 1)
Cr.2 Possible sedimentation areas Sub-catch. Component 2 (Box 1)
Cr.3 Precipitation Lagoon
Cr.4 Waste volume to lagoon capacity ratio Lagoon Component 3 (Box 1)
Cr.5 Lagoon conditions Lagoon
Cr.6 Length of the flow path to surface water bodies Lagoon Component 4 (Box 1)
Cr.7 Copper (Cu) Lagoon
Cr.8 Iron (Fe) Lagoon
Cr.9 Lead (bp) Lagoon
Water quality parameters
Cr.6 Length of the flow path to NATURA sites Lagoon Component 4 (Box 1)
Cr.7 Dilution degree expressed by stream orders Lagoon Component 5 (Box 1)
In order to obtain results about risk in every sub-catchment within the river basin under
investigation, the standardized criterion values have to be combined based on a well-defined
aggregation rule. A set of formulae has been developed in order to arrive at a particular
evaluation of the risk. These formulae are structured in a calculation model which takes into
account the influence of every lagoon on each sub-catchment within the river basin. The
development of the calculation model was based on the comprehension of the relationships
between risk elements (sources, pathways, and receptors) and components (probability and
magnitude, see Box 1) as well as the conceptual model of the problem. Multi-criteria
54 Anas Altartouri, Kalliope Pediaditi, George P. Petropoulos et al.
R wi ci (5)
where:
R = the risk value of specific sub-catchment; wi = the weight of criterion I; and ci = the
score of criterion i.
At this point, every criterion has its standardized values assigned either to a lagoon or a
sub-catchment (analysis units) as well as it relative weight. When aggregating these criteria,
every sub-catchment will have a single value which indicates the combined magnitude
components (Component 1 and 2, Box 1). Likewise, every lagoon will have a single value
indicating the first and the third probability components (Component 3 and 5, Box 1).
Differently, every lagoon will have multi values regarding the second probability component
as many as possible receiving sub-catchments. These values are obtained using the formulae
below, which are basically substitutions of the formula of WLC, and they are applied for both
directions (receptor groups).
The first step is to calculate the two components of magnitude of consequences
Component 1 and 2, Box 1). This should be applied for both directions of the analysis using
the following formulae:
where:
3
1 probabilitylagoon x C P1nlagoon x * WCP1n
st
(8)
n 1
13
3rd probabilit ylagoon x C P3nlagoon x * WCP 3 n (10.a)
n1
(Receptors: Population)
where:
1st probabilitylagoon x
= the probability of hazard occurring from lagoon x
(Component 3, Box 1);
2 nd probabilit y lagoon x
subcatchment y
= the probability of hazard generated from lagoon x to
reach sub-catchment y (Component 4, Box 1);
3 rd probabilitylagoon x
= the probability of harm resulting from the hazard
generated from lagoon x (Component 5, Box 1),
C P1nlagoon x
= the value of criterion n of the 1st probability
component regarding lagoon x;
C P 2 nlagoon x
subcatchme nt y
= the value of criterion n of the 2nd probability
component regarding lagoon x and sub-catchment y;
C P 3nlagoon x
= the value of criterion n of the 3rd probability component
regarding lagoon x;
WC Pmn
= the weight of criterion n from the mth probability
component.
The next step for both directions is to combine the two magnitude components
(Component 1 and 2, Box 1) into one ‗magnitude‘ component and the three probability
56 Anas Altartouri, Kalliope Pediaditi, George P. Petropoulos et al.
components (Component 3, 4, and 5, Box 1) into one ‗probability‘ component. This can be
done using the following formulae:
where:
The application of the previously mentioned formulae results in a magnitude of the risk in
a specific sub-catchment and a probability of a specific lagoon to contribute this risk. Hence,
the risk value in a sub-catchment with a contribution probability of all possible lagoons is
determined as the following:
xn
PROBABILIT Y lagoon x
RISK subcatchment y MAGNITUDE subcatchment y * x 1 subcatchment y (13)
xn
PROBABILIT Ylagoon x
x 1 subcatchment y max
where:
PROBABILIT Y
x 1
lagoon x
subcatchment y
= the production of the probabilities of all lagoons
above
The flowchart of the proposed calculation model is illustrated in Figure 3, 4 (a and b),
and 5. Figure 3 illustrates a flowchart of calculations for each ‗sub-catchment‘ unit within the
river basin which results in a value of risk magnitude in that sub-catchment (combination of
component 1 and 3, Box 1). Figure 4 (a and b), however, illustrates a flowchart of
calculations for each ‗lagoon‘ unit located in the river basin which results in multi values of
probabilities (combination of component 3, 4, and 5, Box 1) of the corresponding lagoon to
cause risk in each ‗sub-catchment‘ unit within the river basin regarding humans or protected
Development of a Decision Support System … 57
areas, respectively. Finally, Figure 5 illustrates the flowchart resulting in a risk value for a
‗sub-catchment‘ unit based on aggregation of the magnitude (resulted from the flowchart in
Figure 3 for the corresponding sub-catchment) and probability (resulting from the flowchart
in Figure 4 (a or b) for every lagoon within the river basin).
Figure 3. Flowchart of calculation model at the sub-catchment level (for both receptor groups).
58 Anas Altartouri, Kalliope Pediaditi, George P. Petropoulos et al.
Lagoon x
2 nd probabilitylagoon x
sucatchment 1
2 nd probabilitylagoon x
WLC subcatchmnt 2 WLC
. .
2 probabilitylagoon x
nd
Figure 4a. Flowchart of calculation model at the lagoon level (for humans as receptors).
Development of a Decision Support System … 59
Lagoon x
2 nd probabilitylagoon x
sucatchment 1
2 nd probabilitylagoon x
WLC subcatchmnt 2 WLC
. .
2 probabilitylagoon x
nd
Figure 4b. Flowchart of calculation model at the lagoon unit (for protected areas as receptors).
60 Anas Altartouri, Kalliope Pediaditi, George P. Petropoulos et al.
The proposed methodology has been implemented in Keritis watershed in western Crete,
Greece (see Figure 6). A 17,855 ha basin, Keritis watershed faces the risk of OMW pollution
as olive mills are the main agricultural industry in the area. Nine sub-catchments have been
delineated and identified within the basin where five OMW lagoons are located (see Figure
6). Having implemented the proposed methodology at a detailed level of analysis, two risk
maps have been produced (see Figure 6). The first map shows the risk of OMW pollution
which may harm the population, while the other map shows the risk of OMW pollution which
may harm the protected areas (NATURA 2000 sites) in the watershed.
The obtained risk maps clearly show that the developed criteria and calculation method
are compatible with risk generating process in the nature. It can be seen from Figure 6 that
some sub-catchments are estimated to have zero-risk value which is due to the absence of
addressed receptors (Figure 6b, sub-catchment 3 and 8), to the absence of connecting
pathways (Figure 6 a and b, sub-catchment 1, 2, and 3), or both (Figure 6b, sub-catchment 8).
Development of a Decision Support System … 61
On the contrary, the study indicated a high risk value in sub-catchments 9 and 6 regarding
population and protected areas, respectively. The reasons behind the former were the
existence of two hazard sources (Component 3, Box1) and their relatively high proximity to
the surface water bodies (Component 4, Box1) as well as the presence of the highest
population (Component 1, Box1), while for the later, the existence of large protected
NATURA sites (Component 1, Box1) as well as the potential of multi sources to contribute to
risk in this sub-catchment (Component 4, Box1).
To conclude, the breakdown of risk into its primary components (Box 1) and the clear
comprehension of risk generating process and its elements (sources, pathways, and receptors)
and controlling factors (represented by the developed criteria) are the main features of this
method resulting in a realistic and unbiased estimation of risk. In other words, not taking into
consideration the aforementioned issues, risk assessment would no longer be an integrated
approach of several components and just a generic description based on a small number of
criteria.
CONCLUSION
This quantitative approach of risk assessment was built based on the risk assessment
framework proposed by DEFRA guidelines. Having analyzed risk elements, components, and
62 Anas Altartouri, Kalliope Pediaditi, George P. Petropoulos et al.
controlling factors as well as the relationships between them, a conceptual model of risk
generating process has been built and a range of criteria has been formulated. This was
followed by MCA which was based on the development of the criteria. The method proposed
different scaling functions in order to standardize the initial raw value of these criteria. Then,
a calculation model, consisting from a set of formulae, was developed in order to obtain a
quantitative assessment of risk in every sub-catchment within the watershed under
investigation. It has been designed in a way to calculate an overall potential risk a sub-
catchment may be exposed to as a result of the contribution of all point-sources, namely
OMW lagoons, located the watershed. This is one of the main strengths of this method as it is
able to estimate risk in every single sub-catchment by considering and analyzing inputs from
all sources of pollution (lagoons) in the whole basin.
This method can be widely applied in the Mediterranean region where water pollution by
OMW is a common environmental problem. It serves as a decision support tools which
inform risk managers and decision makers about the risk assessment results so they can take
the appropriate management measures, e.g. implementing mitigation measures, properly
locating new OMW lagoon, etc. However, site specification should be taken into account in
when applying this approach. The developed method was implemented in the case study of
Keritis watershed (see Section 4). This implementation has shown the applicability of this
method and its potential as a decision supporting tool.
Besides being a replicable method, a strength point of this methodology is its flexibility
to add or remove criteria as well as changing their weights based on the specific needs of
different case studies without affecting the calculation model. This is a very important issue
since the controlling factors of environmental problems are more likely to change spatially
and temporally. In other words, different cases, for the same environmental problem, may be
subjected to different factors which require a modification of the criteria and their weights for
more accurate results. Moreover, this method can be widely applied for other stressor. It can
be effectively refined to address other point-sources of water pollution. In this case, the
calculation model may need some modification in light of the new problem formulation.
However, the generic frame of this method, consisting of the main steps, is still valid. In
addition, this methodology has the potential to be automated and computerized within a GIS
environment. The calculation model can be programmed using a scripting language to be
applied on the related geo-database. This geo-database should contain the layers where
different criteria are assigned and other calculation elements are found.
However, this method has some weaknesses common to all MCA weight assigning
method. In addition, the proposed method does not assess the overall risk on the river basin
under consideration meaning both surface and ground water bodies. Although the risk of
OMW is more likely to be transported via surface water bodies, due to its nature previously
discussed (section 2), investigating its impacts on the groundwater bodies, as a pathway,
would result in more accurate risk map. However, the stated difficulties in modeling the
natural processes associated with the risk generating process needs further investigation
aiming at simulating such processes.
Development of a Decision Support System … 63
ACKNOWLEDGMENTS
This research was carried out within the context of Remotely Accessed Decision Support
System for Transnational Environmental Risk Management, STRiM, INTERREG IIIB
CADSES project which enabled access to the required data and gave the opportunity to
review the research, in its different stages, from experts in several academic and research
institutions within its consortium.
REFERENCES
Allan, I. J., Mills, G. A., Vrana, B., Knutsson, J., Holmberg, A., Guigues, N., Laschi, S.,
Fouillac, A. & Greenwood, R. (2006). Strategic monitoring for the European Water
Framework Diective. Trends in Analytical Chemistry, 25(7):704-715.
Billington, K. (2005). The River Murray and Lower Lakes Catchment Risk Assessment
Project for Water Quality - Concepts and Method. Environment Protection Authority.
DEFRA, Department for Environment, Food and Rural Affaires (2002). Guidelines for
Environmental Risk Assessment and Management.
Donoghue, A. M. (2001). The design of hazard risk assessment matrices for ranking
occupational health risks and their application in mining and minerals processing.
Occupational Medicine, 51(2):118-123.
Eastman, J. R. (2003). IDRISI Kilimanjaro Guide to GIS and Image Processing. Clark Labs,
Clark University, USA.
enHealth, Council and Department of Health and Ageing (2004). Environmental Health Risk
Assessment, Guidelines for assessing human health risks from environmental hazards.
EPA, Environmental Protection Agency (1998). Guidelines for Ecological Risk Assessment.
Heathwaite, A. L., Dils, R. M., Liu, S., Carvalho, L., Brazier, R. E., Pope, L., Hughes, M.,
Phillips, G. & May, L. (2005). Atiered risk-based approach for predicting diffuse and
point source phosphorus losses in agricultural areas. Science of the Total Environment,
344:225-239.
Kapellakis, I. E., Tsagarakis, K. P., Avramaki, Ch. & Angelakis, A. N. (2006). Olive mill
wastewater management in river basins: A case study in Greece. Agricultural Water
Management, 82:354-370.
Karageorgis, A. P., Nikolaidis, N. P., Karamanos, H. & Skoulikidis, N. (2003). Water and
sediment assessment of the Axios River and its coastal environment, Continental Shelf
Research, 23:1929-1944.
Lapucci, A., Lombardo, S., Retri, M. & Santucci, A. (2005). A KDD based multicriteria
decision making model for fire risk evaluation. Association Geographic Information
Laboratories Europe
[http://plone.itc.nl/agile_old/Conference/estoril/papers/23_Alessandra%20Lapucci.pdf]
[Viewed on 15.06.2011].
Maltby, L. (2006). Environmental risk assessment. Environmental Science and Technology,
22:84-101.
64 Anas Altartouri, Kalliope Pediaditi, George P. Petropoulos et al.
Mendoza, G. A. & Macoun, P. (1999). Guidelines for Applying Multi-Criteria Analysis to the
Assessment of Criteria and Indicators. (Jakarta: Center for International Forestry
Research (CIFOR)).
Mendoza, G. A., Anderson, A. B. & Gertner, G. Z. (2002). Integrating multi-criteria analysis
and GIS for land condition assessment: Part I – Evaluation and restoration of military
training areas. Journal of Geographic Information and Decision Analysis GIDA, 6(1):1-
16.
Morrison, G., Fatoki, O. S., Persson, L. & Ekberg., A. (2001). Assessment of the impact of
point source pollution from the Keiskammahoek Sewage Treatment Plant on the
Keiskamma River - pH, electrical conductivity, oxygen- demanding substance (COD)
and nutrients. Water SA, 27:475-480.
Niaounakis, M. & Halvadakis, C. P. (2006). Olive Processing Waste Management, Literature
Review and Patent Survey. (Oxford: Elsevier)
Ogunfowokan, A. O., Okoh, E. K., Adenuga, A. A. & Asubiojo, O. I. (2005). An assessment
of the impact of point source pollution from a university sewage treatment oxidation
pond on a receiving stream: a preliminary study. Journal of Applied Sciences, 5(1):36-
43.
Paliatziki, A. (2006). Analysis of Environmental Pressures and Impacts in the Koiliaris River
Watershed. Dissertation, Technical University of Crete.
Pediaditi, K., Wehrmeyer, W. & Chenoweth, J. (2005). Brownfield redevelopment,
integrating sustainability and risk management. In Environmental Health Risk III,
Brebbia C A, Popov V, Fayzieva D (eds.), WIT press, 21-30.
Peltonen, L. (2006). Recommendations for a risk mitigation oriented European spatial policy.
Geological Survey of Finland, Special Paper, 42:153-167
Power, M. & McCarrty, L.S. (1998). A comparative analysis of environmental risk
assessment/risk management frameworks. Environmental Science and Technology,
32:224A-231A.
Rekolainen, S., Kämäri, J. & Hiltunen, M. (2003). A conceptual framework for identifying
the need and role of models in the implementation of the Water Framework Directive.
International Journal on River Basin Management, 1(4):347–352.
Saaty, T. L. (1990). How to make a decision: The analytic hierarchy process. European
Journal of Operational Research, 48(1):9-26.
WFD, EU Water Framework Directive (2000). Official Journal of the European communities.
[http://eur-lex.europa.eu/LexUriServ/LexUriServ.do?uri=CELEX:32000L0060:EN:NOT]
[Viewed on 15.06.2011]
In: Ecological Modeling ISBN: 978-1-61324-567-5
Editor: WenJun Zhang, pp. 65-81 © 2012 Nova Science Publishers, Inc.
Chapter 4
L. V. Nedorezov*
The Research Center for Interdisciplinary Environmental Cooperation (INENCO) of
Russian Academy of Sciences, Kutuzov seafront 14, Saint Petersburg 191187, Russia.
ABSTRACT
Publication is devoted to the problem of population time series analysis with various
discrete time models of population dynamics. Applications of various statistical
criterions, which are normally used for determination of mathematical model parameters,
are under the discussion. With a particular example on green oak leaf roller (Tortrix
viridana L.) population fluctuations, which had been presented in publications by
Rubtsov (1992), and Korzukhin and Semevskiy (1992) for three different locations in
Europe, the possibilities of considering approach to the analysis of population dynamics
are demonstrated. For approximations of empirical datasets the well-known models of
population dynamics with a discrete time (Kostitzin model, Skellam model, Moran –
Ricker model, Morris – Varley – Gradwell model, and discrete logistic model) were
applied. For every model the final decision about the possibility to use the concrete
model for approximation of datasets are based on analyses of deviations between
theoretical (model) and empirical trajectories: the correspondence of distribution of
deviations to Normal distribution with zero average was checked with Kolmogorov –
Smirnov and Shapiro – Wilk tests, and existence/absence of serial correlation was
determined with Durbin – Watson criteria. It was shown that for two experimental
trajectories Kostitzin model and discrete logistic model give good approximations; it
means that population dynamics can be explained as a result of influence of intra-
population self-regulative mechanisms only. The third considering empirical trajectory
needs in use more complicated mathematical models for fitting.
*
E-mail address: l.v.nedorezov@gmail.com.
66 L. V. Nedorezov
1. INTRODUCTION
At present in ecological modeling there is a huge number of various mathematical
models, which are used for the description of dynamics of separated populations and
elementary ecosystems with small number of interacting species (Gause, 1934; Kostitzin,
1937; Kot, 2001; Begon, Mortimer, 1981; Varley et al, 1975; Maynard Smith, 1976;
Poluektov et al, 1980; Nedorezov, 1986 and others). At the same time a small number of
models was compared with experimental time series on a quantitative level. In most cases
authors confine themselves by the limits of qualitative comparison of theoretical and real
datasets, which does not allow talking about the adequacy of used models to observations.
Analysis of empirical time series needs in dynamical model using. It is important for
creating forecasts of population size changing in time, for estimation of the probability of pest
population outbreak beginning and so on. It can also be important for finding optimal
methods for population management that leads to the necessity of deep knowledge of all basic
elements of population phase portrait structure (Isaev et al, 1980, 2001, 2009; Nedorezov,
1986, 1999; Berryman, 1981, 1990, 1991).
Abundance of mathematical models, which can be found in modern ecological literature,
leads to appearance of serious problems in selection process – first of all, in finding model,
which gives best fitting for empirical time series. Partly it correlates with absence of the
respective criterions for selection of mathematical expressions, which give adequate
description of one or another biological mechanism affecting population size (Isaev et al,
2001, 2009).
There exists one more important problem in modern ecological modeling – this is the
problem of selection of statistical criteria for estimation values of dynamic model parameters.
In literature there is a big number of various using criterions (Wood, 2001 a, b; Turchin et al.,
2003; Nedorezov, 1986 and others).
But what kind of criteria we have to choose in one or another situation and why – in most
cases this is a puzzle. In current paper we analyze these marked problems – the problem of
selection of mathematical model (and analysis of suitability of selected model to fitting real
trajectories of population dynamics) and the problem of selection the best statistical criteria
for dynamic model parameter estimations. Obtained results are applied to analysis of datasets
on green oak leaf roller (Tortrix viridana L.) population dynamics (Rubtsov, 1992;
Korzukhin, Semevskiy, 1992; Nedorezov, Sadykova, 2005, 2008; Nedorezov, Sadykov,
Sadykova, 2010).
x k 1 x k F ( x k , ) , (1)
Analysis of Green Oak Leaf Roller Population Dynamics in Various Locations 67
where x k is population size (or population density) in k -th year (or in k -th moment of
observation), F is a population birth rate (Isaev et al, 1980, 2001, 2009),
x k 1
yk F ( xk , ) , (2)
xk
and is a vector of (unknown) model parameters, then the natural question arises – why we
cannot choose and use one of the models from the table 1? It is well-known that among the
recursive equations from this table one can find the models with very rich set of dynamical
regimes. Moreover, part of these models were applied to the description of dynamics of
various natural populations with success (see, for example, May, 1975; Hassell, 1975, 1978;
Varley et al, 1975; Nedorezov, 1986; Nedorezov, Nedorezova, 1995; Kot, 2001). Moreover,
if simple model of type (1) gives us a sufficient good description of population dynamics
(more precisely, applied statistical criterions don‘t allow us to reject the hypothesis about
suitability of model for fitting of considering time series), then we really haven‘t a
background for creating new and more difficult mathematical model, which contains the first
model as a particular case. It is obvious, that for more complicated model we will also have
satisfactory results.
Taking it into account, we have to note that in our opinion at every time we have to start
the process of population dynamics analysis with a group of simplest mathematical models,
which describe the influence on population the minimal set of regulative mechanisms. For
example, we can start with models, which are presented in table 1: all these models describe
the influence on population the intra-population self-regulative mechanisms only. If we can
prove that it is impossible to explain analyzing datasets as a result of influence of these self-
regulative mechanisms only, we have to use more complicated models, which contain
additional equations for population regulator dynamics – for predators, parasites, climatic
factors etc.
Let‘s assume that with the help of any statistical criteria and for given sample the model
parameters were estimated. The following question arises – how can we check the
correspondence between theoretical population values (which can be obtained with the help
of mathematical model) and empirical time series?
There exists a unique way for solution of this problem – we have to analyze the time
series, which is organized by the differences between theoretical (model) and empirical
values. Following the common ideas we‘ll assume that model gives us a good approximation
of empirical sample if the next requirements are truthful: there are no reasons for rejecting the
hypothesis that average for residuals is equal to zero and distribution of residuals is Normal
(that can be checked, for example, by the Kolmogorov – Smirnov test, and Shapiro – Wilk
test etc.; Bolshev, Smirnov, 1983; Shapiro et al, 1968); there is also the absence of serial
correlation in a sequence of residuals (Durbin – Watson test; Draper, Smith, 1986, 1987). If
these conditions are truthful altogether it means that there are no reasons for rejecting a
hypothesis about suitability of considering model for fitting of empirical time series.
If statistical criterions didn‘t allow us to reject the hypothesis about suitability of model
for fitting of analyzing time series we have a possibility to solve another one problem – we
can try to identify a population dynamics regime. We have to note that there are no reasons to
68 L. V. Nedorezov
say that population dynamics corresponds to dynamical regime, which is realized in model
with obtained (estimated) parameters. The initial sample is a sequence of stochastic values,
and, respectively, estimated model parameters are the stochastic values too. These parameters
don‘t equal to real population characteristics. It means that there exists a serious problem in
determination of real population dynamics type. On the other hand, it is possible to estimate
the probabilities that observed trajectory corresponds to population extinction, or to
population asymptotic stabilization at non-zero level etc. In other words, analysis of initial
sample can allow us to obtain a distribution of dynamical regimes, which can be realized for
population.
Estimation of probabilities of realization for population one or another dynamical regime
can be provided by the following way. In a space of model parameters the confidence
domains can be obtained with well-known methods (Draper, Smith, 1986, 1987). In the same
space of model parameters there is a set of bifurcation surfaces that depends on selected
model. These bifurcation surfaces cut confidence domains onto sub-domains, which
correspond to one or another dynamical regime. Respectively, we can estimate desired
probabilities with the help of standard Monte-Carlo methods.
Thus, the general diagram of population dynamics analysis must include the following
basic stages:
If for all selected models statistical criterions show negative results we have to choose a
group of more complicated mathematical models. These models from a new group may
describe the influence of some additional regulators onto population dynamics (parasites,
quality of food, weather factors etc.), may take into account the existence of time lag in a
reaction of self-regulative mechanisms onto population size changing, or may have additional
variables for sex groups of individuals, age groups or other intra-population structures.
N
Q( ) x k* x k*1 F ( x k*1 , )
2
min
(3)
k 2
Analysis of Green Oak Leaf Roller Population Dynamics in Various Locations 69
It is possible to point out various modifications of criteria (3). For example, when we use
expression (2) for birth rates in (3), the expression for minimizing functional has the
following form:
N
Q( ) y k*1 F ( x k*1 , )
2
min
, (4)
k 2
where y k* are the values of birth rates calculated for empirical time series. When we use log-
transformed values of population size and respective log-transformed expressions in (1)
(sometimes it can lead to strong simplification of final formula) expression (3) can be
presented as follows:
2
N
Q( ) log( xk* ) log( xk*1 ) log(F ( xk*1 , )) min
(5)
k 2
These expressions (3)-(5) have one common property: model (1) is used in these
*
formulas as non-linear regression curve but not as dynamical model. Real trajectory {x k }
compares with set of values, which don‘t belong to any model trajectory altogether (more
precisely, these points belong to one and the same trajectory in unique ideal variant, which is
unrealistic for real datasets). In other words, in expressions (3)-(5) we compare the objects of
two qualitatively different types.
In this situation the following natural question can arise – why we decided to use one-
step ahead regression curve for estimation model parameters? May be, two-step ahead or
three-step ahead can give us better results? What kind of criteria gives us the best results and
why?
Moreover, in expressions (3)-(5) there are the ―doubl e standards‖: if we use (3), (4), or
(5) we assume that elements in initial empirical sample have two qualitatively different
properties. For example, within the limits of first element of sum (3) we assume that x 2* is
any constant, which was measured with Normal distributed error. But in the next element of
sum (3) we assume that expression F ( x2* , ) gives us a theoretical value of population, we
*
have to observe in ecosystem (if amount x 3 was measured without any errors). It means that
in second bracket in sum (3) we assume that amount x 2* was estimated without errors.
These are the reasons we think that criterions of the type (3)-(5) cannot be applied for the
estimation of real population parameters. On the other hand, these criterions can be applied,
for example, for constructing forecasts of population size changing.
One of most perspective ways for model parameter‘s estimations is in using of ―gl obal
fitting‖ (Wood S, 2001a, b), when we try to find the best trajectory (in a set of all model
trajectories), which gives us the best fitting for real trajectory. Let ~
x k ( , x1 ) be a solution of
model (1) for given vector and initial value of population size x1 (we assume that this
70 L. V. Nedorezov
value is unknown model parameter too). Then criteria can be presented in the form (Tonnang
et al., 2009; Nedorezov, Lohr, Sadykova, 2008):
N
Q( , x1 ) ( x k* ~
x k ( , x1 )) 2 min
(6)
, x1
k 1
Note, that in (6) there are no ―doubl e standards‖ and all elements in the sample have the
similar properties. It is possible to point out some other important properties of criteria (6).
For example, if we have the same sample but in model we have more equations (in particular,
for predators, and respective variable in model is invisible) we can use (6) for the estimation
of all model parameters without modification of expression (6). At the same time criteria (3)
needs in additional information and in serious modification.
Below we illustrate this approach to population dynamics analysis on an example of
green oak leaf roller fluctuations (Rubtsov, 1992; Korzukhin, Semevskiy, 1992; Nedorezov,
Sadykova, 2005, 2008; Nedorezov, Sadykov, Sadykova, 2010). Choosing of this object
depends on the existence of several good empirical time series and diversity of biological
opinions about green oak leaf roller dynamics type. Partly it correlates with absence of
common opinion about main population regulators (Hunter et al., 1997; Hassell et al., 1998;
Rubtsov, 1983).
A big number of various publications are devoted to the problems of green oak leaf roller
population dynamics and its mathematical models (see, for example, Varley et al., 1975;
Korzukhin, Semevskiy, 1992; Hunter et al., 1997; Rubtsov, 1983, 1992; Rubtsov, Shvytov,
1980 and others). But a set of serious problems (in particular, the problem about main
population regulators) is open up to current moment and under the discussion (Hunter et al.,
1997; Hassell et al., 1998; Nedorezov, Sadykova, 2005, 2008).
Analysis of phase portrait structures, which had been provided in Rubtsov‘s publications
(Rubtsov, 1983, 1992; Rubtsov, Shvytov, 1980), allowed him to describe the basic laws of
population dynamics. In particular, the author had been showed that in some locations tortrix
could realize an outbreak (Isaev et al., 1980, 2001, 2009; pulse eruptive outbreak in
Berryman‘s classification of insect population dynamic types; Berryman, 1990, 1991). This
regime can be realized within the framework of predator – prey system dynamics, and
characterizes by the existence of three non-zero stationary states in a phase space. Following
the basic stages of population dynamics analysis described above, before using difficult
mathematical models including equations for several interacting populations we have to
check the possibilities of simpler models (table 1). First of all, we have to be sure that it is
impossible to explain observed trajectories from the stand point of influence of self-regulative
intra-population mechanisms only.
Analysis of Green Oak Leaf Roller Population Dynamics in Various Locations 71
Use the models from table 1 means that we a‘priori assume that phase portrait of
population dynamics can be characterized by the existence of one non-zero stationary state (if
population doesn‘t eliminate for every initial values of population size). This stationary state
can be stable or unstable point and in last case we can observe cyclic or chaotic population
fluctuations. In Isaev – Khlebopros classification of insect population dynamics (Isaev et al.,
2001, 2009) such kind of species belong to a group of prodromal species (but not to group of
eruptive species, which can realize an outbreak). This hypothesis will have a good
background if and only if for all analyzed real trajectories (Figure 1) we find at least one
model from the table 1, which gives a good fitting for observed trajectory and all statistical
tests give us positive results.
All models we used for fitting of considering trajectories are presented in table 1.
Parameters of discrete logistic model (table 1) must satisfy the condition ab 4 (Maynard
Smith, 1976; Poluektov et al., 1980; Nedorezov, Nedorezova, 1995). But for fitting of real
time series we have used modified discrete logistic model:
ax (b xk ), xk b
xk 1 k
0, xk b
This model has no additional limits for its non-negative parameters a and b . But the
origin (if ab 4 ) becomes a complicated stationary state.
This additional assumption that product ab can be bigger then 4 can be interpreted as
follows: we assume that population size can intersect the limit level b . But it leads
Analysis of Green Oak Leaf Roller Population Dynamics in Various Locations 73
If we assume that dynamics of green oak leaf roller on Figure 1a corresponds to pure
stochastic fluctuations near any stationary level x s , then x s 13.671 with SE 2.243
(standard error). Minimum value for functional form (6) is equal to 2324.1. This value is
major for all minimum values of functional form (6) for all models from the table 1 for first
time series. Applications of Kolmogorov – Smirnov test ( d 0.1247 with p 0.2 ) and
Shapiro – Wilk test ( W 0.9145 with p 0.0584) show that there are no reasons for
rejecting the hypothesis about normality (with zero average) of the distribution of residuals
(Bolshev, Smirnov, 1983; Shapiro et al., 1968); Durbin – Watson criteria shows
( d 1.5619; for sample size 22 and for one predictor variable the realization of inequality
d d L 1.24 means that there is a negative serial correlation in a sequence of residuals; if
1.43 d U d 2 there is no serial correlation; critical values d L and d U are presented
for 5% level of significance; for 1% level of significance the values for critical levels are
equal to 1.00 and 1.17 respectively), that there is no serial correlation in a sequence of
residuals (Draper, Smith, 1986, 1987). Thus, there are no reasons for rejecting the hypothesis
that on Figure 1a we have pure stochastic fluctuations near average value.
Realization of last hypothesis means that population regulators are very weak on the
considering interval of population size changing. If we assume, that population dynamics can
be described by the Kostitzin model the value of loss-function Q (6) is less than in previous
case (table 2). At the same time there are no reasons for rejecting the hypothesis that Kostitzin
model is suitable for fitting of experimental time series (table 3). Note that coefficient b ,
which describes in model the influence of self-regulative mechanisms on population size
changing, is small enough.
74 L. V. Nedorezov
Table 2. Model parameter’s estimations and values of functional form Q (6) for all time
series on green oak leaf roller changing in time
Models x0 a b Q
Estimations for first time series (Figure 1a)
1 21.89 1.394 0.032 2190.8
2 15.472 0.174 24.4 1238.0
3 21.636 27.755 0.048 2194.8
4 0.874 40.516 1.467 1476.0
5 8.728 27.04 0.232 1515.6
Estimations for second time series (Figure1b)
1 0.0074 28603.0 1436.2 3571.6
2 2.19 0.116 37.0 1261.6
3 0.682 19.92 3.514 3571.9
4 20.08 1.414 0.119 3940.7
5 0.972 24.26 0.158 3337.6
Estimations for third time series (Figure1c)
1 2.7∙10-12 731489.6 60263.0 6442.3
2 0.103 0.0896 50.8 1310.9
3 0.00014 12.138 8.24 6442.3
4 7.85∙10-258 10.78 0.95 6456.2
5 0.05 147.1 0.708 4578.5
Left Right
Average±SE KS2 SW3 DW4
limit1 limit1
Results for first time series
1 -0.034±2.178 -4.562 4.495 0.1187/p>0.2 0.9311/p=0.129 1.7064
2 0.87±1.626 -2.511 4.252 0.1476/p>0.2 0.9361/p=0.165 1.6634
3 -0.038±2.18 -4.571 4.495 0.1203/p>0.2 0.9302/p=0.124 1.7026
4 0.448±1.785 -3.263 4.16 0.1131/p>0.2 0.9376/p=0.177 0.7935
5 -0.454±1.809 -4.215 3.307 0.1027/p>0.2 0.95/p=0.315 0.7027
Results for second time series
1 0.0042±2.78 -5.778 5.786 0.1673/p>0.2 0.9253/p=0.098 0.995
2 0.7093±1.645 -2.712 4.131 0.1468/p>0.2 0.9793/p=0.905 2.263
3 -0.025±2.781 -5.808 5.757 0.167/p>0.2 0.9259/p=0.101 0.995
4 -0.0005±2.92 -6.074 6.073 0.18/p>0.2 0.9305/p=0.126 0.9972
5 -0.126±2.69 -5.715 5.463 0.136/p>0.2 0.951/p=0.331 1.0197
Results for third time series
1 0.1392±3.148 -6.344 6.622 0.2222/p<0.1 0.7442/p=0.00002 0.935661
2 3.1161±1.276 0.488 5.744 0.2379/p<0.1 0.8524/p=0.0016 1.961968
3 0.1379±3.148 -6.346 6.622 0.2222/p<0.1 0.7441/p=0.00002 0.935704
4 0.1011±3.152 -6.39 6.592 0.2233/p<0.1 0.7426/p=0.00002 0.935025
5 4.06±2.527 -1.144 9.264 0.3405/p<0.01 0.6281/p<10-5 1.824773
1
Limits for 95% confidence interval
2
KS – values and probabilities of Kolmogorov – Smirnov criteria
3
SW – values and probabilities of Shapiro – Wilk criteria
4
DW – values of Durbin – Watson criteria.
Analysis of Green Oak Leaf Roller Population Dynamics in Various Locations 75
The best result is observed for discrete logistic model (table 2); good results are also
observed for Morris – Varley – Gradwell model and Moran – Ricker model. For all models
we can conclude that the estimated values of population parameters belong to ―bi ological‖
zone. From the table 3 we can see that for all models we cannot reject the hypotheses that the
distributions of residuals are Normal with zero averages. At the same time for residuals for
Morris – Varley – Gradwell model and for Moran – Ricker model the negative serial
correlations are observed. It allows us to say that these models cannot be applied for fitting of
considering time series. For other models the Durbin – Watson criteria shows that there are no
serial correlations.
On Figure 2 empirical trajectory of tortrix fluctuations compares with theoretical
trajectories, which were obtained for estimated parameters (table 2) for Kostitzin model and
discrete logistic model. It is important to note that for estimated parameters discrete logistic
model predicts population elimination in 1992; it doesn‘t correspond to reality and can be
interpreted as additional limit for prognostic properties of this model.
On Figure 3 there are the 99% confidence domain for parameters a and b of discrete
logistic model (with fixed value x 0 15.472 ) together with some bifurcation curves.
Taking into account that limits of confidence domain don‘t intersect curve ab 3 we can
conclude that probability of realization of the regime of population stabilization or the regime
of population elimination (within the framework of this model) is less than 0.01 . The
observed minimum for functional (6) is occupied in intersection of straight lines L1
( x 0.174) and L2 ( y 24.4 ) (Figure 3).
Figure 2. Population density changing in time: curve 1 corresponds to empirical time series (Figure 1a),
curve 2 is the trajectory of Kostitzin‘ model, and curve 3 is the trajectory of discrete logistic model.
76 L. V. Nedorezov
Figure 3. Domain on the plane (a, b) where values of functional (6) is less than 2217.3 (with fixed
value x 0 15.472 ) for discrete logistic model. ab 2 , ab 3 , and ab 4 are the bifurcation
curves. Intersection of lines L1 and L2 gives a point of estimated minimum for loss-function (6).
On Figure 4 there are the limits of confidence domains for Kostitzin model parameters on
the plane (a, b) at fixed value of initial population size x 0 ( x 0 21.89 , table 2). As one
can see from this picture, boundaries k of confidence domains intersect bifurcation line
a 1 , but the probability of asymptotic population extinction is very small.
Finally, the provided analysis and comparison of theoretical trajectories with observed
time series (Figure 1a) show that population fluctuations can be explained as a result of
influence of self-regulative intra-population mechanisms only. We have no reasons to reject
the hypothesis that analyzed empirical trajectory corresponds to asymptotic stabilization at
non-zero level (Kostitzin model). Analysis of modified discrete logistic model shows (Figure
3) that with big probability there are the periodic fluctuations in population dynamics.
Figure 4. Curves 1 , 2 3 are the boundaries of confidence domains for 90%, 95% and 99%
and
respectively for Kostitzin‘ model. Bifurcation line a 1 is the boundary for domains of population
extinction ( a 1) and its asymptotic stabilization at non-zero level ( a 1 ).
Analysis of Green Oak Leaf Roller Population Dynamics in Various Locations 77
If we assume that dynamics of green oak leaf roller on Figure 1b corresponds to pure
stochastic fluctuations near any stationary level x s , then x s 18.95 with standard error for
average 2.922. Minimum of functional form (6) is equal to 3943.9. Applications of
Kolmogorov – Smirnov test ( d 0.1861 with p 0.2 ) and Shapiro – Wilk test
( W 0.9197 with p 0.0749) show that there are no reasons for rejecting the hypothesis
about normality of the distribution of the residuals between theoretical and empirical values
(Bolshev, Smirnov, 1983; Shapiro et al., 1968); Durbin – Watson test shows ( d 0.9927 )
that there is the negative serial correlation (Draper, Smith, 1986, 1987) in a sequence of
residuals. Consequently, the hypothesis that on Figure 1b there is the stochastic fluctuations
near average must be rejected.
Note, that obtained estimations for Kostitzin model (table 2) belong to non-biological
zone for population parameters (maximum value of birth rate a is much bigger than
maximum fecundity of individuals). Respectively, it can be one of the reasons for rejecting
the hypothesis about suitability of this model for fitting of empirical time series.
Analysis of the sequence of residuals shows (table 3) that there is no negative serial
correlation for discrete logistic model only. It means that we have reasons for rejecting the
hypotheses about suitability of considering models for fitting of empirical time series. On the
other hand all used tests don‘t allow us to reject the same hypothesis for discrete logistic
model.
Thus, like in previous case population fluctuations can be explained as a result of
influence of self-regulative intra-population mechanisms only. On the other hand, exploitation
of discrete logistic model (table 1) meets with serious problems. In particular, for estimated
values we cannot determine asymptotic regime of population dynamics: model shows that
population must extinct after 14 years (in 2002) that doesn‘t correspond to reality. On Figure
5 real trajectory of tortrix fluctuations compares with theoretical trajectory, which was
obtained for estimated parameters (table 2) for discrete logistic model.
Figure 5. Population density changing in time: curve 1 corresponds to empirical time series (Figure 1b),
and curve 2 is the trajectory of discrete logistic model.
78 L. V. Nedorezov
CONCLUSION
In population dynamics analysis it is possible to mark the following important steps:
1) At the beginning we have to select a model or group of models, which have the
similar properties (for example, all models describe the influence of intra-population
self-regulative mechanisms on population size changing only in simplest variant,
there are no time lags in reactions of self-regulative mechanisms etc.), or we must
construct a new model if existing models don‘t describe some important elements of
population structure, regulative mechanisms etc. In other words, we have to construct
a new model if all existing models don‘t correspond to considering situation. During
the selection process it is very important to take into account that every selected
model has its own limits for application to real datasets. In particular, the choosing of
the Kostitzin model means that a‘priory we assume that population size can change
monotonously only. Respectively, all deviations of empirical values from this law
can be (and must be) explained as a result of influence of external stochastic factors
(for example, weather conditions).
2) On the next step we have to select a statistical criterion for determination of model‘s
parameters. If model is only assumed for obtaining forecasts of population size
changing in time it is expedient to use criterions of the type (3)-(5). But if the model
isn‘t assumed for the solution of especially practical problems only, and the main
goal of time series analysis is in determination of population dynamics type, it is
much better to use criterions of the type (6).
Analysis of Green Oak Leaf Roller Population Dynamics in Various Locations 79
If all used statistical criterions don‘t allow rejecting the hypothesis about suitability of
considering model for approximation of time series, it gives respectively to present correct
solutions for some other problems. For example, it allows us to solve the problem on
identification of population dynamics type, or at least to describe the set of mechanisms
influenced the population size. Also it can give the solution of the problem of absent
datapoint estimation (―hol es‖ in time series), give a correct (in mathematical sense)
description of the problem of optimal management of population etc.
Application of several mathematical models (table 1) with discrete time for the
approximation of green oak leaf roller population dynamics (Figure 1) showed that in the first
case (Figure 1a) we couldn‘t reject the hypothesis that observed dynamical regime
corresponds to pure stochastic fluctuations near any stationary level. The influence of
regulative mechanisms is considerably weak.
For the second case (Figure 1b) it was obtained that observed fluctuations can be
explained as a result of influence of intra-population self-regulative mechanisms only. In the
last case it was impossible to find good fitting for all models. It gives us a background for the
following hypothesis: in this situation the dynamics of green oak leaf roller population cannot
be explained as a result of influence of self-regulative mechanisms only. It means also that for
fitting of this time series we have to use more difficult mathematical models, which describe
the influence of some other regulators on population dynamics or take into account some
additional intra-population effects.
REFERENCES
Begon M., Mortimer M. Population Ecology: a united study of animals and plants. Oxford
etc.: Blackwell sci. publ., 1981
80 L. V. Nedorezov
Berryman AA. Population systems: a general introduction. New York: Plenum Press, 1981
Berryman AA. 1990. Identification of Outbreaks Classes. Math. Comput. Modelling, 13: 105–
116.
Berryman AA. 1991. Population theory: an essential ingredient in pest prediction,
management and policy making. Am. Ent., 37: 138-142.
Beverton RJ., Holt SJ. 1957. On the dynamics of the exploited fish populations. Great Brit.
Min. Agr. Fish, Food, Fish. Invest., 2-19: 1-533.
Bolshev LN., Smirnov NV. Tables of Mathematical Statistics. Moscow: Nauka, 1983. (in
Russian)
Draper NR., Smith H. Applied Regression Analysis. V.1. Moscow: Finance and Statistics,
1986. (in Russian)
Draper NR., Smith H. Applied Regression Analysis. V.2. Moscow: Finance and Statistics,
Moscow, 1987. (in Russian)
Gause GF. The Struggle for Existence. Baltimore: Williams and Wilkins, 1934.
Hassell MP. 1975. Density-dependence in single-species populations. J. Anim. Ecol., 44: 283-
295.
Hassell MP. The Dynamics of Arthropod Predator-Prey Systems. Princeton: Princeton
University Press, 1978.
Hassell MP., Crawley MJ., Godfray HCJ., Lawton JH. 1998. Top-down versus bottom-up and
Ruritanian bean bug. Proc. Nat. Acad. Sci USA, 95: 10661-10664.
Hunter MD., Varley GC., Gradwell GR. 1977. Estimating the relative roles of top-down and
bottom-up forces on insect herbivore populations: A classic study revisited. Proc. Natl.
Acad. Sci. USA, 94: 9176-9181.
Isaev AS., Khlebopros RG., Kondakov YP., Kiselev VV., Nedorezov LV., Soukhovol‘sky
VG. 2009. Forest Insect Population Dynamics. Euroasian Entomol. J., 8: 3-115.
Isaev AS., Khlebopros RG., Nedorezov LV., Kondakov YP., Kiselev VV., Soukhovol‘sky
VG. Forest Insect Population Dynamics. Moscow: Nauka, 2001. (in Russian).
Isaev AS., Nedorezov LV., Khlebopros RG. 1980. Qualitative Analysis of the
Phenomenological Model of the Forest Insect Number Dynamics. Pest and Pathogen
Control, 9: 1-44.
Korzukhin MD., Semevskiy FN. Forest sinecology. Saint Petersburg: Gidrometeoizdat, 1992.
(in Russian)
Kostitzin VA. La Biologie Mathematique. Paris: A.Colin, 1937.
Kot M. Elements of Mathematical Ecology. Cambridge: Cambridge University Press, 2001.
May RM. 1975. Biological populations obeying difference equations: stable points, stable
cycles and chaos. J. Theor. Biol., 51: 511-524.
Maynard Smith JM. Models in ecology. Moscow: Mir, 1976. (in Russian)
Maynard Smith JM, Slatkin M. 1973. The stability of predator-prey systems. Ecology, 54:
384-391.
Moran PAP. 1950. Some remarks on animal population dynamics. Biometrica, 6: 250-258.
Morris RF. 1959. Single-factor analysis in population dynamics. Ecology, 40: 580-588.
Nedorezov LV. Modeling of Forest Insect Outbreaks. Novosibirsk: Nauka, 1986. (in Russian)
Nedorezov LV. 1999. Restoration of Phase Portrait Structure for the Dynamics of a Forest
Pest, the Pine Moth (Dendrolimus pini L.). Ecol. Modell., 115: 35-44.
Analysis of Green Oak Leaf Roller Population Dynamics in Various Locations 81
Nedorezov LV., Lohr BL., Sadykova DL. 2008. Assessing the importance of self-regulating
mechanisms in diamondback moth population dynamics: Application of discrete
mathematical models. J. Theor. Biol., 254: 587–593.
Nedorezov LV., Nedorezova BN. 1995. Correlation between Models of Population Dynamics
in Continuous and Discrete Time. Ecological Modelling, 82: 93-97.
Nedorezov LV., Sadykov AM., Sadykova DL. 2010. Population dynamics of green oak leaf
roller: applications of discrete-continuous models with non-monotonic density-dependent
birth rates. J. Gen. Biol., 71: P. 41-51.
Nedorezov LV., Sadykova DL. 2005. Toward a problem of selection of mathematical model
of population dynamics (on an example of green oak leaf roller). Euro-Asian Ent. J., 4:
263-272.
Nedorezov LV., Sadykova DL. 2008. Green oak leaf roller moth dynamics: An application of
discrete time mathematical models. Ecological Modelling, 212: 162-170.
Poluektov RA., Pykh JuA., Shvitov IA. Dynamic models of ecological systems. Leningrad:
Gidrometeoizdat, 1980. (in Russian)
Ricker WE. 1954. Stock and recruitment. J. Fish.Res. board of Canada, 11: 559-623.
Rubtsov VV. 1983. Mathematical model for development of leaf-eating insects (oak leaf
roller taken as an example). Ecological Modelling, 18: 269-289.
Rubtsov VV. Models of oscillating processes in forest ecosystems. Moscow: Forest Institute,
1992. (in Russian)
Rubtsov VV., Shvytov IA. 1980. Model of the dynamics of the density of forest leaf-eating
insects. Ecological Modelling, 8: 39-47.
Shapiro SS., Wilk MB., Chen HJ. 1968. A comparative study of various tests of normality. J.
of the American Statistical Association, 63: 1343-1372.
Skellam JG. 1951. Random dispersal in theoretical populations. Biometrika, 38: 196-218.
Tonnang H., Nedorezov LV., Owino J., Ochanda H., Löhr B. 2009. Evaluation of discrete
host –parasitoid models for diamondback moth and Diadegma semiclausum field time
population density series. Ecological Modelling, 220: 1735-1744.
Turchin P., Wood SN. Ellner SP. et al. 2003. Dynamical effects of plant quality and
parasitism on population cycles of larch bud moth. Ecology, 84: 1207-1214.
Varley GC., Gradwell GR. 1960. Key factors in population studies. J. Anim. Ecol., 29: 399-
401.
Varley GC., Gradwell GR. 1970. Recent advances in insect population dynamics. Ann. Rev.
Ent., 15: 1-24.
Varley GS., Gradwell GR., Hassell MP. Insect Population Ecology. An analytical approach.
London: Blackwell Scientific Publications, 1975.
Wood SN. 2001a. Minimizing model fitting objectives that contain spurious local minima by
bootstrap restarting. Biometrics, 57: 240-244.
Wood SN. 2001b. Partially specified ecological models. Ecol. Monographs, 71: 1-25.
In: Ecological Modeling ISBN: 978-1-61324-567-5
Editor: WenJun Zhang, pp. 83-96 © 2012 Nova Science Publishers, Inc.
Chapter 5
ABSTRACT
In the last decades, numerical modelling has gained increasing consensus in the
scientific world, and particularly in the framework of behavioural and population
ecology. Through numerical models it is possible to reconstruct what is observed in the
environment or in the laboratory and to get a more in-depth comprehension of the factors
regulating the phenomena under examination.
Numerous approaches have been developed in this framework, but probably one of
the most promising is the individual-based modelling. With this type of approach it is
relatively straightforward to investigate aspects related to the ecology of a population
starting from the characterisation of processes taking place at the scale of the individual
organism.
This contribution is intended to provide a general view of the main features of the
individual-based models and of their peculiarities in comparison to other modelling
strategies. Special emphasis will be given to applications in the field of phyto- and
zooplankton ecology and behaviour, and results from the available literature on this topic
will be used as examples.
1. INTRODUCTION
Natural processes are quite often the result of complex and hardly predictable interactions
among the components of a system. Experimental studies, either in situ or in the laboratory,
provide important insights into some of these mechanisms; however, owing to the intrinsic
complexity of these processes and of their interactions, only a few of them can be investigated
at a given time. Numerical modelling often represents an affordable approach to get a more
holistic view of natural systems and of the processes acting in them. As discussed in Mac
Nally (1997), experimental methodology is typically based on the Popperian paradigm of the
hypothesis-deduction approach. Numerical models come to the aid in understanding
ecological processes: in a correct approach, the scope of numerical simulations is to
reproduce natural phenomena and provide new elements to understand their underlying
dynamics, rather than a blind acceptance of model results (Wissel, 1992; Grimm, 1999).
Models are thus ―pur poseful representations‖ (Starfield et al., 1990) and great problem-
solving tools to make the main properties of the considered system emerge and to explain the
observed phenomenon (Grimm, 1999). In a numerical model, the dynamics of several
variables are integrated through interactions of processes (Wroblewski, 1983). As a general
rule of thumb, when modelling a complex system the simplest components should be
identified and the interactions among them and with the environment investigated
(Wroblewski, 1983).
As underlined by Judson (1994), two fundamental aspects make ecology different from
other natural sciences: the lack of first principles but on the other hand the strong presence of
the unifying principle of Darwinian evolution. Since May (1974 and 1976), increasing
awareness has been accumulated about the possible development of chaotic dynamics in even
simple systems. Models of population dynamics are not an exception to this rule (Łomnicki,
1999), and for this reason ecological modelling is still nowadays in a continuous progress.
Over the years several modelling approaches have been developed to address key ecological
topics. Among them, individual-based models (IBMs) have emerged as a promising
framework to relate the individual behaviour with the patterns observed at community and
population levels (Grimm, 1999).
Classically, ecological state-variable models describe a population in terms of bulk
properties averaged over a large number of individuals, without considering the variability
among them, and use continuous functions changing in time and space (Fennel and Osborn,
2005). But within a population individuals are not all the same, and their differences are
reflected in the structure and dynamics of the population itself (Łomnicki, 1999). In an IBM
the focus of the interest is the individual, considered as the central elemental component of
the system (with an approach equivalent to that of experimental biology), and a population is
assumed as made up of individuals differing in their properties (Uchmański and Grimm,
1996). IBMs belong to the family of the agent-based models (or multi agent-based models), a
modelling approach designed to investigate the interactions among numerous components of
a system. These models are a natural extension of the Ising model (Ising, 1925) and of the
cellular-automata like models (Wolfram, 1994). In an IBM the modelled ―agent ‖ is simply an
individual, treated as a unique and discrete entity with at least one property evolving through
time. IBMs explicitly include heterogeneity among individuals (e.g., spatial location, body
size, physiological parameters, etc.), and this makes them particularly suitable to link the
Individual Based Modelling of Planktonic Organisms 85
individual with aggregated levels (e.g., population dynamics, community structure, spatial
distribution, etc.), while at the same time stabilising the model itself (Cope, 2005). By
modelling the probabilistic behaviour of individual organisms and averaging over a
reasonably high number, IBMs use a ―bot tom-up‖ approach (Souissi et al., 2005) and are
capable of delineating population-level dynamics as emergent properties due to the
interactions among the individuals and between the individual and its environment
(Railsback, 2001).
Since the rules governing the individual can account for several processes, IBMs permit
more realistic assumptions than those used for state variable models (Souissi et al., 2004). In
addition, using observed biological entries IBMs do not introduce any mathematical artifact
in the numerical representation (Scheffer et al., 1995). Some IBMs are defined as spatially
explicit, when the individual is associated with a position in space which can be either
continuous or discrete, and may also include mobility if the individual is allowed to move
inside its environment (as for simulations of animal behaviour). IBMs can also be considered
as i-state configuration models (Metz and Diekmann, 1986; Caswell and John, 1992; Maley
and Caswell, 1993), where for each individual a set of i-states (e.g., age, size, weight, etc.) is
defined at each time step.
The first models using individuals as basic units date to the early 1980s (e.g., DeAngelis
et al., 1980; Beyer and Laurence, 1980), but it was only after the work by Huston et al. (1988)
that the IBM approach has been unequivocally defined. The possibility of considering IBMs
as a unifying ecological theory was discussed in Huston et al. (1988) and then reviewed a
decade later by Grimm (1999), while Fennel and Osborn (2005) proposed an alternative
framework to relate state variables and individuals.
In the literature a number of seminal reviews about IBMs are available, focusing on the
potentials and on the applications of this modelling framework to ecological issues (e.g.,
DeAngelis et al., 1990 and 1994; DeAngelis and Gross, 1992; Judson, 1994; Uchmański and
Grimm, 1996; Grimm, 1999; Grimm et al., 1999; Łomnicki, 1999). Of course, all that glitters
ain‘t gold and some proviso must be mentioned. Since IBMs can virtually model all the
individuals of a population, a downside is the costly demand of computational and storage
resources to let the model run fluently. While the computing performances of present personal
computers have made giant leaps and are still evolving very rapidly, some resampling
techniques are commonly adopted. Since realistic numbers of individuals need to be
represented in the model for a reliable representation of natural phenomena, reducing the
number of modelled entries does not represent a suitable solution. This procedure would in
fact decrease the variability in the variables modelled, with possible development of irregular
dynamics (Scheffer et al., 1995). A common procedure consists in assuming the individual
modelled as a group of individuals sharing some common characteristic. This is the root of
the ―L agrangian-ensemble method‖ (Woods and Onken, 1982) and of the ―s uper-individual‖
approach (Scheffer et al., 1995). In this way, every individual represents a varying number of
specimens sharing the same destiny, and for each of them a number of variables and
processes can be simulated and their time-dependent variation be studied. Such methodology
may however modify the spatial and temporal model dynamics, especially when a large
number of individuals is aggregated (Parry and Evans, 2008). Parallel computing can provide
a helpful solution to overcome this limitation maintaining the original model structure (e.g.,
Parry and Evans, 2008).
86 Daniela Cianelli, Marco Uttieri and Enrico Zambianchi
2. PHYTOPLANKTON
Phytoplankton is composed by autotrophic, photosynthetic organisms belonging to
different taxa and living in all aquatic ecosystems (Mann and Lazier, 1996). They are moved
by the water currents both horizontally and vertically, even though some species are capable
of some degree of autonomous motility. Most organisms are too small to be seen at naked
eye, they often form large aggregates in the form of blooms. Phytoplankton represents almost
the 1–2 % of the total biomass of the world ocean but at global scale these organisms are able
to fix at least 30-60% of the total organic carbon (Falkowski et al., 1994).
Light and nutrients are the basic resources used by phytoplanktonic organisms as a source
of energy to perform their biosynthetic processes. As light intensity and nutrient
concentration show both diel and seasonal changes, phytoplankton has adapted to these
resource variations that regularly occur in the aquatic environment. In particular
phytoplankton has developed a photosynthetic apparatus, which adapts to the changes in light
intensity frequently observed in the water column. Phytoplanktonic organisms respond to
light variations both by adjusting the rates of biochemical processes and by changing the
organization, composition and functioning of the photosynthetic apparatus (Falkowski and
LaRoche, 1991). Different phytoplankton organisms exposed to the same resource
distributions will show different bio-chemical composition and photosynthetic performances
(Falkowski and LaRoche, 1991). This in turn affects the growth rate of the entire
phytoplankton population in the water column, modulating the species abundances and
altering their distributions and occurrence.
In marine ecosystems the physical-biological conditions continuously change supporting
the wide range of photophysiological responses adopted by phytoplanktonic organisms.
Among the others, the high variability in the light regime and nutrient availability
experienced by the cells as well as the frequent vertical displacements in the water column
Individual Based Modelling of Planktonic Organisms 87
induced by turbulent mixing and convection are crucial factors inducing the different
responses of organisms (e.g., Lewis et al., 1984; Figueiras et al., 1999).
Nevertheless, until recently the phytoplankton community has been frequently described
in the marine ecosystem models through ensemble averages, treating the organisms like a
continuum. For example, using these bulk (Eulerian) models the primary production has been
simulated computing the depth-averaged light intensity experienced by phytoplankton
population and then calculating the growth over time. As showed by Woods and Onken
(1982) such an approach, averaging non-linear equations before integration, causes
inaccuracy in the model results.
In order to reproduce realistic dynamics of phytoplankton populations the individual
physiology and the interactions with other organisms and the environment have to be taken
into account (e.g., DeAngelis and Gross, 1992).
The IBMs currently represents the most suitable tool for reconstructing the time evolution
of a phytoplankton community in terms of temporal and spatial histories of the individual
organisms. The individual-based approach allows treating populations as composed of a large
number of organisms, whose individual histories both determine the physiological response to
environmental conditions and the species composition in the water column.
Modelling the primary production by means of the IBM approach, the light perceived by
each individual is firstly computed, then the growth of each organisms is integrated. The
emergent property of the growth of the entire population produces the primary production
estimate of the water column, which may significantly differ from the estimates obtained
through bulk Eulerian models.
The numerical approach to the study of phytoplankton behaviour in the water column
was first introduced in the late 1970‘s (e.g., Marra, 1978a and 1978b; Kamykowski, 1979;
Falkowski and Wirick, 1981), but it was only after the work by Woods and Onken (1982) that
the IBM description for phytoplankton received wider attention. Here we briefly summarize
some of the most relevant studies applying the IBM approach to simulate a large number of
complex physical-biological processes acting simultaneously and at different scales in the
water column.
The first numerical study conducted by Marra (1978a and 1978b) investigated the
interaction between mixing and light regimes showing the different photophysiological
responses to variable light regime experienced by phytoplankton. Kamykowski (1979) firstly
modelled the interplay between individual phytoplankters and a variable flow field associated
with a semidiurnal internal tide, while the distribution of phytoplanktonic organisms in
Langmuir circulations was simulated by Evans and Taylor (1980). In their paper, Falkowski
and Wirick (1981) analysed the effects of variations in the light regime due to vertical mixing
on primary productivity. Phytoplankton cells were allowed to light-shade adapt on a fixed
time scale by varying their Chla:C ratios in response to variations in the light regime. Their
results showed that despite the physiological adaptation to light, vertical mixing may have
little effect on the integrated water column primary productivity.
However the utility of the IBM approach has been fully recognized only after the
reference paper by Woods and Onken (1982), who developed a Lagrangian ensemble model
coupled with a one-dimensional model of the upper ocean. The authors used a Lagrangian
biological model to study how turbulent transport of cells through the diurnal light gradient
affected the depth distribution and energy uptake of a phytoplankton population. The model
showed that as the surface mixed layer deepened during the night, phytoplankton cells and
88 Daniela Cianelli, Marco Uttieri and Enrico Zambianchi
particles produced in a shallow mixed layer during one day were mixed downward at rates far
larger than their still-water settling rates.
Several additional contributions subsequently appeared in the literature addressing
specific aspects of phytoplankton photophysiology in a rapidly changing environment. Lande
and Lewis (1989) questioned the rationale for using the time consuming Lagrangian approach
as opposed to the traditional Eulerian one, while a more recent attempt to overcome the limit
imposed by handling average quantities within an Eulerian framework has been proposed by
Janowitz and Kamykowski (1999). Yamazaki and Kamykowski (1991) and Kamykowski et
al. (1994) investigated the interactions between vertical swimming and photo-adaptation
response by a random walk representation of dinoflagellates swimming in the surface mixed
layer. The results highlighted the interactions of individual organisms with the simulated
turbulent mixing. Subsequently Kamykowski et al. (1994) also analysed by means of the IBM
approach the effect of the photoinhibition process on primary production estimates. The
model clearly demonstrated that vertical mixing regime strongly determined how the
individual light history affected each population‘s statistical characteristics. More recently
Nagai et al. (2003) coupled the Lagrangian photoresponse model of Kamykowski et al.
(1994) with a 2nd-order turbulence closure model (Mellor and Yamada, 1982). In this study
Nagai et al. (2003) investigated the effects of wind mixing and diurnal photoresponse on the
daily phytoplankton production in a realistic water column. Their results suggested that
intense wind mixing in a lower-transparency water column determined greater phytoplankton
production. Consequently, vertical mixing was not a relevant factor for the photoresponse in
open-ocean water, while in a coastal condition it played a more relevant role.
Cianelli et al. (2004) developed an individual-based model describing the spatial and
temporal evolution of phytoplankton organisms moving in the Antarctic mixed layer during
summer. While previous studies used simplified descriptions of cell photo-response, this
study firstly combined the dynamic photoacclimation of the pigment content with a
mechanistic description of photoinhibition process. Moreover, in order to simulate in detail
the phytoplankton photo-physiology the model explicitly considered the dynamics of organic
carbon (C) and chlorophyll a (Chl a) along with the vertical structure of the water column. As
the paper focused on the role of light variability on phytoplankton growth in the Antarctic
mixed layer, the authors simulated both a nutrient-replete and an iron-replete scenario, thus
reproducing the conditions frequently observed in coastal areas (Martin et al., 1990) or at the
onset of the growth season in Antarctica. The effect of different turbulent regimes and mixed
layer depths on the integrated primary production was investigated using in situ measured
parameters and kinetic constants consistent with the measured photosynthetic rates. The
coupling of different mixing levels with photoacclimation strategies led to a wide range of
photophysiological responses which underlined the role of the individual physiological
histories in determining the growth of the entire population. The results showed that the
highest rate of cell accumulation was reached when the vertical mixing compensated for
photoinhibition, thus suggesting that photoacclimation to low irradiance and strong mixing
regimes represented a crucial factor in the photosynthetic performance of Antarctic
phytoplankton.
In their recent paper Esposito et al. (2009) used the IBM approach to analyse how a
fluctuating light environment may influence the carbon assimilation by phytoplankton cells.
In particular this model included the dynamics of photoacclimation and photodamage-repair
mechanisms. Different light regimes (steady, square wave, sinusoidal light–dark cycles and
Individual Based Modelling of Planktonic Organisms 89
sustained the optimal growth conditions of both populations, while a variable turbulent
mixing limited their growth reducing the photophysiological differences between the species.
Cianelli et al. (2009a) also investigated how the photophysiological responses of an
individual species to the environmental forcings were affected by the concomitant presence of
the other one. The comparison between individual species (each species developing alone in
the same environmental conditions) and coexisting species simulations suggested that the two
species mutually affected their photosynthetic capability in idealized environmental
conditions. On the other hand, in a more realistic scenario the turbulent mixing might support
the diversity of phytoplankton species composition in the water column.
Being aware that an exhaustive review of all the IBM studies applied to phytoplankton
organisms is outside of the aim of the present work, we have here illustrated some of the most
representative IBMs analyzing the influence of individual variance and photosynthetic
responses on the growth of bulk phytoplankton populations.
3. ZOOPLANKTON
Water column ecosystems are populated by a great variety of microscopic metazoans
collectively named zooplankton. They comprise the majority of taxa, covering a wide size
spectrum, from 10-6 m (microzooplankton) to 101 m (megazooplankton) (Sieburth et al.,
1978). In marine environments the most abundant zooplanktonic organisms are copepods,
while in freshwaters cladocerans are the most represented taxon; they both are crustaceans
with an average body length of 0.2-20 mm (mesozooplankton). Besides their numerical and
geographical importance, freely swimming zooplankters are primary actors in the functioning
of pelagic ecosystems. Zooplanktonic organisms are crucial for the transfer of matter and
energy between lower (e.g., phytoplankton) and higher (e.g., fish) trophic levels (Fowler and
Knauer, 1986), as well as for linking the inertial and the viscous realms (Naganuma, 1996).
The predation upon phytoplankton provides the energy for metabolic requirements, but
determines also the egestion of fecal pellets which, by passive sinking, enhance the vertical
fluxes of carbon from the upper layers towards the deep ocean. In addition, these small
inhabitants of aquatic systems can be efficient indicators of global scale climate changes
(Richardson, 2008). For these reason, it is clear that understanding the ecology of these small
organisms can improve the current knowledge of the functioning of aquatic ecosystems and
their criticality with respect to global scale warming issues. The function played by
zooplankton in large-scale dynamics, developing over spatial scale in the order of tens and
hundreds of meters and over time periods of hours and days, is mediated by the behaviour
displayed at the individual level, which instead occurs at millimetre scale and over time scales
in the order of seconds. Despite their gap, these two scales are intimately correlated: small-
scale interactions with other organisms (prey, predators and mates) are regulated by the
individual behaviour and by the environmental abiotic factors (e.g., light, temperature), which
in turn affect the patterns observed at larger scales.
In the last two decades, the use of numerical models in zooplankton ecology has
burgeoned, as summarised by Carlotti et al. (2000). In this framework, models are mainly
used for three objectives (Carlotti et al., 2000):
Individual Based Modelling of Planktonic Organisms 91
Depending on the task addressed, different typologies of models can be used (Carlotti et
al., 2000). Despite the great potentialities, IBMs in this research area are still underexploited.
While the number of applications to fisheries research is substantial (as reviewed, e.g., in
Carlotti et al., 2000 and Neuheimer et al., 2010), the use of these models for copepods and
cladocerans is much less developed. In the following we will review applications of the
individual-based approach to copepods only, though several IBMs have been developed for
cladocerans also (e.g., Mooij and Boersma, 1996; Zadereev et al., 2003).
The first application of the individual-based approach to copepods is the work by
Batchelder and Miller (1989), who modelled the growth, development, reproduction and
death of Metridia pacifica over a one-year simulation. This model was subsequently refined
(Batchelder and Williams, 1995) to explain the consequence of the vertical distribution of
food on copepod‘s growth.
IBMs can be coupled with other models, such as water circulation or ecosystem ones.
Miller et al. (1998) modelled the life history and population dynamics of the copepod
Calanus finmarchicus in the Georges Bank region. The IBM included sex- and stage-
dependent biological traits, while the circulation model was used to move the modelled
copepods in the fluid, both horizontally and vertically. Such integrated investigation allowed
the authors to identify aggregative hot-spots in the region and to relate them to possible
restocking mechanisms.
Carlotti and Wolf (1998) implemented an IBM of C. finmarchicus based on the
― Lagrangian-ensemble method‖ (Woods and Onken, 1982) and integrated it with a food
supply deriving from a one-dimensional Eulerian NPZD ecosystem model. The individual-
based part also included a realistic description of individual swimming over the water column
(vertical migrations). The results of this model matched the observed dynamics of C.
finmarchicus in the Norwegian Sea.
Souissi et al. (2004) built an IBM to simulate the whole life cycle of Centropages
abdominalis, while Souissi et al. (2005) used an IBM to study the population dynamics of
Eurytemora affinis. Their results indicated specific space-time patterns to describe the
dynamics of the copepods, with a good match between simulated and in situ observed
blueprints.
Gentleman et al. (2008) compared different model typologies for copepod development
and proposed an alternative approach, the stage-based model of individuals, to represent the
progressive development through stages rather than using growth equations. The benefits of
this approach were demonstrated by replicating the development time for C. finmarchicus as
derived from laboratory trials. This approach was then used by Neuheimer et al. (2009) to
study the time-varying mortality in the nauplii of C. finmarchicus and in Neuheimer et al.
(2010) to model the recruitment of C. finmarchicus in the North-Western Atlantic, evaluating
the effects of temperature and chlorophyll-a on the physiological traits of this species.
Dur et al. (2009) modelled the reproduction of E. affinis using physiological parameters
such as female longevity, clutch size and interclutch duration. The model, validated through
92 Daniela Cianelli, Marco Uttieri and Enrico Zambianchi
Theoretical IBMs by Uttieri et al. (2007) and Cianelli et al. (2009b) demonstrated the role
of zooplankton motion behaviour and pattern of distribution of resources in determining the
probability of encountering a prey. The results of these works demonstrated that swimming
movement characterised by higher morphological complexity tallied more encounters than
smoother trajectories in uniform (Uttieri et al., 2007) and patchy (Cianelli et al., 2009) prey
distributions. Uttieri et al. (2010) developed an IBM to compare the search strategy and
encounter success of two co-occurring marine copepods (C. furcatus and O. plumifera)
showing different motion rules and sensory performances. This model included the numerical
description of the swimming motion of the two copepods (C. furcatus: Mazzocchi and
Paffenhöfer, 1999; Uttieri et al., 2008 – O. plumifera: Paffenhöfer and Mazzocchi, 2002), as
well as a realistic reconstruction of their perceptive fields (C. furcatus: Uttieri et al., 2008 –
O. plumifera: Paffenhöfer and Mazzocchi, 2002). The encounter success of the two copepods
was tested in homogeneous and patchy distributions of prey, showing that C. furcatus scored
more encounters than O. plumifera which however recorded higher search efficiencies.
CONCLUSION
Plankton populations are composed by microscopic organisms that, even when belonging
to the same species, differ at individual level from each other and interact with the aquatic
environment in a unique way. Planktonic organisms, being a product of evolutionary
processes, have developed adaptation strategies to better exploit the environmental resources.
The inter-individual and individual-environment interactions determine the emergent
properties of the entire population; as a consequence the investigation of the processes taking
place at the level of the individual provides a deeper comprehension of the functioning of
aquatic systems.
An important contribution to the understanding of plankton behaviour and ecology is thus
provided by the individual-based numerical approach (IBMs) which is particularly
appropriate to reproduce the complexity of plankton ecosystems. This promising tool has
been extensively used for a number of applications, revealing important aspects about
population dynamics and behavioural adaptations. In this work we have briefly summarized
the applications of IBMs to the field of phyto- and zooplankton ecology and behaviour. The
results here synthesized highlight that the dynamics of the individual physiological responses
is often non-linear and that extreme behaviour may play a crucial role. In such a framework,
an approach based on individuals is obviously the best candidate for a realistic numerical
description of plankton ecology.
REFERENCES
Batchelder, H. P.; Miller, C. B. Ecol. Model. 1989, 48, 113-136.
Batchelder, H. P.; Williams, R. ICES J. Mar. Sci. 1995, 52, 469-482.
Benenson, I.; Martens, K.; Birfir, S. Comput. Environ. Urban. 2008, 32, 431-439.
Berger, U.; Piou, C.; Schiffers, K.; Grimm, V. Perspect. Plant. Ecol. 2008, 9, 121-135.
Beyer, J. E.; Laurence, G. C. Ecol. Model. 1980, 8, 109-132.
94 Daniela Cianelli, Marco Uttieri and Enrico Zambianchi
Souissi, S.; Ginot, V.; Seuront, L.; Uye, S.-I. In Handbook of Scaling Methods in Aquatic
Ecology - Measurements, Analysis, Simulation; Seuront, L.; Strutton, P. G. (Eds.); CRC
Press: Boca Raton, 2004; pp 523-542.
Souissi, S.; Seuront, L.; Schmitt, F. G.; Ginot, V. Nonlinear Anal.-Real. 2005, 6, 705-730.
Souissi, S.; Seuront, L.; Schmitt, F. G.; Ginot, V. Nonlinear Anal.-Real. 2005, 6, 705-730.
Starfield, A. M.; Smith, K. A.; Bleloch, A. L. How to Model It: Problem Solving for the
Computer Age; McGraw-Hill: New York, 1990; pp 206.
Strickler, J. R. Limnol. Oceanogr. 1977, 22, 165-170.
Tiselius, P.; Jonsson, P.; Verity, P. G. Bull. Mar. Sci. 1993, 53, 247-264.
Uchmański, J.; Grimm, V. TREE. 1996, 11, 437-441.
Uttieri, M.; Paffenhöfer, G.-A.; Mazzocchi, M. G. Mar. Biol. 2008, 153, 925-935.
Uttieri, M.; Nihongi, A.; Mazzocchi, M. G.; Strickler, J. R.; Zambianchi, E. J. Plankton Res.
2007, 29, i17-i26.
Uttieri, M.; Sabia, L.; Cianelli, D.; Strickler, J. R.; Zambianchi, E. J. Mar. Syst. 2010, 81,
112-121.
Uttieri, M.; Mazzocchi, M. G.; Nihongi, A.; Ribera d'Alcalà, M.; Strickler, J. R.; Zambianchi,
E. J. Plankton Res. 2004, 26, 99-105.
Wiggert, J. D.; Hofmann, E. E.; Paffenhöfer, G.-A. ICES J. Mar. Sci. 2008, 65, 379-398.
Wiggert, J. D.; Haskell, A. G. E.; Paffenhöfer, G.-A.; Hofmann, E. E.; Klinck, J. M. J.
Plankton Res. 2005, 27, 1013-1031.
Wissel, C. Ecol. Model. 1992, 63, 1-12.
Wolfram, S. Cellular Automata and Complexity: Collected Papers; Addison-Wesley:
Reading, 1994; pp 596.
Woods, J.; Onken, R. J. Plankton Res. 1982, 4, 735-756.
Woods, J.; Barkmann, W. Phil. Trans. R. Soc. London B. 1994, 343, 27-31.
Wroblewski, J. S. Ocean Sci. Eng. 1983, 8, 245-285.
Yamazaki, H.; Kamykowski, D. Deep Sea Res. I. 1991, 38, 219-241.
Zadereev, E. S.; Prokopkin, I. G.; Gubanov, V. G.; Gubanov, M. V. Ecol. Model. 2003, 162,
15-31.
In: Ecological Modeling ISBN: 978-1-61324-567-5
Editor: WenJun Zhang, pp. 97-114 © 2012 Nova Science Publishers, Inc.
Chapter 6
ABSTRACT
The larval phase of most blowfly species is considered a critical developmental
period in which intense limitation of feeding resources frequently occurs. Furthermore,
such a period is characterised by complex ecological processes occurring at both
individual and population levels. These processes have been analysed by means of
traditional statistical techniques such as simple and multiple linear regression models.
Nonetheless, it has been suggested that some important explanatory variables could well
introduce non-linearity into the modelling of the nutritional ecology of blowflies. In this
context, dynamic aspects of the life history of blowflies could be clarified and detailed by
the deployment of machine learning approaches such as artificial neural networks
(ANNs), which are mathematical tools widely applied to the resolution of complex
problems. A distinguishing feature of neural network models is that their effective
implementation is not precluded by the theoretical distribution of the data used.
Therefore, the principal aim of this investigation was to use neural network models
(namely multi-layer perceptrons and fuzzy neural networks) in order to ascertain whether
these tools would be able to outperform a general quadratic model (that is, a second-order
regression model with three predictor variables) in predicting pupal weight values
(outputs) of experimental populations of Chrysomya megacephala (F.) (Diptera:
*
Email address: drebianconi@yahoo.com.br
98 Michael J. Watts, Andre Bianconi, Adriane Beatriz S. Serapiao et al.
Calliphoridae), using initial larval density (number of larvae), amount of available food,
and pupal size as input variables. These input variables may have generated non-linear
variation in the output values, and fuzzy neural networks provided more accurate
outcomes than the general quadratic model (i.e. the statistical model). The superiority of
fuzzy neural networks over a regression-based statistical method does represent an
important fact, because more accurate models may well clarify several intricate aspects
regarding the nutritional ecology of blowflies. Additionally, the extraction of fuzzy rules
from the fuzzy neural networks provided an easily comprehensible way of describing
what the networks had learnt.
Keywords: regression models; life history; neural algorithms; larval phase; pupal mass.
1. INTRODUCTION
It is a well-known fact that several species of blowflies can be mechanical vectors of
pathogenic microorganisms (Zumpt, 1965; Guimarães et al., 1978; Furlanetto et al., 1984;
Laurence, 1986; Lima and Luz, 1991). In addition, blowflies have been increasingly utilised
in forensic studies with the purpose of determining post-mortem intervals (Greenberg, 1991;
Catts and Goff, 1992; Arnaldos et al., 2005; Gomes and Von Zuben, 2005; Shiao and Yeh,
2008; Cammack and Nelder, 2010). From an ecological perspective, detailed studies of the
dynamics of these insects are also very useful because some blowflies may have important
implications for the field of invasion ecology (Cammack and Nelder, 2010). Therefore,
comprehending every developmental stage of blowflies may well be considered essential both
for medico-criminal analyses as well as for ecological research as a whole (Bianconi et al.,
2010a, 2010b).
Chrysomya megacephala (Fabricius) (Diptera: Calliphoridae), for example, represents a
blowfly of well-known medical and veterinary importance that is able to cause facultative
myiasis in humans and animals (Zumpt, 1965; Guimarães et al., 1978; Furlanetto et al., 1984;
Laurence, 1986; Lima and Luz, 1991; Gabre et al., 2005). This species is native to the
Oriental zoogeographic region, but it is now established in parts of South America following
accidental introduction (Guimarães et al., 1978; Laurence, 1981; Wells, 1991). Moreover, in
the field of forensic entomology, the genus Chrysomya has been utilised as a useful biological
indicator (Greenberg, 1991; Catts and Goff, 1992; Arnaldos et al., 2005; Gomes and Von
Zuben, 2005; Shiao and Yeh, 2008; Cammack and Nelder, 2010). Owing to the medico-
criminal, biological, and ecological utility of this species, several studies have been conducted
for analysing its bionomic traits in a detailed manner (e.g. Von Zuben et al., 1993, 2000,
2001; Bianconi et al., 2010a, 2010b; Hu et al., 2010).
With respect to the larval phase of C. megacephala, this developmental stage is
considered a critical period in which intense limitation of resources may well occur (Levot et
al., 1979; Goodbrod and Goff, 1990; Reis et al., 1994). This limitation can lead to dynamic
competitive processes (Shiao and Yeh, 2008), wherein each larva attempts to feed off the
available resources, scrambling to exploit the feeding substrate before the depletion of the
food resource (Ullyett, 1950; de Jong, 1976; Lomnicki, 1988; Von Zuben et al., 2001).
Therefore, such exploitative events are characterised by interrelated processes that take place
at both the individual and population levels (Wijesundara, 1957; Herzog et al., 1992; Von
Zuben et al., 2001; Tammaru et al., 2004). Moreover, it has been suggested that both larval
The Effectiveness of Artificial Neural Networks … 99
density and availability of food affect the competition for food in a concurrent manner.
Hence, it is very useful and important to investigate the crowding level of immature
individuals on the feeding resources by means of simultaneous variations of larval densities
and amounts of food (Von Zuben et al., 2000; Ireland and Turner, 2006).
The outcomes of exploitative competition for food resources may determine population
parameters such as survival, fecundity, weight and size of the resultant adults, so that the
variation of these bionomic features could well be influenced by the immatures‘ population
density (Von Zuben et al., 2000, Ireland and Turner, 2006; Shiao and Yeh, 2008). Von Zuben
et al. (1993), for instance, regard the number of emerging adults as a variable that tends to
decrease with an increase in the number of immature individuals of C. megacephala. Hence,
profitable mass-production techniques usually demand the determination of feasible cost-
benefit relationships between larval density and amount of available food (Papandroulakis et
al., 2000; Von Zuben et al., 2001).
Regarding the analysis of bionomic features of insects, pupal weight is considered to be
an important variable for comprehending potential relationships between larval and pupal
weight of Cochliomyia hominivorax (Coquerel) (Calliphoridae) (Peterson II and Candido,
1987). Investigating the thermal requirements for development of nymphalid species, Bryant
et al. (1997) took pupal weight into consideration with the purpose of determining possible
connections between relative performance and distribution. Tammaru et al. (1996) analysed
the pupal weight of Epirrita autumnata (Borkhausen) (Lepidoptera: Geometridae) in order to
assess the relationship between body size and realised fecundity, and Tammaru et al. (2004)
stated that the pupal weight of this same species could be affected by starvation treatments.
Pupal mass was also utilised in analysing dietary specialisation aspects of a micro-
lepidopteran culture that had been maintained on an artificial diet for approximately 350
generations (Warbrick-Smith et al., 2009). Furthermore, pupal weight has been widely
utilised in several works concerning use of different substrates for rearing dipteran larvae
(Brewer, 1992; Friese, 1992; Chaudhury and Alvarez, 1999; Tachibana and Numata, 2001;
Chang et al., 2004).
Unsuitable or inaccurate predictive models may hinder the effective comprehension of
the underlying principles that govern several biological processes, including the nutritional
ecology of blowflies. For example, mass rearing of larvae and pupae with the purpose of
using these immatures as food or bait for other animals constitutes a process in which
adequate models could effectively ameliorate its implementation (Von Zuben et al., 1993,
2000, 2001). Additionally, the suitable conditions which yield pupae containing higher
concentrations of energy (higher biomass) could be assessed and established, using analytical
tools that permit the implementation of mathematical models with higher prediction
capability (Von Zuben et al., 1993, 2000; Bianconi et al., 2010a, 2010b).
In this context, the complexities of the nutritional ecology of blowflies could be clarified
and detailed by the deployment of appropriate modelling techniques such as artificial neural
networks (ANNs), which are mathematical tools widely applied to the resolution of complex
biological problems. A notable feature of artificial neural networks is their independence
from any assumptions about the theoretical distribution of the data used (Bishop, 1995;
Haykin, 1999; Bryant and Shreeve, 2002; Pearson et al., 2002; Zhang and Barrion, 2006).
Moreover, if the dimensional features of a specific system are too complex for a conventional
regression statistical model, artificial neural networks may represent a more effective
100 Michael J. Watts, Andre Bianconi, Adriane Beatriz S. Serapiao et al.
modelling tool (Schultz and Wieland, 1997; Haykin, 1999; Schultz et al., 2000; Zhang and
Wei, 2009).
Neural network algorithms are founded on the construction of models that may possess a
large number of simple processing units (that is, neurons or nodes) that contain several
connections between them and are usually lined up in layers. The number of neurons in the
input layer represents the variables that will be used to feed the neural network and should be
the most relevant variables for the problem in question (Bishop, 1995; Haykin, 1999). ANNs
were conceived with the aim of imitating the functionality of the human brain (Haykin, 1999;
Huang, 2009; Huang et al., 2010). Thus, part of the terminology used in the area of artificial
neural networks, namely neurons, synapses, learning, layers, etc., is due to such a fact.
However, it is important to emphasise that those terms are only associated with mathematical
functions or the method of implementing them.
Studies of process-based neural network models are not as frequent in ecological and
environmental areas as they are in engineering applications (Zhang and Zhang, 2008; Huang,
2009). However, considerable improvements have been achieved in recent years (Zhang and
Zhang, 2008; Huang, 2009; Zhang and Wei, 2009). In the broader context of ecosystem
dynamics, neural networks have been successfully utilised. For example, the abundance of
selected water insects in a small stream was predicted by means of neural models (Obach et
al., 2001). Environmental data collected as part of a study of microhabitat use by butterflies
were utilised with the aim of evaluating the potential for using neural modelling in developing
predictive models of microhabitat temperature (Bryant and Shreeve, 2002). Furthermore,
other multidisciplinary studies have described the development of scale-independent models,
based on coupling artificial neural networks with climate-hydrological process models in
order to simulate species‘ distribution, including insect species (Pearson et al., 2002; Pearson
and Dawson, 2003; Harrison et al., 2006).
Worner and Gevrey (2006) used a self-organising map, which is an artificial neural
network model, with the purpose of identifying global pest species assemblages and potential
invasive insects, including dipteran species. Zhang and Barrion (2006) conducted function
approximation and documentation on sampling data using neural networks, based on
invertebrate data sampled in an irrigated rice field. Howe et al. (2007) demonstrated the value
of using neural networks to predict body temperature and activity of insects by means of
modelling the body temperature and activity of a widespread butterfly species in relation to
weather. With the purpose of improving the prediction accuracy of potential species invasion,
Watts and Worner (2008) deployed two types of biotic factors in ANN models to predict
global establishment of selected phytophagous insect species, and such predictions were then
combined with those derived from an ANN model based on abiotic (climate) factors.
With respect to mass-production techniques, the adequate development of feasible
automatic feeding devices for rearing larval individuals could well benefit from the use of
neural network-based models (Papandroulakis et al., 2000), and effective control strategies of
insect pests usually rely on the knowledge derived from basic research on oviposition rate,
adult emergence, larval development until pupation, etc (Von Zuben et al., 2000; Köppler et
al., 2009). In this context, neural networks may be useful to establish the best possible cost-
benefit relationships between larval density and amount of food, with the aim of producing
heavier pupae and minimising production costs.
Zhang and Zhang (2008) analysed the effectiveness of neural networks in modelling
survival process and mortality distribution of Spodoptera litura (Fabricius) (Lepidoptera:
The Effectiveness of Artificial Neural Networks … 101
Adult blowflies were captured around the campus of the Universidade Estadual de
Campinas-Unicamp in Campinas, São Paulo state, Brazil (22º 49‘ 9.52‖ S and 47º 4‘ 12.54‖
W). The specimens were then identified and kept in nylon net cages (30 x 30 x 48 cm). Prior
to identifying the blowflies, the individuals were anaesthetised in a freezer at -18º C for 30 s.
These insects were provided with water and refined sugar ad libitum and taken as the parental
generation of this investigation. The cages were kept at room temperature (25 + 1º C), 60 +
10% relative humidity, and light:dark regime of 12:12 h. In order to induce the development
of the gonotrophic cycle, females were supplied with beef liver. For the formation of the next
generations, ovipositions were obtained using small pots containing decaying beef that were
put into the cages in order to stimulate egg laying.
Five proportions of larvae to amount of food were considered (i.e. 5, 10, 20, 30, and 40
larvae/g) in the current work, and they were obtained as follows: 75, 150, 300, 450, and 600
larvae were put into pots that contained 15g of food; 150, 300, 600, 900, and 1200 larvae in
pots containing 30g of food; 300, 600, 1200, 1800, and 2400 larvae in pots containing 60g of
102 Michael J. Watts, Andre Bianconi, Adriane Beatriz S. Serapiao et al.
food; and 450, 900, 1800, 2700, and 3600 larvae in pots with 90g of an artificial diet
proposed by Leal et al. (1982). Glass pots (8 cm in height x 7 cm in diameter) containing four
amounts (i.e. 15, 30, 60, and 90g) of the artificial diet were utilised. As to the pupal stage,
pupae were individually weighed, and pupal sizes were taken on the eighth day after the peak
of the larval hatching period. Additional experimental details regarding the formation of the
larval densities and amounts of food are provided in Von Zuben et al. (2000) and Bianconi et
al. (2010a).
2.2. Variables
Pupal weight was the dependent variable (output), and the other three were considered
the independent or explanatory variables (inputs), namely larval density (that is, initial
number of larvae), amount of available food (in grams), and pupal size (in mm). These
variables were chosen based on their biological significance for the comprehension of the
nutritional ecology of blowflies. Combinations of the three input variables provided 1114
output values.
In several practical situations, neural networks and statistical models can both be utilised,
because they may respond to the same question, albeit with important differences in the
accuracy of the outcomes. Therefore, the performance of neural models is usually compared
with that derived from statistical methods (Schultz and Wieland, 1997; Bianconi et al.,
2010b). Hence, a general quadratic regression method was used as the statistical ‗counterpart‘
of the neural models deployed in the current investigation.
Bianconi et al. (2010a; 2010b) utilised a first-order linear regression method in order to
compare the outcomes derived from implementing well-known neural network models to the
results obtained by using the statistical model. Nonetheless, these same authors suggested that
each one of the independent variables might have produced distinct output responses. Overall,
such explanatory attributes might have generated non-linear variation in the output variable
(i.e. pupal weight), and the accuracy of conventional regression methods such as the first-
order regression models may be significantly reduced in the presence of non-linearity (Neter
et al., 1996; Schultz and Wieland, 1997). Therefore, in the present investigation, instead of
using the traditional first-order regression model, we utilised a second-order regression model
(i.e. a general quadratic model) in order to compare the performance of three types of
artificial neural network (ANN) algorithms to that derived from implementing the general
quadratic regression model (i.e. a second-order regression model).
It is important to emphasise that the general quadratic model (GQM) represents a special
case of the traditional linear regression model deployed by Bianconi et al. (2010a, 2010b).
Nonetheless, the term ‗linear model‘ refers to the fact that this model may be linear in the
parameters, but this linearity is not associated with the shape of the response surface. In
simple terms, a general quadratic regression (GQM) may be described in the following form:
ŷ = a + b1x1 + b2x2 + b3x3 + b12x1x2 + b13x1x3 + b23x2x3 + b11x12 + b22x22 + b33x32 (1)
The Effectiveness of Artificial Neural Networks … 103
in which ŷ represents the output (i.e. predicted pupal weight values); x1= initial larval density,
x2= amount of available food (g), x3 = pupal size (mm); a is a constant; and bi, bii, and bij
represent regression coefficients. The regression coefficients bij (i.e. b12, b13, and b23) may be
termed ‗interaction effect coefficients‘ for interactions between pairs of predictor variables. A
general quadratic regression such as (1) is considered to be a second-order regression model
because the input variables may be expressed in the model to the first and second powers
(Kutner et al., 2004). On the other hand, the traditional regression model implemented by
Bianconi et al. (2010a, 2010b) is deemed to be a first-order regression model in which the
effects of the explanatory variables on the mean response of the output variable do represent
additive effects that do not interact with one another (Kutner et al., 2004). Hence, it is
expected that the second-order model described in the present work would be capable of
providing more accurate outcomes than the traditional first-order regression model deployed
by Bianconi et al. (2010b) in predicting pupal weight values.
The entire data set was utilised in implementing the general quadratic model because the
division of the data set into training and test subsets is suitable for neural network modelling.
That is, if entomologists were to deploy this quadratic statistical model in conducting
experiments, they would utilise the entire data set because a general quadratic model is
derived from the whole data set.
In the current paper, the coefficient of determination (R2) and the root mean square error
(RMSE) were utilised in assessing the performance of the statistical and neural network
models. Apart from these metrics, a residual plot (i.e. plot of residuals against fitted values)
was utilised in order to examine the appropriateness of the regression model in relation to the
constancy of the variance of the error terms. The statistical assumptions of the general
quadratic model (GQM) followed the detailed descriptions in Kutner et al. (2004). The ANN
models used in this work will be described in the following section.
Artificial neural networks are able to model complex non-linear systems, even when the
exact nature of any relationships is unknown (Schultz and Wieland, 1997; Schultz et al.,
2000; Bryant and Shreeve, 2002; Howe et al., 2007; Zhang et al., 2008a). The framework for
deploying neural network models is underpinned by the concept of artificial neurons (that is,
processing elements) that are usually laid out in a parallel fashion (Haykin, 1999). On the
whole, interconnected layers in combination with their neurons constitute the architecture of
neural networks, in which each neuron in one specific layer is only fully connected to neurons
of other layers. Neurons are usually interconnected by means of ‗synaptic weights‘ (Cheng
and Titterington, 1994; Haykin, 1999). In outline, the training of a neural network comprises
the act of presenting the neural network model with input and/or output data; the creation and
implementation of connections that are able to recognise patterns; and the learning of such
patterns based on the relationship between input and output data sets via the adaptation of
specific synaptic weights to varying input data (Cheng and Titterington, 1994; Bishop, 1995;
Haykin, 1999).
A widely used ANN is the multi-layer perceptron, or MLP (Crick, 1989). This model
consists of three layers of artificial neurons: an input layer, where there is one neuron for each
input variable; a hidden neuron layer, where each neuron has incoming connections from the
104 Michael J. Watts, Andre Bianconi, Adriane Beatriz S. Serapiao et al.
input layer; and the output layer, where there is one neuron for each output layer, and each
output neuron has incoming connections from the hidden layer neurons. The hidden layer gets
its name from the fact that it is not directly exposed to either the input or output variables: it is
‗hidden‘ from the outside world. Training a MLP involves setting the values of the
connection weights, and the most commonly used method of training MLP is
backpropagation of errors (Rumelhart et al., 1986). This is a gradient-descent error
minimisation algorithm, where errors at the output layer are propagated back through the
structure of the MLP, with the errors at each layer being used to adjust the incoming
connection weights of each neuron.
An alternative method of training MLP is via evolutionary programming (EP) (Fogel et
al., 1965). This is an evolutionary algorithm (Fogel et al., 1997) whereby populations of
solution attempts are evolved over generations; the performance of each solution attempt in
the population is evaluated over the problem, and the solution attempts with the best
performance in each generation are used to generate the next generation. When applied to
training MLP, the populations consist of MLP, the performance of each MLP is evaluated
over the training data, and new solutions are generated by applying normally distributed
changes to the connection weights. We measured the performance of each MLP as the mean
squared error (MSE) over the training data set. For both BP and EP training of MLP, one
hundred trials were carried out over the selected training parameters. The MLP that had the
best performance over the validation data partition was used to predict over the test partition.
For each trial with MLP, the contributions of each input neuron to the output of the
network were also determined. Many methods have been proposed for determining the
importance of each of the input neurons of a MLP. These include the methods of Garson
(1991), Milne (1995), Gevrey et al. (2003) and Olden and Jackson (2002). Since it was
desirable to identify features that negatively affect pupal weight and the methods of Milne
(1995) and Gevrey et al. (2003) return unsigned values, these methods were rejected. Of the
remaining methods, the work of Olden et al. (2004) has shown that the method of Olden and
Jackson (2002) is the least biased, and it has been previously used in ecological modelling
applications (Joy and Death, 2004). Thus, this method was selected.
A more advanced ANN is the Fuzzy Neural Network FuNN (Kasabov et al., 1997).
FuNN was designed to provide an easy method of combining the advantages of fuzzy logic
(Zadeh, 1965) with the advantages of neural networks. FuNN are ANN with fuzzy logic
elements embedded within them, specifically fuzzy membership functions attached to their
input and output variables, and are trained using BP. Fuzzy logic allows concepts that are not
crisply defined, such as ‗Low‘, ‗Medium‘ or ‗High‘, to be expressed in fuzzy rules that are
more easily understood, are more succint and are more robust than rules that rely on crisply
defined concepts. A chief advantage of FuNN is the ability to extract and insert fuzzy rules
from and into a FuNN structure. Fuzzy rules are extracted from a trained FuNN, and are
useful for explaining in a comprehensible manner the knowledge that the network has
captured.
A potential disadvantage of ANN is the network inability to suitably respond to data sets
which were not previously shown (lack of generalisation ability or overfitting). Normally, to
avoid overfitting of the network, the number of neural connections should be lower than the
number of training examples. However, there are not definite rules or methods of establishing
the suitable number of hidden layers and neurons (Watts and Worner, 2008; Zhang and
Zhang, 2008; Huang, 2009; Huang et al., 2010). Therefore, cross-validation (CV) may be a
The Effectiveness of Artificial Neural Networks … 105
useful way of avoiding such drawbacks by means of inserting the cross-validation subset into
the network in order to verify its performance over the training. In regard to the use of neural
modelling in entomology, Zhang and Zhang (2008), for example, utilised cross-validation
procedures.
3. RESULTS
Table 1 shows the coefficients of determination (R2) and root mean square errors that
were derived from the data subsets utilised for testing the neural network models. Regarding
the statistical model (i.e. general quadratic regression), the entire data set (i.e. 1114 examples)
was utilised. The following equation was derived by fitting the entire data set to (1):
in which ŷ represents the estimated quadratic model derived from fitting the entire data set to
(1).
Table 1. Coefficients of determination (R2) and root mean square errors (RMSE)
derived from test subsets of each predictive model. MLP-BP means a multi-layer
perceptron (MLP) ANN trained using backpropagation. MLP-EP means a MLP trained
with evolutionary programming (EP). FuNN is the results of the Fuzzy Neural Network
FuNN. With respect to the statistical model (GQM), the entire data set (n=1114) was
utilised in obtaining the metrics. GQM: general quadratic model
Test
Neural models R2 RMSE
MLP-BP 0.5659 5.8509
MLP-EP 0.5680 6.2415
FuNN 0.6394 5.1099
The training parameters that were found to be the most effective for BP-trained MLP
were 15 hidden neurons, trained for 100 epochs with a learning rate and momentum of 0.5.
The optimal parameters for EP-trained MLP were 12 hidden neurons, a population size of 200
MLP, over 7500 generations. The best 40 MLP of each generation were used as the parents of
the next generation. Finally, for FuNN, the optimal parameters were 15 hidden neurons,
trained for 100 epochs with a learning rate and momentum of 0.5. There were three fuzzy
membership functions attached to each input neuron and the output neuron.
106 Michael J. Watts, Andre Bianconi, Adriane Beatriz S. Serapiao et al.
The mean and standard deviations of the input contributions derived from BP-trained
MLP were x1 = -6.6061±0.5057, x2 = 3.7892±0.4177, and x3 = 12.6191±0.3450. The input
contributions derived from EP-trained MLP were x1 = -8.7523±2.8487, x2 = 4.9177±3.6100,
and x3 = 9.5995±3.9692.
Three fuzzy rules were extracted from the trained FuNN and are presented below. The
numbers following the labels Low, Medium and High are confidence factors:
A threshold value was applied to the rule extraction process, which eliminated rules and
rule elements that did not contribute strongly to the model: thus, not all variables are included
in the rule antecedents.
The R2 obtained by means of the general quadratic model (GQM) was slightly higher (R2
= 0.5761, and RMSE = 7.7530) than that derived from the conventional first-order multiple
linear regression model (i.e. R2 = 0.5720, and RMSE = 7.8320) implemented by Bianconi et
al. (2010b). On the other hand, the fuzzy neural network FuNN provided both a higher R2
(0.6394) and lower RMSE value (5.1099) than the general quadratic model (GQM) utilised in
the present investigation, while the BP-trained MLP and EP-trained MLP both yielded lower
RMSE values (5.8509 and 6.2415 respectively), but slightly lower R2 values (0.5659 and
0.5680).
Figure 1. Residuals (i.e. the difference between predicted and observed values) as a function of the
predicted values. The whole data set is represented (n = 1114). Each point represents the difference
between the observed pupal weight (actual value) and the pupal weight predicted by the general
quadratic model (Equation 2).
The Effectiveness of Artificial Neural Networks … 107
In addition, it is important to note that the residuals derived from the general quadratic
model (Figure 1) are not ‗biased‘. That is, the residuals neither increase nor decrease with the
increase in the predicted output values.
DISCUSSION
Bryant and Shreeve (2002) compared the predictive power of a feed-forward
backpropagation neural network with that provided by a conventional multiple regression
model in estimating microhabitat temperature, and the neural network method was found to
exhibit a higher correlation between predicted and observed values. Howe et al. (2007)
modelled the body temperature and activity of a widespread butterfly species (Polyommatus
icarus) by means of a multilayer feed-forward backpropagation network, and this neural
model was deemed superior to a generalised linear modelling approach to predicting body
temperature.
Similarly, MLP trained using backpropagation and evolutionary programming and the
fuzzy neural network FuNN deployed in the current study exhibited better performances than
a second-order regression-based model (i.e. general quadratic model). Furthermore, Watts and
Worner (2008) successfully used multilayer perceptrons (Cascaded MLP) in order to improve
the prediction accuracy of potential insect species invasions, and improved test accuracy was
obtained through the utilised neural models.
Simple first-order linear regression models (that is, one single output as a function of one
single input variable) have been utilised in studies concerning the nutritional ecology of
blowflies (Von Zuben et al., 1993, 2000). Gomes et al. (2009), for instance, examined the
burrowing behaviour of blowflies in response to different conditions of temperature by means
of a linear regression method in order to determine the relationship between increase in
temperature and decrease in pupal weight. Bianconi et al. (2010a, 2010b) deployed a
conventional multiple linear regression model in analysing the nutritional ecology of C.
megacephala, and this statistical regression technique was less accurate than neural networks.
Bianconi et al. (2010b) suggested that the input variables (i.e. initial number of larvae,
amount of available food, and pupal size) could have introduced some non-linearity into the
output response. Such non-linearity may well hinder the implementation of first-order
multiple linear regression models, because first-order models should be linear both in the
parameters as well as in the response surface. On the other hand, second-order models such as
the general quadratic model utilised in the current work (Equation 1) are not restricted to
linear responses only. Thus, this type of model could be more appropriate for predicting pupal
weight values than the first-order model utilised by Bianconi et al. (2010b).
Nonetheless, the R2 value derived from the general quadratic model implemented in the
current investigation (R2 = 0.5761, Table 1) was slightly higher than that (i.e. R2 = 0.5720)
obtained by means of a traditional first-order regression model (Bianconi et al., 2010b).
Therefore, it could be concluded that traditional multiple regression models such as first- and
second-order models were not capable of incorporating the non-linearity present in the input
variables into the fitted model. Moreover, the RMSE derived from FuNN was lower than all
of the other outcomes (i.e. Bianconi et al., 2010b, and the current investigation).
108 Michael J. Watts, Andre Bianconi, Adriane Beatriz S. Serapiao et al.
The input contribution analysis of both the BP and EP-trained MLP indicated that the
amount of food available (x2) was the least important variable, as this had the lowest
contribution score for both sets of results. This interpretation agrees with the results of the
statistical model, which assigned relatively small coefficients to that variable and from the
fuzzy rules extracted from FuNN, where this variable was included in only one of the three
extracted rules. Pupal size x3 had a large positive contribution, that is, a large value for pupal
size would contribute to a large value for pupal weight (y1). To some extent, this
interpretation agrees with the statistical model, where a relatively large positive coefficient
was assigned to the squared term (i.e. 0.7071, Equation 2), and by the fuzzy rules, as pupal
size was included in each of the three extracted rules. Finally, initial larval density (x1) had a
large negative contribution for MP and EP-trained MLP. This means that a large value for
initial larval density would contribute to a small value for pupal weight. In the statistical
model very small coefficients were assigned to this variable, and in the third extracted fuzzy
rule, a low value of this variable (if x1 is Low) led to a high pupal weight (then y1 is high). In
summary, the neural models utilised in predicting the output variable revealed that a large
value of larval density x1 contributes to a small pupal weight, a large value of pupal size x3
contributes to a high pupal weight, and the amount of food available x2 contributes the least to
the pupal weight.
Zhang et al. (2008a) investigated the spatial distribution pattern of grassland insects by
means of neural models. They concluded from their results that neural networks were more
flexible than a conventional model. Additionally, these authors indicated that further research
based on more complex distribution patterns should be conducted with the aim of obtaining
conclusions that are more reliable. Similarly, neural networks were deemed to be superior to a
conventional model (i.e. general quadratic model) in the current study. Nevertheless, the
present work utilised simple experimental designs and only three explanatory variables
(inputs). Therefore, more complex experimental designs should be implemented in order to
explain the portion of the total variance that was not accounted for by the models.
Furthermore, larger sample sizes are highly desirable, because the number of examples
(sample size) that is usually utilised in nutritional ecology may prevent the full utilisation of
the potential of neural networks (Bianconi et al., 2010a, 2010b).
The sample size used in the present investigation (n = 1114) may be sufficient to conduct
most ecological and biological approaches to analysing the nutritional ecology of blowflies
(Bianconi et al., 2010a, 2010b). Nonetheless, it is important to note that two of the input
variables (i.e. larval density and amount of food available) provided the utilised methods with
few distinct values, because only 11 larval densities and four distinct amounts of available
food were used (see Section 2 for details). Therefore, the larger number of distinct values
measured for the output variable (i.e. pupal weight), using the same combination of values of
those input variables, may have caused a lack of fit between input data and predicted output
values that may have decreased the predictive capability of the statistical and neural network
methods (Bianconi et al., 2010b).
Nonetheless, the number of distinct values of larval densities deployed in the present
work was considerably larger than those widely used in experimental designs on the
nutritional ecology of blowflies (Von Zuben et al., 2000; Bianconi et al., 2010b). Moreover,
such studies would be impractical if an even larger number of larval densities were used (Von
Zuben et al., 1993, 2001; Bianconi et al., 2010a, 2010b). Hence, the utilisation of artificial
The Effectiveness of Artificial Neural Networks … 109
neural networks is justified in this case, since the neural algorithms coped with the lack of fit
better than the general quadratic model (Table 1).
Future studies may consider other complex variables that were not assessed in the current
work. Ambient temperature, for example, was utilised by Zhang et al. (2008b). These authors
deployed neural network algorithms (functional link artificial neural networks) in order to
model the food intake dynamics of larvae of S. litura (Lepidoptera). Six different
temperatures were used for measuring the food intake, and the neural network approach was
deemed accurate. Moreover, temperature is regarded as an important variable in analysing the
dynamics of blowflies (Shiao and Yeh, 2008, Hwang and Turner, 2009; Cammack and
Nelder, 2010; Hu et al., 2010).
The utilisation of a simple linear regression may well require a non-trivial amount of
statistical expertise. The deployment of a multiple non-linear regression model such as a MLP
does require more knowledge and experience (Sarle, 1994). Therefore, it is important to
highlight the usefulness of multidisciplinary studies with the aim of conducting effective
investigations of bionomic parameters, because it is possible that unresolved problems
concerning the bionomics of immature and adult individuals of blowflies could be
disentangled and clarified by the use of neural models (Bianconi et al., 2010a, 2010b).
The employment of these complex analytical tools may well help entomologists to
feasibly schematise the sort of practical situations in which the utilisation of artificial
networks is able to provide new insights into the nutritional ecology of blowflies (Bianconi et
al., 2010a, 2010b). Additionally, the extraction of fuzzy rules from the fuzzy neural networks
provided an easily comprehensible way of describing what the networks had learnt.
Therefore, the outcomes of the present investigation may stimulate the use of artificial neural
networks in entomological research as a whole.
ACKNOWLEDGEMENTS
The Conselho Nacional de Desenvolvimento Científico e Tecnológico provided A.
Bianconi and C.J. Von Zuben with financial support.
REFERENCES
Arnaldos MI, García MD, Romera E, Presa JJ, Luna A. 2005. Estimation of postmortem
interval in real cases based on experimentally obtained entomological evidence. Forensic
Science International, 149:57-65.
Bianconi A, Von Zuben CJ, Serapião ABS, Govone JS. 2010a. Artificial neural networks: A
novel approach to analysing the nutritional ecology of a blowfly species, Chrysomya
megacephala. Journal of Insect Science, 10:1-18 (Article 58).
Bianconi A, Von Zuben CJ, Serapião ABS, Govone JS. 2010b. The use of artificial neural
networks in analysing the nutritional ecology of Chrysomya megacephala (F.) (Diptera:
Calliphoridae), compared with a statistical model. Australian Journal of Entomology,
49:201-212.
110 Michael J. Watts, Andre Bianconi, Adriane Beatriz S. Serapiao et al.
Bishop CM. Neural Networks for Pattern Recognition. Oxford: Oxford University Press,
1995.
Brewer FD. 1992. Gel extenders in larval diet of Cochliomyia hominivorax (Diptera:
Calliphoridae). Journal of Economic Entomology, 85:445-450.
Bryant SR, Thomas CD, Bale JS. 1997. Nettle-feeding nymphalid butterflies: temperature,
development and distribution. Ecological Entomology, 22:390-398.
Bryant SR, Shreeve TG. 2002. The use of artificial neural networks in ecological analysis:
estimating microhabitat temperature. Ecological Entomology, 27:424-432.
Cammack JA, Nelder MP. 2010. Cool-weather activity of the forensically important hairy
maggot blow fly Chrysomya rufifacies (Macquart) (Diptera: Calliphoridae) on carrion in
Upstate South Carolina, United States. Forensic Science International, 195:139-142.
Catts EP, Goff ML. 1992. Forensic entomology in criminal investigations. Annual Review of
Entomology, 37:253-272.
Chang CL, Caceres C, Jang EB. 2004. A novel liquid larval diet and its rearing system for
melon fly, Bactrocera cucurbitae (Diptera: Tephritidae). Annals of the Entomological
Society of America, 97:524-528.
Chaudhury MF, Alvarez LA. 1999. A new starch-grafted gelling agent for screwworm
(Diptera: Calliphoridae) larval diet. Journal of Economic Entomology, 92:1138-1141.
Cheng B, Titterington DM. 1994. Neural Networks: a review from a statistical perspective.
Statistical Science, 9:2-30.
Crick F. 1989. The recent excitement about neural networks. Nature, 337:129–132.
de Jong G. 1976. A model of competition for food. I. Frequency-dependent viabilities. The
American Naturalist, 110:1013-1027.
Fogel DB, Wasson EC, Boughton EM, Porto VW. 1997. A step toward computer-assisted
mammography using evolutionary programming and neural networks. Cancer Letters,
119:93–97.
Fogel LJ, Owens AJ, Walsh MJ. Artificial intelligence through a simulation of evolution. In
Maxfield M, Callahan A, Fogel LJ (editors). Biophysics and Cybernetic Systems:
Proceedings of the 2nd Cybernetic Sciences Symposium, pages 131–155. 1965.
Friese DD. 1992. Calf milk replacers as substitutes for milk in the larval diet of the
screwworm (Diptera: Calliphoridae). Journal of Economic Entomology, 85:1830-1834.
Furlanetto SMP, Campos MLC, Harsi CM, Buralli GM, Ishihata GK. 1984. Microorganismos
enteropatogênicos em moscas africanas pertencentes ao gênero Chrysomya (Diptera:
Calliphoridae) no Brasil. Brazilian Journal of Microbiology, 15:170-174.
Gabre RM, Adham FK, Chi H. 2005. Life table of Chrysomya megacephala (Fabricius)
(Diptera: Calliphoridae). Acta Oecologica, 27:179-183.
Garson GD. 1991. Interpreting neural-network connection weights. AI Expert, 6:47-51.
Gevrey M, Dimopoulos I, Lek S. 2003. Review and comparison of methods to study the
contribution of variables in artificial neural network models. Ecological Modelling,
160:249-264.
Gomes L, Von Zuben CJ. 2005. Postfeeding radial dispersal in larvae of Chrysomya albiceps
(Diptera: Calliphoridae): implications for forensic entomology. Forensic Science
International, 155:61-64.
Gomes L, Gomes G, Von Zuben CJ. 2009. The influence of temperature on the behavior of
burrowing in larvae of the blowflies, Chrysomya albiceps and Lucilia cuprina, under
controlled conditions. Journal of Insect Science, 9:1-5 (Article 14).
The Effectiveness of Artificial Neural Networks … 111
Goodbrod JR, Goff ML. 1990. Effects of larval population density on rates of development
and interactions between two species of Chrysomya (Diptera: Calliphoridae) in laboratory
culture. Journal of Medical Entomology, 27:338-343.
Greenberg B. 1991. Flies as forensic indicators. Journal of Medical Entomology, 28:565-577.
Guimarães JH, Prado AP, Linhares AX. 1978. Three newly introduced blowfly species in
Southern Brazil (Diptera, Calliphoridae). Revista Brasileira de Entomologia, 22:53-60.
Harrison PA, Berry PM, Butt N, New M. 2006. Modelling climate change impacts on
species‘ distributions at the European scale: implications for conservation policy.
Environmental Science and Policy, 9:116-128.
Haykin S. Neural Networks: A Comprehensive Foundation, 2nd edition. Upper Saddle River:
Prentice Hall, 1999.
Herzog JDA, Milward-Azevedo EMV, Ferreira YL. 1992. Observações preliminares sobre o
ritmo horário de oviposição de Chrysomya megacephala (Fabricius) (Diptera,
Calliphoridae). Anais da Sociedade Entomológica do Brasil, 21:101-106.
Howe PD, Bryant SR, Shreeve TG. 2007. Predicting body temperature and activity of adult
Polyommatus icarus using neural network models under current and projected climate
scenarios. Oecologia, 153:857-869.
Hu Y, Yuan X, Zhu F, Lei C. 2010. Development time and size-related traits in the oriental
blowfly, Chrysomya megacephala along a latitudinal gradient from China. Journal of
Thermal Biology, 35:366-371.
Huang Y. 2009. Advances in artificial neural networks – methodological development and
application. Algorithms, 2:973-1007.
Huang Y, Lan Y, Thomson SJ, Fang A, Hoffmann WC, Lacey RE. 2010. Development of
soft computing and applications in agricultural and biological engineering. Computers
and Electronics in Agriculture, 71:107-127.
Hwang CC, Turner BD. 2009. Small-scaled geographical variation in life-history traits of the
blowfly Calliphora vicina between rural and urban populations. Entomologia
Experimentalis et Applicata, 132:218-224.
Ireland S, Turner B. 2006. The effects of larval crowding and food type on the size and
development of the blowfly, Calliphora vomitoria. Forensic Science International,
159:175-181.
Joy MK, Death RG. 2004. Predictive modelling and spatial mapping of freshwater fish and
decapod assemblages using GIS and neural networks. Freshwater Biology, 49:1036-
1052.
Kasabov NK, Kim J, Watts MJ, Gray AR. 1997. FuNN/2 – a fuzzy neural network
architecture for adaptive learning and knowledge acquisition. Information Sciences –
Applications. 101(3-4): 155-175.
Köppler K, Kaffer T, Vogt H. 2009. Substantial progress made in the rearing of the European
cherry fruit fly, Rhagoletis cerasi. Entomologia Experimentalis et Applicata, 132:283-
288.
Kutner MH, Nachtsheim CJ, Neter J. Applied Linear Regression Models, 4th edition. New
York: Irwin, 2004.
Laurence BR. 1981. Geographical expansions of the range of Chrysomya blowflies.
Transaction of the Royal Society of Tropical Medicine and Hygiene,75: 130-131.
Laurence BR. 1986. Old World blowflies in the New World. Parasitology Today, 2:77-79.
112 Michael J. Watts, Andre Bianconi, Adriane Beatriz S. Serapiao et al.
Leal TTS, Prado AP, Antunes AJ. 1982. Rearing the larvae of the blowfly Chrysomya
chloropyga (Wiedemann) (Diptera, Calliphoridae) on oligidic diets. Revista Brasileira de
Zoologia, 1:41-44.
Levot GW, Brown KR, Shipp E. 1979. Larval growth of some calliphorid and sarcophagid
Diptera. Bulletin of Entomological Research, 69:469-475.
Lima MLPS, Luz E. 1991. Espécies exóticas de Chrysomya (Diptera, Calliphoridae) como
veiculadoras de enterobactérias patogênicas em Curitiba, Paraná, Brasil. Acta Biológica
Paranaense, 20:61-83.
Lomnicki A. Population ecology of individuals. Princeton: Princeton Press, 1988.
Milne LK. 1995. Feature selection with neural networks with contribution measures. In:
Proceedings of the Australian Conference on Artificial Intelligence AI’95, Canberra.
Neter J, Kutner MH, Nachtsheim CJ, Wasserman W. Applied Linear Statistical Models, 4th
edition. New York: Irwin, 1996.
Obach M, Wagner R, Werner H, Schmidt HH. 2001. Modelling population dynamics of
aquatic insects with artificial neural networks. Ecological Modelling, 146:207-217.
Olden JD, Jackson DA. 2002. Illuminating the ―bl ack box‖: a randomization approach for
understanding variable contributions in artificial neural networks. Ecological Modelling,
154: 135-150.
Olden JD, Joy MK, Death RG. 2004. An accurate comparison of methods for quantifying
variable importance in artificial neural networks using simulated data. Ecological
Modelling, 178:389-397.
Papandroulakis N, Markakis G, Divanach P, Kentouri M. 2000. Feeding requirements of sea
bream (Sparus aurata) larvae under intensive rearing conditions. Development of a fuzzy
logic controller for feeding. Aquacultural Engineering, 21:285-299.
Pearson RG, Dawson TP. 2003. Predicting the impacts of climate change on the distribution
of species: are bioclimate envelope models useful? Global Ecology & Biogeography,
12:361-371.
Pearson RG, Dawson TP, Berry PM, Harrison PA. 2002. SPECIES: A spatial evaluation of
climate impact on the envelope of species. Ecological Modelling, 154:289-300.
Peterson II RD, Candido AO. 1987. Larval and pupal weight relationships of six strains of
screwworm (Diptera: Calliphoridae) reared in the laboratory and in wounds. Journal of
Economic Entomology, 80:1213-1217.
Reis SF, Stangenhaus G, Godoy WAC, Von Zuben CJ, Ribeiro OB. 1994. Variação em
caracteres bionômicos em função da densidade larval em Chrysomya megacephala e
Chrysomya putoria (Diptera: Calliphoridae). Revista Brasileira de Entomologia, 38:33-
46.
Rumelhart DE, Hinton GE, Williams RJ. 1986. Learning representations by back-propagating
errors. Nature, 323:533–536.
Sarle WS. 1994. Neural networks and statistical models. In: Proceedings of the Nineteenth
Annual SAS Users Group International Conference. pp. 1538-1550. SAS Institute Inc.
Schultz A, Wieland R. 1997. The use of neural networks in agroecological modelling.
Computers and Electronics in Agriculture, 18:73-90.
Schultz A, Wieland R, Lutze G. 2000. Neural networks in agroecological modelling – stylish
application or helpful tool? Computers and Electronics in Agriculture, 29:73-97.
The Effectiveness of Artificial Neural Networks … 113
Shiao SF, Yeh TC. 2008. Larval competition of Chrysomya megacephala and Chrysomya
rufifacies (Diptera: Calliphoridae): behavior and ecological studies of two blow fly
species of forensic significance. Journal of Medical Entomology, 45:785-799.
Tachibana SI, Numata H. 2001. An artificial diet for blow fly larvae, Lucilia sericata
(Meigen) (Diptera: Calliphoridae). Applied Entomology and Zoology, 36:521-523.
Tammaru T, Kaitaniemi P, Ruohomäki K. 1996. Realized fecundity in Epirrita autumnata
(Lepidoptera: Geometridae): relation to body size and consequences to population
dynamics. Oikos, 77:407-416.
Tammaru T, Nylin S, Ruohomäki K, Gotthard K. 2004. Compensatory responses in
lepidopteran larvae: a test of growth rate maximisation. Oikos, 107:352-362.
Ullyett GC. 1950. Competition for food and allied phenomena in sheep-blowfly populations.
Philosophical Transactions of the Royal Society of London, 234:77-174.
Von Zuben CJ, Reis SF, do Val JBR, Godoy WAC, Ribeiro OB. 1993. Dynamics of a
mathematical model of Chrysomya megacephala (Diptera: Calliphoridae). Journal of
Medical Entomology, 30:443-448.
Von Zuben CJ, Stangenhaus G, Godoy WAC. 2000. Larval competition in Chrysomya
megacephala (F.) (Diptera: Calliphoridae): effects of different levels of larval
aggregation on estimates of weight, fecundity and reproductive investment. Revista
Brasileira de Biologia, 60:195-203. (in Portuguese)
Von Zuben CJ, Von Zuben FJ, Godoy WAC. 2001. Larval competition for patchy resources
in Chrysomya megacephala (Dipt., Calliphoridae): implications of the spatial distribution
of immatures. Journal of Applied Entomology, 125:537-541.
Warbrick-Smith J, Raubenheimer D, Simpson SJ, Behmer ST. 2009. Three hundred and fifty
generations of extreme food specialisation: testing predictions of nutritional ecology.
Entomologia Experimentalis et Applicata, 132:65-75.
Watts MJ, Worner SP. 2008. Comparing ensemble and cascaded neural networks that
combine biotic and abiotic variables to predict insect species distribution. Ecological
Informatics, 3:354-366.
Watts MJ, Worner SP. 2009. Estimating the risk of insect species invasion: Kohonen self-
organising maps versus k-means clustering. Ecological Modelling, 220:821-829.
Wells JD. 1991. Chrysomya megacephala (Diptera: Calliphoridae) has reached the
continental United States: review of its biology, pest status, and spread around the world.
Journal of Medical Entomology, 28:471-473.
Wijesundara DP. 1957. The life-history and bionomics of Chrysomya megacephala (Fab.).
Ceylon Journal of Science, 25:169-185.
Worner SP, Gevrey M. 2006. Modelling global insect pest species assemblages to determine
risk of invasion. Journal of Applied Ecology, 43:858-867.
Zadeh LA. 1965. Fuzzy sets. Information and Control, 8:338-353.
Zhang WJ, Barrion A. 2006. Function approximation and documentation of sampling data
using artificial neural networks. Environmental Monitoring and Assessment, 122:185-
201.
Zhang WJ, Zhang XY. 2008. Neural network modeling of survival dynamics of
holometabolous insects: a case study. Ecological Modelling, 211:433-443.
Zhang WJ, Zhong XQ, Liu GH. 2008a. Recognizing spatial distribution patterns of grassland
insects: neural network approaches. Stochastic Environmental Research and Risk
Assessment, 22:207-216.
114 Michael J. Watts, Andre Bianconi, Adriane Beatriz S. Serapiao et al.
Zhang WJ, Liu GH, Dai HQ. 2008b. Simulation of food intake dynamics of holometabolous
insect using functional link artificial neural network. Stochastic Environmental Research
and Risk Assessment, 22:123-133.
Zhang WJ, Wei W. 2009. Spatial succession modeling of biological communities: a multi-
model approach. Environmental Monitoring and Assessment, 158:213-230.
Zumpt F. Myiasis in Man and Animals in the Old World. London: Butterworths, 1965.
In: Ecological Modeling ISBN: 978-1-61324-567-5
Editor: WenJun Zhang, pp. 115-172 © 2012 Nova Science Publishers, Inc.
Chapter 7
Peter F. Newton1*
1
Canadian Wood Fibre Centre, Canadian Forest Service,
Natural Resources Canada, Sault Ste. Marie, Ontario, Canada, P6A 2E5.
ABSTRACT
An ecological-based decision-support system and corresponding algorithmic
analogue for managing natural black spruce (Picea mariana (Mill) BSP.) and jack pine
(Pinus banksiana Lamb.) mixed stands was developed. The integrated hierarchical system
consisted of six sequentially-linked estimation modules. The first module consisted of a
key set of empirical yield-density relationships and theoretically-based functions derived
from allometry and self-thinning theory that were used to describe overall stand dynamics
including temporal size-density interrelationships and expected stand development
trajectories. The second module was comprised of a Weibull-based parameter prediction
equation system and an accompanying composite height-diameter function that were used
to recover diameter and height distributions. The third module included a set of species-
specific composite taper equations that were used to derive log product distributions and
volumetric yields. The fourth module was composed of a set of species-specific
allometric-based composite biomass equations that were used to estimate mass
distributions and associated carbon-based equivalents for each above-ground component
(bark, stem, branch and foliage). The fifth module incorporated a set of species-specific
end-product and value equations that were used to predict chip and lumber volumes and
associated monetary equivalents by sawmill type (stud and randomized length mill
configurations). The sixth module encompassed a set of species-specific composite
equations that were used to derive wood and log quality metrics (specific gravity and
mean maximum branch diameter, respectively). The stand dynamic and structural
recovery modules were developed employing 382 stand-level measurements derived
*
Correspondence: peter.newton@nrcan.gc.ca
116 Peter F. Newton
from 155 permanent and temporary sample plots situated throughout the central portion
of the Canadian Boreal Forest Region, the taper and end-product modules were
developed employing published results from taper and sawmill simulation studies, and
the biomass and fibre attribute modules were developed using data from density control
experiments.
The potential of the system in facilitating the transformative change towards the
production of higher value end-products and a broader array of ecosystem services was
exemplified by simultaneously contrasting the consequences of density management
regimes involving commercial thinning treatments in terms of overall productivity, end-
product yields, economic efficiency, and ecological impact. This integration of
quantitative relationships derived from applied ecology, plant population biology and
forest science into a common analytical platform, illustrates the synergy that can be
realized through a multi-disciplinary approach to forest modeling.
1. INTRODUCTION
Density management within even-aged forest stands consists of regulating species
composition, density-stress levels and structural characteristics, by manipulating initial
planting densities at the time of establishment (initial espacement; IE) and (or) reducing stand
densities during subsequent stages of stand development (e.g., precommercial thinning (PCT)
at the sapling stage, and (or) commercial thinning (CT) at the semi-mature stage).
Ecologically, these treatments redistribute the finite environmental resources on a given site
(e.g., solar radiation, moisture, nutrients, and physical growing space) to selected crop trees
by controlling the intensity and frequency of symmetrical and asymmetrical competitive
interactions. Regulating site occupancy through density management has been a cornerstone
of silvicultural practice since it was first introduced in forestry by Reventlow in 1879
(Pretzsch, 2009). Density management continues to be an important component of intensive
forest management as evident by current efforts: over 500,000 ha of productive forest land
receive IE, PCT or CT treatments every year in both Canada and Finland (CCFM (2009) and
Peltola (2009), respectively).
A broad array of benefits at the tree, stand and landscape scale can be obtained when
implementing proper density management treatment protocols or prescriptions. Productivity
related benefits include increased growth and resultant yields leading to enhanced end-
products (e.g., Kang et al., 2004), early stand operability (e.g., Erdle, 2000), reduced self-
thinning rates and thus lower mortality losses (e.g., Pelletier and Pitt, 2008), spatial and
structural uniformity resulting in lower extraction, processing and manufacturing costs (e.g.,
Tong et al., 2005), and increased carbon sequestration rates (e.g., Nilsen and Strand, 2008).
Other tangible benefits include the production of coarse woody debris (e.g., Sturtevant et al.,
1996) and provision of hiding requirements (e.g., Smith and Long, 1987) for wildlife habitat,
controlling successional pathways in order to prevent the establishment and development of
ericaceous shrub species (e.g., Lindh and Muir, 2004), and enhanced biological diversity
(e.g., Verschuyl et al., 2011). Conversely, however, density management can result in
negative outcomes if regimes have not been optimally designed for a given management
Development and Utility of an Ecological-based Decision-Support System … 117
objective. Thinning treatments which remove too many trees resulting in large inter-tree
distances can promote the development of large branches and resultant knots thus lowering
lumber grades during the manufacturing stage (e.g., Zhang et al., 2005), extend the period of
juvenile wood production resulting in lower specific gravity and associated reductions in end-
product quality and value (e.g., Tong et al., 2009), and (or) reduce merchantable volume
productivity due to prolonged periods of insufficient site occupancy (e.g., Fleming et al.,
2005). Therefore, in order to attain the maximum benefits of density management treatments,
minimize exposure to the serious negative reunifications of incorrect prescriptions, and
provide an objective and transparent framework for justifying treatment decisions, decision-
support tools are required.
The stand density management diagram (SDMD) is a proven density management
decision-support tool which has been utilized by resource managers throughout many of the
world‘s forest regions. For example, SDMDs have been developed and utilized in managing
Japanese red pine (Pinus densiflora Siebold and Zucc.) plantations in Japan and South Korea
(Ando (1962, 1968) and Kim et al. (1987), respectively), Monterey pine (Pinus radiata D.
Don.) plantations in New Zealand and Spain (Drew and Flewelling (1977) and Castedo-
Dorado et al. (2009), respectively), Douglas fir (Pseudotsuga menziesii (Mirb.) Franco.)
plantations in the Pacific Northwest (Drew and Flewelling (1979) and Long et al. (1988)) and
Spain (López-Sánchez and Rodríguez-Soallerio, 2009), lodgepole pine (Pinus contorta var.
latifolia Engelm.) stands in the western USA (Flewelling and Drew (1985), McCarter and
Long (1986), and Smith and Long (1987)), black spruce plantations in central Canada
(Newton and Weetman, 1994), loblolly pine (Pinus taeda L.) plantations in the southern USA
(Dean and Baldwin, 1993), Scots pine (Pinus sylvestris L.) and Austrian black pine (Pinus
nigra Arn.) plantations in eastern Europe (Stankova and Shibuya, 2007), and Merkus pine
(Pinus merkusii Jungh. et de Vriese) plantations in Indonesia (Heriansyah et al., 2009). These
ecological-based decision-support tools are built upon (1) quantitative functional relationships
derived from applied ecology, plant population biology and forest production theory which
include the reciprocal equations of the competition–density and yield–density effect (Kira,
1953; Shinozaki and Kira, 1956), self-thinning rule (Yoda et al., 1963), and site occupancy-
production relationships (Drew and Flewelling, 1979), and (2) empirical allometric
relationships which include size–density relationships for describing the effect of population
density-stress on tree dimensions such as quadratic mean diameter, mean volume, and mean
live crown ratio (e.g., Drew and Flewelling, 1977)). These relationships represent the
cumulative effect of density-dependent resource competition processes on productivity,
allometry and survivorship patterns at both the individual and population levels.
Historically, the analytical development of SDMDs has been characterized by a sequence
of continuous incremental advancements in which increasingly complex and innovative
model variants have been proposed. The first model forms, static SDMDs, where initially
developed in Japan by Ando (1962) and later introduced into North America by Drew and
Flewelling (1977, 1979). However, the lack of mortality submodels to describe stand
dynamics during the self-thinning stage limited their use in density management decision-
making. In response, dynamic SDMDs in which an embedded mortality submodel was
explicitly incorporated within the SDMD model structure, were proposed in the mid 1990s
(e.g., Newton and Weetman 1993, 1994). Later, acknowledging the paradigm shift in
management focus from volumetric yield maximization to end-product recovery and value
118 Peter F. Newton
maximization (e.g., Barbour and Kellogg, 1990; Emmett 2006), and realizing the limitations
of both the static and dynamic variants in terms of addressing these new objectives, the
structural SDMD was introduced (Newton et al., 2004, 2005). Specifically, the structural
model incorporated a parameter prediction equation system for recovering diameter
distributions within the dynamic SDMD model structure and hence enabled the estimation of
size-dependent end-product and value attributes.
The most recent iteration of the SDMD modeling framework is represented by the
integrated modular-based structural stand density management model (SSDMM) developed
for natural and managed jack pine (Pinus banksiana Lamb.) stand-types (Newton, 2009). The
rationale for developing this model was to provide resource managers with a comprehensive
density management decision-support tool that could address volumetric, end-product,
economic and ecological objectives, simultaneously. While this modeling platform has been
successfully applied to monospecific stand-types, the mixed stand analogue has yet to be
developed. Consequently, the first objective of this study was to develop a modular-based
SSDMM and associated algorithmic analogue for natural (naturally regenerated stands
without a history of density regulation) black spruce (Picea mariana (Mill) BSP.) and jack
pine mixed stands. The second objective was to demonstrate the utility of the resultant model
by contrasting operationally plausible density control regimes using a broad array of
productivity, economic and ecological performance metrics (e.g., volumetric yield outcomes,
log-product distributions, biomass production and carbon yields, recoverable products and
associated values, economic efficiency, duration of optimal site occupancy, structural
stability, fibre attributes, and operability status). Refer to Newton (2010) for a general
overview of SDMDs in terms of their history, modeling approach, foundation studies, and
current modeling activities, and to Drew and Flewelling (1979), Newton and Weetman (1993)
and Newton et al. (2004) for specific details regarding the quantitative foundation of the
static, dynamic and structural model variants, respectively.
2. METHOD
The hierarchical designed SSDMM consisted of six sequentially-linked estimation
modules which were denoted according to the following nomenclature: Module A - Dynamic
SDMD; Module B - Diameter and Height Recovery; Module C - Taper Analysis and Log
Estimation; Module D - Biomass and Carbon Estimation; Module E - Product and Value
Estimation; and Module F - Fibre Attribute Estimation Module. Analytically, Module A
involved the development of a dynamic SDMD via the parameterization and integration of a
set of static and dynamic yield–density relationships. Module B consisted of the development
of a Weibull-based parameter prediction equation system and an associated composite height-
diameter prediction equation for recovering diameter and height distributions. Module C was
developed through the employment of species-specific dimensional compatible taper
equations derived from the literature which were used to predict log products (number of
pulplogs and sawlogs) and stem volumes. Module D utilized species-specific allometric-
based composite biomass equations derived from both the literature (jack pine) and density
control experimental data (black spruce) to predict oven-dried masses and associated carbon
equivalents for all four above-ground components, bark, stem, branch and foliage. Module E
Development and Utility of an Ecological-based Decision-Support System … 119
employed species and sawmill (stud and randomized length mill) specific end-product and
value equations derived from the literature to predict the volume and monetary worth of the
manufactured end-products (wood chips and dimensional lumber). Module F involved the
development of composite equations for estimating wood density and mean maximum branch
diameter. An algorithmic analogue of the resultant modular-based SSDMM was also
developed and its utility exemplified by simultaneously contrasting a set of complex density
management regimes in terms of productivity metrics, log-product distributions, biomass pro-
duction and carbon yields, quantity and quality of recoverable end-products, economic
efficiency, duration of optimal site occupancy, and structural stability.
In presenting the methods, it should be noted that in cases where the required
relationships were previously developed in a concurrent investigation (i.e., parameter
prediction equation system, asymptotic size–density relationship, and the composite height-
diameter function) or reported in the literature (e.g., taper, end-product and value functions),
only an abridged version of the pertinent methods and results are included. In other cases
where the required relationships could not be derived from the core calibration data set
(described below) nor from the literature, supplemental data sets were acquired and analyzed
from which the required functions were developed (e.g., data from Nelder plots were used to
develop mean live crown ratio and mean maximum branch diameter prediction functions, and
data from other types of density control experiments were used to parameterized the biomass
and wood density prediction functions). Although similarities exist in regards to the
monospecific model presented by Newton (2009) for jack pine stand-types, the mixed model
variant presented in this study expands, modifies and extends the modeling platform through
the introduction of species invariant relationships for bivariate mixtures, a response delay
function for accounting for post-thinning effects on stand dynamics, and new performance
metrics. In order to facilitate replication and application to other species, the principal
analytical methods, model structure, and the computational sequence utilized, are provided in
their entirety.
Stands belonging to the natural even-aged black spruce and jack pine mixed stand-type,
denoted PImPNb(N), were defined as those which had the following attributes (Table 1): (1)
comprised principally of black spruce and jack pine in which the (i) density percentage had to
constitute a minimum of 90% and 10% on a collective and individual species basis,
respectively, for stands at the establishment stage of development, or alternatively, (ii) basal
area percentage had to constitute a minimum of 90% and 10% on a collective and individual
species basis, respectively, for stands past the establishment stage of development; (2)
situated on mineral soils and regenerated naturally following a stand-replacing disturbance
(e.g., wildfire or forest harvesting); and (3) a silvicultural history absent of artificial
regeneration, PCT or CT treatments.
The calibration data for Modules A and B were derived from measurements obtained
from temporary and permanent sample plots (denoted TSPs and PSPs, respectively) located
throughout the central region of the Canadian Boreal Forest Region. In total, 382 tree-list
measurements derived from 155 sample plots were utilized. Geographically, these plots were
largely concentrated within Forest Sections B-4, B-7, B-8, B-9, B-10, B-11 and B-14 (Rowe,
120 Peter F. Newton
1972) which fall within the northeastern, northcentral and northwestern administrative
regions of the Province of Ontario. Historically, the plots were initially established by various
forest-based industrial corporations and government agencies during the 1930-1990 period
and were distributed across the landscape using a stratified pseudo-random sampling design
(n., strata were based on the age-class structure and site quality variation within a given
region or tenured landbase). The plots were circular, square or rectangular in shape with a
mean overall area of 0.082 ha (standard deviation (SD) = 0.021 ha; minimum/maximum =
0.040/0.120 ha) and density of 189 trees/plot (SD = 104 trees/plot; minimum/maximum =
42/926 trees/plot). For the purposes of this study, the plots were also differentiated based their
utility in describing temporal change (stand dynamics): static plots were those that were only
measured once whereas dynamic plots were those that were measured more than once and
hence capable of providing change estimates in terms of surviver growth, ingress, ingrowth
and mortality. Table 1 summarizes the plot measurements according to the source
organization, series and type (static or dynamic) along with the number and frequency of
measurements obtained.
For each plot, its geographic location, disturbance history and the measurement
techniques utilized, were known. Generally, the measurements on each plot included species-
specific tree-lists consisting of diameter measurements at breast-height (1.37 m or 1.3 m) -
outside bark (D; ±0.25 cm) for all biotic trees greater than 2.54 cm in D, and D and total
height (H; ±0.30 m) measurements on a subset of approximately 10 black spruce and jack
pine sample trees, usually selected from across a plot‘s diameter range.
Combining the individual tree values (D and H measurements or estimates derived from
species-specific allometric-based height-diameter functions) with regional volume equations
and the plot area information, the following stand-level variables were calculated: (1) mean
dominant height (Hd; m), defined as the mean height of the trees within the largest height
quintile; (2) quadratic mean diameter (Dq; cm); (3) basal area (G; m2/ha); (4) total volume (Vt
(m3/ha), computed as the sum of individual tree total volumes (vT; m3/tree) as determined
from the D and H values inputted into regional-wide standardized volume equations (Honer et
al., 1983); (5) merchantable volume (Vm (m3/ha), computed as the sum of the individual tree
merchantable volumes (vM; m3/tree) for all trees greater than 9 cm in D as determined from
the D and H values, vT estimate and specified merchantability limits (i.e., 0.15 m stump-
height and a 7.62 cm top diameter (inside-bark); Honer et al., 1983); (6) total density (N
(stems/ha)); and (7) relative density index (Pr (%/100) as defined as the ratio of N to the
maximum N attainable in a stand with the same mean volume (Drew and Flewelling, 1979;
Newton and Weetman, 1993; Newton, 2006a). The mensurational characteristics of the
datasets are summarized in Table 2.
Table 1. Defining attributes of the natural black spruce and jack pine mixed
stand-type and plot data sources, types and their measurement frequency
Stand-type Principal Defining Attributes Plot Series and Number of Plots by Measurement Type Measurement Sequence and Number
Denotation
Seriesa Typeb # of Consecutive Total
# of Plots
Measurements #
Static Dynamic
PImPNb(N) Even-aged black spruce and jack pine mixed Kimberly Clark (PSPs) 20 1 102 382
stands situated on mineral soils which regenerated
American-Can (PSPs) 33 2 1
naturally following a stand-replacing disturbance
(e.g., wildfire or forest harvesting) with no prior Boreal-Growth
66 3 2
history of artificial regeneration or density (OMNR/TSPs)
management treatments (e.g., precommercial or PGP (TSPs) 36 4 9
commercial thinning). Black spruce and jack pine 5 17
collectively constituted >90% of the total density
6 19
at the time of establishment, or >90% of the basal
area of established stands, with each species 7 3
constituting a minimum of 10% of either measure. 8 2
Similar to the JP2 working group analog (Watt et
al., 2001).
a - Plot series denoted according to Ontario-centric nomenclature where the host organization responsible for initial plot establishment is
acknowledged: corporation or governmental agency where OMNR refers to the Ontario Ministry of Natural Resources. Note, TSPs and PSPs
denote temporary and permanent sample plots, respectively.
b - Dynamic plots were those for which a remeasurement was obtained whereas static plots were those for which only a single measurement was
obtained.
122 Peter F. Newton
As described in detail below, the development of the dynamic SDMD consisted of the
parameterization of the following relationships: self-thinning line and the derived relative
density index function; yield–density relationships and the associated isolines for quadratic
mean diameter, mean dominant height and mean live crown ratio; mean volume–density
relationships at the time of initial crown closure and those delineating the zone of maximum
production; and net density change function for predicting post-crown-closure size–density
trajectories. Subsequent integration of these relationships within the traditional modeling
framework resulted in the dynamic SDMD for mixed stands.
where 0 and 1 are intercept and slope (self-thinning) coefficients, respectively, and is
an error term. The parameter estimates for self-thinning mixed stands, as previously reported
by Newton (2006a), were used in this study (Table 3). The function for calculating relative
density index (Pr; Drew and Flewelling, 1979), defined as the ratio of a stand‘s observed
density ( N o ) to that of the maximum density (Nmax) attainable in a stand with the same mean
volume vo , was derived from the self-thinning rule (Eq. (2); Table 3).
1
vo 1
Pr N o 100 (2)
Thus solving Eq. (2) for log10 v yielded the isoline function for a given Pr value (Eq.
(3)).
100
log10 v log10 1 1 log10 N (3)
Pr
Table 2. Mensurational characteristics of the sample trees and stands
utilized to develop the modular-based SSDMM for mixed stands
log10 N for a given Dq value (Eq. (5)) was derived by solving Eq. (4) for log10 v .
log10 Dq 0 2 log10 N
log10 v (5)
1
Table 3. Parameter estimates and associated statistics for the regression relationships developed in Module A (Dynamic SDMD)
Pr N o / vo / 106.145 .
0.805
(4) Derived relative density index (Pr (%/100)) equation (Eq. (2)):
(5) Source: Newton (2006b).
Quadratic mean diameter = (1) Parameter estimates obtained from OLS regression analysis.
̂0 0.5155
f(mean volume and density) (2) The intercept parameter estimate includes a correction factor for the bias introduced via the logarithmic
(Eq. (4)) ˆ1 0.3674 transformation (Baskerville, 1972; Sprugel, 1983).
(3) Regression statistics: nreg = 2; nres = 358; multiple coefficient of determination (R2) = 0.981; standard error of
̂ 2 -0.0377 the estimate (SEE; log10(cm)) = 0.0147; and F-ratio = 9430* where * denotes a significant (p 0.05)
relationship.
Mean volume = f(mean ̂ 0 -1.3715 (1) Parameter estimates obtained from OLS regression analysis.
dominant height and (2) The intercept parameter includes a correction factor as described above.
quadratic mean diameter) ̂1 0.8576 (3) Regression statistics: nreg = 2; nres = 356; R2 = 0.995; SEE (log10(dm3)) =
(Eq. (6)) 0.0183; and F-ratio = 38547*.
̂ 2 2.0495
Mean live crown ratio =
ˆ0
3.4027 (1) Parameter estimates obtained from OLS regression analysis.
f(mean volume and density) (2) The intercept parameter includes a correction factor as described above.
(Eq. (8)) ˆ1 -0.3684 (3) Regression statistics: nreg = 2; nres = 1960; R2 = 0.545; SEE (log10(%)) = 0.1139; and F-ratio = 1170*.
Net density change model ̂ 0 -0.1010 (1) Parameter estimates obtained from nonlinear regression analysis.
(Eq. (10)) (2) Regression statistics: nreg = 3; nres = 215; SEE (log10(stems/ha)) = 18.28; and F-ratio = 1164004*.
̂1 0.2756
̂ 2 -0.3055
̂ 3 -0.3716
126 Peter F. Newton
The Hd isoline function which defines the relationship between log10 v and
log10 N for a given Hd value (Eq. (7)) was derived by expressing v as function of Hd and
Dq using Eq. (6), substituting Eq. (4) into Eq. (6), and then solving for log10 v .
0 2 0 1 log10 ( H d ) 2 2 log10 ( N )
log10 (v ) (7)
1 2 1
actual parameter estimates and associated regression statistics for Eq. (8) are listed in Table 3.
In terms of deriving the isolines for Lr values of 35, 40, 50, 60, 70 and 80%, Eq. (8) was
rearranged with respect to log10 v log10 v log10 Lr 0 2 log10 N 1
and log10 v calculated across the density range, for the specified Lr value.
data set consisting of multiple v N data pairs for the initial crown closure condition. These
data pairs were then used to parameterized Eq. (9).
where i , 0,1 are parameters estimated employing OLS regression analysis, and is an
error term. Employing the same detection procedures and evaluation protocol as that specified
for Eq. (4), regression results suggested that relationship was adequate in describing the size-
density condition at the time of initial crown closure (i.e., the relationship was significant (p ≤
0.05), explained a moderate proportion of the variation, and was in compliance with the constant
error variance, correct model specification and normality assumptions). Table 3 lists the
parameter estimates and associated regression statistics for Eq. (9).
1 2 3
EXP 0 Pr A S I
Nt
Nt 1 N m (t ) (10)
N
min(t )
Development and Utility of an Ecological-based Decision-Support System … 129
where Nt+1 is the density at time t+1 (t = 1,…, T-1; T = rotation age), Nt is the density at time
t, N min(t ) is the minimum asymptotic density at which density-dependent processes initiate as
log10 ( v )0 1
defined as the density at the time of crown closure ( Nmin(t ) 10 as derived from
Eq. (9)), i , i 0,...,3 are model parameters, and is an error term. Computationally, mean
annual net density change estimates were derived from the dynamic sample plots as follows.
Initial and final densities for each successive plot measurement that had attained crown
closure status were calculated from which the mean annual net density change estimate was
derived: ΔN = (N2-N1)/(A2-A1) where ΔN is in units of stems/ha/yr, and N2 and N1 are the
densities (stems/ha) at stand ages A2 and A1, respectively. These change estimates (Nt (i.e., N1)
and Nt+1 (Nt + ΔN)) along with the corresponding Pr, A and SI values were used in
combination with nonlinear regression analysis to parameterize Eq. (10). Using the same
evaluation protocol as that described for Eq. (4), the regression results suggested that the
relationship was significant (p ≤ 0.05) and was in compliance with the constant error variance,
correct model specification and normality assumptions. Table 3 lists the resultant parameter
estimates and associated regression statistics for Eq. (10).
c D a c 1 D a c
exp if a D
f ( D; a, b, c) b b b (11)
0 if D < a
where a is the location parameter and is less than or equal to the minimum D value (Dmin), b
is the scale parameter which reflects the range of the distribution, and c is the shape parameter
which reflects the degree of skewness within the D distribution. These parameter estimates
were then expressed as direct or indirect functions of a set of stand-level variables, as shown
in Eqs. (12-15), using multiple regression analysis and cumulative density function regression
(CDFR; Cao, 2004) analysis.
D a c
F ( D; a, b, c) 1 exp
b
(16)
Table 4. Parameter estimates and associated statistics for the regression relationships used in Module B (Diameter and Height
Recovery): parameter prediction equation (PPE) system and the composite height-diameter function
where l (i ) , l 0,...,7 are parameters, and ε is an error term. The resultant parameter
estimates and associated regression statistics are given in Table 4.
Due to data limitation which prevented the development of a taper equation for the
natural mixed stand-type, composite variable exponent taper equations developed previously
for monospecific black spruce (Sharma and Parton, 2009) and jack pine (Newton, 2009)
stands, were utilized (Eq. (18a) and (18b), respectively).
2
h h G
1 2 3 4 2
h H H D H h
d D 0 (18a)
1.3 H 1.3
h G
1 2 4 2
h H D H h
d D 0
1.3 (18b)
H 1.3
where d is the inside-bark diameter (cm) at height h (m), i , i = 0,...,4 are model
parameters, and ε is an equation-specific error term. These equations arose from a concurrent
study in which the objective was to develop dimensional compatable taper equations for a
suite of boreal species (e.g., Newton and Sharma, 2008; Sharma and Parton, 2009). Briefly,
the equations were parameterized using stem profile data obtained from percent-height
destructive stem analysis sampling procedures on 113 jack pine and 1189 black spruce trees.
The sample trees were selected using a size-based stratified random sampling protocol within
9 jack pine and 25 black spruce stands. The actual parameter estimates are given in Table 5.
134 Peter F. Newton
The lack of available biomass data for mixed stands negated the development of stand-
specific equations. Alternatively, however, composite biomass functions previously
developed for estimating bark (periderm) mass per tree (Mp; oven-dry g), stem mass per tree
(Ms; oven-dry g), branch mass per tree (Mb; oven-dry g) and foliage mass per tree (Mf; oven-
dry g) within monospecific black spruce (Newton, unpublished) and jack pine (Newton,
2009) stands, were employed. Briefly, the composite models were based on a multivariate
extension of the simple equation of allometry in which stand-level measures of relative
density stress (Pr) and stage of development (Hd) were explicitly incorporated (e.g., Jolliffe et
al., 1988; Newton, 2006b). The resultant equations predict the oven-dried mass of each
component per tree using D, H, Pr and Hd as predictor variables (Eq. (19)).
m(i ) 0(i ) D 2 H
1( i ) 2( i ) Pr 3( i ) H d 4( i ) Pr H d
Pr H d
5( i ) 6( i ) 7 ( i )
Pr Hd (i ) (19)
where m(i) is the mass of the ith biomass component, j (i ) , j 0,...,7 are parameters specific
to the ith component estimated via OLS regression analysis (all-possible best-subset
regression procedure employing Mallows (1973) Cp selection criterion) following a double-
logarithmic transformation, and ( i ) is the error term specific to the ith component. For black
spruce, the parameterization data set consisted of mp, ms, mb, mf, D, H, Pr and Hd data derived
from 161 destructively sampled black spruce trees located in 52 variable-sized (minimum of
100 trees/plot) area plots which were established within 18 monospecific even-aged stands
located throughout Forest Section B28b (Rowe, 1972)). The stands were selected according to
a stratified pseudo-random sampling approach in which an approximately equal number of
stands within each of 5 age classes (15, 30, 45, 60 and 75 year) were sampled. Similarly, for
jack pine, the parameterization procedure consisted of obtaining estimates via OLS regression
analysis using mp, ms, mb, mf, D, H, Pr and Hd data derived from 186 semi-mature jack pine
trees situated within 6 stands located in northern and northeastern Ontario. The sample trees
were selected for destructive sampling according to a size-based stratified random sampling
protocol. Table 6 lists the resultant component-specific parameter estimates and associated
regression statistics.
Table 5. Parameter estimates and associated statistics for the composite
variable-exponent taper equations used in Module C (Taper Analysis and Log Estimation)
Table 6. Parameter estimates and associated statistics for the regression relationships
used in Module D (Biomass and Carbon Estimation): composite biomass equations by component (Eq. (19))
Black Bark 4.8 0.9524 - - - - - -0.1507 2 156 2.39 0.992 0.2261 4613*
Spruce Stem 14.2 0.9325 - - - 0.0006 - 0.2611 3 154 1.90 0.998 0.1265 11996*
Branch 4.1 1.2048 - -0.0127 - - - -0.5282 3 154 2.34 0.898 0.5409 454*
Foliage 16.6 0.9611 0.1522 - -0.0121 - - -0.6127 4 154 3.96 0.892 0.4964 318*
Jack Bark 17.1 0.5787 0.2083 0.0136 -0.0175 - - - 4 181 3.74 0.808 0.2425 191*
Pine Stem 66.0 1.0096 -0.1471 - - 0.6895 - - 3 182 4.12 0.930 0.1716 802*
Branch 330.3 - 1.1513 0.0570 -0.0503 - - -2.6615 4 181 3.95 0.863 0.3191 286*
Foliage 345.1 - 1.1635 0.0551 -0.0498 - - -2.8438 4 181 3.91 0.836 0.3448 231*
a - Component-specific parameter estimates obtained from OLS regression analysis where the independent variables were selected according to an all-
possible best-subset regression procedure employing Mallows‘ (1973) Cp criterion. Note, the intercept parameter estimates, denoted ̂0' , includes
a correction factor for the bias introduced via the logarithmic transformation (Baskerville, 1972; Sprugel, 1983), and Comp. denotes Component.
b - As defined in Table 3.
136 Peter F. Newton
1( i ) 2( i )
Y(i ) 0(i ) D H (i ) (20a)
Y(i ) 0(i ) D2 H
1( i ) 2( i ) Pr 3( i ) H d 4( i ) Pr H d
Pr H d
5( i ) 6( i ) 7( i )
Pr Hd (i ) (20b)
where Y(i) is either the (1) total lumber volume per tree (vl(s) (dm3)), total lumber value per tree
(pl(s) (CAN$(2002))), or total product value per tree (pt(s) (CAN$(2002))) recovered under a
stud sawmill processing protocol, or (2) total lumber volume per tree (vl(r) (dm3)), total lumber
value per tree (pl(r) (CAN$(2002))) or total product value per tree (pt(r) (CAN$(2002)))
recovered under a randomized length sawmill processing protocol, j (i ) , j 0,..., 7 and
(i ) are equation-specific parameters and error terms associated with the ith dependent
variable. These equations were parameterized employing simulation results derived using
virtual taper profiles from tree measurements obtain from sample plots distributed throughout
the central region of the Canadian Boreal Forest. Specifically, the Optitek sawing simulator
(Forintek Canada Corp. 1994) software, employing conventional stud sawmill and
randomized length sawmill processing protocols, was used to derive product volumes and
value estimates. The stud sawmill processing protocol consisted of bucking each virtual stem
of a given diameter, length, taper and wane into 2.44 m sections and then sawing each section
into dimensional lumber products according to an algorithm which maximized lumber value
recovery. The randomized length sawmill processing protocol consisted of optimally bucking
the virtual stem into sections of variable length (4.88 to 1.22 m by approximately 0.6 m
intervals) and then sawing each section into dimensional lumber products employing a sawing
algorithm which similarly maximized lumber value recovery. The resultant sawdust, chip and
lumber volumes in combination with market-based economic values for the calendar year
2002, were used to generate tree-level estimates of lumber value and total product value by
species and sawmill-type. Combining these estimates along with the stand-level variables,
simple (Eq. (20a)) and composite (Eq. (20b)) equations were developed by sawmill-type.
Table 7 lists the resultant component-specific parameter estimates, regression statistics and
predictive performance measures for each equation.
Table 7a. Parameter estimates and associated statistics for the regression relationships used in Module E
(Product and Value Estimation): sawmill-specific product recovery and value functions (Eq. (20a))
Table 7b. Parameter estimates and associated statistics for the regression relationships used in Module E (Product and Value
Estimation): composite sawmill-specific product recovery and value functions (Eq. (20b))
Jack vl(s) 0.00054 1.2528 0.2017 - -0.0069 -0.5462 0.2597 - 5 946 5.00 0.947 0.1561 3367*
Pine pl(s) 0.00005 1.5637 - -0.0030 - - - -0.0461 3 948 2.30 0.959 0.1561 7291*
pt(s) 0.00002 1.5305 0.0435 -0.0088 - -0.2880 0.5576 - 5 946 5.01 0.975 0.1133 7438*
vl(r) 0.00004 1.5929 - -0.0156 - -0.0993 - 1.2756 4 947 5.34 0.952 0.1525 4703*
pl(r) 0.00001 1.6632 - -0.0127 - -0.1313 0.9862 - 4 947 3.04 0.969 0.1319 7335*
pt(r) 0.00001 1.6130 - -0.0125 - -0.1248 0.9748 - 4 947 3.45 0.973 0.1180 8595*
a - Component-specific parameter estimates obtained from OLS according to an all-possible best-subset regression procedure employing Mallows‘
(1973) Cp criterion. The intercept parameter estimate ( ̂ 0' ) includes a logarithmic-based correction factor (Baskerville, 1972; Sprugel, 1983).
b - As defined in Table 3; SEE in units of loge(dm3) for dependent variables vl(s) and vl(r) and loge(CAN$(2002)) for dependent variables pl(s), pt(s), pl(r)
and pt(r).
138 Peter F. Newton
The composite wood density and mean maximum branch diameter functions previously
developed for black spruce (Newton, unpublished) and jack pine (Newton, 2009) were
utilized. Briefly, species-specific cross-sectional area-weighted mean wood density per tree
( wD (g/cm3)) was estimated using Eq. (21), and species-specific mean maximum branch
diameter within the first 4.9 m sawlog per tree ( bD (cm)) was estimated using Eq. (22).
wD 0 D 2 H
1 2 Pr 3 H d 4 Pr H d
Pr 5 H d 6 Pr H d 7
(21)
d B 0 D2 H
1 2 Pr 3 H d 4 Pr H d
Pr5 H d6 Pr H d 7
(22)
where j , j 0,...,7 and j , j 0,...,7 are model parameters estimated via OLS
regression analysis following a double-logarithmic transformation using an all-possible best-
subset regression procedure based on Mallows‘ (1973) Cp selection criterion, and is an
equation-specific error term. Table 8 lists the resultant parameter estimates and associated
regression statistics.
Composite mean ˆ0' 3.6137 0.4901 (1) Parameter estimates obtained from OLS regression analysis where
maximum branch the independent variables were selected according to an all-possible
ˆ - 0.2976
diameter 1 best-subset regression procedure employing Mallows‘ (1973) Cp
equation - -0.0338 criterion.
(Eq. (22))
ˆ2 (2) The intercept parameter estimates includes a correction factor for
ˆ - - the bias introduced via the logarithmic transformation (Baskerville,
3
- 0.0024 1972; Sprugel, 1983).
ˆ4 (3) Regression statistics: nreg = 1, nres = 117, R2 = 0.053, SEE
ˆ5
-0.0427 - (loge(cm)) = 0.1147, CP statistic = 2.92, and F-ratio =7*, for black
spruce; and nreg = 4, nres = 510, R2 = 0.483, SEE (loge(cm)) = 0.1440,
ˆ6 - -
CP statistic = 3.38, and F-ratio =119*, for jack pine.
ˆ7
- -0.2021
Module C - Taper Analysis and Module D - Biomass and Carbon Module E – Product and Value Estimation Module F - Fibre Attribute Estimation
Log Estimation Estimation
Species-specific sawmill-specific product Species-specific composite wood density and
Species-specific composite taper equations Species-specific composite biomass equations recovery and value equations mean maximum branch diameter equations
(Figure 2(c)) (Figure 2(d)) (Figure 2(e)) (Figure 2(f))
The function was based on the difference in live crown ratios between trees within the
pre-thinned and post-thinned stand condition as initially conceptualized by Newton (2003).
To illustrate this idea, it is instructive to consider the empirical results derived from an IE
black spruce experiment as they are reported in the literature (i.e., Table 1 in McClain et al.,
1994). The differential in mean live crown ratio between a high density-stressed stand
(surrogate for the thinned stand just before the time of thinning (Lr = 44%)) and a low
density-stressed stand (surrogate for the thinned stand immediately following thinning (Lr =
80%)), situated on the same site and of equal age, was 36% after 41 years. If the high density-
stressed stand was thinned at an age of 41, SDMD-based models would assume there was no
differential between the stands and erroneously describe the post-treatment size-density
trajectory as being identical to that of a stand which was always managed at the lower
density. Thus in this example, the temporal duration of the response delay is approximated by
the length of time required for trees within the thinned stand to achieve a mean live crown
ratio of 80%. More precisely, by extending this concept and accounting for the intrinsic
change in live crown ratio within the lower density stand in the years immediately following
thinning, the exact number of years required for the live crown ratios in both stands to
converge, can be determined. Specifically, the number of years that the live crown ratio of the
tree of mean dominant height took to rebuild its live crown ratio (based on a static live crown
base), to be equivalent to that predicted for a similar tree growing at the lower density, was
defined as the response delay period. Given that trees within recently thinned stand would not
be fully occupying their newly allocated space and hence not competing nor incurring
density-dependent mortality until their live crown ratios recovered, stand densities were
conditionized to remain constant during this period (i.e., the net density change function was
disabled during this adjustment period).
The temporal rate at which the size–density trajectory tracks through the SDMD is site-
dependent and hence described by the mean values derived from species-specific height-age
functions (Eq. (23a) for black spruce (Carmean et al., 2006) and Eq. (23b) for jack pine
(Carmean et al., 2001)).
0.6167( S I 1.3)0.3116
0.1136 AB
H d 1.3 16.95( S I 1.3) 1 K 50
(23a)
1
( S I 1.3) 0.6167( S I 1.3)0.3116
where K 1 0.1136
16.95( S I 1.3)
1.3723( S I 1.3)0.0802
0.6224
AB
H d 1.3 4.1459( S I 1.3) 1 K
50
1
(23b)
( S I 1.3) 1.3723( SI 1.3) 0.0802
where K 1 0.6224
4.1459( S I 1.3)
Pˆr (t ) ) using the dynamic SDMD and its embedded functions. Module B (Figure 2(b))
recovers the corresponding location, scale and shape parameter estimates for the 3-parameter
Weibull PDF using the PPE system (Eqs. (12)-(15)) in combination with the stand-level
variable estimates derived from Module A, from which density estimates are derived for each
recovered diameter class using Eq. (16) and a 10% truncation threshold rule (Clutter et al.,
1983). Height estimates for each recovered diameter class are then calculated employing the
stand-level and diameter-class-specific variable estimates ( Pˆr ( t ) and Hˆ d (t ) (from Module A)
ˆ , respectively) via the composite height-diameter function (Eq. (17)). Module C
and D(t , j )
(Figure 2(c)) computes tree and diameter-class estimates of merchantable and total stem
volumes, log-type distributions and residual merchantable tip volumes at each time step via
the taper equations (Eqs. (18a) and (18b)) using stand-level and diameter-class-specific
ˆ from Module A, and Dˆ , Hˆ ˆ
predictor variable estimates ( G(t ) (t , j ) ( t , j ) and N ( t , j ) from Module
B, respectively). Module D (Figure 2(d)) provides bark, stem, branch, foliage and total
biomass estimates at the tree, diameter-class and stand levels for each time step using the
composite biomass equations (Eq. (19)) in combination with the stand-level and diameter-
class-specific variable estimates ( Pˆr ( t ) and Hˆ d (t ) from Module A and D
ˆ , Hˆ
(t , j ) ( t , j ) and
level and diameter-class-specific variable estimates ( Pˆr ( t ) and Hˆ d (t ) from Module A, and
mean wood density for the merchantable tree population, and mean maximum branch
diameter for the sawlog-sized tree population, are subsequently calculated using a diameter-
class density-based weighing factor.
The above sequence represent the essential computations required for deriving volumetric
yields, diameter distributions, tree heights, log assortments, components-specific biomass and
carbon outcomes, sawmill-specific products and associated values, and fibre attributes, for a
given density management regime, site quality and rotation age.
(a) Module A - Dynamic SDMD
Hˆ d ( t )
Input Variables A(t )
(t )
Nˆ f β S , N I , Nˆ (t 1) , Pˆr (t ) , A(t ) , S I , T2 N ( i ) , T3N ( i )
Nˆ ( t )
S
I
β S
ˆ ˆ
v(t ) f β v , H d (t ) , N (t )
ˆ ˆ
v
( t )
N I
β Dq
β
Dˆ q (t ) f β Dq , vˆ(t ) , Nˆ ( t ) Dˆ
T2 A ( i ) q ( t )
2N ( i )
T
v
β
Gˆ ( t ) f Dˆ q ( t ) , Nˆ ( t )
Gˆ
(t )
Pr
Vˆt ( t ) f vˆ( t ) , Nˆ ( t ) Vˆ
T3A ( i ) β Lr t(t)
T β
N ( i )
3 S I ( PIm )
Pˆ f β , vˆ , Nˆ
r(t) Pr (t)
(t) Pˆ
r(t)
β ˆ Lˆr ( t )
SI ( PNb ) L
r ( t )
f β Lr , vˆ ( t) , Nˆ
( t)
(1) Estimating upper stem diameter at 2.59 m height intervals from stump height
Computation: upper stem diameter
Input Variables and Constants dˆ
f βT( PIm ) , h(t , j ) , Dˆ (t , j ) , Hˆ (t , j ) , Gˆ ( t ) f βT( PNb ) , h( t , j ) , Dˆ ( t , j ) , Hˆ ( t , j ) , Gˆ ( t )
Fixed (t , j )
2
Input
Variables Constants Hˆ ( t , j ) 0.3 Hˆ ( t , j ) 0.3
ˆ
where h( t , j ) =0.3, 0.3+1
100
,0.3+2
100
,..., H ( t , j )
Dˆ (t , j )
d
s
dp
ˆ
Hˆ if d d then increment sawlog count per tree ( ˆ
n ) by 1
(t, j) s ls ( t , j )
(t , j ) dt
Nˆ if dˆ d and dˆ = d then increment pulplog count per tree (nˆ
lp (t , j ) ) by 1
( t , j ) dm
( t , j ) s ( t , j ) p
Nˆ (t ) if dˆ d p and dˆ = d t then calculate residual tip volume per tree (vˆr (t , j ) )
βT( PIm )
(t , j ) (t , j )
Gˆ (t ) β
T( PNb )
ˆ
if d (t , j ) = d m then calculate merchantable stem volume per tree (vˆm (t , j ) )
ˆ
if d ( t , j ) = 0 then calculate total stem volume per tree (vˆt (t , j ) )
(2) Calculating diameter-class and stand-level estimates of the number of
sawlogs and pulplogs, residual tip volumes, and merchantable and total stem
volumes
J Output Variables
Nˆ ls (t , j ) Nˆ (t , j ) nˆls (t , j ) Nˆ ls (t ) Nˆ ls (t , j ) ˆ
j 5 N ls (t )
ˆ
Nˆ
J
N
lp (t )
lp (t , j ) N ( t , j ) nˆlp ( t , j ) N lp ( t ) Nˆ lp (t , j )
ˆ ˆ
ˆ
j 5
Vr ( t )
J ˆ
Vr ( t , j ) N ( t , j ) vˆr ( t , j ) Vt ( t ) Vˆr ( t , j )
ˆ ˆ ˆ Vm (t )
j 5 vˆ
m ( t , j )
J
Vm ( t , j ) N ( t , j ) vˆm ( t , j ) Vm ( t ) Vˆm (t , j )
ˆ ˆ ˆ '
ˆ Vˆ '
j 5
ˆ ) m(t )
J
V m ( t ) V t ( t
Vˆ '
Vˆt ( t , j ) Nˆ ( t , j ) vˆt ( t , j ) Vˆ ' Vˆt (t , j ) t (t )
t(t)
j 1
(d) Module D - Biomass and Carbon Estimation
s( t )
mˆ p( t , j ) mˆ s( t , j ) mˆ b( t , j ) mˆ f( t , j )
t( t , j )( PIm ) ( PIm ) ( PIm ) ( PIm ) ( PIm )
(t , j )
2
j 1
ˆ ˆ
mˆ p( t , j )( PNb ) f β M p ( PNb ) , D(t , j ) , H (t , j ) , H d (t ) , Pr (t )
ˆ ˆ J mˆ b( t , j )( PIm ) mˆ b( t , j )( PNb )
Output Variables
Input Variables and Constants Mˆ Nˆ Mˆ
b( t )
Fixed
ˆ ˆ
mˆ s( t , j ) ( PNb ) f β M s ( PNb ) , D(t , j ) , H (t , j ) , H d (t ) , Pr (t )
ˆ ˆ
j 1
(t , j )
2
p( t )
Mˆ
Input
Constants mˆ
b( t , j ) ( PNb )
f β M b ( PNb ) , Dˆ (t , j ) , Hˆ (t , j ) , Hˆ d (t ) , Pˆr (t ) J
Mˆ f Nˆ (t , j )
mˆ f( t , j )( PIm ) mˆ f (t , j )( PNb )
s( t )
Mˆ b( t )
( t ) j 1 2
Variables β
β m p ( PIm )
ˆ
mˆ f( t , j ) ( PNb ) f β M f ( PNb ) , D(t , j ) , H (t , j ) , H d (t ) , Pr (t )
ˆ ˆ ˆ
Mˆ
ˆ J mˆ t(t , j )( PIm ) mˆ t(t , j )( PNb ) f( t )
mˆ t( t , j ) ( PNb ) mˆ p( t , j )( PNb ) mˆ s( t , j ) ( PNb ) mˆ b( t , j ) ( PNb ) mˆ f( t , j ) ( PNb )
m
Mˆ t( ) Nˆ (t , j )
s ( PIm )
(t , j )
D β t
Mˆ t( t )
2
Hˆ mb ( PIm )
j 1
( t , j ) β ˆ ˆ ˆ
C p( t )
Nˆ m f ( PIm ) cˆ ˆ c c
0.5 mˆ p( t , j )
J
ˆ
C p(t ) N (t , j )
p ( , ) p( , )
( PIm ) ( PNb )
t j t j
(t , j ) β m p( t , j )( PIm ) ˆ
Cs( t )
( PIm )
2
Hˆ d (t )
p ( PNb )
cˆ 0.5 mˆ s( t , j ) j 1
β ms ( PNb ) ˆ
s( t , j )( PIm ) ( PIm )
cˆs( t , j )( PIm ) cˆs( t , j )( PNb ) Cb( t )
Pˆr (t ) cˆ ˆ Cˆ Nˆ
J
0.5
s( t )
m
β mb ( PNb ) b( t , j )( PIm ) b( t , j )
( PIm ) (t , j )
2 Cˆ f
cˆ 0.5 mˆ
j 1
(t )
β m f ( PNb ) f( t , j )( PIm ) f( t , j )
Cˆt
cˆb( t , j )( PIm ) cˆb( t , j )( PNb )
( PIm )
J
cˆt Cˆ Nˆ
b( t )
ˆ
c cˆs( t , j ) cˆb( t , j ) cˆ f( t , j ) (t )
( t , j )( PIm ) (t , j )
p
( t , j )( PIm ) ( PIm ) ( PIm ) ( PIm )
j 1
2
cˆ p( t , j ) 0.5 mˆ p( t , j )
( PNb ) ( PNb )
J cˆ f( t , j )( PIm ) cˆ f( t , j )( PNb )
cˆs 0.5 mˆ s( t , j ) Cˆ f Nˆ (t , j )
( t , j )( PNb ) ( PNb )
( t ) j 1 2
cˆb( t , j )
0.5 mˆ b( t , j )
( PNb ) ( PNb )
J cˆt( t , j )( PIm ) cˆt( t , j )( PNb )
cˆ f 0.5 mˆ Cˆt( t ) Nˆ (t , j )
( t , j )( PNb )
f (t , j )
( PNb ) 2
j 1
cˆt ˆ
c cˆs( t , j ) cˆb( t , j ) cˆ f( t , j )
( t , j )( PNb ) p ( t , j )( PNb ) ( PNb ) ( PNb ) ( PNb )
(e) Module E - Product and Value Estimation
β pl ( s )
j 5 ˆ
Vl ( s ) ( t )
( PIm )
ˆ J
Pl ( s ) ( t ) N (t , j ) pˆ l ( s ) ( t , j )
Input β ˆ
f β pl ( s ) , Dˆ (t , j ) , Hˆ (t , j ) f β pl ( s ) , Dˆ (t , j ) , Hˆ (t , j ) , Hˆ d (t ) , Pˆr (t )
ˆ
Pl ( s ) ( t )
pt ( s )
Variables pˆ j 5
( PIm )
( PIm ) ( PNb )
Dˆ β v l ( s )(t , j ) 2 ˆ J
ˆ
Pt ( s )
Pt ( s ) ( t ) N (t , j ) pˆ t ( s ) ( t , j )
( PIm ) ˆ
( )
l r
(t , j )
(t )
f β Pt ( s ) , Dˆ (t , j ) , Hˆ (t , j ) f β Pt ( s ) , Dˆ (t , j ) , Hˆ ( t , j ) , Hˆ d ( t ) , Pˆr ( t ) j 5 ˆ
Hˆ (t , j ) β pl ( r ) ( PIm ) Pc ( s ) ( t )
pˆ t ( s ) (t , j )
( PIm ) ( PNb )
ˆ
J
Nˆ (t , j ) β 2 Pc ( s ) ( t ) Nˆ (t , j ) pˆ c ( s )
(t , j ) Vˆc ( r )
t ( r ) ( PIm ) pˆ c ( s ) pˆ t ( s ) pˆ l ( s )
p
j 5
(t )
ˆ
Pr (t ) β (t, j) (t, j) (t, j)
ˆ J
Vˆ
Vc ( r ) ( t ) N ( t , j ) vˆc ( r ) ( t , j )
vl ( s )( PNb )
ˆ
f β vl( r ) , Dˆ ( t , j ) , Hˆ ( t , j ) f β vl( r ) , Dˆ ( t , j ) , Hˆ ( t , j ) , Hˆ d ( t ) , Pˆr ( t ) l (r )(t )
Hˆ d (t ) β p j 5
Pˆ
l ( s ) ( PNb ) vˆl ( r ) ( t , j )
( PIm ) ( PNb )
2 J
l (r )(t )
vˆm (t , j )
β pt ( s ) ( PNb ) vˆ Vˆl ( r ) ( t ) Nˆ ( t , j ) vˆl ( r ) ( t , j ) ˆ
c( r ) vˆm ( t , j ) vˆl ( r ) j 5
P
t (r )(t )
β (t, j) (t, j)
ˆ
J
Pˆl ( r ) Nˆ ( t , j ) pˆ l ( r )
vl ( r )
( PNb )
f β pl ( r ) , Dˆ (t , j ) , Hˆ (t , j ) f β pl ( r ) , Dˆ (t , j ) , Hˆ (t , j ) , Hˆ d (t ) , Pˆr (t ) Pc ( r ) ( t )
β ( t ) j 5 (t , j )
pˆ l ( r )
( PIm ) ( PNb )
pl ( r )
( PNb )
(t, j)
2 J
Pˆt ( r ) Nˆ ( t , j ) pˆ t ( r )
β
pt ( r ) ( PNb ) f β pt ( r ) , Dˆ (t , j ) , Hˆ (t , j ) f β pt ( r ) , Dˆ (t , j ) , Hˆ (t , j ) , Hˆ d (t ) , Pˆr (t ) ( t ) j 5 (t , j )
ˆ
( PIm ) ( PNb )
p Pˆ
J
c ( r ) ( t )
t ( r ) (t , j )
2 Nˆ ( t , j ) pˆ c ( r )
(t , j )
pˆ pˆ t ( r ) pˆ l ( r ) j 5
c ( r ) ( t , j ) (t , j ) (t , j )
(f) Module F - Fibre Attribute Estimation
Estimating wood density and mean maximum branch diameter per tree
Input Variables and Constants
Input
Variables
Fixed
Computations
Constants
Dˆ (t , j )
β wD
wˆ D( t , j )
f β wD , Dˆ (t , j ) , Hˆ (t , j ) , Hˆ d (t ) , Pˆr (t ) f β wD( PNb ) , Dˆ (t , j ) , Hˆ (t , j ) , Hˆ d (t ) , Pˆr (t )
( PIm )
ˆ ( PIm )
H 2
( t , j ) β
Nˆ
(t , j )
bD( PIm )
β
ˆ
bD (t , j )
f βbD , Dˆ (t , j ) , Hˆ (t , j ) , Hˆ d (t ) , Pˆr (t )
( PIm )
f β bD( PNb ) , Dˆ (t , j ) , Hˆ (t , j ) , Hˆ d (t ) , Pˆr (t )
wD( PNb )
Hˆ d (t )
2
βb
ˆ D( PNb )
P r ( t )
Calculating mean wood density and
mean maximum branch diameter per
stand
J Output Variables
ˆ
j 5
ˆ ˆ
wD ( t , j ) N ( t , j )
ˆ
WD ( t ) J
if D( t , j ) 10 WD ( t )
ˆ
Nˆ ( t , j )
B
j 5 D (t )
ˆ
J
bD (t , j ) Nˆ (t , j )
Bˆ D (t )
j 8
if D(t , j ) 16
J
Nˆ ( t , j )
j 8
Figure 2. Schematic illustration of the computational framework utilized in the modular-based SSDMM: (a) Module A - Dynamic SDMD; (b) Module
B - Diameter and Height Recovery; (c) Module C - Taper Analysis and Log Estimation; (d) Module D - Biomass and Carbon Estimation; (e) Module E
- Product and Value Estimation; and (f) Module F - Fibre Attribute Estimation. Refer to Table 9 for variable definitions and additional computational
details.
Table 9. Variable definitions and associated computational details associated
with the SSDMM are schematically illustrated in Figure 2
associated vector of parameter estimates ( βDq ˆl where l 0,1,2 ; Table 3).
Gˆ ( t ) Predicted basal area (m2/ha) at time t: Gˆ (t ) 0.00007854 Dˆ q2(t ) Nˆ (t )
Vˆt (t ) Predicted total volume (m3/ha) at time t: Vˆt ( t ) 1000 vˆ(t ) Nˆ (t )
Pˆr ( t ) Predicted relative density index (%/100) at time t. Estimated employing Eq. (2) where β Pr denotes the associated vector of
parameter estimates ( β Pr ˆ 0 ˆ1 ; Table 3).
Lˆr ( t ) Predicted mean live crown ratio (%) at time t. Estimated employing Eq. (8) where β Lr denotes the associated vector of
Hˆ ( t , j ) Predicted height (m) of the trees within the jth two-centimetre-wide diameter class at time t. Estimated employing Eq. (17)
where β H denotes the associated vector of parameter estimates ( β H ˆl where l 0,...,7 ; Table 4).
volume summation between the height at the top of the last recovered pulp or saw log and the height (h(j,t)) at which the dˆ( t , j )
value is equivalent to the specified merchantability limit for merchantable stems (minimum inside bark diameter of 10.0 cm
(dm)).
vˆm(t , j ) Predicted merchantable stem volume per tree per tree (m3/tree) within the jth merchantable-sized diameter class at time t.
Estimated employing Eqs. (18a) and (18b) according to specified dimensional requirements for merchantable volume via
numeric integration techniques: sectional volume summation between a stump height of 0.30 m and the height (h(j,t)) at which
the dˆ value is equivalent to the specified merchantability minimum inside bark diameter (dm; 10.0 cm).
(t , j )
vˆt (t , j ) Predicted total stem volume per tree per tree (m3/tree) within the jth diameter class at time t. Estimated employing Eqs.
(18a) and (18b): (1) numeric integration via summation of sectional volumes between a stump height of 0.30 m and the height
(h(j,t)) at which a zero dˆ value first occurs; (2) calculating the volume of the 0.30 m stump; and (3) summing the both
(t , j )
Nˆ lp (t , j ) Predicted number of pulplogs within the jth merchantable-sized diameter class (pulplogs/jth diameter class) at time t.
Vˆr (t , j ) Predicted residual log tip volume within the jth merchantable-sized diameter class (m3/jth diameter class) at time t.
Vˆr ( t ) Predicted residual log tip volume per unit area (m3/ha) at time t.
Vˆm ( t , j ) Predicted merchantable stem volume within the jth merchantable-sized diameter class (m3/jth diameter class) at time t.
Vˆm' ( t ) Approximated merchantable volume per unit area (m3/ha) at time t (estimate derived from the taper equations).
Vˆt ( t , j ) Predicted total stem volume within the jth diameter class (m3/jth diameter class) at time t.
Vˆt '(t ) Approximated total stem volume per unit area (m3/ha) at time t (estimate derived from the taper equations).
k p , s , b, f , t
βmk , k p, s, b, f denotes the parameter estimate vectors specific to the kth component and nth species
(n)
( β mk ˆl where l 0,...,7 ; Table 6). Total mass calculated as the arithmetic sum of the individual component masses.
(n)
cˆk( t , j ) Predicted bark (periderm; k=p), stem (k=s), branch (k=b), foliage (k=f) and total (k=t) carbon equivalents per tree (g/tree)
(n)
within the jth diameter class at time t for the nth species. Calculated as a fixed proportion of the component mass estimates.
k p, s, b, f , t
Mˆ Predicted bark (k=p), stem (k=s), branch (k=b), foliage (k=f) and total oven-dry mass (k=t) per unit area (g/ha) at time t.
k( t )
k p , s , b, f , t
Cˆ k( t ) Predicted bark (k=p), stem (k=s), branch (k=b), foliage (k=f) and total carbon (k=t) per unit area (g/ha) at time t.
k p , s , b, f , t
k s, r
(20b) where βvl ( k ) ˆl ( k ) where l = 0,1,2 and βvl ( k ) ˆl ( k ) where l = 0,...,7 denotes the black spruce and jack pine
( PIm ) ( PNb )
parameter estimate vectors for lumber volume, respectively, specific to the kth processing protocol (Table 7).
vˆc ( k ) Predicted chip volume recovered by a stud (k=s) and randomized length (k=r) sawmill processing protocol per tree
(dm3/tree) within the jth merchantable-sized diameter class at time t for the nth species. Note, calculated as the arithmetic
( t , j )( n )
k s, r difference between the taper-based merchantable volume per tree estimate and lumber volume per tree estimate.
pˆ l ( k ) Predicted value of dimensional lumber recovered by a stud (k=s) and randomized length (k=r) sawmill processing protocol
per tree (CAN$(2002)/tree) within the jth merchantable-sized diameter class at time t for the nth species. Estimated employing
( t , j )( n )
k s, r
Eqs. (20a) and (20b) where β pl ( k ) ˆl ( k ) where l = 0,1,2 and β pl ( k ) ˆl ( k ) where l = 0,...,7 denotes the black spruce and
( PIm ) ( PNb )
jack pine parameter estimate vectors for lumber value, respectively, specific to the kth processing protocol (Table 7).
pˆ t ( k ) Predicted value of lumber and chip products recovered by a stud (k=s) and randomized length (k=r) sawmill processing
protocol per tree (CAN$(2002)/tree) within the jth merchantable-sized diameter class at time t for the nth species. Estimated
( t , j )( n )
k s, r
employing Eqs. (20a) and (20b) where β pt ( k ) ˆl ( k ) where l = 0,1,2 and β pt ( k ) ˆl ( k ) where l = 0,...,7 denotes the
( PIm ) ( PNb )
black spruce and jack pine parameter estimate vectors for total value, respectively, specific to the kth processing protocol (Table
7).
pˆ c ( k ) Predicted mean value of chips recovered by a stud (k=s) and randomized length (k=r) sawmill processing protocol per tree
(CAN$(2002)/tree) within the jth merchantable-sized diameter class at time t for the nth species. Note, calculated as the
( t , j )( n )
k s, r arithmetic difference between the total and lumber mean value estimates.
Vˆc ( k ) Predicted chip volume recovered by a stud mill (k=s) and randomized length (k=r) sawmill processing protocol per stand for
(t )
all merchantable-sized trees at time t (dm3/ha).
k s, r
Vˆl ( k ) Predicted lumber volume recovered by a stud mill (k=s) and randomized length (k=r) sawmill processing protocol per stand
(t )
for all merchantable-sized trees at time t (dm3/ha).
k s, r
Pˆl ( k ) Predicted value of dimensional lumber recovered by a stud mill (k=s) and randomized length (k=r) sawmill processing
(t )
protocol per stand for all merchantable-sized trees at time t (CAN$(2002)/ha).
k s, r
Pˆt ( k ) Predicted value of all products (lumber and chip) recovered by a stud mill (k=s) and randomized length (k=r) sawmill
(t )
processing protocol per stand for all merchantable-sized trees at time t (CAN$(2002)/ha).
k s, r
Pˆc ( k ) Predicted value of chips recovered by a stud mill (k=s) and randomized length (k=r) sawmill processing protocol per stand
(t )
for all merchantable-sized trees at time t (CAN$(2002)/ha).
k s, r
Module F - Fibre Attribute Estimation (Figure 2(f))
wˆ D( t , j ) Predicted stem cross-sectional area-weighted mean wood density per tree within the jth diameter class at time t (g/cm3) for
the nth species. Estimated employing Eq. (21) where βwD denotes the parameter estimate vector specific to the nth species
(n)
(n)
Predicted mean maximum branch diameter (cm) within the first 4.9 m sawlog per tree within the jth sawlog-sized diameter
bˆD (i , j )( n )
class Dˆ
16 at time t. Estimated employing Eq. (22) where β denotes the parameter estimate vector specific to the nth
(t , j ) bD( n )
Wˆ D ( t )
Predicted diameter-class-weighted mean wood density within the merchantable-sized tree population Dˆ (t , j ) 10 at time t
3
(g/cm ).
Bˆ D ( t )
Predicted diameter-class-weighted mean maximum branch diameter within the sawlog-sized tree population Dˆ (t , j ) 16 at
time t (cm).
Development and Utility of an Ecological-based Decision-Support System … 155
A standardized set of stand-level performance indices were derived from the model‘s
output in order to simplify the decision-making process in terms of comparing alterative
density management regimes. These indices reflect overall productivity, type of products
produced, economic efficiency, degree of optimal site occupancy, structural stability, and
fibre quality attributes, over the rotation. Indices representing merchantable volume
production (mean annual merchantable volume increment; RMAI (Eq. (24)), biomass
accumulation rate (mean annual biomass increment; RBMI (Eq. (25)), and carbon sequestration
potential (mean annual carbon increment; RCAI (Eq. (26)) were used to quantify overall
productivity for a given regime.
K
RMAI Vm Vm ( k ) A(T ) (24)
k 1
where Vm is the standing merchantable volume (m3/ha) at rotation (A(T)) and Vm( k ) is the
merchantable volume removed during the kth thinning entry (k = 1,…,K; K = 4),
K
RBMI M t M t ( k ) A(T ) (25)
k 1
where M t is the standing total aboveground biomass (t/ha) at rotation and M t ( k ) is the total
aboveground biomass (t/ha) removed during the kth thinning entry (k = 1,…,K; K = 4),
K
RCAI Ct Ct ( k ) A(T ) (26)
k 1
where C t is the standing total aboveground carbon (t/ha) at rotation and Ct ( k ) is the total
aboveground carbon (t/ha) removed during the kth thinning entry (k = 1,…,K; K = 4).
Relative indices reflecting the type and quality of logs produced (percentage of sawlogs
produced; RSL; (Eq. (27)) and the resultant end-products manufactured (percentage of lumber
volume recovered by sawmill type; RLV(m) (Eq. (28)) were used to differentiate each regime in
terms of its potential to produce commercial-grade end-products.
K
Nls Nls ( k )
RSL 100 k 1 (27)
K
K
Nls Nls ( k ) N lp N lp ( k )
k 1 k 1
156 Peter F. Newton
where N ls and N lp are the total number of sawlogs (logs/ha) and pulplogs (logs/ha) at
rotation, respectively, and Nls ( k ) and Nlp ( k ) are the total number of sawlogs (logs/ha) and
pulplogs (logs/ha) removed during the kth thinning entry (k = 1,…,K; K = 4), respectively,
K
Vl ( m) Vl ( k , m)
RLV ( m ) 100 k 1 (28)
K
K
Vl ( m ) Vl ( k ,m ) Vc ( m ) Vc ( k ,m )
k 1 k 1
where Vl ( m) and Vc ( m ) are the lumber and chip volumes (m3/ha) recovered employing the mth
sawmill processing protocol (m = 1 (stud mill) or 2 (random length mill)) from the
merchantable-sized trees at rotation, respectively, and Vl ( k ,m ) and Vc ( k ,m) are the lumber and
chip volumes (m3/ha) recovered employing the mth sawmill processing protocol from the
merchantable-sized trees removed during the kth thinning entry (k = 1,…,K; K = 4),
respectively.
Economic efficiency was measured by the land expectation value (i.e., the maximum an
investor could pay for bare land to achieve a specified rate of return (discount rate)) of the
manipulated regimes, relative to the control regime (E(m); Eq. (29)).
LTE ( m ) LCE ( m )
E( m ) 100 where (29)
L C
E ( m )
K
PtT(m ) (1 I r ) A(T ) CFE (1 I r ) A(T ) CFT T (k ) (1 I r ) A(T ) A( k )
k 1
K
K T
k 1
Pt (k ,m ) (1 I r ) 1 I r
A( k )
A( T ) A( k )
CV T (k ) (1 I r )
T A( T ) A( k )
CVT H
LT
k 1
E (m )
(1 Dr ) (T ) 1
A
T Ys 2002
Pt ( m ) Pt ( m ) (1 I r )
T Ys 2002
Pt ( k ,m ) Pt ( k ,m ) (1 I r )
where CFT T ( k ) cFT T ( k ) (1 I r ) ( k )
A
T
CV T ( k ) cV T ( k ) (1 I r )
T A( k )
Vˆm ( k )
CVT H cVT H (1 I r ) A( T ) Vˆm (T )
C
Pt C( m) (1 I r )
A( T )
CFE (1 I r )
A( T )
CVC H where P C
t ( m) Pt ( m) (1 I r )Ys 2002
C
L
Vˆm (T )
E ( m)
(1 Dr ) 1 CV H cV H (1 I r )
A( T ) C A( T )
Development and Utility of an Ecological-based Decision-Support System … 157
where LE ( m ) and LE ( m ) are the land expectation values at rotation attained employing the
T C
mth sawmill processing protocol within the density manipulated and control stands,
respectively, Pt ( m ) and Pt ( m ) are the inflation-adjusted (to the year of simulation; Ys) total
T C
product values ($/ha) recovered employing the mth sawmill processing protocol from the
merchantable-sized trees within the density manipulated and control stands at rotation,
respectively, Pt ( k , m ) is the inflation-adjusted total product value ($/ha) recovered employing
T
the mth sawmill processing protocol from the merchantable-sized trees removed during the
kth thinning entry (k = 1,…,K; K = 4), CFE is the fixed cost ($/ha) incurred at the time of stand
establishment (e.g., regeneration assessment or vegetation management expenses), cF T ( k )
T
and CF T ( k ) are the inflation unadjusted and adjusted fixed costs ($/ha) incurred during the
T
kth thinning entry, respectively (e.g., logistical costs such as those associated with
transporting thinning equipment), cVC H and CVC H are the inflation unadjusted and adjusted
variable costs (dollars per cubic metre of merchantable volume harvested ($/m3)),
respectively, associated with crown charges (stumpage and renewal fees), harvesting,
transportation, processing and manufacturing at the time harvest within the control stand,
cVT H and CVT H are the inflation unadjusted and adjusted variable costs ($/m3), respectively,
at the time of harvest within the treated stand, cV T ( k ) and CV T ( k ) are the inflation
T T
unadjusted and adjusted variable costs ($/m3), respectively, at the time of the kth thinning
entry within the treated stand, Ir and Dr are the inflation and discount rate, respectively, and
A( k ) and A(T) are the stand ages at the time of the kth thinning entry and at rotation,
respectively.
In order to evaluate the regimes to the degree to which each optimally occupied the site
during the rotation, the percentage of years in which the size-density trajectory was within the
optimal density management window, was calculated (SO; Eq. (30)).
Y
SO 100 O (30)
YN
where YO is the number of years in which the size-density trajectory was within the
conceptual optimal relative density management zone as delineated by lower and upper
relative density index thresholds of 0.32 and 0.45, respectively, and YN is the rotation length
in years. The weighted mean height/diameter ratio for trees within the dominant crown
classes (SS; Eq. (31)) was used to differentiate the regimes in terms of stand stability.
J H (t , j )
N ( t , j )
1 j 1 100 D(t , j )
SS J (31)
T
if D(t , j ) D80
T t i
N (t , j )
j 1
158 Peter F. Newton
where D80 is the 80th percentile of the recovered diameter frequency distribution as
calculated from the Weibull (1951) scale and shape parameters (Bailey and Dell, 1973):
D80 bˆ log e 0.20 . Similarly, the weighted mean wood density for trees within the
1/ cˆ
merchantable tree population ( WD ; Eq. (32)) and the weighted mean maximum branch
diameter ( BD ; Eq. (33)) within the sawlog-sized tree population were used as indices of
wood and log quality, respectively.
J
T D (t , j ) (t , j )
w N
1
WD
j 5
(32)
T t 1 J
N (t , j )
j 5
where wD (t , j ) is the stem cross-sectional area-weighted mean wood density at time t of trees
within the jth merchantable size diameter class D(t , j ) 10 .
J
T D (t , j ) (t , j )
b N
1
BD j J
8
(33)
T t i
N (t , j )
j 8
where bD ( t , j ) is the mean maximum branch diameter within the first 4.9 m sawlog at time t
within the jth sawlog-sized diameter-class D(t , j ) 16 . Note, in the case of the thinned
stands, WD and BD values are calculated for the period since the last treatment, exclusively.
interest and discount rate information. Essentially, the program carries out all the required
computations as described in Figure 2, calculates the performance indices, and presents the
results in both graphical and tabular formats.
Although the computational requirements are mitigated with the development of the
software program, designing the optimal density regime for a given objective is still a major
challenge considering the number of decision variables involved. For example, a silviculturist
must decide which sites should be treated, the number, type, intensity (removable densities)
and timing of thinning treatments, when to conduct the final harvest (rotation length), and
what economic values are most appropriate (fixed and variable costs estimates, and interest
and discount rates). Furthermore, the candidate regimes must comply with the local statutory
framework which may introduce a broad array of additional conditions that must be met
before treatments can be implemented. Thus in order to exemplify the complexity of density
management decision-making and illustrate the potential utility of the SSDMM, the software
was used to determine and evaluate the consequences of implementing CT treatments within
density-stressed mixed stands situated on sites of moderate productivity.
In this example, CT is used as a stand improvement practice in order to realize a diverse
set of stand-level objectives. These included (1) avoidance of density-dependent mortality
within the merchantable-sized classes during the later stages of stand development so that
potentially valuable crop trees are retained on the site until final harvest, (2) provision of
interim fibre supplies via the mid-rotational partial harvests (thinning yields), (3) reduction in
vertical and horizontal structure heterogeneity by increasing spatial pattern uniformity and
decreasing size variation within the crop tree population thus reducing variable costs
associated with harvesting, transportation and manufacturing, and (4) enhancing carbon
sequestration potential by increasing biomass production and reducing the generation of
abiotic masses.
Preferable, CT density management regimes are those which increased mean tree size
without incurring declines in merchantable volume production, do not unacceptably increase
the risk of volume losses to wind, snow, insects, and disease, and minimize the occurrence of
density-dependent mortality within the merchantable-sized diameter classes. Although
legislative and operational constraints placed on CT treatments vary by stand type and
jurisdiction, the ones proposed for coniferous stands within the central portion of the
Canadian Boreal Forest Region are used in this example (McKinnon et al., 2006): (1) CT
treatments should be implemented when density-dependent mortality is occurring or
imminent within the merchantable-sized classes; (2) CT should occur within stands where the
mean live crown ratio exceeds 35% and the stand basal area exceeds 25 m2/ha; and (3) CT
treatments should reduce basal areas by no more than 30-35% per entry. Furthermore, if the
goal is to enable stands to achieve an optimal level of production before thinning commences,
stands should be allowed to attain the upper threshold relative density value of at least 0.45,
before treatment.
Thus within the context of implementing CT treatments within density-stressed mixed
stands, a control regime (untreated) consisting of a stand with an initial density of 5000
stems/ha (NI) which naturally established on a moderately good site quality (SI = 17)
following forest harvesting was contrasted with two stands with identical initial conditions,
but subject to the following CT treatments. The first of two CT treatments was equivalent in
both stands: removing 1500 stems/ha from stands which had achieved all of the above CT
criteria in addition to having attained merchantable status (i.e., having a minimum quadratic
160 Peter F. Newton
mean diameter of 10 cm). Similarly, once the treated stands had re-achieved the criteria for
CT treatments, they were subsequently thinned again. Specifically, at a stand age of 65 yr,
30% of the basal area in the first CT stand, and 35% of the basal area in the second CT stand,
was removed. The control and treated stands were denoted Regime 1, 2 and 3 where Regime
3 was the stand that received the heaviest second CT treatment. The rotation age was set at 85
yr for all 3 regimes. The economic and operability variables were specified as follows: (1) a
cost of $100/ha were applied to all 3 regimes to cover the expense of a regeneration
assessment survey; (2) the fixed costs associated with equipment transport and basic
administration for each CT treatment was set at $100/ha; (3) the variable costs associated with
stumpage, renewal, harvesting, transportation and manufacturing in units of dollars per cubic
metre of merchantable volume harvested at rotation, was set at $100/m3 for the control
regime, and $60/m3 for the regimes involving the CT treatments; note, this cost differential is
principally due to the reduction in harvesting and manufacturing costs due to increased spatial
pattern uniformity and decreased piece-size variation commonly attributed to thinning
treatments; (4) similar to (3), the variable costs associated with the first and second thinning
yields were set at $100/m3 and $70/m3, respectively; (5) interest and discount rates were set at
2 and 4%, respectively; (6) a quadratic mean diameter of 16 cm was set as an operability
target; and (7) the simulation year was set to 2011.
Figure 3 illustrates the resultant mean volume-density trajectories for each regime within
the context of the traditional SDMD graphic. The stands at the time of the first thinning (45
yr) were at the midway point between the 14 and 16 m dominant height isolines, slightly
above the 10 cm quadratic mean diameter isoline, midway between the 0.6 and 0.7 relative
density isolines, and just above the 40% mean live crown ratio isoline. The interpolated
values for these and other values were as follows: mean dominant height of 15.2 m, quadratic
mean diameter of 10.3 cm, relative density of 0.64, mean live crown ratio of 41%, mean
volume of 52 dm3, stand density of 3575 stems/ha, and basal area of 29.7 m2/ha. Thus the
stands had met the minimum conditions for CT.
The first thinning reduced density stress levels by approximately 30% (relative density
indices declining from 0.64 to 0.45) and placed the residual stands exactly on the upper
threshold of the optimal density management window (i.e., 0.45 relative density index). The
treatment removed approximately 31% of the basal area and resulted in a merchantable
volume recovery of 33 m3/ha. At the time of the second thinning (65 yr), both of the treated
stands were almost on the 19 m dominant height isoline, midway between the 14 and 16 cm
quadratic mean diameter isolines, just above the 0.7 relative density isoline, and
approximately intersecting the 35% mean live crown ratio isoline, which is the minimum
value allowed for implementing a CT treatment. At this stage, the interpolated values for
these and other values were as follows: mean dominant height of 18.9 m, quadratic mean
diameter of 15.1 cm, relative density of 0.71, mean live crown ratio of 36%, mean volume of
138 dm3, stand density of 1750 stems/ha, and basal area of 31.4 m2/ha. For Regimes 2 and 3,
the second CT treatment removed 30% and 35% of the basal area, respectively, which
resulted in a merchantable volume recovery of 52 m3/ha and 61 m3/ha, respectively. Table 10
provides a complete list of the rotational and thinning yield estimates by regime. The derived
stand-level performance indices are given in Table 11.
Development and Utility of an Ecological-based Decision-Support System … 161
Figure 3. Graphical illustration of the dynamic SDMD for black spruce and jack pine mixtures. Note
the following principal components of the SDMD: (1) isolines for mean dominant height (Hd; 6-24 m
by 2 m intervals), quadratic mean diameter (Dq; 4-24 cm by 2 cm intervals), mean live crown ratio (Lr;
35, 40, 50,…, 80%), relative density index (Pr; 0.1-1.0 by 0.1 intervals); (2) self-thinning rule at a Pr =
1.0 (solid diagonal line delineating the outer boundary of the size-density region); (3) lower and upper
Pr values delineating the optimal density management window (Dm; 0.32 ≤ Pr ≤ 0.45); (4) crown
closure line (solid diagonal line delineating the inner boundary of the size-density region); and (5)
expected 85 yr size-density trajectories with 1 yr intervals denoted for 3 user-specified density
management regimes for stands situated on sites of moderate productivity (SI = 17). Regime 1 consisted
of a regeneration density of 5000 stems/ha with no thinning. Regime 2 consisted of a regeneration
density of 5000 stems/ha with 2 commercial thinnings: one at 45 yr in which 1500 stems/ha were
removed and a subsequent one at 65 yr in which 700 stems/ha were removed. Similarly, Regime 3
consisted of a regeneration density of 5000 stems/ha with 2 commercial thinnings: one at 45 yr in
which 1500 stems/ha were removed and a subsequent one at 65 yr in which 805 stems/ha were
removed. Source: SSDMM algorithm.
Evaluating the regimes in terms of the realization of the multiple objectives under
consideration, it is evident that commercial thinning is a viable treatment when managing
mixed stands. Density-dependent mortality rates within the thinned stands were considerably
less than those within the unthinned stand during the time from the first treatment to the time
of the second treatment: total densities declined by 1049 stems/ha within the control stand
versus 325 stems/ha within the thinned stands. Similar trends were observed from the time of
the second thinning to rotation age: total densities declined by 712 stems/ha within the control
stand versus 60 stems/ha for Regime 2, and 17 stems/ha for Regime 3.
162 Peter F. Newton
Attributea Regimeb
(t = T) 1 2 3
A(t ) (yr) 85 85 85
Gˆ ( t ) (m /ha)2
43 28 27
a - As defined in Table 9.
Development and Utility of an Ecological-based Decision-Support System … 163
b - Regime 1 - NI = 5000 (control); Regime 2 - NI = 5000 with T2 A (1) = 45/ T2 N (1) = 1500 + T2 A ( 2 ) =
65/ T2N ( 2 ) = 700; Regime 3 - NI = 5000 with T3A (1) = 45/ T3N (1) = 1500 + T3 A ( 2 ) = 65/ T3N ( 2) = 805.
Table 11. Stand-level performance indices for black spruce and jack pine mixtures
subjected to mid-rotation commercial thinning treatments
Indexa Regimeb
1 2 3
RMAI (m3/ha/yr) 3.7 3.5 3.5
RBMI (t/ha/yr) 3.2 3.5 3.6
RCAI (t/ha/yr) 1.6 1.8 1.8
RSL (%) 46 37 37
RLV(s) (%) 59 58 58
RLV(r) (%) 68 68 69
EP(s) (%) - 15 16
EP(r) (%) - 5 5
SO (%) 7 8 8
SS (m/m) 102 95 94
WD (g/cm3) 0.45 0.47 0.47
BD (cm) 2.75 2.79 2.79
a - As defined in the text.
b- As defined in Table 10.
In terms of site occupancy, the thinned stands had fully reoccupied their sites by rotation
age as measured by their relative densities (0.65 and 0.62 for Regimes 2 and 3, respectively;
Table 10). Crop trees within the thinned stands had fully adjusted to their newly given space
from the second thinning just prior to rotation age (i.e., stand ages of 80 and 83 yr for Regime
2 and 3, respectively). Based on a 16 cm quadratic mean diameter threshold, the thinned
stands achieved operability status 10 years earlier than the control regime (66 versus 76 yr).
Although merchantable volumetric yields were slightly less for the thinned stands, biomass
productivity and carbon sequestration potential were greater (Table 11). The percentage of
lumber volume recovered over the rotation was approximately equivalent among the regimes,
irrespective of sawmill-type. The economic-based metrics suggested that CT resulted in an
approximate 10% gain in efficiency relative to the control regime. This differential is partially
due to the time-sensitive nature of the net revenues recovered from the end-products obtained
from the CT treatments. The lower costs of harvesting, transportation and manufacturing also
contributed to this result. The duration of optimal site occupancy did not vary much among
the regimes and with the exception of the years immediately following the CT treatments
(response delay period), all the regimes had fully occupied their sites once they had attained
initial crown closure status. Stand stability and branch diameters increased slightly with
thinning whereas mean wood density decreased slightly. Although these results present CT in
a positive light in terms of capturing expected mortality, increasing biomass productivity and
carbon sequestration potential, improving economic efficiency and enhancing stand stability,
the results are dependent on the specified regime chosen and the economic assumptions
employed. Apart from the specific applicability of this example, the demonstration clearly
164 Peter F. Newton
illustrates the utility of the SSDMM in terms of managing for multiple objectives within the
black spruce and jack pine mixed stand-type.
A large number of relationships derived from applied ecology, plant population biology
and forest science, are used in the development of SDMDs (Drew and Flewelling, 1977; Jack
and Long, 1996; Newton, 1997). These include the reciprocal equations of the competition–
density and yield–density effect (Kira, 1953; Shinozaki and Kira, 1956), self-thinning rule
(Yoda et al., 1963), relative density indices (Ando 1962, Drew and Flewelling, 1979), and
forest production theories (Langsaeter, 1941; Drew and Flewelling, 1979). The integration of
these biologically-based constructs within the modular-based SSDMM provides an ecological
foundation for density management decision-making.
One of the principal relationships employed is the self-thinning rule which is used to
represent the asymptotic size-density relationship within stands undergoing density-dependent
mortality. Quantitatively, the power exponent of this relationship for black spruce and jack
pine mixtures was found to be -1.2 which is significantly (p ≤ 0.05) different from the
theoretical expected value of -1.5, as postulated under the rule‘s traditional geometric
deviation (Yoda et al., 1963). However, the self-thinning exponent is much closer to the value
of -1.3 which was proposed for a wide range of tree species based on a mechanistic
reformation (Enquist et al., 1998) employing the universal scaling law (West et al., 1997).
This derivation is applicable to plant species which have a fractal-like biological resource
distribution network (West et al., 1997). Assuming that (1) individuals compete for spatially
limited resources, (2) resource use per individual Q scales with their mass (m) according
3/ 4
to Q m , and (3) and growth continues until all available resources have been utilized,
the maximum number of plants which can be supported per unit area (Nmax) can be related to
the rate of resource supply per unit area (R) and the average rate of resource use per
individual Q , according to the relationship, R Nmax Q N max m3/4 . As a population
reaches full site occupancy and the rate of resource use and the rate of resource supply
approaches equivalency within a given environment, R becomes a constant. Hence,
Nmax m3/4 , or more generally, m 0 Nmax
4/3
where 0 is a constant of proportionality.
Therefore by extension, self-thinning black spruce and jack pine mixed stands may be at a
stationary equilibrium condition in which the rate of resource supply is equivalent to the rate
of resource.
Self-thinning deviations proposed by Yoda et al. (1963) and Enquist et al. (1999) assume
that asymmetric competition for finite resources is the principal determinate underlying
density-dependent mortality within plant populations. Although this assumption has been
widely supported by numerous studies, other factors may also be at play within density-
stressed coniferous tree populations. Physical competition for space in which neigbhouring
crowns collide during high wind events commonly results in spatially noncontiguous
canopies due to the loss of branches and associated leaf area (Rudnicki et al., 2001). Over
time, such wind-induced crown abrasions and resultant loss of foliar mass for trees within the
Development and Utility of an Ecological-based Decision-Support System … 165
lower size classes, may produce a mortality pattern analogous to that observed for a light-
based dominance-suppression competitive relationship (i.e., resource pre-emption
competition process in which the smaller individuals do not receive sufficient solar radiation
due to shading from the larger individuals which eventually results in declining growth rates
and subsequent mortality (e.g., Newton and Jolliffe, 2003)). Newton (2006a) derived a
mechanical-based reformation of the self-thinning rule for monospecific jack pine stands
based on this concept. Given the sway dynamics, crown collisions and the noncontiguous
nature of crown cover (crown shyness) commonly observed within self-thinning mixed
coniferous stands at the later stages of stand development, and the concordance between the
empirical self-thinning exponent for mixed stands ( ̂ 1 = -1.2) and that predicted by Newton‘s
(2006a) reformulation ( 1.3 1 1.0 ), suggest this mechanical reformation and its
assumptions (e.g., competition for physical space is partially responsible for the observed
self-thinning pattern) may be applicable to mixed stands as well.
Another important linkage between the SSDMM and ecological theory relates to the
relationship between net production and site occupancy, which is used to define the optimal
density management window. Conceptually, net production increases with increasing site
occupancy until an asymptote is reached for a given species, site quality and stage of stand
development (Langseter, 1941; Mar:Möller, 1954; Assmann, 1970). Additional increases in
site occupancy results in a decline in net production, principally due to density-dependent
mortality arising from the self-thinning process. Thus if the objective is to maximize biomass
productivity and by extension carbon sequestration potential, then stands should be allowed to
achieve full occupancy. Once attained, they should be maintained at the asymptotic
occupancy-productivity condition via density management treatments. For black spruce and
jack pine mixed stands, this condition is represented by the optimal density management
window as delineated by relative densities indices between 0.32 and 0.45. Hence,
conceptually, mixed stands managed below the 0.32 relative density threshold are not
optimally occupying the site and thus not achieving their full productivity potential.
Conversely, stands in which density-stress levels are above the 0.45 relative density threshold
are likely to incur substantial density-dependent mortality which may result in productivity
losses and negative carbon balances. Operationally, however, maintaining stands within this
optimal window is currently impractical given the necessity to implement numerous light and
costly thinning treatments. Nevertheless, the maturing bio-economy along with developing
carbon markets may make this management proposition more economically amiable in the
future.
Ecologically, density management treatments directly affect the competitive
interrelationships within forest tree populations. Treatments such as PCT and CT within
density-stressed stands immediately reduce the intensity and occurrence of symmetrical and
asymmetrical competitive interactions. The temporary reduction or elimination of these
competition pressures increases the availability of moisture, nutrients, solar radiation and
physical growing space, for the remaining crop trees. Thinning from below in which the
smallest trees are removed provides the residual crop trees with an immediate influx of
moisture and nutrients which are shared equally. This competition process is also known as a
resource depletion process in which all competitors passively acquire an equal share of the
available below ground resources on a per-unit size basis (Newton and Jolliffe, 2003).
Conversely, selection thinning in which the trees with the greatest potential to respond to
166 Peter F. Newton
treatment are left and all others are removed, commonly results in a residual population
comprised mostly of trees from the co-dominant crown classes. Essentially, selection thinning
results in an immediate increase in the availability of both below and above ground resources.
In contrast to the passive nature of the resource depletion process, competition for solar
radiation and physical space is an asymmetrical process in which larger-sized competitors
acquire a disproportionate share of the these resources at the expense of the smaller-sized
competitors on a per-unit size basis (Newton and Jolliffe, 2003). The self-thinning stage of
stand development in which the smallest individuals incur mortality is largely driven by a
resource pre-emption process. Thus understanding resource competition processes and their
effects is an essential prerequisite to sound density management decision-making.
In essence, the modular-based SSDMM presented in this study for mixed stands provides
forest managers with an ecological-based decision-support tool for manipulating competitive
relationships within forest tree populations in order to realize single or multiple stand-level
management objectives. In relation to the evolving bio-economy and carbon markets, the
SSDMM can also be used to derive optimal density management regimes for these objectives.
For example, in a general sense, stands should be managed so that they (1) do not excessively
self-thin themselves and thereby do not generate large amounts of abiotic plant material that
could contribute to a negative carbon budget, (2) allocate a greater proportion of their
resources to foliage production in order to increase CO2 sequestration rates, and (3) produce
large volumes of dimensional lumber products so that the sequestrated carbon is stored
infinitely.
ACKNOWLEDGEMENTS
The author expresses his appreciation to: (1) Dave Wood, Forest Ecosystem Science Co-
operative Inc, Thunder Bay, Ontario, Canada, for access to the temporary and permanent
sample plot measurements and records used for model calibration; (2) Dr. I. Amponsah,
Development and Utility of an Ecological-based Decision-Support System … 167
Sustainable Resource Development, Alberta, Canada, Dr. D. Reid, Research Scientist, Centre
for Northern Forest Ecosystem Research, Lakehead University, Thunder Bay, Ontario,
Canada, Dr. M. Sharma, Research Scientist, Ontario Forest Research Institute, Sault Ste.
Marie, Ontario, Canada, and John Parton, Provincial Growth and Yield Coordinator, Ontario
Ministry of Natural Resources (OMNR), Timmins, Ontario, Canada, for analytical advice; (3)
NaturaLogic Inc., North Bay, Ontario, Canada, for VB.Net software design assistance; and
(4) Forestry Research Partnership and the Canadian Wood Fibre Centre for fiscal support.
REFERENCES
Ando T. 1962. Growth analysis on the natural stands of Japanese red pine (Pinus densiflora
Sieb. et. Zucc.). II. Analysis of stand density and growth (in Japanese; English summary).
Government of Japan, Bulletin of the Government Forest Experiment Station (Tokyo,
Japan), No. 147.
Ando T. 1968. Ecological studies on the stand density control in even-aged pure stands (in
Japanese; English summary). Government of Japan, Bulleton of the Government Forest
Experiment Station (Tokyo, Japan) No. 210.
Assmann E. 1970. The Principles of Forest Yield Study. 1st English Edition, Pergamon Press
Ltd. Oxford, England.
Bailey R.L., Dell T.R. 1973. Quantifying diameter distributions with the Weibull function.
Forest Science, 19: 97-104.
Barbour R.J., Kellogg R.M. 1990. Forest management and end-product quality: A Canadian
perspective. Canadian Journal of Forest Research, 20: 405-414.
Baskerville G.L. 1972. Use of logarithmic regression in the estimation of plant biomass.
Canadian Journal of Forest Research, 2: 49-53.
Bella I.E. 1967. Crown width/diameter relationship of open-growing jack pine on four site
types in Manitoba. Canadian Department of Forest and Rural Development, Bi-monthly
Research Notes, 23: 5-6.
Bella I.E., De Franceschi J.P. 1974. Early results of spacing studies of three indigenous
conifers in Manitoba. Environment Canada, Forestry Service, Northern Forestry Research
Center, Edmonton, Alberta. Information Report, NOR-X-113.
Cao Q.V. 2004. Predicting parameters of a Weibull function for modeling diameter
distribution. Forest Science, 50: 682-685.
Carmean W.H., Hazenberg G., Deschamps K.C. 2006. Polymorphic site index curves for
black spruce and trembling aspen in northwest Ontario. Forest Chronicle, 82: 231-242.
Carmean W.H., Niznowski G.P., Hazenberg G. 2001. Polymorphic site index curves for jack
pine in Northern Ontario. Forestry Chronicle, 77: 141-150.
Castedo-Dorado F., Crecente-Campo F., Álvarez-Álvarez P., Barrio A.M. 2009. Development
of a stand density management diagram for radiate pine stands including assessment of
stand stability. Forestry, 82: 1-16.
CCFM (Canadian Council of Forest Ministers). 2009. National forestry database.
http://nfdp.ccfm.org/silviculture/national_e.php.
Clutter J.L., Fortson J.C., Pienaar L.V., Brister G.H., Bailey R.L. 1983. Timber Management:
A Quantitative Approach. 1st Edition, John Wiley and Sons, New York, NY, USA.
168 Peter F. Newton
Dean T.J., Baldwin Jr. V.C. 1993. Using a density management diagram to develop thinning
schedules for loblolly pine plantations. Government of the United States of America,
Department of Agriculture, Forest Service, Southern Forest Experiment Station, New
Orleans, Louisiana. Research Paper SO-275.
De Vos S. 1973. The use of nearest neighbour methods. Tijdschrift voor Economische en
Sociale Geografie, 64: 308-319.
Di Lucca C.M. 1999. TASS/SYLVER/TIPSY: systems for predicting the impact of
silvicultural practices on yield, lumber value, economic return and other benefits. In: C.
Barnsey, Editor, Stand Density Management Planning and Implementation Conference,
Edmonton, Alberta. Clear Lake Publishing Ltd., Edmonton, Alberta, Canada. pp. 7-16.
Drew T.J., Flewelling J.W. 1977. Some recent Japanese theories of yield-density relationships
and their application to Monterey pine plantations. Forest Science, 23: 517-534.
Drew T.J., Flewelling J.W. 1979. Stand density management: an alternative approach and its
application to Douglas-fir plantations. Forest Science, 25: 518-532.
Emmett B. 2006. Increasing the value of our forest. Forestry Chronicle, 82: 3-4.
Enquist B.J., Brown J.H., West G.B. 1998 Allometric scaling of plant energetics and
population density. Nature, 395: 163-166.
Erdle T. 2000. Forest level effects of stand level treatments: using silviculture to control the
AAC via the allowable cut effect. In: P.F. Newton, Editor, Expert Workshop on the
Impact of Intensive Forest Management on the Allowable Cut, Canadian Ecology Centre,
Mattawa (2000), Ontario, Canada. pp. 19-30.
http://www.forestresearch.ca/Projects/fibre/IFMandACE.pdf.
Fleming R.L., Mossa D.S., Marek G.T. 2005. Upland black spruce stand development 17
years after cleaning and precommercial thinning. Forestry Chronicle, 81: 31-41.
Flewelling J.W., Drew T.J. 1985. A stand density management diagram for lodgepole pine. In
D.M. Baumgarter, R.G. Krebill, J.T. Arnott, and G.F. Weetman (Editors), Lodgepole
pine: the species and its management, Pullman, Washington, USA: Washington State
University.
Heriansyah I., Bustomi S., Kanazawa Y. 2009. Density effects and stand density management
diagram for merkus pine in the humid tropics of Java, Indonesia. Journal of Forestry
Research, 5: 91-113.
Honer T.G., Ker M.F., Alemdag I.S. 1983. Metric timber tables for the commercial tree
species of central and eastern Canada. Government of Canada, Department of
Agriculture, Canadian Forestry Service, Maritimes Forest Research Centre, Fredericton,
New Brunswick. Information Report M-X-140.
Hyink D.M., Moser J.W. 1983. A generalized framework for projecting forest yield and stand
structure using diameter distributions. Forest Science, 29: 85–95.
Hynynen J., Ahtikoski A., Siitonen J., Sievanen R., Liski J. 2005. Applying the MOTTI
simulator to analyse the effects of alternative management schedules on timber and non-
timber production. Forest Ecology and Management, 207: 5-18.
Isobe T., Feigelson E.D., Akritas M.G., Babu G.J. 1990. Linear regression in astronomy. I.
Astrophysical Journal, 364: 104-113.
Jack S.B., Long J.N. 1996. Linkages between silviculture and ecology: an analysis of density
management diagrams. Forest Ecology and Management, 86: 205–220.
Jolliffe P.A., Eaton G.W., Potdar M.V. 1988. Plant growth analysis: allometry, growth and
interference in Orchardgrass and Timothy. Annals of Botany, 62: 31-42.
Development and Utility of an Ecological-based Decision-Support System … 169
Kang K.Y., Zhang S.Y., Mansfield S.D. 2004. The effects of initial spacing on wood density,
fibre and pulp properties in jack pine (Pinus banksiana Lamb.). Holzforschung, 58: 455-
463.
Khil‘mi G.F. 1957. Theoretical Forest Biogeophysics. Academy of Sciences of the USSR
(Translated from Russian by the National Science Foundation, Washington, D.C.).
Kim D.K., Kim J.W. Park S.K. Oh M.Y., Yoo J.H. 1987. Growth analysis of natural pure
young stand of red pine in Korea and study on the determination of reasonable density (in
Korean; English abstract). Government of Korea, Research Reports of the Forestry
Institute (Seoul, Korea), 34: 32-40.
Kira T., Ogawa H., Sakazaki N. 1953. Intraspecific competition among higher plants. I.
Competition-yield-density interrelationship in regularly dispersed populations. Journal of
the Institute of Polytechnics (Osaka City University, Japan), Series D, 4: 1-16.
Kotze H., Malan F. 2007. Further progress in the development of prediction models for
growth and wood quality of plantation-grown Pinus patula sawtimber in South Africa.
In: Dykstra, D.P., Monserud, R.A. (Tech. Eds), Forest Growth and Timber Quality:
Crown Models and Simulation Methods for Sustainable Forest Management, Proceedings
of an International Conference, August 7-10, 2007, Portland, OR, USA. Department of
Agriculture, Forest Service, Pacific Northwest Research Station, General Technical
Report, PNW-GTR-791: 113-123.
Langsaeter A. 1941. Om tynning i enaldret gran- og furuskog (About thinning in even-aged
stands of spruce, fir and pine). Meddel. F. d. Norske Skogforsoksvesen, 8: 131-216.
Lindh B.C., Muir P.S. 2004. Understory vegetation in young Douglas-fir forests: does
thinning help restore old-growth composition. Forest Ecology and Management, 192:
285-296.
Liu C., Zhang S.Y. 2005. Models for predicting product recovery using selected tree
characteristics of black spruce. Canadian Journal of Forest Research, 35: 930-937.
Long J.N., McCarter J.B., Jack S.B. 1988. A modified density management diagram for
coastal Douglas-fir. Western Journal of Applied Forestry, 2: 6-10.
López-Sánchez C., Rodríguez-Soalleiro R. 2009. A density management diagram including
stand stability and crown fire risk for Pseudotsuga Menziesii (Mirb.) Franco in Spain.
Mountain Research and Development, 29: 169-176.
Mallows C.L. 1973. Some comments on Cp. Technometrics, 15: 661-675.
Mar:Möller C.M. 1954. The influence of thinning on volume increment. I. Results of
investigations. In: C. Mar:Möller, J. Abell, T. Jagd, and F. Juncker, Editors, Thinning
problems and practices in Denmark, State University of New York, College of Forestry,
Syracuse, NY, USA, Technical Publication, 76: 5-32.
McCarter J.B., Long J.N. 1986. A lodgepole pine density management diagram. Western
Journal of Applied Forestry, 1: 6-11.
McClain K.M., Morris D.M., Hills S.C., Buse L.J. 1994. The effects of initial spacing on
growth and crown development for planted northern conifers: 37-year results. Forestry
Chronicle, 70: 174–182.
McKinnon L.M., Kayahara G.J., White R.G. 2006. Biological framework for commercial
thinning evenaged single-species stands of jack pine, white spruce, and black spruce in
Ontario. Ontario Ministry of Natural Resources, Science and Information Branch,
Northeast Science and Information Section. Technical Report, TR-046.
170 Peter F. Newton
Meyer J.S., Ingersoll C.G., McDonald L.L., Boyce M.S. 1986. Estimating uncertainty in
population growth rates: jackknife vs. bootstrap techniques. Ecology, 67: 1156-1166.
Nelder J.A. 1962. New kinds of systematic designs for spacing experiments. Biometrics, 18:
283-307.
Neter J., Wasserman W., Kutner M.H. 1990. Applied Linear Statistical Models, 3rd Edition,
Irwin, Boston, MA, USA.
Newton P.F. 1997. Stand density management diagrams: review of their development and
utility in stand-level management planning. Forest Ecology and Management, 98: 251-
265.
Newton P.F. 2003. Yield prediction errors of a stand density management program for black
spruce and consequences for model improvement. Canadian Journal of Forest
Research, 33: 490-499.
Newton P.F. 2006a. Asymptotic size–density relationships within self-thinning black spruce
and jack pine stand-types: Parameter estimation and model reformulations. Forest
Ecology and Management, 226: 49-59.
Newton P.F. 2006b. Forest production model for upland black spruce stands—Optimal site
occupancy levels for maximizing net production. Ecological Modelling, 190: 190–204.
Newton P.F. 2009. Development of an integrated decision-support model for density
management within jack pine stand-types. Ecological Modelling, 220: 3301-3324.
Newton P.F. 2010. Stand Density Management Diagrams. SciTopics. Retrieved November
30, 2010, from http://www.scitopics.com/Stand_Density_Management_Diagrams.html.
Newton P.F., Amponsah I.G. 2005. Evaluation of Weibull-based parameter prediction
equation systems for black spruce and jack pine stand types within the context of
developing structural stand density management diagrams. Canadian Journal of Forest
Research, 35: 2996-3010.
Newton P.F., Amponsah I.G. 2007. Comparative evaluation of five height–diameter models
developed for black spruce and jack pine stand-types in terms of goodness-of-fit, lack-of-
fit and predictive ability. Forest Ecology and Management, 247: 149–166.
Newton P.F., Jolliffe P.A. 2003. Aboveground dry matter partitioning responses of black
spruce to directional-specific indices of local competition. Canadian Journal of Forest
Research, 33: 1832-1845.
Newton P.F., Sharma M. 2008. Evaluation of sampling design on taper equation performance
in plantation-grown Pinus banksiana. Scandinavian Journal of Forest Research, 23: 358-
370.
Newton P.F., Weetman G.F. 1993. Stand density management diagrams and their utility in
black spruce management. Forestry Chronicle, 69: 421-430.
Newton P.F., Weetman G.F. 1994. Stand density management diagram for managed black
spruce stands. Forestry Chronicle, 70: 65-74.
Newton P.F., Lei Y., Zhang S.Y. 2004. A parameter recovery model for estimating black
spruce diameter distributions within the context of a stand density management diagram.
Forestry Chronicle, 80: 349-358.
Newton P.F., Lei Y., Zhang S.Y. 2005. Stand-level diameter distribution yield model for
black spruce plantations. Forest Ecology and Management, 209: 181-192.
Nilsen P., Strand L.T. 2008. Thinning intensity effects on carbon and nitrogen stores and
fluxes in a Norway spruce (Picea abies (L.) Karst.) stand after 33 years. Forest Ecology
and Management, 256: 201–208.
Development and Utility of an Ecological-based Decision-Support System … 171
Pelletier G., Pitt D.G. 2008. Silvicultural responses of two spruce plantations to midrotation
commercial thinning in New Brunswick. Canadian Journal of Forest Research, 38: 851-
867.
Peltola A. 2009. Finnish Statistical Yearbook of Forestry. http://www.metla.fi/julkaisut/
metsatilastollinenvsk/index-en.htm.
Pretzsch H. 2009. Forest Dynamics, Growth and Yield. Springer, Verlag, Berlin and
Heidelberg.
Pretzsch H., Biber P., Dursky J. 2002. The single tree-based stand simulator SILVA:
construction, application and evaluation. Forest Ecology and Management, 162: 3–21.
Rowe J.S. 1972. Forest regions of Canada. Government of Canada, Department of
Environment, Canadian Forestry Service, Ottawa, Ontario. Publication No. 1300.
Rudnicki M., Silins U., Lieffers V.J., Josi G. 2001. Measure of simultaneous tree sways and
estimation of crown interactions among a group of trees. Trees, 15: 83-90.
Sharma M., Parton J. 2009. Modeling stand density effects on taper for jack pine and black
spruce plantations using dimensional analysis. Forest Science, 55: 268-282.
Shinozaki K., Kira, T. 1956. Intraspecific competition among higher plants. VII. Logistic
theory of the C-D effect. Journal of the Institute of Polytechnics (Osaka City University,
Japan), Series D, 12: 69-82.
Smith F.W., Long J.N. 1987. Elk hiding and thermal cover guidelines in the context of
lodgepole pine stand density. Western Journal of Applied Forestry, 2: 6-10.
Smith N.J. 1989. A stand-density control diagram for western red cedar, Thuja plicata. Forest
Ecology and Management, 27: 235-244.
Sprugel D.G. 1983. Correcting for bias in log-transformed allometric equations. Ecology, 64:
209-210.
Stankova T.V., Shibuya M. 2007. Stand Density Control Diagrams for Scots pine and
Austrian black pine plantations in Bulgaria. New Forests, 34: 123-141
Sturtevant B.R., Bissonette J.A., Long J.N. 1996. Temporal and spatial dynamics of boreal
forest structure in western Newfoundland: silvicultural implications for marten habitat
management. Forest Ecology and Management, 87: 13-25.
Tong Q.J., Zhang S.Y., Thompson M. 2005. Evaluation of growth response, stand value and
financial return for pre-commercially thinned jack pine stands in Northwestern Ontario.
Forest Ecology and Management, 209: 225-235.
Tong Q.J., Fleming R.L., Tanguay F., Zhang S.Y. 2009. Wood and lumber properties from
unthinned and precommercially thinned black spruce plantations. Wood and Fibre
Science, 41: 168-179.
Verschuyl J., Riffell S., Miller D., Wigley T.B. 2011. Biodiversity response to intensive
biomass production from forest thinning in North American forests – A meta-analysis.
Forest Ecology and Management, 261: 221-232.
Vezina P.E. 1963. More about the crown competition factor. Forestry Chronicle, 39: 313-
317.
Watt W.R, Parton J., Chen H., Lucking G., Houle N., Levesque S., Luke A. 2001. Standard
forest units for northeastern Ontario boreal Forests. Ontario Ministry of Natural
Resources, Northeast Science and Technology Unit, Information Report (Draft dated
14/02/01).
Weibull W. 1951. A statistical distribution function of wide applicability. Journal of Applied
Mechanics, 18: 293-297.
172 Peter F. Newton
West G.B., Brown J.H., Enquist B.J. 1997. A general model for the origin of allometric
scaling laws in biology. Science, 276: 122-126.
Yoda K., Kira T., Ogawa H., Hozumi K. 1963. Self-thinning in overcrowded pure stands
under cultivated and natural conditions. Journal of Biology, (Osaka City University,
Japan) 14: 107-129.
Zhang S.Y., Lei Y.C., Bowling C. 2005. Quantifying stem quality characteristics in relation
to initial spacing and modeling their relationship with tree characteristics in black spruce
(Picea mariana). Northern Journal of Applied Forestry, 22: 85-93.
Zhang S.Y., Liu C., Jiang Z.H. 2006. Modeling product recovery in relation to selected tree
characteristics in black spruce using an optimized random sawing simulator. Forest
Products Journal, 56: 93-99.
In: Ecological Modeling ISBN: 978-1-61324-567-5
Editor: WenJun Zhang, pp. 173-204 © 2012 Nova Science Publishers, Inc.
Chapter 8
ABSTRACT
We present a review of the concepts and methods associated to ecological niche
modeling illustrated with the published works on amphibians and reptiles of the
Mediterranean Basin, one of the world's biodiversity hotspots for conservation priorities.
We start by introducing ecological niche models, analyzing the various concepts of niche
and the modeling methods associated to each of them. We list some conceptual and
practical steps that should be followed when modeling, and highlight the pitfalls that
should be avoided. We then outline the history of ecological modeling of Mediterranean
amphibians and reptiles, including a variety of aspects: identification of the ecological
niche; detection of common distribution areas (chorotypes) and other biogeographical
patterns; analysis and prediction of species richness patterns; analysis of the expansion of
native and invasive species; integration of molecular data with spatial modeling;
174 A. Márcia Barbosa, Neftalí Sillero, Fernando Martínez-Freiría et al.
Several definitions of ecological niche have been proposed over time. The first one is due
to Grinnell (1917), who understood the ecological niche as a subdivision of the habitat
containing the environmental conditions that allow the individuals of a species to survive and
reproduce. This concept is based on variables for which the species compete (climatic or
scenopoetic variables according to Soberón, 2007; see also Hirzel and Le Gay, 2008, Wiens
et al., 2009). On the other hand, Elton (1927) emphasized the functional role of a species in a
community, especially its position in the food chain, depending on variables that can be
consumed by the species (nutrients or bionomic variables in Soberón, 2007). Finally,
Hutchinson (1957) defined mathematically the fundamental and the realized niche (Figure 1).
The fundamental niche is an n-dimensional volume of environmental space within which a
species can maintain a viable population and persist over time without immigration. Each
dimension is an environmental variable that influences the niche. The realized niche is a part
of the fundamental niche where the species is not excluded by competition. The main
difference between Grinnell‘s and Elton's niche concepts relative to Hutchinson‘s one is that
the former two used the term niche to refer to places in the environment that can
Ecological Niche Models in Mediterranean Herpetology 175
accommodate the species, while for Hutchison, species, and not the environment, have
niches.
Jackson and Overpeck (2000) replaced the concept of realized niche by that of potential
niche, which is the part of the fundamental niche that is available for the species; some parts
are not available because not all possible combinations of variables under which the species
could survive currently exist in the environment. Similarly, Colwell and Rangel (2009) and
Soberón and Nakamura (2009) considered that there are three different niches: the
fundamental niche, the potential niche (i.e. the existing part of the fundamental niche), and
the realized niche. Finally, Pearson (2007) introduced the concept of occupied niche, to which
species distributions are limited by historical, geographical, and biotic factors (dispersal
ability, competition, predation, parasitism, symbiosis).
Figure 1. The easiest way to visualize the different ecological niches is the BAM (biotic, abiotic,
movement) diagram (see Soberón and Peterson, 2005, Soberón, 2007), which represents the theoretical
environmental space divided into the three main factors that limit the distribution of a species. The
suitable habitat corresponds to the area common to all three factors, which represents the occupied
niche (ON; sensu Pearson, 2006). The area shared by A and M represents Grinnell's niche (GN). The
area shared by A and B is Elton's niche (EN). The whole A area is Hutchinson's fundamental niche
(FN). A species can live in climatically favorable regions to which it has been able to disperse and from
which it is not excluded by biotic interactions. Regions that fail to meet all these conditions are not
suitable for the species' presence.
176 A. Márcia Barbosa, Neftalí Sillero, Fernando Martínez-Freiría et al.
To these concepts we must add two other important ones: the source-sink theory
(Pulliam, 1988, Pulliam, 2000) and dispersal limitation (Holt, 2003). According to the source-
sink theory, some populations may occupy unsuitable habitats (sinks) due to immigration
from healthier nearby populations (sources). Although individuals in the sinks may die due to
the adverse conditions, they are replaced by new immigrants. Here, the realized distribution
goes beyond the fundamental niche, as the species occupies habitats that are inadequate and
not contained in the niche (Pulliam 1988, Pulliam 2000). With dispersal limitation, a species
can be absent from suitable habitats for historical reasons or due to limitations in its ability to
disperse to those habitats (Holt, 2003).
relationships can be very helpful, as long as the constraints and limitations inherent to the
statistical analyses are taken into account (MacNally, 2000).
The correlative methods of modeling species' niches can be classified into three main
groups: presence/absence methods, profile methods, and presence-only methods. The first
kind relates a binary dependent variable (i.e., with only two possible values, such as presence
and absence, one and zero) to a series of independent variables, so it induces the conditions
that make a species present rather than absent. This kind of methods includes generalized
linear models (GLM), such as discriminant analysis (Lachenbruch, 1975), logistic regression
(Hosmer and Lemeshow, 1989), and the favorability function (Real et al., 2006); and
generalized additive models (GAM; Hastie and Tibshirani, 1990), which are more complex
and usually fit the data better, but may be less general, i.e., less applicable to other data sets.
Profile methods compare the environmental conditions in the observed presence areas
with the conditions available in the whole study area, thus outlining presence against a
background. These methods include ecological niche factor analysis (ENFA; Perrin, 1984,
Hirzel et al., 2002), the genetic algorithm for rule-set production (GARP; Stockwell and
Noble, 1992), and maximum entropy (Maxent; Phillips et al., 2004). The definition of these
methods as "presence-only" is incorrect, as they compare presence areas with all the
environmental space analyzed (the so-called background), which includes both presence and
non-presence areas (e.g. Phillips et al., 2009). From this background, some authors select
pseudo-absences (e.g. Chefaoui and Lobo, 2008).
Presence-only data can be modeled, for example, with overlap analysis (Brito et al., 1999,
Arntzen and Teixeira, 2006), which overlays the species‘ presence area to the environmental
variables to derive the range of environmental conditions under which the species can live.
Mahalanobis distance (Etherington et al., 2009), the multidimensional envelope (MDE) used
in BIOCLIM (Busby, 1991), and the HABITAT (Walker and Cocks, 1991) and DOMAIN
models (Carpenter et al., 1993) are other examples of truly presence-only methods. Another
way to analyze only presence data is to use binary models such as logistic regression to
confront, rather than presence and absence of a species, the presence of one species or
subspecific variant against the presence of another one (Romero and Real, 1996, Brito and
Crespo, 2002, Arntzen and Alexandrino, 2004, Real et al., 2005, Arntzen and Espregueira
Themudo, 2008).
There are more complex methods, such as random forests, classification and regression
trees (CART), multivariate adaptive regression splines (MARS), and artificial neural
networks (ANN). Moisen and Frescino (2002) compared the effectiveness of various methods
in modeling the distribution of simulated and actual data on forest variables. The models built
with more sophisticated techniques showed better results with simulated data, but for real data
the difference were not significant and a simple linear model worked almost as well as
complex models. Complex methods may describe species‘ distributions more accurately, but
produce models more difficult to interpret from an ecological point of view. Moreover, they
seldom provide intelligible information on which and how environmental variables are related
to species‘ distributions, an information that can be valuable from a conservation standpoint.
178 A. Márcia Barbosa, Neftalí Sillero, Fernando Martínez-Freiría et al.
A compromise between complexity and intelligibility may be desirable in ENMs meant for
use in conservation and management.
Some absence records may actually correspond to presences that were not detected for
various reasons, including insufficient or no surveying effort. This affects not only
presence/absence modeling methods, but also profile methods, which will include these false
absences in the background but outside the presence area; and presence-only methods, which
will fail to include these undetected presences in the analysis. When the survey is not too
deficient or spatially biased (Reese et al., 2005), explicitly including absence data can
improve the ENM by providing information on locations that may be less suitable for the
species (Hirzel et al., 2001; see also section 1.5) due to historical (e.g. barriers to dispersal),
biotic (competition, predation) or human restrictions (Guisan and Zimmermann, 2000,
Anderson et al., 2002). Even when the data include false absences, these are often associated
to low local abundance of the species, so their inclusion can improve the relationship between
ENM predictions and actual species abundance (Real et al., 2009, Barbosa et al., 2009).
If the aim is to obtain the fundamental niche of the species, the use of absence data may
have an undesirable effect by excluding suitable environmental areas where the species is not
present due to historical limitations or biotic interactions. But if the goal is to approach the
realized niche and the data are of generally good quality, it is advisable to explicitly include
absences, even if they are not all correct. In any case, the quality of absence data is as
important for the models as the quality of the presence data.
The application of profile or presence-only methods may be more desirable in the case of
very limited or dispersed presence data, such as those taken from herbaria or museum
collections (Elith et al., 2006) or from samples identified through molecular analysis (Real et
al., 2005). Presence/absence models based on distribution atlases have been widely used with
success and have shown good relationships with independent data on species abundance (Real
et al., 2009) and good extrapolation ability, both to contiguous geographic areas (Barbosa et
al., 2009) and to finer resolution scales (Barbosa et al., 2010). To be able to model presences
and absences with more reliable data, it would be useful to publish information on areas that
have been surveyed but where the target species were not found.
Abundance data can also be modeled (e.g. Anadón et al. 2010). Abundance provides
more information than the simple presence or absence of species, but it also contains more
noise and is more costly to measure, so it is rarely available. Abundance data can be analyzed
with generalized linear models that assume a Poisson distribution, suitable for count data, or a
negative binomial distribution, when there is over-dispersion of these data (i.e., when the
variance is greater than the mean). When the data contain many more zeros than would be
expected according to any of these distributions, zero-inflated models are more appropriate
(Zuur et al., 2009).
Not all ENMs represent the same niche: the result varies depending on the method and
the type of variables used (Soberón and Nakamura, 2009; Figures 1 and 2). It is widely
accepted that mechanistic ENMs predict the fundamental niche (Pearson and Dawson, 2003,
Kearney and Porter, 2004, Kearney and Porter, 2009, Rodder et al., 2009) and correlative
ENMs are closer to the realized niche, since the recorded presences are determined by biotic
Ecological Niche Models in Mediterranean Herpetology 179
and abiotic factors (Pearson and Dawson, 2003, Araújo and Guisan, 2005, Guisan and
Thuiller, 2005, Soberón and Peterson, 2005, Kearney, 2006, Morin and Lechowicz, 2008,
Pearman et al., 2008, Colwell and Rangel, 2009, Lobo et al., 2010). Within correlative ENMs,
those calculated with presence/absence data yield the probability of finding the species in
each portion of the study area. Above a chosen probability threshold, these models can be
considered to represent the spatial distribution of the habitats that are both suitable and
occupied by the species (Guisan and Zimmermann, 2000; Pearson, 2007). Models based only
on presence data provide an indication of the suitability of the habitat, not necessarily
implying that the species will be found there.
Correlative methods forecast the distribution of the species through correlations among
environmental variables. The values of these variables are determined by the geographical
positions of the species‘ records. In other words, correlative ENMs are sensitive to the
topology of presences (their geographical positions and the relationships between them). If
we want to calculate a species‘ potential distribution without considering its geographical
records, we should use a mechanistic modeling method.
Figure 2 The idea of habitat suitability actually corresponds to a gradient (Soberón, 2010) with
extremely favorable habitats on one end, and completely unfavorable ones (where the species simply
cannot survive) on the other. The position of the species in this gradient depends on the intensity with
which climate, dispersal and biotic interactions act. In habitats with unfavorable climate, the species can
survive for some time if there is immigration and there are no competing species. In contrast, the
species may be absent from a climatically favorable area if it is excluded by other species or if
geographical barriers prevent its access. Intermediate situations represent small populations that subsist
despite unfavorable climatic conditions and biotic interactions. See abbreviations in Figure 1.
180 A. Márcia Barbosa, Neftalí Sillero, Fernando Martínez-Freiría et al.
However, not all authors acknowledge that ENMs predict the ecological niche. For
Kearney (2006), the niche is a mechanical rather than descriptive concept. In this way,
correlative ENMs model only habitats. In contrast, mechanistic ENMs model the niche and
are the only ones that establish a mechanical connection between the model and the species.
Kearney (2006) considers that the niche is determined by the factors that allow a species to
survive, which implies a mechanical relationship between the species and these factors, while
the habitat is simply the place where the species lives. Jiménez-Valverde et al. (2008) and
Lobo (2008) also argue that correlative ENMs predict habitats rather than niches: if the
absences of the species are recorded in areas where they are excluded by biotic factors, the
model calculates the realized distribution; if they are taken from areas where the species is
excluded only by abiotic factors, the model calculates the potential distribution. However,
these "distributions" do not correspond to the realized and fundamental niches, respectively.
Godsoe (2010) goes further by proving mathematically that it is impossible to calculate the
niche: the only thing that can be determined is whether the set of variables used to calculate
an ENM belongs to the niche or not.
The first step in modeling the distribution of a species is to decide which is the motive for
the work (to know the potential distribution, the limiting factors…) and which are the most
appropriate modeling parameters (distribution data, environmental variables, study area,
modeling method, resolution scale). Once these are defined, we must gather chorological (i.e.
location) data of the species and create a database, for example from distribution atlases
(Sillero et al., 2009), museum collections (Brito et al., 2008) or systematic surveys (Hirzel
and Guisan, 2002, Martínez-Freiría et al., 2008, Sillero, 2009). Such surveys can be oriented
to establish only the presence of the species (Martínez-Freiría et al., 2008) or to record also
their absence (Anadón et al., 2006). The spatial distribution of absences determines the type
of ecological niche that will be predicted and, therefore, can significantly alter the outcome of
the models (Figure 1). It is also important that the database has no geographical errors (Sillero
et al., 2005). Even if the ENM is to be produced with a low (i.e., coarse) spatial resolution, it
is recommended that the records of the species are taken with a GPS. This will minimize
errors in the geographical coordinates and allow the data to be used for ENMs with different
spatial resolutions (Carretero et al., 2008, Kaliontzopoulou et al., 2009).
The chorological records should be independent of one another, i.e. should not be
spatially autocorrelated (Koenig, 1999, Dormann, 2007, Dormann et al., 2007). This
condition can only be fulfilled with systematic surveys. The autocorrelation of the surveying
effort must be uniform. Thus, the distribution of the records should correspond as much as
possible to the actual distribution of the species, or at least to the one observed in the field.
For example, the distribution of a species can be strongly aggregated or fragmented. This
variation in the degree of clustering observed in the sample of distribution records must have
a correspondence with reality. An appropriate sampling design with a consistent surveying
effort is the best way to minimize the problem of autocorrelation in the data.
Sample size also influences the results of ENMs (Hirzel and Guisan, 2002, McPherson et
al., 2004, Pearson et al., 2007, Wisz et al., 2008). The accuracy of the ENM first increases
Ecological Niche Models in Mediterranean Herpetology 181
substantially with sample size, and then stabilizes, reaching an asymptote (Stockwell and
Peterson, 2002). Teixeira and Arntzen (2006) study the variation in the number of records of
Chioglossa lusitanica throughout history and its effect on ENMs: from a certain number of
records, the ENM does not improve significantly. There is a minimum sample size below
which it is not possible to calculate an ENM, which depends on the modeling method used.
The types of modeling algorithms that can be used depend on the chorological data
available. Not all methods are equivalent or useful in all situations. To build a mechanistic
ENM, the knowledge on the biology and physiology of the species must be substantial
(Kearney and Porter, 2004, Kearney and Porter, 2009). For correlative ENMs, the
chorological data available will determine the type of modeling method that can be used. If
we have both presence and absence data, we can apply a variety of methods such as GLM or
GAM. If we have presence data within a wider background with environmental data, we can
apply profile methods such as Maxent (Phillips et al., 2004, Phillips et al., 2006), ENFA
(Hirzel et al., 2002) or GARP (Stockwell and Noble, 1992). When data are available only for
the presence area, we can apply presence-only methods such as Mahalanobis distance
(Etherington et al., 2009), BIOCLIM (Busby, 1991), HABITAT (Walker and Cocks, 1991) or
DOMAIN (Carpenter et al., 1993). In each case, we will get an approximation to a different
ecological niche (Soberón and Nakamura, 2009). Moreover, depending on the position of
absences in the BAM diagram (Figure 1), i.e., which factor they derive from, the ENM will
also be different (Lobo et al., 2010).
Lobo et al. (2010) classify absences into three different types: contingent absences, which
correspond to environmentally suitable areas that are not occupied for historical or biotic
reasons; environmental absences, when the environment is in fact unsuitable for species
presence; and methodological absences, caused by survey deficiency. Contingent absences
are outside the realized niche but inside the fundamental one; environmental absences are
outside both niches; and methodological absences are included in both niches (Lobo et al.,
2010). The latter can be very important in amphibian and reptile species, as it is difficult to
ensure that a species is really absent from a place where it has not been detected. This is why
profile and presence-only modeling methods are largely used in herpetology. For comparison,
in the Mediterranean Basin in the last three years, thirteen modeling works were published in
which profile methods were used (Ficetola et al, 2007, Brito et al., 2008, Carretero et al.,
2008, Kaliontzopoulou et al., 2008, Martínez-Freiría et al., 2008, Ficetola et al., 2009, Santos
et al., 2009, Martínez-Freiría et al., 2009, Ribeiro et al., 2009, Rödder and Lötters, 2009,
Sillero, 2009, Sillero et al., 2009, Sillero, 2010), against four works with presence/absence
models (Arntzen and Espregueira Themudo, 2008, Real et al., 2008, Bombi et al., 2009, Real
et al., 2010).
There are differences in the way each method works. For example, ENFA cannot work
with categorical variables: they must be converted to quantitative values. GLM, GAM, GARP
and Maxent models can be projected to other geographical areas or sets of variables; ENFA,
on the other hand, does not allow the projection of the ENM. All methods mentioned here are
affected by correlations between variables, except for ENFA, which previously transforms the
variables into a set of uncorrelated factors, similar to the axes of principal component analysis
(Hirzel et al., 2002). The ability of many methods to discriminate between presences and
absences can be assessed, for example, with the Area Under the ROC (receiver operating
characteristic) Curve (AUC), although Lobo et al. (2008) warn against equating higher
discrimination with higher accuracy when models differing in species prevalence are
182 A. Márcia Barbosa, Neftalí Sillero, Fernando Martínez-Freiría et al.
compared. It is thus not advisable to compare the AUC of models for the same species in
different areas, or for different species in the same area (VanDerWal et al., 2009). ENFA does
not make it possible to calculate the AUC for comparison with other modeling methods and
algorithms. In some cases (GLM, GAM, ENFA) the result is unique, as there is not a random
process associated (e.g. Arntzen, 2006, Espregueira Themudo and Arntzen, 2007, Soares and
Brito, 2007, Arntzen and Espregueira Themudo, 2008, Santos et al., 2006, Ribeiro et al.,
2009, Santos et al., 2009; Sillero et al., 2009). Maxent, on the other hand, produces a different
ENM each time, requiring the calculation of multiple replicas (e.g. Martínez-Freiría et al.,
2008, Martínez-Freiría et al., 2009, Sillero, 2009, Sillero, 2010) and a final averaged ENM
(ensemble model; Araújo and New, 2007). In conclusion, each modeling method has its own
peculiarities, which must be taken into account before undertaking any modeling project.
The next step is the selection of variables that can theoretically influence and limit the
species‘ distribution. This selection is a subjective process and depends on the ecological
relevance of these variables for the modeled species and on their availability, particularly in a
spatially geo-referenced format (see Sillero and Tarroso, 2010 for free environmental data
sources on the Internet). However, ENMs are not necessarily spatial. Actually, in some
methods the last step is the spatialization of the ENM, which is just a mathematical formula
that relates several variables. When the variables are available in a spatial format that can be
incorporated in a Geographic Information System (GIS), the ENM can be represented in
space (e.g. Brito and Crespo, 2002). To comply with statistical theory, the variables should
not be correlated (Keitt et al., 2002, Diniz-Filho et al., 2003, Betts et al., 2006, Segurado et
al., 2006; see section 1.7), which is rarely possible in the real world. At least, we should
remove one variable from each pair of variables that have a high correlation, for example
greater than 70-75% (e.g. Martínez-Freiría et al., 2008), and point to the same causal factor.
However, we cannot tell if we are eliminating the most causal variable (if it is one of the two)
or the variable that correlates with it.
The use of a large number of variables in relation to the number of localities can increase
the risk of over-parameterization of the model, i.e. of including variables whose relationship
with the species is due to chance, and that can only describe the species‘ distribution in the
study area. It is also important to note that the factors determining the ecological niche in the
ENM are expressed differently depending on the scale: historical and abiotic factors are more
relevant on coarse scales (i.e., large areas), while biotic factors act mainly on local scales
(Peterson, 2006).
Once the variables are selected, we should ensure that the study area is adequate. This is
an important step, though it is rarely taken into account, and should be decided at the
beginning of the process along with the modeling aims. Ideally, the study area should include
the entire ecological range of the species: if the area is too small, part of the ranges of some
variables may be left out of the analysis and lead to incorrect results. However, Thuiller et al.
(2004) found no difference between ENMs when they reduced the size of the study area, but
only when they projected ENMs outside their geographic boundaries (see also Barbosa et al.,
2009). In addition, environmental data are often not available for the entire distribution area
of a species. Albert and Thuiller (2008) recommend not using large areas if the distribution of
the species is small. Stockwell and Peterson (2002) propose to divide the area, since ENMs
may react differently to different sample sizes, showing that the species‘ response to
environmental variables is not uniform throughout the study area. Other authors consider that
the definition of the study area should respond to biogeographical criteria (e.g. Sillero et al.,
Ecological Niche Models in Mediterranean Herpetology 183
2009). Perhaps the best method is to check that the response curves of the variables are not
truncated (Guisan and Thuiller, 2005). If a curve is truncated, the study area should be
increased until a normal curve is reached (and then the correlations between variables should
be measured again). In any case, we should avoid to define study areas with political criteria
such as national divisions when they do not correspond to natural limits (e.g. Brito et al.,
1996, Teixeira et al., 2001, Teixeira and Arntzen, 2002, Arntzen, 2006, Arntzen and Teixeira,
2006 ), although data availability is often limited by political boundaries.
ENMs assume an equilibrium between the species and the environment, i.e., that the
species occupies all available favorable habitats and is absent from all unfavorable ones
(Araújo and Pearson, 2005, Wiens et al., 2009). However, it is very unlikely that this
condition is met. If a species is still expanding its distribution, chorological records do not
represent the breadth of its ecological niche, and the ENM may not identify all potentially
favorable areas. One solution is to increase the size of the study area; another is to divide the
data into groups that simulate the dispersion and calculate the respective ENMs over the same
area (Saddler, 2010): if the ENMs are similar, they can be considered robust. We can also
limit the chorological sample to proven breeding sites (Ficetola et al., 2008, Ficetola et al.,
2009), which correspond to populations and not to dispersing individuals (which are not in
equilibrium). In the case of introduced species, the best approach may be to model the
distribution in their original area and then project the ENM to the introduced area (Ficetola et
al., 2007, Pearman et al. 2008, Beaumont et al., 2009, Rodder and Lötters, 2009).
Once the modeling method and the study area are defined, the species‘ chorological data
are gathered, and the predictor (and uncorrelated) variables are selected, we can calculate the
ENM. Although each modeling method provides a different type of information, an ENM
should be composed primarily of three items: a map of the predictions (as long as it is a
spatial model), the importance with which each variable contributes to the model, and a
measure of its accuracy. The map depicts the favorable habitats for the species, and is usually
composed of continuous values, bounded between zero (completely unfavorable) and one
(completely favorable). In profile models, this result is often called habitat suitability map
(HSM; Hirzel et al., 2002). However, in some cases, it is of interest to convert the HSM into a
(potential) distribution map, with only values of "present" and "absent". To do this, we have
to choose a threshold within the continuous range of the HSM to distinguish unfavorable
habitats (below the threshold) from favorable ones (above the threshold). The choice of
threshold is up to the researcher and depends on the modeling aims (for reviews see Liu et al.,
2005, Jiménez-Valverde and Lobo, 2007).
When there is correlation between the variables but it is lower than 1, which is what
occurs in the real world, the exclusion of correlated variables produces 1) a decrease in the
predictive ability of the ENM which is greater when the correlation between the variables is
lower, and 2) a loss of explanatory power of the variable kept in the ENM that increases with
increasing correlation. Therefore, in correlative ENMs, the elimination of highly correlated
variables simplifies the ENM with little loss of predictive power. However, ENMs can be
seriously affected in their explanatory power if the hypothetically relevant variables are not
all included, even when they are, and precisely because they are, correlated.
For example, Cartron et al. (2000) point out that if some of the relationships between
variables are negative, the effect that is operating on the variables may be weakened by
another, stronger mechanism, due to the relationships between them. Thus, in a system with
three variables, if two of the correlations are positive and one is negative, the predicted
relationships may not all be seen in the relationships between each pair of variables, since
there are effects operating in different directions (Bárcena et al., 2004). If, for example, a
species is favored by high temperatures and rainfall, but these variables are negatively
correlated in the analyzed territory (i.e., precipitation is higher where the temperature is
lower), the effect of each variable is weakened by the effect of the other. Only the inclusion
of the two correlated variables in the same ENM will allow detecting the actual effect of each
variable.
ENMs obtained inductively from the recorded distribution of a species follow certain
rules that guarantee their consistency with that distribution, and therefore do not require
validation with respect to their starting data. However, it is possible to validate whether an
ENM remains accurate when applied to similar chorological data sets in other geographical
areas or moments in time, or when it is used to determine other population parameters of the
species. An ENM should thus be validated with data different from those used to build it,
according to, and specifically for, the purpose for which it was built. For example, an ENM
made to be transferred to the future, such as those that predict species‘ potential distributions
under climate change scenarios (e.g. Araújo and Pearson, 2005, Araújo et al., 2006, Araújo et
al. 2008, Sillero, 2010) cannot be validated in the present.
We can, however, determine if the ENM is spatially transferable within the analyzed
area. To do this, we can divide the territory into sub-areas of recalibration and pseudo-
validation. The general ENM, obtained with data from the complete study area, can be
recalibrated in the first sub-area to then check if it works equally well in the second sub-area.
The two sub-areas can be defined either randomly or by applying the formula proposed by
Fielding and Bell (1997):
[1+(p-1)1/2]-1,
where p is the number of predictors in the ENM. The degree of agreement between the results
of recalibration and pseudo-validation can be estimated by comparing their respective values
of Kappa (Cohen, 1960), sensitivity, specificity, correct classification rate, or AUC (Fielding
and Bell 1997, Manel et al., 2001). However, we must keep in mind that a discrepancy
Ecological Niche Models in Mediterranean Herpetology 185
between recalibration and pseudo-validation does not imply that the recalibrated ENM is
incorrect, and certainly not that the general ENM is incorrect. It simply indicates that the
factors that act predominantly in the area of recalibration differ from those that do in the area
of pseudo-validation. The ENM to be used should, in any case, be the one based on the entire
study area, as general models tend to work better than those based on subsets of data (e.g.
Barbosa et al., 2009).
1.9. Are We Really Modeling the Ecological Niche, or Can We Overcome the
Concept of Niche When Modeling?
Environmental processes are inherently complex and vary with spatial and temporal
scales. This complexity is often based on the definition of laws or assumptions about the way
these processes work, often expressed in the form of mathematical or logical relationships,
which are the core of ENMs. Mechanistic ENMs, as correlative ones, are based on the
previous acceptance of relationships between a species and the factors that supposedly
determine its distribution. A considerable degree of idealization is therefore necessary to
describe the distributions of species with mechanistic ENMs.
Among other things, a mechanistic ENM requires accepting a mechanism on which to
base it. This approach is based on Newtonian mechanics, called into question since the early
Twentieth Century by quantum mechanics, which is formulated on the basis of subatomic
phenomena. Biological phenomena are also subject to uncertainty, which is one of the pillars
of quantum mechanics. This uncertainty arises not only from the inability to identify all the
factors involved in the process, but also from the inability to determine the final outcome of
the process even when all relevant factors are controlled. The different degrees of coherence
between phenomena are mathematically translated to correlations. Correlative ENMs may
thus represent coherence between natural phenomena more adequately than mechanistic
Newtonian models. Furthermore, living organisms not only respond to environmental factors:
environmental factors are also strongly conditioned by the action of living organisms. The
classic formulation of logic and mathematics is not well equipped to handle this formulation
on physical systems, and even less on live systems and the relationships between them. A
more realistic interpretation of the complex ecological and biological systems can come from
the application of fuzzy logic.
Fuzzy logic is a form of multi-valued logic that allows for several values of truth, but also
takes into account that these values are inaccurate. Fuzzy logic and its applications have their
origin in the theory of fuzzy sets proposed by Zadeh (1965), who established that a fuzzy set
is characterized by a membership function that assigns to each object in the set a degree of
membership ranging (with continuous values) from zero to one. The need for fuzzy sets arises
in situations where it is difficult to determine if an element belongs to a set or not. Classical
sets are special cases of fuzzy sets in which only two degrees of membership (0 and 1) are
allowed. Fuzzy logic has been used to predict species‘ distributions (Robertson et al., 2004),
to detect favorable areas for species (Real et al., 2005, Real et al., 2008, Real et al., 2009), to
analyze gaps in the protection of biodiversity using distribution models (Estrada et al., 2007,
2008, Real et al., 2006b), and to assess the impact of climate change on species‘ distributions
(Levinsky et al., 2007, Real et al., 2010).
186 A. Márcia Barbosa, Neftalí Sillero, Fernando Martínez-Freiría et al.
The theory of complexity can also be used in the interpretation of ENMs. Historical and
ecological factors, together with the idiosyncratic response of each species to these factors,
determine the current configuration of species‘ distributions. These factors dynamically
delimit the biogeographic responses of biodiversity in time and space. In this type of complex
scenario, distributions are the geographical response of the species to the past and present
factors that act(ed) in a particular area. The interpretation of ENMs as the degree of
membership of each locality to the fuzzy set of suitable areas for a species allows obtaining a
measure of how much the species has been influenced towards each location by the various
factors.
One of the most important applications of ENMs is the identification of the species'
ecological niche. Various studies have specifically aimed at identifying the potential
distribution of species and the variables that determine it within the Mediterranean Basin, for
example in Portugal (Brito et al., 1996, Sá-Sousa, 2000, Teixeira et al., 2001, Teixeira and
Ferrand, 2002, Arntzen, 2006), in Spain (Real et al., 2005, Anadón et al., 2006, Roman et al.,
2006, Carretero et al., 2010) and, recently, in Morocco (Beukema et al., 2010). Among these
works, Brito et al.‘s (1996) is the first one on ecological niche modeling within the
Mediterranean Basin, and uses logistic regression to calculate the potential distribution of
Lacerta schreiberi in Portugal. Using similar methodologies, Sá-Sousa (2000) and Teixeira et
al. (2001) analyze, respectively, the biogeography of Podarcis hispanica and Chioglossa
lusitanica in Portugal. Sá-Sousa (2000) separates the two forms of P. hispanica (type 1 and
type 2), confirming that their distributions are parapatric and influenced by different factors.
Teixeira et al. (2001) calculate an ENM of C. lusitanica in Portugal and project it to Spain.
Like Sá-Sousa (2000), Real et al. (2005) use logistic regression to distinguish the niches of
two cryptic species with parapatric distributions: Discoglossus galganoi and D. jeanneae.
Anadón et al. (2006) and Roman et al. (2006) provide examples of modeling applications at
the local level; both studies use GLM, the former for Testudo graeca in the Spanish region of
Ecological Niche Models in Mediterranean Herpetology 187
Murcia, and the latter for Podarcis carbonelli in Doñana (Southern Spain), although without
developing the spatial component of the ENM. Carretero et al. (2010) use Maxent to model
the distribution of the endemic lizard Algyroides marchi at three scales in southern Iberia, and
identify suitable areas and environmental factors related to its presence. Beukema et al.
(2010) combine distributional and genetic data of Salamandra algira in Morocco for studying
its phylogeny and biogeography. They model the distributions of viviparous and oviparous
populations of these species using Maxent, and compare both niches using ENMtools
(Beukema et al., 2010).
Ecological modeling is also used to identify chorotypes, i.e., distinct distribution patterns,
often shared by several species and significantly different from other distribution patterns.
Chorotypes can be determined from the observed distributions of various taxa (e.g. Real et
al., 1992, 1997) or by previously modeling these distributions (e.g. Sillero et al, 2009). Real
et al. (1997) document a gradual longitudinal replacement of reptile species in the eastern part
of the Rif region (northern Morocco). They attribute this to the northward movement of the
Saharan boundaries, which have not yet reached biogeographical equilibrium. Thus, Saharan
reptiles enter the Rif from the east, through the lower basin of the River Moulouya. Seven
reptile chorotypes were identified in the western part of the Rif, and these comprise
Mediterranean species and others endemic to the Maghreb (the region that spans most of
North-western Africa, excluding the Sahara). These chorotypes are segregated from one
another according to altitude. Historical and ecological processes can account for the
distributions shared by these species, which have inhabited the Rif for longer than eastern
reptiles. Sillero et al. (2009) analyze the biogeography of Iberian herpetofauna and identify
seven chorotypes for amphibians and seven for reptiles from the classification of a
presence/absence matrix calculated from ENFA models and environmental variables obtained
from satellite images. These chorotypes separate the species of Atlantic and Mediterranean
affinity. Flores et al. (2004) and Real et al. (2008) use ecological modeling to characterize
environmentally chorotypes previously identified from observed distributions. Aragón et al.
(2010) compare the influence of climatic and non-climatic factors on the distribution of
Iberian species of endotherms and ectotherms. They use GAM and find that amphibians and
reptiles are more influenced by precipitation and temperature than birds and mammals
(Aragón et al., 2010). Rueda et al. (2010) generate analytically derived regionalizations for
multiple groups of European plants and animals and explore potential influences on the
regions for each taxonomic group. They use GLM for modeling the obtained coherent clusters
and identify a discernable biogeographic structure in the European biota, mainly influenced
by climate (Rueda et al., 2010).
ENMs can also be used to study and identify other biogeographical patterns, such as
spatial variations in species' morphology and genetics. The vast majority of species show
geographic differences in morphological traits in response to changing selective pressures of
188 A. Márcia Barbosa, Neftalí Sillero, Fernando Martínez-Freiría et al.
the environments in which they live (Stearns and Hoekstra, 2000, West-Eberhard, 2003), both
biotic (e.g. predation, competition) and abiotic (e.g. temperature, precipitation). The study of
geographic variation in phenotypic traits has been a recurring theme during the second half of
the Twentieth Century (see Thorpe, 1987) and, currently, has resurfaced under a new
approach owing to the use of GIS and ENMs. Within the Mediterranean area, two articles
address the geographical variation in the morphology of Iberian vipers using ENMs as a tool:
Brito et al. (2008) study the morphological variation of Vipera latastei throughout its
distribution range, and Martinez-Freiría et al. (2009) study the morphological variation and
convergence of V. aspis and V. latastei, both along their whole contact area in northeastern
Spain and, on a more local scale, at the contact zone of the upper Ebro River (northern Spain).
Both works employ the same methodology: a combination of geo-statistics and GIS to obtain
uni- and multivariate patterns of geographic variation in morphology. They then use ENMs to
analyze correlations between these patterns and environmental variables. Tomović et al.
(2010) study the morphological variation of V. ammodytes in its European distribution area,
using a similar approach to that of Brito et al. (2008) and Martínez-Freiría et al. (2009), and
identify three morphologically different groups and their climatic requirements using Maxent.
Luiselli (2006) uses logistic regression to examine which environmental factors have led two
species of whip snakes that are not phylogenetically close, Hierophis viridiflavus in Italy
(Europe) and Psammophis phillipsii in Nigeria (Africa), to converge morphologically and
ecologically. Ficetola et al. (2010b) identify the relationship between bioclimatic variables
and body size predicted a priori by several alternative hypotheses for the newt T. carnifex in
Italy. They explore the correlations among these features and use an information theoretic
approach (Akaike‘s information criterion) to select the best model (Ficetola et al., 2010b). In
a similar way, Romano et al. (2010) analyze the relationships between body size and climate
for two sister species of salamander (Salamandrina perspicillata and S. terdigitata) endemic
to Italy, using GLM and an information theoretic approach.
As with morphology, the study of genetic variability and structuring of species may lead
to the identification of geographical patterns (e.g. Alexandrino et al., 2004). However, the
importance of the geographical component in typical studies of phylogeography has been
recognized only recently (see Manel et al., 2003, Kidd and Ritchie, 2006). This new approach
is known as landscape genetics. Broadly speaking, these studies combine GIS and ENMs to
identify geographic patterns in the variation of genetic markers, to delimit genetically similar
groups, to identify routes of interconnectivity between populations, and to study their
relationships with eco-geographical variables (e.g. Spear et al., 2005, Cushman et al., 2006).
However, to our knowledge, no work has yet been published on Mediterranean herptile
species using the methodologies employed in this field.
Alexandrino et al. (2004) combine the study of both morphological and genetic
variations, comparing ENMs obtained for C. lusitanica (Teixeira et al., 2001) with maps of
genetic and phenetic variation. However, this comparison is made in a descriptive way,
without investigating in depth the environmental factors that can be related to both
geographic variations.
Ecological Niche Models in Mediterranean Herpetology 189
Species richness can also be estimated using ENMs, either through direct modeling
(Nogués-Bravo and Martínez-Rica, 2004, Araújo et al., 2008) or through the addition of
individual species‘ ENMs (Soares and Brito, 2007, Estrada et al., 2007, Estrada et al, 2008,
Sillero et al., 2009). In any case, the arithmetic difference between observed and estimated
species richness shows the areas where there may be a deficit in knowledge, that is, where
there are probably species yet to record (Sillero et al., 2009). Thus, ecological modeling is a
useful tool to manage and plan chorological atlas surveys (Loureiro and Sillero, 2010). ENMs
may also allow identifying environmental factors that influence species richness (Soares and
Brito, 2007, Araújo et al., 2008). However, it is necessary to model separately the
distributions of species with different biogeographic affinities (see Nogués-Bravo and
Martínez-Rica, 2004, Ribeiro et al., 2009).
ENMs can be used to identify areas not yet occupied by expanding species (e.g. Hyla
merdionalis in the Iberian Peninsula; Sillero, 2009, Sillero, 2010) or by invasive species (e.g.
Rana catesbeiana and Trachemys scripta in Italy; Ficetola et al., 2007, Ficetola et al., 2009;
Ficetola et al. 2010a).
Sillero (2009) and Sillero (2010) develop ENMs at local scale (Salamanca province,
Spain) and at continental scale (Europe and North Africa) in which they note that H.
meridionalis occupies almost all suitable habitats available. They conclude that this species
cannot expand its distribution much more and that it is valid to assume that is in equilibrium
with the environment.
Ficetola et al. (2007) and Ficetola et al. (2009) use Maxent to identify suitable areas for
the expansion of R. catesbeiana and the reproductive populations of T. scripta in Italy,
respectively. Also with Maxent, Ficetola et al. (2010a) model the historical and current
distributions of R. catesbeiana in Italy to infer how changes in the landscape are involved in
the expansion of this invasive species. Moreover, they use five scenarios of landscape
variation (derived from the ALARM project) to predict the future expansion of this species
(Ficetola et al., 2010a).
ENMs have also been used to model chorotypes of several invasive species in the Iberian
Peninsula, including the red-eared slider Trachemys scripta, determining the most influential
environmental and human factors (Real et al., 2008). Chorotype modeling can aid
conservation or management measures for the entire set of species considered; it can also help
determine the areas that are most prone to invasions (Real et al., 2008).
The "taxonomic inflation‖ (Isaac et al., 2004, Harris and Froufe, 2005) that has affected
both amphibians and reptiles in recent years has led to the subdivision of species into several
new species whose presence records cannot be distinguished a posteriori. Moreover, many of
these species are cryptic (not distinguishable by morphological characters), making the
190 A. Márcia Barbosa, Neftalí Sillero, Fernando Martínez-Freiría et al.
correct identification of these records difficult even after their definition as new species. The
clear-cut distinction of cryptic species often requires genetic or molecular analyses that can be
very expensive and time-consuming. In cases like these, ENMs may help to distinguish the
distributions of such species from the characteristics that define their areas of occurrence.
Starting from the genetic identification of a sample of individuals, we can build ENMs that
allow inferring to which species or subspecific variety other individuals will belong, based on
their location (e.g. lineages of C. lusitanica in Portugal: Arntzen and Alexandrino, 2004; D.
galganoi and D. jeanneae in Spain: Real et al., 2005). Spatial modeling of genetic data can
also serve to compare the niches of distinct populations of a species, as is done by Beukema
et al. (2010) with viviparous and oviparous forms of Salamandra algira in Morocco (see
section 2.1).
Contact zones, mainly those that occur between phylogenetically close species, are very
important in the study of evolutionary processes (Hewitt, 1988). They usually occur at the
limits of species‘ distributions, in areas of environmental transition (ecotones), where
environmental factors and biotic interactions play a major role in the local distribution and
population dynamics of the species in contact (e.g. Martínez-Freiría et al., 2008, Martínez-
Freiría et al., 2010). ENMs allow identifying environmental factors that affect these species,
the responses of each species to the variations in these factors, and the areas where species
can potentially coexist (areas of potential sympatry).
In the study of contact zones, Brito and Crespo (2002) use logistic regression to calculate
for the first time a single ENM for two species simultaneously, with the aim of identifying
environmental requirements and areas of sympatry between Vipera latastei and V. seoanei in
northern Portugal. Since the distributions of both species are parapatric, they calculate two
separate ENMs where the absence data for each species correspond to the presences of the
other species. Espregueira Themudo and Arntzen (2007) use the same methodology to
determine the factors the influence the local distribution of Triturus marmoratus and T.
pygmaeus in the area of Caldas da Rainha (Portugal). However, they only calculate one ENM
(for T. marmoratus). Arntzen and Alexandrino (2004) use logistic regression to analyze the
distribution of C. lusitanica in northern Portugal and obtain several models that identify the
environmental factors more closely related to the distribution of each of the two genetically
distinct groups. Although they do not represent the likely areas of contact, they do identify the
environmental variables to which the two forms have a similar response, and identify a zone
of contact and eco-morphological transition for the two groups (north of the Mondego River,
Portugal). Martínez-Freiría et al. (2008) analyze the distribution of the three Iberian species of
vipers at their contact zone in the upper Ebro River (northern Spain). They use presence data
and Maxent, with which they obtain the responses of these species to environmental factors
and the areas of potential occurrence and sympatry. By representing and comparing the
responses of these species to certain environmental factors common to them, they identify
those factors for which these species show a similar pattern, and that thus allow their
coexistence; and the factors for which the species show different responses, which thus lead
to habitat segregation.
Ecological Niche Models in Mediterranean Herpetology 191
ENMs can also be used to infer the conservation status of species, to identify factors of
threat, and to propose management measures. The earliest such work with Mediterranean
herpetofauna is the one carried out by Brito et al. (1999) on the identification of priority areas
for conservation, delimitation of areas of high extinction risk, assessment of the degree of
protection, and definition of a conservation strategy for L. schreiberi in Portugal. The authors
combine the results obtained with previous works (ENM, Brito et al., 1996, habitat selection,
Brito et al., 1998) with other variables such as the protected area network in Portugal,
detected density, and electrophoretic data of allozymes of this species. Similarly, Teixeira and
Ferrand (2002) identify important areas for the conservation of the genetic diversity of C.
lusitanica by combining 1) ENMs for the current conditions, made with logistic regression
(Teixeira et al., 1996, Teixeira et al., 2001), discriminant analysis, classification trees, and
overlay analysis; 2) ENMs for the years 2050 and 2080, made with logistic regression and
discriminant analysis (Teixeira and Arntzen, 2002); and 3) analysis of the geographic
variation in molecular data (Alexandrino et al., 2004). Santos et al. (2006) and Santos et al.
(2009) use ENFA to identify biotic and abiotic factors involved in the distributions of V.
latastei and C. austriaca, respectively, in the Iberian Peninsula, and evaluate the conservation
status of both species. Santos et al. (2006) realize that V. latastei, though it should be
widespread in the Mediterranean part of the Iberian Peninsula due to its high environmental
adaptability, is relegated to mountainous regions by human activities. Santos et al. (2009)
combine ENMs of C. austriaca at regional scale in Iberia with local analyses of the isolated
populations of southern Spain (performed through intensive sampling), to infer why this
species has small isolated populations in the southern Iberian Peninsula.
Analyses of species richness are very important to determine the areas that should be
protected to preserve the largest possible number of taxa (Soares and Brito, 2007). For
example, Rey Benayas et al. (2006) evaluate the potential impact of future infrastructures on
Spanish herpetofaunal diversity. Estrada et al. (2007) and Estrada et al. (2008) use ENMs in
Andalusia (southern Spain) to assess the degree of agreement between the protected area
network and the important areas for amphibians and reptiles, according to their species
richness, rarity, endemism, and vulnerability. From another point of view, but similar to the
previous one, Ribeiro et al. (2009) use ENMs to evaluate the impact of human activities on
reptile diversity in Catalonia (north-eastern Spain). The authors correlate the differences
between observed and potential species richness (ENMs calculated from 25 species) with
different types of land use; agricultural areas appear to be the least favorable to reptiles,
having the largest differences between observed and potential richness (Ribeiro et al, 2009).
The NIM can be used to infer the state of conservation of the species, to identify factors
of threat and to propose management measures. The earliest work in this area is conducted by
Brown et al. (1999) on the identification of priority conservation areas, delimitation of areas
at high risk of extinction, assessing the degree of protection and definition of a conservation
strategy for L. schreiberi in Portugal. The authors combined the results of previous work
(ENM, Brito et al., 1996, selection of habitats, Brito et al., 1998), along with other variables
such as the network of protected areas in Portugal, density and electrophoretic data detected
of allozymes of the species. Similarly, Teixeira & Ferrand (2002) identified important areas
for conservation of genetic diversity of C. lusitanica by combining 1) ENM for current
conditions, using logistic regression (Teixeira et al., 1996, Teixeira et al., 2001), discriminant
192 A. Márcia Barbosa, Neftalí Sillero, Fernando Martínez-Freiría et al.
analysis, classification trees and overlay analysis, 2) for the years 2050 ENM and 2080, using
logistic regression and discriminant analysis (Teixeira & Arntzen, 2002), and 3) analysis of
geographic variation in molecular data (Alexandrino et al., 2004). Santos et al. (2006) and
Santos et al. (2009) ENFA used to identify biotic and abiotic factors involved in the
distribution of V. latastei and C. Austrian in the Mediterranean Basin, respectively, and
evaluate the conservation status of both species. Santos et al. (2006) found as V. latastei, even
though it should be present in almost all Mediterranean Iberia due to its high environmental
adaptability, is relegated to the mountainous areas due to human activities. Santos et al.
(2009) combined C. ENM Austrian regional scale in Iberia, along with a more local scale
analysis of isolated populations of southern Spain (performed by intensive sampling), to infer
why this species isolated populations with low effective population in the southern half of
peninsular.
Analyses of species richness are of great importance in determining the areas to be
protected and try to preserve the largest possible number of taxa (Soares & Brito, 2007). For
example, Benaiah Rey et al. (2006) evaluated the potential impact of future infrastructure in
Spain Spanish herpetofaunística diversity. Estrada et al. (2007) and Estrada et al. (2008) used
ENM restricted to Andalusia to assess the degree of agreement between the network of
protected areas and important areas for amphibians and reptiles, according to its richness,
rarity, endemism and vulnerability. From another point of view, but like the previous one,
highlights a recent study by Ribeiro et al. (2009), who used ENM to evaluate the impact of
human activities on the diversity of reptiles in Catalonia. The authors correlated the
differences between observed richness and potential richness (ENM calculated from 25
species of reptiles Catalan) with different land uses: agricultural use areas were less favorable
to the reptiles, having the largest differences between wealth (Ribeiro et al, 2009).
The current availability of climatic data for various possible scenarios of future climate
change (e.g. WorldClim: www.worldclim.org/futdown.htm; World Climate Research
Programme: http://ccr.aos.wisc.edu/model/ipcc10min/index.html) has increased the studies
that predict the response of organisms to this process. Due to the strong dependence of
amphibians and reptiles on environmental conditions, these studies can be a valuable tool to
identify the vulnerability of these species to climate change and to develop effective
conservation measures.
In the Mediterranean Basin, the first published ENM predicting the future range of a
species is the one by Teixeira and Arntzen (2002), which focuses on C. lusitanica in Spain
and Portugal. The modeling methods are logistic regression and discriminant analysis, and the
climatic variables include contemporary conditions and the predictions of the International
Panel for Climate Change from the year 2001, which predicted increases in July temperature
of 2 and 3ºC for the years 2050 and 2080, respectively. Whilst climate change does not
consist simply of an increase in temperature, in 2002 there was not enough information on
how precipitation and other variables would be altered. Araújo et al. (2006) use ENMs,
through a combination of GLM, GAM, classification trees and artificial neural networks, to
determine the potential effects of climate change (WorldClim data for the years 2020 and
2050 at a resolution of 50x50 km) on the distributions of European amphibians and reptiles.
Ecological Niche Models in Mediterranean Herpetology 193
Similarly, Carvalho et al. (2010) model the distributions of 37 Iberian endemics or quasi-
endemics (15 amphibians and 22 reptiles) under current weather conditions, and project them
to the future climatic conditions predicted for the years 2020, 2050, and 2080. This study uses
a finer scale (10x10 km) and combines various modeling methods, such as Maxent, GLM,
GAM, classification trees, artificial neural networks, Generalized Boosting Models (GBM),
random forests, mixed discriminant analysis, and MARS. Real et al. (2010) and Márquez et
al. (2011) evaluate the predicted effect of different greenhouse gas emission scenarios on the
distributions of Alytes dickhilleni and Vipera latastei in continental Spain using two global
circulation models. Ficetola et al. (2010a) forecast the future expansion of invasive R.
catesbeiana in Italy under five landscape variation scenarios (see section 2.5).
ACKNOWLEDGMENTS
A version of this review focused on the Iberian Peninsula was previously published in
Spanish in the Boletín de la Asociación Herpetológica Española (Bulletin of the Spanish
Herpetological Association). We thank its editors Xavier Santos and Alexander Richter-Boix
for permission to reproduce it here in English. A.M.B., N.S. and F.M.-F. are supported by
post-doctoral fellowships (SFRH/BPD/26666/2006, SFRH/BPD/40387/2007 and
SFRH/BPD/69857/2010) from Fundação para a Ciência e a Tecnologia (Portugal), co-
financed by the European Social Fund. The ‗Rui Nabeiro‘ Biodiversity Chair is financed by
Delta Cafés.
REFERENCES
Albert, CH, Thuiller, W. 2008. Favourability functions versus probability of presence:
advantages and misuses. Ecography, 31: 417-422.
Alexandrino, J., Teixeira, J., Arntzen, JW, Ferrand, N. 2004. Historical biogeography and
conservation of the golden-striped salamander ( Chioglossa lustanica ) in northwestern
Iberia: integrating ecological, phenotypic and phylogeographic data. 189-205. In : Weiss,
S., Ferrand, N. (eds.), Phylogeography of Southern European Refugia. Netherlands.
Springer.
Anadon, JD, Gimenez, A., Martinez, M., Martinez, J., Perez, I., Esteve, MA 2006. Factors
determining the distribution of the spur-thighed tortoise Testudo graeca in south-east
Spain: a hierarchical approach. Ecography, 29: 339-346.
Anadón, JD, Giménez, A., Ballestar, R. 2010. Linking local ecological knowledge and
habitat modelling to predict absolute species abundance at large scales. Biodiversity and
Conservation , 19: 1443-1454.
Anderson, RP, Peterson, AT, Gomez Laverde, M. 2002. Using niche-based GIS modeling to
test geographic predictions of competitive exclusion and competitive release in South
American pocket mice. Oikos, 98: 3-16.
Aragón, P, Lobo, JM, Olalla-Tárraga, MA, Rodríguez, MA 2010. The contribution of
contemporary climate to ectothermic and endothermic vertebrate distributions in a glacial
refuge. Global Ecology and Biogeography 19: 40–49.
Araújo, MB, Guisan, A. 2006. Five (or so) challenges for species distribution modelling.
Journal of Biogeography, 33: 1677-1688.
Araújo, MB, New, M. 2007. Ensemble forecasting of species distributions. Trends in
Ecology, Evolution, 22: 42-47.
Araújo, MB, Nogués-Bravo, D., Diniz-Filho, JAF, Haywood, AM, Valdes, PJ, Rahbek, C.
2008. Quaternary climate changes explain diversity among reptiles and amphibians.
Ecography, 31: 8-15.
Araújo, MB, Pearson, RG 2005. Equilibrium of species' distributions with climate.
Ecography, 28: 693-695.
Araújo, MB, Thuiller, W., Pearson, RG 2006. Climate warming and the decline of
amphibians and reptiles in Europe. Journal of Biogeography, 33: 1712-1728.
Ecological Niche Models in Mediterranean Herpetology 195
Araújo, MB, Whittaker, RJ, Ladle, RJ, Erhard, M. 2005. Reducing uncertainty in projections
of extinction risk from climate change. Global Ecology and Biogeography, 14: 529-538.
Araújo, MB, Williams, PH 2000. Selecting areas for species persistence using occurrence
data. Biological Conservation, 96: 331-345.
Arntzen, JW 2006. From descriptive to predictive distribution models: a working example
with Mediterranean amphibians and reptiles. Frontiers in Zoology, 3.
Arntzen, JW, Alexandrino, J. 2004. Ecological modelling of genetically differentiated forms
of the Mediterranean endemic golden-striped salamander, Chioglossa Lusitanica.
Herpetological Journal, 14: 137-141.
Arntzen, JW, Espregueira Themudo, G. 2008. Environmental parameters that determine
species geographical range limits as a matter of time and space. Journal of Biogeography,
35: 1177-1186.
Arntzen, JW, Teixeira, J. 2006. History and new developments in the mapping and modelling
of the distribution of the golden-striped salamander, Chioglossa lusitanica. Zeitschrift für
Feldherpetologie, Supplement : 1-14.
Aspinall, RJ, Matthews, K. 1994. Climate change impact on distribution and abundance of
wildlife: An analytical approach using GIS. Environment and Pollution , 86: 217-223
Austin, MP 2002. Spatial prediction of species distribution: an interface between ecological
theory and statistical modelling. Ecological Modelling, 157: 101-118.
Austin, MP 2007. Species distribution models and ecological theory: a critical assessment and
some possible new approaches. Ecological Modelling, 200: 1-19.
Barbosa, AM, Real, R., Vargas, JM 2009. Transferability of environmental favourability
models in geographic space: The case of the Mediterranean desman ( Galemys
pyrenaicus ) in Portugal and Spain. Ecological Modelling, 220: 747-754.
Barbosa AM, Real R., Vargas JM 2010. Use of coarse-resolution models of species'
distributions to guide local conservation inferences. Conservation Biology , early view.
doi: 10.1111/j.1523-1739.2010.01517.x.
Bárcena, S, Real, R, Olivero, J and Vargas, JM. 2004. Latitudinal trends in breeding
waterbird species richness in Europe and their environmental causes. Biodiversity and
Conservation, 13: 1997-2014
Beaumont, LJ, Gallagher, RV, Thuiller, W., Downey, PO, Leishman, MR, Hughes, L. 2009.
Different climatic envelopes among invasive populations may lead to underestimations of
current and future biological invasions. Diversity and Distributions, 15: 409-420.
Beukema, W, de Pous, P, Donaire, D, Escoriza, D, Bogaerts, S, Toxopeus, AG, de Bie,
CAJM, Roca, J, Carranza, S. 2010. Biogeography and contemporary climatic
differentiation among Moroccan Salamandra algira. Biological Journal of the Linnean
Society, 101: 626–641.
Betts, MG, Diamond, AW, Forbes, GJ, Villard, MA, Gunn, JS 2006. The importance of
spatial autocorrelation, extent and resolution in predicting forest bird occurrence.
Ecological Modelling, 191: 197-224.
Bombi, P.; Salvi, D.; Vignoli, L., and Bologna, M. A. Modelling Bedriaga's rock lizard
distribution in Sardinia: An ensemble approach. Amphibia-Reptilia. 2009; 30:413-424.
Brito, JC, Brito-e-Abreu, F., Paulo, OS, Rosa, HD, Crespo, EG 1996. Distribution of
Schreiber's green lizard ( Lacerta schreiberi ) in Portugal: a predictive model.
Herpetological Journal, 6: 43-47.
196 A. Márcia Barbosa, Neftalí Sillero, Fernando Martínez-Freiría et al.
Brito, JC, Crespo, EG 2002. Distributional analysis of two vipers ( Vipera latastei and V.
seoanei ) in a potential area of sympatry in the Northwestern Mediterranean Basin. 129-
138. In : Schuett, GW, Hoggren, M., Douglas, ME, Greene, HW (eds.), Biology of the
Vipers. Eagle Mountain, Utah. Eagle Mountain Publishing, LC.
Brito, JC, Crespo, EG, Paulo, OS 1999. Modelling wildlife distributions: Logistic Multiple
Regression vs Overlap Analysis. Ecography, 22: 251-260.
Brito, JC, Godinho, R., Luís, C., Paulo, OS, Crespo, EG 1999. Management strategies for
conservation of the lizard Lacerta schreiberi in Portugal. Biological Conservation, 89:
311-319.
Brito, JC, Santos, X., Pleguezuelos, JM, Sillero, N. 2008. Inferring Evolutionary Scenarios
with Geostatistics and Geographical Information Systems (GIS) for the viperid snakes
Vipera latastei and V. monticola. Biological Journal of the Linnean Society, 95: 790-806.
Bulluck, L., Fleishman, E., Betrus, C., Blair, R. 2006. Spatial and temporal variations in
species occurrence rate affect the accuracy of occurrence models. Global Ecology and
Biogeography, 15: 27-38.
Busby, JR 1991. BIOCLIM - A Bioclimatic Analysis and Prediction System. 64-68. In :
Margules, CR, Austin MP (eds.), Nature Conservation: Cost Effective Biological Surveys
and Data Analysis. CSIRO. Canberra.
Carpenter, G., Gillison, AN, Winter, J. 1993. DOMAIN: a flexible modelling procedure for
mapping potential distributions of plants and animals. Biodiversity and Conservation, 2:
667-680.
Carretero, MA, Sillero, N., Ayllón, E., Kaliontzopoulou, A., Lima, A., Hernández-Sastre, PL,
Godinho, R., Harris, DJ 2008. Multidisciplinary approaches for conserving Southern
isolates of Atlantic lizards in the Mediterranean Basin. Herpetologia Sardiniae : 251-255.
Carretero, MA, Ceacero, F, García-Muñoz, E, Sillero, N, Olmedo, MI, Hernández-Sastre, PL,
Rubio, JL 2010. Seguimiento de Algyroides marchi. Informe final. Monografías SARE.
Asociación Herpetológica Española – Ministerio de Medio Ambiente y Medio Rural y
Marino. Madrid.
Cartron, JL. E., Kelly, JF, Brown, JH 2000. Constraints on patterns of covariation: a case
study in strigid owls. Oikos , 90: 381-389.
Cohen, JA 1960. A coefficient of agreement for nominal scales. Educational and
Psychological Measurement, 20: 37-46.
Colwell, RK, Rangel, TF 2009. Hutchinson's duality: The once and future niche. Proceedings
of the National Academy of Sciences, 106: 19651-19658.
Chefaoui, RM, Lobo, JM 2008. Assessing the effects of pseudo-absences on predictive
distribution model performance. Ecological Modelling, 210: 478-486.
Diniz-Filho, JAF, Bini, LM, Hawkins, BA 2003. Spatial autocorrelation and red herrings in
geographical ecology. Global Ecology and Biogeography, 12: 53-64.
Dormann, CF 2007. Effects of incorporating spatial autocorrelation into the analysis of
species distribution data. Global Ecology and Biogeography, 16: 129-138.
Dormann, CF, McPherson, JM, Araújo, MB, Bivand, R., Bolliger, J., Carl, G., Davies, R.,
Hirzel, A., Jetz, W., Kissling, W., Kuhn, I., Ohlemuller, R., Peres-Neto, P., Reineking,
B., Schroder, B., Schurr, FM, Wilson, R. 2007. Methods to account for spatial
autocorrelation in the analysis of species distributional data: a review. Ecography, 30:
609-628.
Ecological Niche Models in Mediterranean Herpetology 197
Elith, J., Graham, C., Anderson, R., Dudik, M., Ferrier, S., Guisan, A., Hijmans, R.,
Huettmann, F., Leathwick, J., Lehmann, A., Li, J., Lohmann, L., Loiselle, B., Manion,
G., Moritz, C., Nakamura, M., Nakazawa, Y., Overton, J., Peterson, AT, Phillips, SJ,
Richardson, K., Scachetti-Pereira, R., Schapire, R., Soberón, J., Williams, S., Wisz, M.,
Zimmermann, NE 2006. Novel methods improve prediction of species' distributions from
occurrence data. Ecography, 29: 129-151.
Elith, J., Graham, CH 2009. Do they? How do they? Why do they differ? On finding reasons
for differing performances of species distribution models. Ecography, 32: 66-77.
Elton, C. 1927. Animal Ecology. London. Sedgwick and Jackson.
Engler, R., Randin, CF, Vittoz, P., Czáka, T., Beniston, M., Zimmermann, NE, Guisan, G.
2009. Predicting future distributions of mountain plants under climate change: does
dispersal capacity matter? Ecography, 32: 34-45 .
Espregueira Themudo GE, Arntzen JW. 2007. Newts under siege: range expansion of
Triturus pygmaeus isolates populations of its sister species. Diversity and Distributions,
13: 580-586.
Estrada A, Márquez AL, Real R, Vargas JM. 2007. Utilidad de los espacios naturales
protegidos de Andalucía para preservar la riqueza de especies de anfibios. Munibe , 25:
74-81.
Estrada A, Márquez AL, Real R, Vargas JM. 2008. ¿En qué medida preservan los espacios
naturales la riqueza de reptiles en Andalucía? 367-373. In : Redondo, MM, Palacios, MT,
López, FJ, Santamaría, T., Sánchez, D. (eds.), Avances en Biogeografía. Departamento
Análisis Geográfico Regional y Geografía Física. Universidad Complutense de Madrid.
Madrid.
Etherington TR, Ward AI, Smith, G. C., Pietravalle, S., Wilson, G. J. 2009. Using the
Mahalanobis distance statistic with unplanned presence-only survey data for
biogeographical models of species distribution and abundance: a case study of badger
setts. Journal of Biogeography, 36:845-853.
Ficetola, GF, Thuiller, W., Miaud, C. 2007. Prediction and validation of the potential global
distribution of a problematic alien invasive species: the American bullfrog. Diversity and
Distributions, 13: 476-485.
Ficetola, GF, Thuiller, W., Padoa-Schioppa, E. 2008. From introduction to the establishment
of alien species: a preliminary analysis of bioclimatic differences between presence and
reproduction localities in Trachemys scripta. Herpetologia Sardiniae : 266-269.
Ficetola, GF, Thuiller, W., Padoa-Schioppa, E. 2009. From introduction to the establishment
of alien species: bioclimatic differences between presence and reproduction localities in
the slider turtle. Diversity and Distributions, 15: 108-116.
Ficetola, GF, Maiorano, L, Falcucci, A, Dendoncker, N, Boitani, L, Padoa-Schioppa, E,
Miaud, C, Thuiller, W 2010a. Knowing the past to predict the future: land-use change
and the distribution of invasive bullfrogs. Global Change Biology, 16: 528–537.
Ficetola, GF, Scali, S, Denoël, M, Montinaro, G, Vukov, TD, Zuffi, MAL, Padoa-Schioppa,
E. 2010b. Ecogeographical variation of body size in the newt Triturus carnifex:
comparing the hypotheses using an information-theoretic approach. Global Ecology and
Biogeography, 19: 485–495.
Fielding, AH, Bell, JF 1997. A review of methods for the assessment of prediction errors in
conservation presence/absence models. Environmental Conservation, 24: 38-49.
198 A. Márcia Barbosa, Neftalí Sillero, Fernando Martínez-Freiría et al.
Flores, T., Puerto, MA, Barbosa, AM, Real, R., Gosalvez, RU 2004. Agrupación en corotipos
de los anfibios de la provincia de Ciudad Real (España). Revista Española de
Herpetologia, 18: 41-53.
Godsoe, W. 2010. I can't define the niche but I know it when I see it: a formal link between
statistical theory and the ecological niche. Oikos, 1: 53-60.
Grinnell, J. 1917. The niche-relationships of the California Thrasher. Auk, 34: 427-433.
Guisan, A., Lehmann, A., Ferrier, S., Austin, M., Overton, J.Mc.C., Aspinall, R., Hastie, T.
2006. Making better biogeographical predictions of species' distributions. Journal of
Applied Ecology, 43: 386-392.
Guisan, A., Thuiller, W. 2005. Predicting species distribution: offering more than simple
habitat models. Ecology Letters, 8: 993-1009.
Guisan, A., Zimmermann, NE 2000. Predictive habitat distribution models in ecology.
Ecological Modelling, 135: 147-186.
Harris, DJ, Froufe, E. 2005. Taxonomic inflation: species concept or historical geopolitical
bias? Trends in Ecology, Evolution, 20: 6-7.
Hastie, TJ, Tibshirani, RJ 1990. Generalized Additive Models. Chapman, Hall. London. New
York.
Hernandez, PA, Graham, CH, Master, LL, Albert, DL 2006. The effect of sample size and
species characteristics on performance of different species distribution modeling
methods. Ecography, 29: 773-785.
Hewitt, GM 1996. Some genetic consequences of ice ages, and their role in divergence and
speciation. Biological Journal of the Linnean Society, 58: 247-276.
Hirzel, AH, Guisan, A. 2002. Which is the optimal sampling strategy for habitat suitability
modelling? Ecological Modelling, 157: 331-341.
Hirzel, AH, Hausser, J., Chessel, D., Perrin, N. 2002. Ecological-niche factor analysis: how to
compute habitat suitability maps without absence-data? Ecology, 83: 2027-2036.
Hirzel, AH, Helfer, V., Metral, F. 2001. Assessing habitat-suitability models with a virtual
species. Ecological Modelling, 145: 111–121.
Hirzel, AH, Le Lay, G. 2008. Habitat suitability modelling and niche theory. Journal of
Applied Ecology, 45: 1372-1381.
Holt, RD 2003. On the evolutionary ecology of species' ranges. Evolutionary Ecology
Research, 5: 159–178.
Hosmer, DWJ, Lemeshow, S. 1989. Applied Logistic Regression. John Wiley, Sons. New
York.
Hutchinson, GE 1957. Concluding remarks. Cold Spring Harbour symposium on quantitative
biology : 415-427.
Isaac, NJB, Mallet, J., Mace, GM 2004. Taxonomic inflation: its influence on macroecology
and conservation. Trends in Ecology, Evolution, 19: 464-469.
Jackson, ST, Overpeck, JT 2000. Responses of Plant Populations and Communities to
Environmental Changes of the Late Quaternary. Paleobiology 26: 194-220.
Jiménez-Valverde, A., Lobo, JM, 2007. Threshold criteria for conversion of probability of
species presence to either–or presence–absence. Acta Oecologica, 31: 361-369.
Jiménez-Valverde, A., Lobo, JM, Hortal, J. 2008. Not as good as they seem: the importance
of concepts in species distribution modelling. Diversity and Distributions, 14: 885-890.
Kaliontzopoulou, A., Brito, JC, Carretero, MA, Larbes, S., Harris, DJ 2008. Modelling the
partially unknown distribution of wall lizards (Podarcis) in North Africa: ecological
Ecological Niche Models in Mediterranean Herpetology 199
Phillips, SJ, Dudík, M., Elith, J., Graham, CH, Lehmann, A., Leathwick, J., Ferrier, S. 2009.
Sample selection bias and presence-only distribution models: implications for
background and pseudo-absence data. Ecological Applications, 19: 181-197.
Phillips, SJ, Dudík, M., Schapire, RE 2004. A maximum entropy approach to species
distribution modeling. Proceedings of the Twenty-First International Conference on
Machine Learning, 655-662.
Pleguezuelos, JM, Márquez, R., Lizana, M. 2002. Atlas de distribución y Libro Rojo de los
Anfibios y Reptiles de España. Dirección General de Conservación de la Naturaleza-
Asociación Herpetológica Española, 2ª impresión. Madrid.
Prinzing, A., Durka, W., Klotz, S., Brandl, R. 2002. Geographic variability of ecological
niches of plant species: are competition and stress relevant? Ecography, 25: 721-729.
Pulliam, HR 1988. Sources, Sinks, and Population Regulation. The American Naturalist, 132:
652-661.
Pulliam, HR 2000. On the relationship between niche and distribution. Ecology Letters, 3:
349-361.
Real, R. 1992. Las tendencias geográficas de la riqueza específica. 85-94. In : Vargas, JM,
Real, R., Antúnez, A. (eds.), Objetivos y método biogeográficos. Aplicaciones en
Herpetología. Monografías Herpetológicas. Asociación Herpetológica Española.
Valencia.
Real, R., Barbosa, AM, Martínez-Solano, I., Garcia-Paris, M. 2005. Distinguishing the
distributions of two cryptic frogs (Anura: Discoglossidae) using molecular data and
environmental modeling. Canadian Journal of Zoology, 83: 536-545.
Real, R., Barbosa, AM, Rodríguez, A., García, FJ, Vargas, JM, Palomo, LJ, Delibes, M.
2009. Conservation biogeography of ecologically interacting species: the case of the
Mediterranean lynx and the European rabbit. Diversity and Distributions, 15: 390-400.
Real, R., Barbosa, AM, Vargas, JM 2006. Obtaining environmental favourability functions
from logistic regression. Environmental and Ecological Statistics, 13: 237-245.
Real, R., Estrada, A., Barbosa, AM, Vargas, JM, 2006b. Aplicación de la lógica difusa al
concepto de rareza para su uso en Gap Analysis : el caso de los mamíferos terrestres en
Andalucía. Serie Geográfica 13, 99-116. http://www.geogra.uah.es/inicio/revista/index-
13.php
Real, R, Pleguezuelos, JM and Fahd, S. 1997. The distribution patterns of reptiles in the Riff
region, Northern Morocco. African Journal of Ecology, 35: 312-325
Real, R., Marquez, AL, Estrada, A., Munoz, AR, Vargas, JM 2008. Modelling chorotypes of
invasive vertebrates in mainland Spain. Diversity and Distributions, 14: 364-373.
Real, R., Márquez, AL, Olivero, J., Estrada, A., 2010. Species distribution models in climate
change scenarios are still not useful for informing policy planning: an uncertainty
assessment using fuzzy logic. Ecography 33, 304-314.
Real, R., Vargas, JM, Guerrero, JC 1992. Análisis biogeográfico de clasificación de áreas y
de especies. 73-84. In : Vargas, JM, Real, R., Antúnez, A. (eds.), Objetivos y método
biogeográficos. Aplicaciones en Herpetología. Monografías Herpetológicas. Asociación
Herpetológica Española. Valencia.
Reese, GC, Wilson, KR, Hoeting, JA, Flather, CH 2005. Factors affecting species distribution
predictions: a simulation modeling experiment. Ecological Applications, 15: 554-564.
202 A. Márcia Barbosa, Neftalí Sillero, Fernando Martínez-Freiría et al.
Rey Beneyas, JM, Montana, E., Belliure, J., Eekhout, XR 2006. Identifying areas of high
herpetofauna diversity that are threatened by planned infrastructure projects in Spain.
Journal of Environmental Management, 79: 3.
Ribeiro, R., Santos, X., Sillero, N., Carretero, MA, Llorente, GA 2009. Biodiversity and Land
Uses: Is Agriculture the Biggest Threat for reptiles' assemblages? Acta Oecologica, 35:
327-334.
Robertson, MP, Villet, MH, Palmer, AR, 2004. A fuzzy classification technique for
predicting species´ distributions: applications using invasive alien plants and indigenous
insects. Diversity and Distributions 10, 461-474.
Román, R., Ruiz, G., Delibes, M., Revilla, E. 2006. Factores ambientales condicionantes de la
presencia de la lagartija de Carbonell Podarcis carbonelli (Pérez–Mellado, 1981) en la
comarca de Doñana. Animal Biodiversity and Conservation, 29: 73-82.
Romano, A, Ficetola, GF. 2010. Ecogeographic variation of body size in the spectacled
salamanders (Salamandrina): influence of genetic structure and local factors. Journal of
Biogeography, 37: 2358-2370.
Romero, J., Real, R. 1996. Macroenvironmental factors as ultimate determinants of the
distribution of common toad and netterjack toad in the south of Spain. Ecography , 19:
305-312.
Rueda, M, Rodríguez, MA, Hawkins, BA. 2010. Towards a biogeographic regionalization of
the European biota. Journal of Biogeography, 37: 2067-2076.
Rushton, SP, Ormerod, SJ, Kerby, G. 2004. New paradigms for modelling species
distributions? Journal of Applied Ecology, 41: 193-200.
Rödder, D., Lötters, S. 2009. Niche shift versus niche conservatism? Climatic characteristics
of the native and invasive ranges of the Mediterranean house gecko ( Hemidactylus
turcicus ). Global Ecology and Biogeography, 18: 674-687.
Sa-Sousa, P. 2000. A predictive distribution model for the Mediterranean wall lizard (
Podarcis hispanicus ) Portugal. Herpetological Journal, 10: 1-11.
Santos, X., Brito, JC, Caro, J., Abril, AJ, Lorenzo, M., Sillero, N., Pleguezuelos, JM 2009.
Habitat suitability, threats and conservation of isolated populations of the smooth snake (
Coronella austriaca ) in the southern Mediterranean Basin. Biological Conservation,
142: 344-352.
Santos, X., Brito, JC, Sillero, N., Pleguezuelos, JM, Llorente, GA, Fahd, S., Parellada, X.
2006. Inferring habitat-suitability areas with ecological modelling techniques and GIS: A
contribution to assess the conservation status of Vipera latastei. Biological Conservation,
130: 416-425.
Segurado, P., Araújo, MB, Kunin, WE 2006. Consequences of spatial autocorrelation for
niche-based models. Journal of Applied Ecology, 43: 433-444.
Shugart, HH 1990. Using ecosystem models to assess potential consequences of global
climatic change. Trends in Ecology and Evolution , 5: 303-307.
Sillero, N. 2009. Potential distribution of the new populations of Hyla meridionalis in
Salamanca (Spain). Acta Herpetologica, 4: 83-98.
Sillero, N. 2010. Modelling new suitable areas for Hyla meridionalis in a current and future
expansion scenario. Amphibia-Reptilia, 31: 37-50.
Sillero, N., Brito, JC, Toxopeus, B., Skidmore, AK 2009. Biogeographical patterns derived
from remote sensing variables: the amphibians and reptiles of the Mediterranean Basin.
Amphibia-Reptilia, 30: 185-206.
Ecological Niche Models in Mediterranean Herpetology 203
Sillero, N., Celaya, L., Martín-Alfageme, S. 2005. Using GIS to Make An Atlas: A Proposal
to Collect, Store, Map and Analyse Chorological Data for Herpetofauna. Revista
Española de Herpetologia, 19: 87-101.
Sillero, N., Tarroso, P. 2010. Free GIS for herpetologists: free data sources on Internet and
comparison analysis of proprietary and free/open source software. Acta Herpetologica, 5:
63-85.
Soares, C., Brito, JC 2007. Environmental correlates for species richness and biogeographic
relationships among amphibians and reptiles in a climate transition area. Biodiversity and
Conservation, 16: 1087-1102.
Soberón, J. 2007. Grinnellian and Eltonian niches and geographic distributions of species.
Ecology Letters, 10: 1115-1123.
Soberón, J. 2010. Niche and area of distribution modeling: a population ecology perspective.
Ecography, 33: 159-167.
Soberón, J., Nakamura, M. 2009. Niches and distributional areas: Concepts, methods, and
assumptions. Proceedings of the National Academy of Sciences, 106: 19644-19650.
Soberón, J., Peterson, AT 2005. Interpretation of Models of Fundamental Ecological Niches
and Species Distributional Areas. Biodiversity Informatics, 2.
Stearns, SC, Hoekstra, R. 2000. Evolution: an introduction. Oxford University Press. Oxford.
Stockwell, DRB, Noble, IR 1992. Induction of sets of rules from animal distribution data: A
robust and informative method of data analysis. Mathematics and Computers in
Simulation, 33: 385-390.
Stockwell, DRB, Peterson, AT 2002. Effects of sample size on accuracy of species
distribution models. Ecological Modelling, 148: 1-13.
Stoms, DM, Davis, FW, Cogan, CB 1992. Sensitivity of wildlife models to uncertainties in
GIS data. Photogrammetric Engineering, Remote Sensing , 58: 843-850.
Sykes, MT, Prentice, IC, Cramer, W. 1996. A bioclimatic model for the potential distribution
of north European tree species under present and future climates. Journal of
Biogeography , 23: 203-233.
Teixeira, J., Arntzen, JW 2002. Potential impact of climate warning on the distribution of the
Golden-striped salamander, Chioglossa lusitanica , on the Mediterranean Basin.
Biodiversity and Conservation, 11: 2167-2176.
Teixeira, J., Ferrand N. 2002. The application of distribution models and Geographical
Information Systems for the study of biogeography and conservation of herpetofauna:
Chioglossa lusitanica as a case study. Revista Española de Herpetologia, Vol. Especial:
119-130.
Teixeira, J., Ferrand, N., Arntzen, JW 2001. Biogeography of the golden-striped salamander
Chioglossa lusitanica : a field survey and spatial modelling approach. Ecography, 24:
614-624.
Thorpe, RS 1987. Geographic variation: a synthesis of cause, data, pattern and congruence in
relation to subspecies, multivariate analysis and phylogenesis. Bollettino di Zoologia, 54:
3-11.
Thuiller, W., Brotons, L., Araújo, MB, Lavorel, S. 2004. Effects of restricting environmental
range of data to project current and future species distributions. Ecography, 27: 165-172 .
Tomović, L, Crnobrnja-Isailović, J, Brito, JC. (2010): Geostatistics and Geographical
Information Systems uncover the evolutionary history of the nose-horned viper (Vipera
ammodytes) on the Balkans. Biological Journal of the Linnean Society, 101: 651-666.
204 A. Márcia Barbosa, Neftalí Sillero, Fernando Martínez-Freiría et al.
VanDerWal, J., Shoo, LP, Graham, C., Williams, SE 2009. Selecting pseudo-absence data for
presence-only distribution modeling: How far should you stray from what you know?
Ecological Modelling, 220: 589-594.
Vieites, DR, Nieto-Román, S., Wake, DB 2009. Reconstruction of the climate envelopes of
salamanders and their evolution through time. Proceedings of the National Academy of
Sciences, 106: 19715-19722.
Walker, PA, Cocks, KD 1991. HABITAT: a procedure for modelling a disjoint environmental
envelope for a plant or animal species. Global Ecology and Biogeography Letters, 1:
108-118.
West-Eberhard M.-J. 2003. Developmental plasticity and evolution. Oxford University Press,
Oxford.
Wiens, JA, Stralberg, D., Jongsomjit, D., Howell, CA, Snyder, MA 2009. Niches, models,
and climate change: Assessing the assumptions and uncertainties. Proceedings of the
National Academy of Sciences, 106: 19729-19736.
Wiens, JJ, Graham, CH 2005. Niche conservatism: Integrating Evolution, Ecology, and
Conservation Biology. Annual Review of Ecology, Evolution, and Systematics, 36: 519-
539.
Wisz, MS, Hijmans, RJ, Li, J., Peterson, AT, Graham, CH, Guisan, A., NCEAS Predicting
Species Distributions Working Group. 2008. Effects of sample size on the performance
of species distribution models. Diversity and Distributions, 14: 763-773.
Woodward, FI, Cramer, W. 1996. Plant functional types and climatic changes: Introduction.
Journal of Vegetation Science, 7: 306-308.
Zadeh, LA 1965. Fuzzy sets. Information and Control, 8: 338–353.
Zuur, AF, Ieno, EN, Walker, N., Saveliev, AA, Smith, GM 2009. Mixed Effects Models and
Extensions in Ecology with R. Springer, New York.
In: Ecological Modeling ISBN: 978-1-61324-567-5
Editor: WenJun Zhang, pp. 205-222 © 2012 Nova Science Publishers, Inc.
Chapter 9
ABSTRACT
Numerical techniques (e.g. correlation, multiple regression and factor analysis, path
analysis, methods of network analysis, and, in particular, simulation modelling) may be
very helpful in investigations of indirect relationships in aquatic ecosystems. Here we
give a brief overview of some examples of the relevant studies, and focus on 1) a case
study of a freshwater eutrophic lake, where statistical analysis of the datasets obtained
within a comprehensive monitoring programme, and sensitivity analysis by a
mathematical model ‗Rostherne‘, helped to reveal the previously overlooked
relationships between Si and P biogeochemical cycles coupled through the dynamics of
primary producers, and 2) give an overview of how the coupling of physical, chemical,
and biological processes in the marine ecosystem models offers a basis for investigations
of indirect interactions in continental shelf seas. Complex aquatic ecosystem models
provide a numerical simulation of biogeochemical fluxes underpinned by coupling
* Corresponding author. Present address: SBE, Institute for infrastructure and environment, Heriot-Watt University,
Edinburgh EH14 4AS, UK.
E-mail: e96kri69@netscape.net.
206 V. Krivtsov and C. F. Jago
physical forcing functions with definitions simulating biological and chemical processes,
and offer a potential for quantitative interpretation of sediment proxies in the stratigraphic
record. Combination of models and sediment proxies, calibrated by training sets, can
provide information on water column structure, surface heating, mixing, and water depth,
thus providing a basis for reconstruction of the past, and predicting the future
environmental dynamics.
INTRODUCTION
Natural ecosystems are complex, and are characterised by a multitude of interconnected
relationships between ecosystem components (Krivtsov, 2008). The understanding of these
complex interactions is paramount for sustainable management of environmental resources.
However, ecological research mainly tends to concentrate on investigations of direct
relationships, whilst indirect interactions (and especially the less obvious, e.g. the delayed
ones) are often overlooked or understudied (NB In this paper all the relations not restricted to
the effects of a direct transaction of matter and energy between the adjacent ecosystem
components will be treated as indirect). Mathematical techniques (e.g. correlation, multiple
regression and factor analysis, simulation modelling, path analysis and methods of network
analysis, etc.) may be very helpful in investigations of indirect relationships in ecosystems.
The origin of algorithms capable of studying indirect interactions within ecological
context can be traced back to the 18th century (Krivtsov, 2004). Here we refer to some
examples of the relevant aquatic studies, and, in particular, give a brief account how
mathematical techniques have been helpful in investigating indirect effects in a freshwater
eutrophic lake (see Krivtsov et al., 1998, 1999a,b,c, 2001, and references therein) where
indirect relationships appeared to occur on (and across) various levels of organisation. We
also give an overview of how the coupling of physical, chemical, and biological processes in
the models of continental shelf seas gives a potential for a breakthrough in the investigations
of indirect effects. Finally, we argue, that due to the importance of aquatic ecosystems, the
existence of comprehensive models linking physical, chemical, and biological processes, and
the experience accumulated in studies of indirect effects in the aquatic environment, complex
aquatic models are likely to continue to play a leading role in the investigations of indirect
effects, in particular within the framework of the comparative theoretical ecosystem analysis
(sensu Krivtsov, 2002, 2004).
Hulot and coauthors (Hulot et al.. 2000) compared the performance of linear food chain
models and an intermediate complexity model, applied to data of a mesocosm experiment
simulating lake nutrient enrichment. The intermediate complexity model (with separation of
trophic levels into functional groups according to size and diet) was the only one which
performed satisfactory, thus highlighting the importance of functional diversity and indirect
interactions.
Another modelling study (Loladze et al., 2000) investigated how the interactions between
phytoplankton and zooplankton change if the Lotka-Volterra model incorporates chemical
heterogeneity for both trophic levels. It was found that indirect competition between two
populations for P can shift the relationship from a usual (+, −) type to an unusual (−, −) type,
leading to a very complex overall dynamics.
Structural equation modelling (a technique combining path, factor, and regression
analyses) was used by Malaeb et al. (Malaeb et al., 2000) to estimate the contribution of
indirect effects of sediment contamination and natural variability on biodiversity and growth
potential in a selection of North American estuaries. They found that a positive indirect effect
of natural variability (mediated through biodiversity) on growth potential exceeded a direct
negative effect, resulting in the overall positive relationship.
A simulation model of the Mediterranean infralittoral rocky bottom was used by
McClanahan and Sala, 1997) to study possible effects of various management options.
Running a number of ‗what–if‘ scenarios they concluded that many of potential changes are
likely to be indirect effects caused by changes in trophic composition. For example, if
invertivorous fish were removed as part of a management scenario, sea urchins would reduce
algal abundance and primary production, leading to competitive exclusion of herbivorous
fish. Although similar interactions were known from tropical seas, these results were not
anticipated by previous field studies in the Mediterranean.
Carrer and Opitz (Carrer and Opitz, 1999) investigated indirect interactions in the Lagoon
of Venice using ‗Ecopath‘, a software implementing methods of network analysis. Among
other interesting relationships, they found that about half of the food of nectonic benthic
feeders and nectonic necton feeders passed through detritus at least once, whilst there was no
direct transfer of such food according to the diet matrix. The paper contains a number of
references to other studies where network analysis was used to analyse indirect relationships
among ecosystem components (see also Patten, 1992; Fath and Patten, 1998; Fath and Patten,
1999, and references therein), as well as a reference to the Ecopath web site, which, in turn,
gives a list of studies (mainly aquatic) where this software was applied. A couple of examples
from this list are reviewed below.
Ortiz and Wolff, 2002a and Ortiz and Wolff, 2002b used Ecopath with Ecosim software
to study benthic communities in Chile. They found that a simulated harvest of the clam
Mulinia generated a complex interplay involving direct and indirect effects, and drastically
changed the properties of the whole system.
Another study utilizing network analysis provided an analysis of the extended path and
flow structure for the well documented oyster reef model (Whipple, 1999). Few simple paths
and large number of compound paths were counted. The study provided structural evidence
for feedback control in ecosystems, and illustrated importance of non-living compartments (in
this case detritus) for the ecosystem‘s functioning. Even for the model with a low cycling
index (i.e. 11%) multiple cyclic passage paths provided a considerable (22%) flow
Some Aspects of Phytoplankton and Ecosystem Modelling … 209
contribution. Therefore, it was envisaged that for ecosystems with higher cycling indexes the
patterns observed should be even more pronounced.
Application of mathematical modelling techniques has been helpful in furthering the
understanding of a number of aspects of marine ecosystem dynamics in the Menai Strait
(Krivtsov et al., 2008a) and in the Irish Sea and Liverpool Bay (see Krivtsov et al., 2008b
and references therein). In particular, these studies helped to elucidate a complex role played
in the ecosystem by suspended particulate matter (SPM), which is arguably thought to be one
of the most important ecosystem constituents affecting the majority of ecological processes
(Krivtsov et al., 2011). At present, most ecosystem models are in dire need of improving the
representation of the SPM dynamics, and it has been argued (Krivtsov et al., 2008b) that
many of the discrepancies between measurements and simulations commonly observed in
aquatic modelling studies may, in fact, have been caused by inadequate representation of the
SPM subsystem.
MODELLING BIOGEOCHEMICAL
FLUXES IN CONTINENTAL SHELF SEAS
Representation of a biogeochemical cycle in an ecological model should involve all
relevant physical forcing functions, biological and abiotic processes, and chemical fractions
associated with biota, dissolved nutrient pool, bottom sediments, and suspended particulate
matter (SPM). Among those, SPM is often the least appreciated, and is, consequently,
underrepresented in ecological models. The discussion below reviews a number of important
aspects related to biogeochemical cycling, with a particular emphasis on SPM.
SPM
1999; Smith and Tett, 2000) capable of describing biogeochemical fluxes in shelf seas have,
albeit to a variable extent, included tidal mixing, water column dynamics, algal production
and loss processes, and SPM benthic fluff settling/resuspension.
Tidal Mixing
In continental shelf seas, daily inputs of turbulent kinetic energy from tidal and wind
stirring, and convective mixing, are opposed by buoyancy inputs due to solar heating. The
balance between these depends on seasonal variability of the surface heat flux. In deeper
regions, seasonal stratification develops whereby a thermocline separates surface and bottom
mixed layers (SML and BML, mixed by wind and tide, respectively – see Figure 1).
Shallower regions remain mixed for most of the time. During summer stratification develops,
and a tidal mixing front (which is the surface expression of the thermocline) is positioned
between stratified and mixed regions, where the intensity of tidal turbulent mixing is
sufficient to overcome the tendency to stratify due to surface heating. By equating turbulent
mixing to the potential energy difference before and after mixing, it has been shown that the
critical determinant of water column structure is tidal stirring, predicted by h/u3, where u is a
depth-mean average tidal current and h is water depth (Simpson and Hunter, 1974). A
reasonable precision of front prediction can be achieved by a combined wind and tide stirring
model (for details see Simpson and Bowers, 1981). Further improvements can be obtained
through formulations of water column structure based on consideration of turbulent kinetic
energy (tke) (e.g. Sharples and Tett, 1994; Simpson et al., 1996; Luytens et al., 1996;
Sharples, 1999). In such models, a turbulence closure scheme (e.g. Mellor and Yamada,
1982) links vertical stratification, driven by surface heating, and turbulence generated by tidal
friction at the seabed and wind stress at the surface; in effect, water column stability is related
to the efficiency of vertical turbulent transport. Thus, changes in tke (q2/2 where q is the
turbulent velocity scale) are given by:
(1)
where Kq is the vertical eddy diffusivity of tke, Nz is the depth-dependent coefficient of eddy
viscosity, Kz is the vertical eddy diffusivity, u and v are the x and y components of current
velocity, g is the acceleration due to gravity, ρ is water density, BI is a constant of the
turbulence closure scheme, and l is the constant turbulent length scale. This formulation
contains the vertical diffusion of tke, shear production of turbulence, work done against
buoyancy, and the dissipation of tke (the terms on the right hand side of Eq. 1, respectively).
Distribution of heat input from solar radiation is mediated by attenuation due to SPM and
algal biomass in the water column. Such models have been very successful in simulating
vertical mixing processes in general, and genesis of the thermocline in particular.
212 V. Krivtsov and C. F. Jago
Figure 1. Conceptual diagram of water column structure and water quality in tidal shelf seas. *=factor
which limits algal growth. 2.7 is a critical transition value of h/u3
It should be noted, that although frontal zones tend to be quite narrow (ca. 5 km wide), a
front is not an abrupt transition. The frontal density structure implies a pressure gradient
perpendicular to the front which drives an along-front current. The geostrophic balance is not
perfect, due to friction, and a small cross-stream flow is generated down the pressure
gradient. Consideration of the cross-frontal dynamical balance, including friction (James,
1978; Garrett and Loder, 1981 ), suggests a weak cross-frontal circulation with a surface
convergence close to the region of maximum horizontal gradient (e.g. Pingree et al., 1974).
The Simpson–Hunter stratification parameter predicts a lateral migration of fronts during the
lunar cycle of 10–20 km. Even after removing tidal advection, Simpson and Bowers (1979)
concluded that fronts migrate laterally by a few kilometers in response to the lunar cycle.
Numerical models (Simpson and Bowers, 1981; Sharples and Simpson, 1996) suggest that the
position of the frontal zone lags the spring-neap cycle by a few days. A major consequence of
lateral migration is that water from the mixed region is incorporated into the stratified region
as the front advances into shallower water on neap tides (Simpson and Hunter, 1974;
Simpson and Bowers, 1979). Furthermore, infrared satellite imagery and direct observations
show that baroclinic instabilities along the front grow into large eddies (typically 25–40 km)
which make a large contribution to cross-frontal mixing (Pingree and Griffiths, 1978). In
addition, some water from the BML may be mixed across the thermocline by spring tide
currents. Finally, some diffusion of the thermocline might occur due to dissipation of internal
tides.
Advection/Resuspension
Observations of SPM in continental shelf seas can be reasonably reproduced (Jago and
Jones, 1998) by the following conceptual model:
Some Aspects of Phytoplankton and Ecosystem Modelling … 213
t
dS
dx 0
S ( t ) S0 U x dt k U x , (2)
where
Values of background concentration, combined function k, and the gradient along the
dS
tidal stream are usually calculated by linear regressions between observations of S(t), and
dx
tidal current displacement and speed.
The formulation presented above proved to be reasonable, and is capable of reproducing
the so-called ‗twin peaks signal‘ in SPM concentration, which results from the
superimposition of the quarter-diurnal signal due to the resuspension of benthic fluff and
semi-diurnal signal caused by longitudinal concentrational gradients in SPM in long-term
suspension. Further improvements in the performance can be gained by introducing a
numerical model which includes horizontal advection of a particulate concentration gradient
and vertical diffusion, and the bottom boundary conditions including fluxes due to
resuspension and diffusion, with entraiment rate being a function of the bed shear stress (for
details see Jago and Jones, 1998, and references therein).
(3)
where X is biomass and is the specific growth rate, Kz is vertical eddy diffusivity, and g is
the loss of algal biomass to grazers.
214 V. Krivtsov and C. F. Jago
It should be noted that in summer 30–80% of the total algal production in the euphotic
zone takes place in the thermocline (Fransz and Gieskes, 1984) and that the greatest
production occurs near tidal mixing fronts (Pingree et al., 1975, 1978; Savidge, 1976;
Holligan, 1981; Holligan et al., 1983; Loder and Platt, 1985. ; Tett et al., 1993; Tett and
Walne, 1995). This could be a passive response to the convergent flows which have been
observed at fronts but it is more likely to result from in situ growth of plankton. This is where
the combination of light and nutrients is optimal (Dufour and Stretta, 1973): nutrient renewal
during the summer, due to mixing by tide and wind, and surface stabilisation and reduction of
h1 during fairweather and neap tides. Thus a chlorophyll maximum is observed at the
thermocline and at fronts. The optimal conditions for rapid algal growth are at the front, as
lateral mixing across the front is greater than vertical mixing across the thermocline (Garrett
and Loder, 1981; Tett, 1981; Tett et al., 1986).
The growth rate of phytoplankton may be nutrient-limited:
(4)
where m is the maximum growth rate, kQ is the subsistence cell nutrient quota, and Q is the
cell nutrient quota (=the ratio of algal cell internal nutrient concentration to chlorophyll
biomass). Unlike in most freshwater ecosystems, in continental shelf seas nitrogen is usually
less available for growth, and is therefore more limiting, than is phosphorus (Tett and Droop,
1988), while silicon may be more limiting than nitrogen for diatoms (Brzezinski, 1985).
In addition to nutrient limitation, phytoplankton growth may be slowed down by low
light and/or temperature. There are numerous definitions how exactly these limitations may
be combined with the limitations imposed by nutrient availability, and the detailed
consideration of the matter would be beyond the scope of this paper. Low light due to high
SPM concentrations is a major control in mixed regions of shelf seas. The insight into one of
the more simple representations may be seen in the definitions of the freshwater ecosystem
model Rostherne presented elsewhere (see Krivtsov et al. 2000b).
biogeochemical fluxes in stratified and mixed regions of tide-driven shelf seas. Importantly,
the coupled models provide numerical relationships between factors such as water
temperature and seabed anoxia (which are potentially recorded in the sediments by biological
and geochemical proxies) and governing variables such as water depth, mixing, and
turbulence.
Another relevant example in this respect is POLCOMS (the Proudman Oceanographic
Laboratory Coastal Ocean Modelling System, see www.pol.ac.uk/home/research/polcoms), a
three-dimensional modelling system whose main elements are a three-dimensional baroclinic
hydrodynamic model (Holt and James, 2001) linked to a surface wave model (Wolf et al.,
2002; Osuna et al., 2007), a sediment resuspension and transport model (Holt and James,
1999) and an ecosystem model (ERSEM, the European regional seas ecosystem model, see
e.g. Baretta et al., 1995; Blackford et al., 2004) with benthic and pelagic components. This
modelling system has been developed primarily to investigate physical–biogeochemical
interactions in shelf seas, see for example Proctor et al. (2003).
Coupled models are capable of reproducing a complex interplay between ecosystem
components, and simulating spatial and temporal variation of algal productivity and the
dynamics in tidal shelf seas. Simulations suggest that plankton are constrained in mixed
regions by the transparency of the water column (high nutrients but high turbidity), and that
productivity seems to be greater in areas where h1 (i.e. the 'mixed layer optical thickness',
where is an attenuation coefficient, h1 is the thickness of the layer through which algae are
transported by vertical turbulence) is small rather than where nutrients are greatest (Tett et al.,
1993). Plankton are also constrained in the BML of stratified regions (high nutrients but low
light). In the SML of stratified regions, productivity is initially high in spring (high nutrients
and good light) but rapidly diminishes during the summer as nutrients are consumed and not
replaced. Grazing by herbivorous zooplankton and settling of algal cells (associated with
SPM aggregates) to the seabed reduce the standing stock in the SML. Subsequent predation
of zooplankton reduces grazing pressure and increase of nutrients by wind-driven mixing
stimulates a secondary autumn algal bloom in the SML. As the spring bloom wanes,
enhanced chlorophyll concentrations (up to 100 mg m−3) and maximum phytoplankton
biomass are generally seen at the thermocline in seasonally stratified regions (e.g. Anderson,
1969; Cullen and Eppley, 1981; Holligan et al., 1983). Turbulent tidal mixing generates new
production by periodically supplying nitrate from the BML. However, turbulent tidal mixing
also entrains algae into the BML where they are lost from the productive regions of the water
column (Sharples et al., 2001).
The nutrient status of algal cells has been shown to affect their cohesiveness (Kiorboe et
al., 1990) and enhanced post-bloom plankton agglutination has been attributed to an increase
in stickiness due to nutrient depletion (Logan and Alldredge, 1989; Smetacek, 1985). Such
stickiness has been linked to SPM aggregation with respect to flagellates and dinoflagellates
(Passow and Wassman, 1994; Jones et al., 1998) as well as to diatoms. Strong biological
mediation of SPM aggregation during the late stages of a flagellate bloom, resulting in
increased particle size and settling flux of SPM to the seabed, has been measured (Jago, et al.,
2007). Links between algal activity and particulate settling flux suggest that the greatest flux
per unit area of particulate matter to the seabed should occur in regions of greatest algal
production per unit volume, i.e. in frontal zones.In the aftermath of blooms, the enhanced
settling flux gives rise to low density, organic-rich, benthic fluff on the bed (Jago et al.,
1993). The correct understanding of the exact nature, and the dynamics of this fluff is crucial,
216 V. Krivtsov and C. F. Jago
because of its links with a number of important biological and chemical processes, and its
overall effect on biogeochemical cycling, and may be greatly facilitated by further
investigations combining experimental approaches, with comprehensive monitoring and
modelling studies.
REFERENCES
Abrams P. A., Menge B. A., Mittelbach G. G., Spiller D. A. & Yodzis P. (1996) The role of
indirect effects in food webs. In: Food webs: integration of patterns and dynamics. (eds.
G. Polis & W. K.) pp. 371 - 395. Chapman and Hall, New York.
Some Aspects of Phytoplankton and Ecosystem Modelling … 217
Anderson, G.C. Subsurface chorophyll maximum in the Northeast Pacific Ocean. Limnol.
Oceanogr. 14 (1969), pp. 386–391.
Baretta J.W. , W. Ebenhoeh and P. Ruardij, The European regional seas ecosystem model, a
complex marine ecosystem model., Neth. J. Sea Res. 33 (3–4) (1995), pp. 233–246.
Bartell S. M., Lefebvre G., Kaminski G., Carreau M. & Campbell K. R. (1999) An ecosystem
model for assessing ecological risks in Quebec rivers, lakes, and reservoirs. Ecological
Modelling, 124: 43-67.
Blackford J.C., J.I. Allen and F.J. Gilbert (2004) Ecosystem dynamics at six contrasting sites:
a generic model study, J. Mar.Syst. 52: 217–234.
Brzezinski, M.A. The Si:C:N ratio of marine diatoms: interspecific variability and the effect
of some environmental variables. J. Phycol. 21 (1985), pp. 347–357.
Carrer S. & Opitz S. (1999) Trophic network model of a shallow water area in the northern
part of the Lagoon of Venice. Ecological Modelling, 124: 193-219.
Cullen, J.J. and R.W. Eppley , Chlorophyll maximum layers of the Southern-California Bight
and possible mechanisms of their formation and maintenance. Oceanol. Acta, 4 (1981),
pp. 23–32.
Dippner J. W. (1998) Competition between different groups of phytoplankton for nutrients in
the southern North Sea. Journal of Marine Systems, 14: 181-198.
Droop , M.R. 25 years of algal growth kinetics - a personal view. Bot. Mar. 26 (1983), pp.
99–112.
Dufour, P., Stretta, J.M., 1973. Fronts thermique et thermohalins dans la région du Cap Lopez
(Golfe de Guinée) juin-juillet 1972: phytoplancton, zooplancton, micronecton et pêche
thonière. Documents Scientifiques, Centre de Recherches Océanographiques, Abidjan
4/99-142.
Fath B. D. & Patten B. C. (1998) Network synergism: Emergence of positive relations in
ecological systems. Ecological Modelling, 107: 127-143.
Fath B. D. & Patten B. C. (1999) Review of the foundations of network environ analysis.
Ecosystems, 2: 167-179.
Fransz, H.G. and W.W.C. Gieskes , The imbalance of phytoplankton and copepods in the
North Sea. Rapp. P.-v Réun. Cons. perm. int. Explor. Mer 183 (1984), pp. 218–225.
Garrett, C.J.R. and J.W. Loder , Dynamical aspects of shallow sea fronts. Phil. Trans. R. Soc.
London A, 302 (1981), pp. 563–581.
Hanratty M. P. & Liber K. (1996) Evaluation of model predictions of the persistence and
ecological effects of diflubenzuron in a littoral ecosystem. Ecological Modelling, 90: 79-
95.
Holligan, P.M. Biological implications of fronts on the northwest European shelf. Phil. Trans.
R. Soc. London A, 302 (1981), pp. 547–562.
Holligan, P.M. , M. Viollier, C. Dupouy and J. Aiken , Satellite studies on the distribution of
chlorophyll and dinoflagellates in the western English Channel. Cont. Shelf Res. 2 (1983),
pp. 81–96.
Holt J.T. and I.D. James (1999) A simulation of the Southern North Sea in comparison with
measurements from the North Sea Project. Part 2. Suspended particulate matter, Cont.
Shelf Res. 19: 1617–1642.
Holt, J.T. and I.D. James (2001) An s-coordinate density evolving model of the north west
European continental shelf. Part 1. Model description and density structure, J. Geophys.
Res. 106 (C7), 14015–14034.
218 V. Krivtsov and C. F. Jago
Hulot F. D., Lacroix G., Lescher-Moutoue F. O. & Loreau M. (2000) Functional diversity
governs ecosystem response to nutrient enrichment. Nature, 405: 340-344.
Hutchinson G. E. (1957) A Treatise on Limnology. Chapman and Hall Ltd., London.
Jago C.F. and S.E. Jones (1998). Observation and modelling of the dynamics of benthic fluff
resuspended from a sandy bed in the southern North Sea. Cont. Shelf Res. 18:1255–1282.
Jago C.F. A.J. Bale, M.O. Green, M.J. Howarth, S.E. Jones, I.N. McCave, G.E. Millward,
A.W. Morris, A.A. Rowden and J.J. Williams (1993) Resuspension processes and seston
dynamics. Phil. Trans. R. Soc. London A, 343: 475–491.
Jago, C.F., Jones, S.E., Kennaway, G., Latter, R.J., McCandliss, R.R., Rippeth, T., Simpson,
J.H., 2002. Mediation of shelf sea suspended particle properties by plankton and
turbulence. EOS Trans. AGU 83(4), Ocean Sciences, Meet. Suppl., Abstract OS22D-213.
Jago, C.F. and Jones, S.E., 2002. Diagnostic criteria for reconstruction of tidal continental
shelf regimes: changing the paradigm. Marine Geology, 191, 95-117.
Jago, C.F., Kennaway G, Novarino G, Jones S E, 2007. Size and settling velocity of
suspended flocs during a Phaeocystis bloom in the tidally-stirred Irish Sea, N W
European shelf. Marine Ecology Progress Series, 345, 51-62
James , I.D. A note on the circulation induced by a shallow sea front. Estuar. Coast. Mar. Sci.
7 (1978), pp. 197–202.
Jones, S.E. , C.F. Jago, A.J. Bale, D. Chapman, R. Howland and J. Jackson , Aggregation and
resuspension of suspended particulate matter at a seasonally stratified site in the southern
North Sea: physical and biological controls. Cont. Shelf Res. 18 (1998), pp. 1283–1310.
Jørgensen S. E. (1980) Lake management. Pergamon, Oxford.
Jorgensen S. E. (1994) Fundamentals of Ecological Modelling (2nd Edition). Elsevier,
Amsterdam.
Kiorboe, T. , K.P. Anderson and H.G. Dam , Coagulation efficiency and aggregate formation
in marine phytoplankton. Mar. Biol. 107 (1990), pp. 235–245.
Krivtsov V. (2001) Study of cause-and-effect relationships in the formation of biocenoses:
Their use for the control of eutrophication. Russian Journal of Ecology, 32: 230-234.
Krivtsov V. (2002) Indirect Effects in Ecosystems: a Review of Recent Modelling Studies
and a Methodological Framework for Comparative Theoretical Analysis. In: Integrated
Assessment and Decision Support. Proceedings of the first biennial meeting of the
International Environmental Modelling and Software Society (also available at
http://www.iemss.org/iemss2002/proceedings/pdf/volume%20uno/410_Krivtsov.pdf)
(eds. A. E. Rizzoli & A. J. Jakeman) pp. 233 - 238, Lugano.
Krivtsov V., Bellinger E. & Sigee D. (2000a) Incorporation of the intracellular elemental
correlation pattern into simulation models of phytoplankton uptake and population
dynamics. Journal of Applied Phycology, 12: 453-459.
Krivtsov V., Bellinger E. & Sigee D. (2003a) Ecological study of Stephanodiscus rotula
during a spring diatom bloom: dynamics of intracellular elemental concentrations and
correlations in relation to water chemistry, and implications for overall geochemical
cycling in a temperate lake. Acta Oecologica, 24: 265-274.
Krivtsov V., Bellinger E. G. & Sigee D. C. (1999a) Modelling of elemental associations in
Anabaena. Hydrobiologia, 414: 77-83.
Krivtsov V., Sigee D., Corliss J. & Bellinger E. (1999b) Examination of the phytoplankton of
Rostherne Mere using a simulation mathematical model. Hydrobiologia, 414: 71-76.
Some Aspects of Phytoplankton and Ecosystem Modelling … 219
Krivtsov V., Tien C., Sigee D. & Bellinger E. (1999c) X-ray microanalytical study of the
protozoan Ceratium hirundinella from Rostherne Mere (Cheshire, UK): Dynamics of
intracellular elemental concentrations, correlations and implications for overall
ecosystem functioning. Netherlands Journal of Zoology, 49: 263-274.
Krivtsov V., Bellinger E., Sigee D. & Corliss J. (1998) Application of SEM XRMA data to
lake ecosystem modelling. Ecological Modelling, 113: 95-123.
Krivtsov V., Bellinger E., Sigee D. & Corliss J. (2000b) Interrelations between Si and P
biogeochemical cycles - a new approach to the solution of the eutrophication problem.
Hydrological Processes, 14: 283-295.
Krivtsov V., Corliss J., Bellinger E. & Sigee D. (2000c) Indirect regulation rule for
consecutive stages of ecological succession. Ecological Modelling, 133: 73-82.
Krivtsov V., Goldspink C., Sigee D. C. & Bellinger E. G. (2001b) Expansion of the model
'Rostherne' for fish and zooplankton: role of top-down effects in modifying the prevailing
pattern of ecosystem functioning. Ecological Modelling, 138: 153-171.
Krivtsov V., Sigee D. & Bellinger E. (2001c) A one-year study of the Rostherne Mere
ecosystem: seasonal dynamics of water chemistry, plankton, internal nutrient release, and
implications for long-term trophic status and overall functioning of the lake.
Hydrological Processes, 15: 1489-1506.
Krivtsov V., Sigee D. & Bellinger E. (2002b) Elemental concentrations and correlations in
winter micropopulations of Stephanodiscus rotula: an autecological study over a period
of cell size reduction and restoration. European Journal of Phycology 37: 27-35.
Krivtsov V., J. Gascoigne and M. W. Skov (2011) Dynamics of suspended particulate matter
in the Menai Strait (UK) and its implications for ecosystem functioning and management.
In A. D. Nemeth (Editor). The Marine Environment: Ecology, Management and
Conservation. NovaScience Publishers, pp 179-197.
Krivtsov V., Walker S. J. J., Staines H. J., Watling R., Burt-Smith G. & A. G. (2004b)
Integrative analysis of ecological patterns in an untended temperate woodland utilising
standard and customised software. Environemntal Modelling and Software: Volume 19,
Issue 3, March 2004, Pages 325-335.
Krivtsov V., Gascoigne J. and Jones S. E. 2008a. Harmonic analysis of suspended particulate
matter in the Menai Strait (UK). Ecological Modelling, 212, 53-67.
Krivtsov V., Howarth M. J., Jones S. E., Souza A. J. and Jago C. F. 2008b. Monitoring and
modelling of the Irish Sea and Liverpool Bay: An overview and an SPM case study.
Ecological Modelling, 212, 37-52.
Krivtsov V. 2004. Investigations of indirect relationships in ecology and environmental
sciences: a review and the implications for comparative theoretical ecosystem analysis.
Ecological Modelling, 174, 37-54
Krivtsov V. 2008. Indirect Effects in Ecology. In: Sven Erik Jorgensen and Brian Fath,
Editors-in-Chief, Encyclopedia of Ecology, Academic Press, Oxford. Pages 1948-1958,
ISBN 978-0-08-045405-4, DOI: 10.1016/B978-008045405-4.00693-5.
Loder, J.W., Platt, T., 1985. Physical controls on phytoplankton production at tidal fronts. In:
Gibbs, P.E. (Ed.), Proc. 19th European Mar. Biol. Symposium, Cambridge University
Press, Cambridge, pp. 3–21.
Logan, B.E. and A.L. Alldredge , Potential for increased nutrient uptake by flocculating
diatoms. Mar. Biol. 101 (1989), pp. 443–450.
220 V. Krivtsov and C. F. Jago
Loladze I., Kuang Y. & Elser J. J. (2000) Stoichiometry in producer-grazer systems: Linking
energy flow with element cycling. Bulletin of Mathematical Biology 62: 1137-1162.
Luytens, P.J. , E. Deleersnijder, J. Oxer and K.G. Ruddick , Presentation of a family of
turbulence closure models for stratified shallow water flows and preliminary application
to the Rhine outflow region. Cont. Shelf Res. 16 (1996), pp. 101–130.
Luytens, P.J., Jones, J.E., Proctor, R., Tabor, A., Tett, P., Wild-Allen, K., 1999. COHERENS
- a coupled hydrodynamical-ecological model for regional and shelf seas: users
documentation. Management Unit of the Mathematical Models of the North Sea,
Brussels, MUMM Internal Document, 911 pp.
Malaeb Z. A., Summers J. K. & Pugesek B. H. (2000) Using structural equation modeling to
investigate relationships among ecological variables. Environmental and Ecological
Statistics, 7: 93-111.
McClanahan T. R. & Sala E. (1997) A Mediterranean rocky-bottom ecosystem fisheries
model. Ecological Modelling, 104: 145-164.
Mellor, G.L. and T. Yamada , Development of a turbulence closure model for geophysical
fluid problems. Rev. Geophys. Space Phys. 20 (1982), pp. 841–875.
Mortimer C. H. (1941) The exchange of dissolved substances between mud and water in
lakes. Journal of Ecology, 29: 280 - 329.
Mortimer C. H. (1942) The exchange of dissolved substances between mud and water in
lakes. Journal of Ecology, 30: 147-201.
Naito W., Miyamoto K., Nakanishi J., Masunaga S. & Bartell S. M. (2002) Application of an
ecosystem model for aquatic ecological risk assessment of chemicals for a Japanese lake.
Water Research, 36: 1-14.
Navarrete S. A. & Menge B. A. (1996) Keystone predation and interaction strength:
Interactive effects of predators on their main prey. Ecological Monographs, 66: 409-429.
Ortiz M. & Wolff M. (2002a) Application of loop analysis to benthic systems in northern
Chile for the elaboration of sustainable management strategies. Marine Ecology-Progress
Series, 242: 15-27.
Ortiz M. & Wolff M. (2002b) Dynamical simulation of mass-balance trophic models for
benthic communities of north-central Chile: assessment of resilience time under
alternative management scenarios. Ecological Modelling, 148: 277-291.
Osuna P., A.J. Souza and J. Wolf (2007) Effects of the deep-water wave breaking dissipation
on the wind-wave modelling in the Irish Sea, J. Mar. Syst. 67: 59–72.
Passow, U. and P. Wassman , On the trophic fate of Phaeocystis pouchetii (Hariot). IV. The
formation of marine snow by P. pouchetii. Mar. Ecol. Progr. Ser. 104 (1994), pp. 153–
161.
Patten B. C. (1985) Energy Cycling in the Ecosystem. Ecological Modelling, 28: 1-71.
Patten B. C. (1992) Energy, Emergy and Environs. Ecological Modelling, 62: 29-69.
Patten B. C., Bosserman R. W., Finn J. T. & Gale W. G. (1976) Propagation of cause in
ecosystems. In: Systems analysis and simulation in ecology (ed. B. C. Patten) pp. 457 -
579. Academic Press, New York.
Pingree, R.D. and D.K. Griffiths , Tidal fronts on the shelf seas around the British Isles. J.
Geophys. Res. 83 (1978), pp. 4615–4622.
Pingree, R.D. , P.M. Holligan, G.T. Mardell and R.N. Head , The effects of vertical stability
on phytoplankton distribution in the summer on the nortwest European shelf. Deep Sea
Res. 25 (1978), pp. 1011–1028.
Some Aspects of Phytoplankton and Ecosystem Modelling … 221
Pingree, R.D., G.R. Forster and G.K. Morrison , Turbulent convergent tidal fronts. J. Mar.
Biol. Assoc. UK. 54 (1974), pp. 469–479.
Pingree, R.D., P.R. Pugh, P.M. Holligan and G.R. Forster , Summer phytoplankton blooms
and red tides along tidal fronts in the approaches to the English Channel. Nature, 258
(1975), pp. 672–677.
Proctor R., J.T. Holt, J.I. Allen and J. Blackford (2003) Nutrient fluxes and budgets for the
north west European Shelf from a three-dimensional model., Sci. Total Environ. 314–
315: 769–785.
Reynolds C. S. (1984) The ecology of freshwater phytoplankton. Cambridge University
Press, Cambridge.
Savidge, G. A preliminary study of the distribution of chlorophyll a in the vicinity of fronts in
the Celtic and western Irish Sea. Estuar. Coast. Mar. Sci. 4 (1976), pp. 617–625.
Sharples, J. and P. Tett , Modelling the effect of physical variability on the midwater
chlorophyll maximum. J. Mar. Res. 52 (1994), pp. 219–238.
Sharples, J. , C.M. Moore, T. Rippeth, P.M. Holligan, D.J. Hydes, N.R. Fisher and J.H.
Simpson , Phytoplankton distribution and survival in the thermocline. Limnol. Oceanogr.
46 (2001), pp. 486–496.
Sharples, J., 1999. Investigating the seasonal vertical structure of phytoplankton in shelf seas.
Prog. Oceanogr. Suppl. S, 3–38.
Sharples, J., Simpson, J.H., 1996. The influence of the springs-neaps cycle on the position of
shelf fronts. In: Aubrey, D.G., Friedrichs, C.T. (Eds.), Buoyancy effects on coastal and
estuarine dynamics, Am. Geophys. Union, Washington, DC, pp. 71–82.
Simpson, and D. Bowers , Models of stratification and frontal movement in shelf seas. Deep
Sea Res. 28A (1981), pp. 727–738.
Simpson, J.H. and D. Bowers , Shelf sea front adjustments revealed by satellite IR imagery.
Nature, 280 (1979), pp. 648–651.
Simpson, J.H. and J.R. Hunter , Fronts in the Irish Sea. Nature, 250 (1974), pp. 404–406.
Simpson, J.H. , W.R. Crawford, T.P. Rippeth, A.R. Campbell and J.V.S. Cheok, The vertical
structure of turbulent dissipation in shelf seas. J. Phys. Oceanogr. 26 (1996), pp. 1579–
1590.
Smetacek, V.S. Role of sinking in diatom life history cycles: ecological, evolutionary and
geological significance. Mar. Biol. 84 (1985), pp. 239–251.
Smith, C.L. and P. Tett , A depth-resolving numerical model of physically forced
microbiology at the European shelf edge. J. Mar. Syst. 675 (2000), pp. 1–36.
Tett, P. Modelling of phytoplankton production at shelf sea fronts. Phil. Trans. R. Soc.
London A, 302 (1981), pp. 605–615.
Tett, P. and A. Walne , Observations and simulations of hydrography, nutrients and plankton
in the southern North Sea. Ophelia, 42 (1995), pp. 371–416.
Tett, P. , I. Joint, D. Purdie, M. Baars, S. Oosterhuis, G. Daneri, F. Hannah, D.K. Mills, D.
Plummer, A. Pomroy, A.W. Walne and H.J. Witte , Biological consequences of tidal
stirring gradients in the North Sea. Phil. Trans. R. Soc. London A, 340 (1993), pp. 493–
508.
Tett, P., A. Edwards and K.J. Jones , A model for the growth of shelf sea phytoplankton in
summer. Estuar. Coast. Shelf Sci. 23 (1986), pp. 641–672.
222 V. Krivtsov and C. F. Jago
Tett, P., Droop, M.R., 1988. Cell quota models and planktonic primary production. In:
Wimpenny, J.W.T. (Ed.), Handbook of Laboratory Model Systems for Microbiol
Ecosystems, vol. 2. CRC Press, Boca Raton, FL, pp. 177–233.
Wardle D. A. (2002) Communities and ecosystems. Linking the aboveground and
belowground components. Princeton University Press, Princeton.
Whipple S. J. (1999) Analysis of ecosystem structure and function: extended path and flow
analysis of a steady-state oyster reef model. Ecological Modelling, 114: 251-274.
Wolf, S.L. Wakelin and J.T. Holt (2002) A coupled model of waves and currents in the Irish
Sea, Proceedings of the 12th International Offshore and Polar Engineering Conference
Kitakyushu, Japan, vol. 3 May 26–31(2002), pp. 108–114.
Wootton J. T. (1994) Predicting Direct and Indirect Effects - an Integrated Approach Using
Experiments and Path-Analysis. Ecology, 75: 151-165.
Wootton J. T. (2002) Mechanisms of successional dynamics: Consumers and the rise and fall
of species dominance. Ecological Research, 17: 249-260.
In: Ecological Modeling ISBN: 978-1-61324-567-5
Editor: WenJun Zhang, pp. 223-265 © 2012 Nova Science Publishers, Inc.
Chapter 10
ABSTRACT
In the evolution of social insects, the colony and not the (often sterile) individual
worker should be considered the major unit of selection. Thus, social insect colonies are
considered to be 'super-organisms', which have – like all other organisms – to perform
behaviors which affect their outside environment and which alter their own future
internal status. The way these behaviors are coordinated is by means of communication,
which is either direct or indirect and which involves information exchange either by
transmitting signals or by exploiting cues. Therefore, social insect colonies perform
information processing in a rather similar way as multicellular organisms do, where
behaviors result from the exchange of information among their sub-modules (cells). In
many cases, self-organization allows a colony to evaluate massive amounts of
information in parallel and to decide about the colony's future behavioral responses.
Many feedback systems that govern self-organization of workers have been investigated
empirically and theoretically. Here, we discuss models which have been proposed to
explain division of labor and task selection in social insects. We demonstrate how the
collective regulation of labor in eusocial insect colonies is studied by means of top-down
modeling and by bottom-up models, often analyzed with multi-agent computer
simulations.
Keywords: social insects, division of labor, task selection, colony integration, multi-agent
modeling, honeybees.
*
Tel: +43 316 380 8759, Fax: +43 316 380 9875.
224 Thomas Schmickl and Karl Crailsheim
1. INTRODUCTION
For any living organism, choosing an appropriate behavior is a crucial issue to survive
and to reproduce. Usually, the actions that compose behaviors are attributed to certain
functionality, like 'foraging for food', 'reproduction', or 'self-defense'. In most cases, a
successful performance of such behaviors significantly alters the constitution and the mode of
operation of this organism. This happens for example by increasing the activity of specific
organs in multicellular organisms or of specific organelles in unicellular organisms.
In our field of study, which is social insects, similar behaviors are observed in insect
colonies. Such colonies are usually conceived as 'super-organisms' (Moritz & Fuchs 1998,
Hölldobler & Wilson 2008). This is reasoned by the fact that in social insects, whole colonies
– and not individual workers – can be seen as units of selection, as was reviewed by Tarpy et
al. (2004) for honeybees. In most cases, female workers are sterile and colonies perform
reproduction by division of whole colonies (e.g., swarming in honeybees) or by producing
queens and males for founding new colonies (e.g., most ants, termites, wasps, bumblebees).
We adapt this super-organism approach and discuss colony-level decision-making and self-
regulation as a result from division of labor and from communication/interaction among
workers: We focus here on those eusocial insects which exhibit division of labor (DOL) and
partially also task partitioning (TP). The phenomenon of DOL refers to the fact that the
working duties in the colony are not distributed at random. In many social insect species,
specific groups of workers perform specific sets of tasks within the colony's collective
physiology, thus they have many similarities to organs in metazoa and to organelles in
protozoa and microbes. These analogies were already pointed out by Wilson (1985) as well as
by Robinson (1992), who showed the importance of adaptability and of self-organizational
aspects in DOL.
The predictions of our mathematical models of honeybees, which are discussed in this
chapter, suggest that the processes, which allow a social insect colony to perform collective
decision making and collective homeostasis, are influenced by the colony-wide physiological
network, where specific pathways (flow of nutrients, flow of workforce, flow of information
...) exist. In turn, colony-level decision making affects these colony-level physiology
significantly, making the collective physiology an important component of significant
feedback loops. Colony-level physiology is regulated again by feedback loops that establish
colony-level homeostasis. However, these feedback loops involve in most cases specific
behaviors of individual workers, thus modeling these networks quickly leads to models that
cannot be solved in mathematical closed form. Thus numerical simulation becomes
important. Often the behavior of individual agents that constitute a colony as well as their
environmental conditions are found to be so complex that individual-based modeling is
needed to generate plausible models of these focal systems. However, loss of generality is
often the price that has to be paid, whenever individual-based models are used. We are well
aware of that in our scientific approach, as our models tend to describe the modeled system in
a very detailed manner. However, this level of detail is needed to allow our models to answer
our focal scientific questions. We try to model the physiology of the whole colony by
modeling the physiology of the individuals with as much precision as is feasible. Our concept
is to fuse many physiological building blocks (worker physiology) to one single higher-level
physiology (colony). In turn, behavioral decisions of individual agents are mostly results of
Modeling Population Dynamics, Division of Labor … 225
their physiological status and their local environment, which are summarized as 'proximate
mechanisms' (Mayr 1961, Tinbergen 1963). These individual behaviors ultimately constitute
the collective behavior of the colony. Our modeling approach allows us to investigate how
such colony-level decision making and self-regulation affects energy economics and brood
production of insect colonies. Thus, these studies ultimately allow us to discuss also 'ultimate
causation' of colony-level behaviors (Mayr 1961, Tinbergen 1963).
Also in our work, the main scientific questions are focusing on proximate mechanisms
and on ultimate causation of colony-level reactions (or group-level behavior), which are a
response of a colony to a specific environmental stimulation. These questions can be grouped
into two distinct sets of questions:
Q1: How does the composition of the colony (different specialists) change in
response to these stimuli? How does colony structure (e.g. age structure) change?
How does the flow of nutrients and energy through the colony change? The answers
to this set of questions build the 'substrate' for investigating the second set of
questions, as the colony status restricts the availability of nutrients, building material,
energy and workforce which is available for the colony.
Q2: How does individual worker decision making affect the colony's global
physiological status and vice versa? How is the physiology of individual workers
correlated with the global physiology of the collective? How is the ratio of gain to
costs of specific worker actions related to the colony-wide economics of energy and
nutrients? Which information is spread, stored and filtered in the colony by specific
actions performed by workers? This set of questions is investigating the proximate
mechanisms involved in DOL, but – as we model also the physiology of workers –
the studies aimed to answer these questions deliver also predictions of colony-level
fitness parameters, which are survival and fecundity of colonies. Thus, these studies
link back also to the ultimate causation of observed behaviors. In this article we will
describe how studies of DOL can be performed with a set of models of honeybee
population dynamics and nectar energetics.
The degree of specialization found in social insect workers differs among species: The
phenomenon of specialization refers to the fact that workers are more likely to perform one
task over the other compared to other workers in similar situations. In contrast to that,
plasticity refers to the fact that one worker engages in a variety of tasks or that it changes the
performed task over time. Specialization of workers will be advantageous if the specialist
worker performs its specialized task more efficient than an average worker is able to (Jeanne
1986a). This is most obvious in ants and termites, where there exist sometimes huge
morphological differences among worker castes (Wilson 1985). In addition to morphological
differences between workers, drones and queens, there are also significant physiological
differences observable between worker specialists in honeybees. Due to this physiological
specialization a worker bee can be more efficient in its specialized task. Seeley (1982)
mentions additionally the fact that age-related groups are better in locating specific age-
226 Thomas Schmickl and Karl Crailsheim
related tasks in their local environment than the hypothetical generalist worker could do: For
example young bees most prominently find brood cells to clean nearby, as they just have
emerged from such cells a few hours ago. But specialization could be also disadvantageous: It
makes a colony vulnerable to losses of one specific caste and it maximizes searching time of
specialists to find the next suitable working site.
In honeybees, the observed correlation between task specialization and worker age is
called 'age polyethism' (AP). DOL, TP, and AP are very prominent features in eusocial
insects, as we will describe in more detail in the upcoming sections.
Wilson (1984, 1985) reports the case of the ants Pheidole guilelmimuelleri and Pheidole
pubiventris where even huge morphological differences between small workers (minors) and
huge workers (majors) do not prevent one caste to engage in a task that is typical for the other
caste in times of experimental removal of one caste. Workers with sufficient plasticity might
find the next spots to work sooner than specialists. We assume that natural selection found a
near-optimal trade-off between these two features for every species, according to the species
ecological constraints. Computational models help us to understand these ultimate reasons,
why an according mode of operation in DOL emerged in specific species.
In addition to specialization and plasticity, some social insects show also the
phenomenon of task partitioning. In contrast to DOL, which refers to splitting the workers
into task-specific groups, TP refers to splitting a task among several workers. In most cases of
task partitioning found in social insects, it was observed that the foraging task was
partitioned: Searching, cutting, transporting and storing of food is often performed by
specialist workers. In leaf-cutter ants, TP was reported for almost all tasks that involve
handling of material (also waste removal, see Hart et al., 2002). Ratnieks and Anderson
(1999) reviewed this issue in detail and discussed several important aspects of TP: (1) It
allows workers to specialize in specific parts of a whole task, thus possibly increasing
efficiency. (2) By doing so, reliability of a single subtask increases, as specialized workers are
more reliable to accomplish the sub-tasks, but (3) the reliability of the whole task decreases,
because if one sub-task fails, the whole task cannot be performed. In addition (4) food-
transfer times have to be short enough so that benefits of sub-task specialization are not
wasted. Again here, computational models are able to support ultimate reasoning of TP.
The organs in non-eusocial metazoan organisms show significant changes in activity
during specific behaviors. Also the task-specific groups in eusocial insects change their group
sizes and the activity levels of their members in association with specific colony-level
behaviors. It was found that those mechanisms that allow a social insect colony to decide
collectively for nesting sites or for important foraging targets are in most cases regulated in a
decentralized and self-organized way (Camazine et al., 2001), often also referred to by the
term 'swarm intelligence'. In such systems, a network of feedbacks regulates DOL and TP,
which arises from the flow of substances (food, pheromones) within the colony, as well as
from worker-to-worker interactions and communication. However, this is not totally different
from non-eusocial metazoan organisms, where hormones and other cell-to-cell interactions
also contribute to selection and shaping of the organisms' behaviors.
When reflecting on an observed behavior, scientists should concentrate on two distinct
sets of questions (Mayr 1961): The first set of questions refers to the individual local
mechanisms that cause the behavior. For example ―H ow does it work?‖ and ―Whe n and how
is it triggered?‖. The second set of questions refers to ultimate reasoning of the behavior, like
―Whyhas natural selection favored this behavior?‖. Computational models support scientists
Modeling Population Dynamics, Division of Labor … 227
in elaborating on both sets of questions, as we will discuss in the following sections of this
article. In the following sections we describe two major approaches how DOL and TP are
investigated by means of computational modeling: On the one hand, there are top-down
models, which often consider the population structure of the modeled colony or predict the
population dynamics that arise from DOL. These models often aim for an ultimate reasoning
of DOL. On the other hand, there exist many individual-based models, which often aim on
investigating the proximate mechanisms. By describing our own model 'TaskSelSim' in
detail, we show how an individual-based model is used for addressing questions of ultimate
reasoning.
The system of feedbacks that allows DOL in an insect society can be well studied by top-
down-approaches, for example by stock&flow modeling. This quantity-based modeling
approach was used to investigate the interplay between age demography and DOL in social
insects only rarely. However, we think that this modeling technique is an excellent way to
describe social insect populations: Stock&flow models are graphical representations of
coupled differential-equations. Besides the benefit of being easily understandable for the
broad public due to their graphical nature, they implicitly ensure conservation of mass, as
they track quantities in 'stocks' and 'flows' separately from other informational variables. In
such a stock&flow model of DOL in eusocial insects, available workers are the quantities that
reside in a 'stock' and which flow to other 'stocks' which represent worker groups of specific
tasks. Thus, these 'flows' represent the recruitment results of the recruitment system of the
modeled species. Another flow of workers connects back from the task-related stock to the
stock of un-recruited workers. In addition, birth of new animals increases the available
workforce over time, a process which might be affected by successful performance of specific
tasks. In parallel, death decreases the workforce, whereby mortality rates might be task-
related as well. This scheme represents a sort of system-dynamics skeleton that demonstrates
how DOL and population dynamics are easily described from a top-down perspective. As
DOL should be explained by some ultimate reasoning (e.g. by a fitness enhancement of the
colony), such a point of view is advantageous, as it helps to investigate how DOL potentially
enhances colony growth or stabilizes a colony's population dynamics.
We develop here first a simple basic skeleton of a top-down model of DOL. Figure 1
shows a simple stock&flow model of DOL, which we use as a starting point for our
elaboration: Birth is a flow from a source into a stock, as it produces new workers. In parallel,
workers are subject to death, depicted as flows from stocks into sinks. Workers that engage
into a task are tracked in another stock which exchanges quantities of workers with the stock
of un-recruited workers via two flows, called 'recruitment' and 'abandonment'. Abandonment
is proportional to the amount of recruited workers (fixed decay rate). Recruitment is
proportional to the number of recruited workers (active worker-to-worker recruitment), to the
number of unrecruited workers and to the current workload (stimulus). Birth is steadily
producing new workload but creates also new workforce. In contrast to that, task performance
is reducing the workload and, in parallel, is enhancing birth rates. This simplified sketch of a
model clearly demonstrates how population dynamics inside of a social insect colony affects
228 Thomas Schmickl and Karl Crailsheim
DOL and how DOL in turn affects population dynamics by influencing birth rates and by
task-specific mortality rates.
+
births
+
abandonment
Recruited Unrecruited
workers workers
+ recruitment +
death
of task
+ deaths
normal
group + +
Work to
+ be done
work done new work +
Figure 1. A stock&flow diagram of DOL that is embedded in a simple model of population dynamics.
Boxes indicate quantities of workers or quantities of work. Solid arrows indicate flows (transitions) of
quantities. Dashed arrows indicate causal relationships. Cloud symbols indicate sinks and sources.
From this basic modeling sketch, not only the recruitment and abandonment mechanisms
could be elaborated. In addition, the depicted age-structure should be modeled in higher
resolution for depicting DOL in a specific species. In some social insect species, workers are
morphologically (and possibly also physiologically) in-discriminable. Thus the above
depicted model holds for these species. In other species – especially in ants and termites –
there often exist strong morphological differences, which predispose worker castes for
specific jobs. Figure 2 sketches the basic principles of a model that describes DOL in a
species (of ants) that has two distinct morphological castes. As described by Wilson (1984,
1985), the huge 'majors' engage in brood caring only seldom. However, when 'minors' were
experimentally removed, they start with brood care already one hour after the experimental
disturbance of the colony's age structure. From the colony level perspective, the stock&flow
diagram depicted in figure 2 describes the observed flows of workers, as well as the change in
population structure. For the sake of simplicity, we omitted most of the arrows that indicate
causal relationships and depicted just those that are most significant for DOL. Thickness of
the arrows indicates the strength of the causal relationship.
Modeling Population Dynamics, Division of Labor … 229
births births
abandonment abandonment
<births>
Jobs to be
done inside
of the nest
work done new work
Figure 2. A stock&flow diagram of DOL in a species having morphological castes. The model of DOL
is embedded in a simple model of population dynamics. Boxes indicate quantities of workers or
quantities of work. Solid arrows indicate flows (transitions) of quantities. Dashed arrows indicate causal
relationships. Cloud symbols indicate sinks and sources. Only relationships significant for DOL are
drawn.
foraging foraging
getting
births hatching older deaths
Figure 3. A stock&flow diagram of DOL in a species that shows AP concerning two tasks (foraging,
nursing). The model of DOL is embedded in a simple model of population dynamics. Boxes indicate
quantities of workers or quantities of work. Solid arrows indicate flows (transitions) of quantities.
Dashed arrows indicate causal relationships. Cloud symbols indicate sinks and sources. Only
relationships significant for DOL are drawn. Mortality of brood and age classes are omitted, to keep the
sketch simple.
Wakano et al. (1998) modeled a social insect colony with age structure, in which one
vector holds all populations of age classes. They modeled two tasks (in-nest and foraging).
The degree of specialization to foraging or in-nest work for every age-class n is defined by
setting one parameter xn. Their model incorporates task specific mortality rates. Brood
production is limited by performance of both tasks: with no brood care or with no foraging at
all, no brood is produced. Using this model, and by simulating several parameterizations, the
authors reached balancing between the two modeled tasks. They showed life time expectancy
as an ultimate reasoning of the fact that the risky foraging task is always performed by the
older workers in social insects with age polyethism. The model was used to test three regimes
of AP: Hard: age fixed the job; Soft: every age has a certain probability to perform the job;
Not: no age polyethism at all. The authors found that each kind of AP is predicted to be
advantageous under specific environmental and ecological conditions. In case those
environmental fluctuations affect the performance of both, inside and outside jobs, soft AP
was found to be most adaptive. If fluctuations affect only outside tasks (foraging) then hard
AP was found to be most adaptive. The model of Wakano et al. (1998) is not addressing a
specific species, it is a general modeling approach of AP in social insects.
The first issue to be solved in studying DOL in eusocial insects is to know which workers
are available for specific work. In connection with AP, this means to know how many
workers of each age class are available. In case of morphological different workers, it is
important to know how many workers of each morphe exist. As honeybees show an adaptive
regime of AP quite prominently (e.g., Seeley 1982), the age demography of honeybees was
Modeling Population Dynamics, Division of Labor … 231
studied intensively with classical top-down models and only rarely with bottom-up models. In
our work, we followed both modeling approaches. Empiric studies showed that a honeybee
colony can change its own colony structure in reaction to environmental conditions (e.g.
seasonal changes), for example by changing the egg laying behavior of the queen, the nursing
intensity of larvae or by cannibalizing brood (Schmickl & Crailsheim 2001). Thus, studying
these dynamics is an important aspect in the investigation of DOL.
AP is most prominent and most complex in honeybees, thus many demographic models
have been made in this field. Some of them also incorporate aspects of AP, in a way that
either age structure affects AP or vice versa. Omholt (1986, 1988, 1992) suggested several
closely related honeybee models that predict the egg laying of the queen, the workers deriving
from these eggs and how these workers in turn affect the egg laying. These models predict
plausible population dynamics, they predict foraging intensity (honey yields) and they show
that worker longevity could be a response of the nursing load a bee has encountered.
DeGrandi-Hoffman et al. (1989) published a model called BEEPOP, which models the
queen's laying of female (workers) and male (drones) eggs, and the resulting populations
(with age structure). The model assumed that bees up to a specific age work inside of the
colony and that older bees work outside. Thus, BEEPOP implements 'hard AP', according to
the definitions of Wakano et al. (1998). The model is parameterized by weather profiles and it
predicts population dynamics as well as the foraging workforce throughout the year.
Makela et al. (1993) described an object-oriented model called 'AHBsim', which models
the population dynamics and AP of africanized honeybee colonies. What makes this work
outstanding is the fact, that parameters were coded in a sort of 'artificial genome'. Also colony
level reproduction (swarming) and 'drone production' were part of this model. Thus, Makela
et al. (1993) programmed maybe the first 'Artificial Life'-like model of social insects, in
which evolution could be studied on the colony level. In AHBsim, there is also a hierarchical
order, in which worker bees are associated to specific tasks.
In Schmickl & Crailsheim (2007), we presented a model called 'HoPoMo' (honeybee
population model), which models the population dynamics of a honeybee colony from a given
egg laying pattern of the queen. This model, consisting of more than 60 equations,
incorporates a system of 'soft AP', in which adult workers are recruited on a daily basis for the
tasks of 'nursing', 'nectar handling', 'pollen foraging' and 'nectar foraging'. This is done based
on a hierarchical system of task priorities and according to the current status of supply and
demand on the current day. The model incorporates weather data, a seasonal environmental
resource profile and age demography. After elaborating on the basic model, the reproductive
act of a colony (swarming) was simulated, demonstrating that the population model is able to
predict even significant disturbances of colony structure in a qualitatively and quantitatively
plausible way. In addition, the model is able to predict the effects of changes in the colony's
age demography on the emerging DOL and resource dynamics inside of the colony. In turn,
these variables affect the future intra-colonial population dynamics, thus establishing a time
delayed feedback loop between population dynamics and DOL. Figure 4 depicts the most
important feedback loops that are modeled in HoPoMo.
In our HoPoMo model, DOL is based on a hierarchical system of task priorities, which
considers also the current status of supply and demand on the current day: The model assigns
the available pool of worker bees (workforce) to tasks according to the current colony
demand (nutrient flow, brood nursing demands …). As there is modeled a complex system of
interwoven feedback loops, the modeled colony achieves homeostatic regulation of DOL and
232 Thomas Schmickl and Karl Crailsheim
nutrient stores (see figure 4). After elaborating on the basic model, which was validated
against many sets of empiric measurements on honeybees and after exhaustive sensitivity
analysis of the model, the reproductive act of a colony (swarming) was simulated (see Figure
5, Schmickl & Crailsheim 2007). In future, the model will be used to interpret data from
empiric colony-level experiments on honeybees and to develop mesoscopic honeybee models,
as we have described them in the discussion section.
Figure 4. Major feedback loops regulating DOL and population dynamics in HoPoMo. Reprinted from
Schmickl & Crailsheim (2007). Boxes represent stocks that hold quantities (workers, resources). Grey
arrows represent flows. Black arrows represent causal relationships. Circular shapes represent sources
and sinks of quantities. ELR: daily egg laying rate of the queen. SUPcombs: Egg laying suppression
due to little space available on the combs.
Flows of nutrients have a regulating effect on dynamic regulation of DOL not only in
bees but also in other species: Karsai & Balazsi (2002) showed that the inflow of water into a
wasp colony (Polybia and Metapolybia wasps) dynamically regulates the pulp foraging and
thus regulates the nest building behavior. This was shown by a differential equation model,
yielding results that correspond well to empirical results (Jeanne 1986b, Karsai & Wenzel
1998, 2000). A corresponding individual-based model was published recently by (Karsai &
Runciman 2008).
Models that combine mechanisms of DOL in social insects with colony demography are
not frequently produced. Especially since computers became faster and faster in the last
decades, there is a trend of investigating DOL with microscopic, individual-based, bottom-up
approaches. Most of these studies focus on investigating proximate mechanisms of DOL and
TP, as is described in the upcoming section.
Modeling Population Dynamics, Division of Labor … 233
Figure 5. Application of the model HoPoMo. Using this model we simulated the population dynamics
and the change in colony weights (bees, nutrient stores and comb) during colony reproduction. Top
row: Predicted Intra-colonial population dynamics and weight development of a colony that does not
swarm. Lower two rows: Same predictions for a colony that duplicates by swarming. Middle row:
Colony that stays in the hive. Bottom row: Colony that is formed by the swarm that left the hive for a
new home. All: Day 0 corresponds to the 1 st of January in a year.
threshold triggers a behavior (Lorenz 1978). Beshers and Fewell (2001) give an overview
which classes of models have been proposed to explain the observed models of DOL by
different proximate mechanisms working on the individual worker level. These models are
not necessarily incompatible to each other, as mechanisms might differ with species or with
the kind of regulated tasks.
Bonabeau et al. (1996) showed that a model consisting of two distinct castes ('minors' and
'majors'), having different thresholds to perform the task of brood care, suffices to simulate
the colony-level plasticity in Pheidole ants reported by Wilson (1984, 1985). In this
individual-based model, the probability of each individual to engage in the brood caring task
is modeled by the following equation
st2
Pi ,t , (1)
st2 i2
where Pi denotes the probability of an individual of caste i to engage in the nursing task. The
variable t refers to the current time step in the simulation, s refers to the current stimulus
strength (workload to be performed) and Θi refers to a fixed threshold of caste i that has to be
met by the stimulus to trigger the task. In their simulations they set Θ1=8.0 and Θ2= 1.0.
Equation 1 has been used several times in literature to model behavioral response thresholds.
It became a quasi-standard, although a simple step-like function ( 'if s>Θ then act()')
would suffice in most cases. According to the parameter values used in Bonabeau et al.
(1996), the probabilities to engage in the brood-care task will scale with stimulus intensity as
depicted in figure 6. In their model, individuals abandoned a task with a fixed rate, assuming
a fixed time period for task performance. The dynamics of the stimulus were modeled as
follows: The stimulus increased by a linear term and was decreased proportionally to the
fraction of active workers of both castes in the number of all available workers. Thus, the
authors assumed equal efficiency of both castes in performing the task.
0,9
0,8
Probability of task engagement
0,7
0,6
0,5
0,4
0,3
0,2 minors
majors
0,1
0
0,1 1 10 100
Stimulus intensity
Figure 6. Probabilities of task performance of 'majors' and 'minors' in the model of Bonabeau et al.
(1996). It needs a higher stimulus intensity to trigger the nursing task in a 'major' than in a 'minor'.
Modeling Population Dynamics, Division of Labor … 235
Using this model of fixed thresholds, Bonabeau et al. (1996) could predict the dynamics
of experimental removal of 'minors' in Pheidole ants by assigning different values of Θi for
'minors' and 'majors' in a way that was comparable to empiric observations (Wilson 1984,
1985). This model was extended to a system with two tasks, where each caste is specialized to
one of the two tasks. The authors called the fixed-threshold model ―...t he arguably simplest
model that connects flexibility at the worker level with the resiliency observed at the colony
level...‖ (Bonabeau et al. 1998). Although the model is simple, they report in their extensive
paper that they have been able to simulate several aspects of dynamic DOL: Not only the
plasticity observed in Pheidole ants, but also spatial specialization of workers, and AP could
be modeled. The authors introduced also genetic (threshold) variability. For modeling AP,
they did not vary the individual's threshold over time, but the exposure or sensitivity to a
certain stimulus was depending on a worker's age. The fact that different ages of workers
occupy usually different spatial locations in the colony and are thus exposed to different sets
of stimuli, was also considered by the 'foraging-for-work' hypothesis (Tofts 1993, Tofts &
Franks 1992, Franks & Tofts 1994): This hypothesis is based on the assumption that spatial
displacement of workers, as they grow older, moves them from the central brood nest to the
periphery of the colony, as new workers constantly hatch in the brood nest. In consequence,
the encountered bouquet of local stimuli changes with this displacement. Thus workers
engage sequentially in tasks that correlate with their age, without requiring any intrinsic
specialization mechanism that explicitly reflects a worker's age.
The above mentioned models assumed fixed thresholds that stay constant for every
individual through time. Lorenz (1978) discussed the importance of such behavioral
thresholds and also described their adaptation: Both, the exposure to stimuli as well as the
successful performance of a stimulus-triggered behavior modulate such thresholds. The
phenomena of 'sensitization' and 'facilitation' refer to the fact that once a stimulus has
triggered a behavior, the animal is more likely to repeat this behavior again, even at a lower
stimulus level. This is equivalent to lowering behavioral thresholds. After some time without
such a stimulus, these effects disappear, what is equivalent to increasing these thresholds.
These mechanisms were incorporated into the previously described threshold-reinforcement
models, which assume that the behavioral threshold Θi,j,t of individual i to engage in task j at
time t is decreased each time a worker engages in a task. Thus the worker gets more
specialized to that task j. The same threshold is increased in each time step when the task is
not performed. Variants of this model were analyzed and published by Deneubourg et al.
(1987), by Plowright & Plowright (1988), and by Theraulaz et al. (1991, 1998): In these
models, the probability to engage in a task i for individual j is described by
si2, j ,t
Pi , j ,t (2)
si2, j ,t i2, j ,t
This is a variant of equation 1 in which si,j,t is the associated stimulus to perform task j, as it is
sensed by worker i at time t.
Like in the fixed-threshold models, performance of a task leads to a local decrease of the
associated stimulus strength, but stimuli accumulate again as new workload emerges
constantly. Using this mechanism, simulations showed (Theraulaz et al. 1998) that the
236 Thomas Schmickl and Karl Crailsheim
modeled colony members quickly split up into two groups: one specialized on the first task
and one specialized on the second task. It was also shown that removal of one worker group
leads to an increase in stimulus strength of the associated task and this fact ultimately allows
specialists of another task group to re-specialize. Theraulaz et al. (1998) modeled this
mechanism in a time-continuous way, considering the fractions of time spent by agent i with
and without engaging in task j. In contrast to that, Gautrais et al. (2002) modeled this process
in an individual-based model. They showed that such effects occur more prominently in
bigger simulated colonies compared to small colonies. Equation 3 gives the details for the
threshold adaptation mechanism:
The self-reinforcement model incorporates a strong positive feedback, which drives the
system into a state in which individuals are quickly caught by one of several basins of
attraction. If there are enough other individuals around performing the other task, they will
stay in this basin (of specialization) for a long time. Merkle and Middendorf (2004) identified
several shortcomings of the Gautrais et al. (2002) model: The observed non-specialization in
small colonies was shown to be an effect of initial conditions, because in the initial phase of
the simulation runs, workload (demand) was growing much slower in small colonies than in
bigger colonies. As only frequent task-performance leads to specialization of a worker, these
colonies developed mainly generalists. The authors also identified the maximum height of
thresholds as a crucial parameter: The bigger this maximum value is, the more often an
individual has to perform a task within a short window of time to become a perfect specialist.
In addition to changes of parameterization and simulation procedures, Merkle &
Middendorf (2004) introduced also extensions to the original model: They modeled a limited
life-span of workers and introduced new (grown) workers into the system. In addition, they
varied maximum thresholds with age (comparable to suggestions in Bonabeau et al., 1998),
thus extended the model to depict also AP. Their model extensions contained also a
'competition-for-work' aspect, where those individuals with the lowest thresholds for a given
task were selected to perform the task. Based on such elaborations, Jeanson et al. (2007)
demonstrated that colony-size dependent specialization is observable also without self-
reinforcement of behavioral thresholds, as was already proposed by Bonabeau et al. (1998).
In addition to these IBMs and monte-carlo models that are based on thresholds, there
have been suggested also other mechanisms for explaining dynamical task selection in social
insects. These hypotheses were also investigated by mathematical models to analyze how
DOL emerges within such systems. It is very likely that several of these mechanisms act in
social insect colonies, as there might be different mechanisms working in different species
and various mechanisms probably act in parallel within one colony.
Modeling Population Dynamics, Division of Labor … 237
averaging over several of such partial transfers, the information quality has improved, to the
cost that several storing bees had to be searched.A further elaboration of this model (Gregson
et al. 2003) showed that multiple transfers arise automatically as a by-product of crop-filling
mismatch between returning foragers and waiting storing bees. Thus, such multiple transfers
could be a result of the variance of crop fillings. These multiple unloading acts are probably
not a selected adaptation to improve performance of foraging.
Seeley (1992) reports that queuing delays are the heart of forager regulation in
honeybees: The longer a forager waits for a storing bee, the more likely the forager will
perform a tremble dance, which is a signal for other bees to engage in the task of storing. In
contrast to that, if the searching time was short, the forager is very likely to perform a waggle
dance, which recruits more bees to exploit the nectar source the forager returned from. It is
clear that here two feedback loops regulate DOL concerning foraging and storing. Thus, these
two feedback loops regulate the colony's collective foraging behavior. Anderson (1998)
constructed a continuous-time stochastic model, which incorporates a simplified version of
the probability curves depicted in Seeley (1992). The queuing was modeled similar to the
models in Anderson& Ratnieks (1999a,b). In a simulated colony of 100 workers, they could
switch from storing to foraging or vice versa, depending on the experienced queuing delays.
The simulation shows that the simulated colony adjusts its task-partitioning in reaction to
environmental changes like a sudden change in the time foragers need for their trip.
Anderson (1998) points out the importance of the negative feedback loop that arises in
the system through the existence of storing bees. Foragers depend on storing bees and thus
they represent a limited resource. The waggle dance provides a positive feedback on the
foragers, increasing the forager group to the cost of the storing group. And the tremble dance
recruits new storing bees to the cost of the forager group. Fewer forager bees lead to a
decrease in the forager return rate at the hive entrance, what in turn decreases average
queuing delays of foragers. This finally leads to less tremble dancing. This self-referencing
chain of causal relationships establishes a classical negative feedback loop which balances the
system around a set point. One fact that was not incorporated in the model of Anderson
(1998) was the existence of other bees in the hive (e.g., nurse bees) and also age demography
was not reflected: In honey bees, storing bees are usually recruited from mid-aged bees and
forager bees are usually older bees (Lindauer 1952, Seeley 1982), thus – in future – such
models should be implemented with an underlying age demography model, as it is discussed
at the end of this chapter. Such an underlying demographic model will predict for every point
in time, how many mid-aged bees there will be available and unemployed, given the fact that
these bees engage also in other tasks, like, for example, wax building.
It is interesting, that the empirically found probability-curves for tremble dances and
waggle dances under different conditions of searching time well reflect the threshold curve
modeled by variants of equation 1:
, (4)
where st refers to the experienced searching time. With values of Θwaggle=10 and n=2.5, a
good agreement to empirical data is found (Schmickl & Crailsheim 2004). The probability to
perform a tremble dance is well modeled by
Modeling Population Dynamics, Division of Labor … 239
, (5)
where values of Θtremble=40 and n=5 model good agreement (Schmickl & Crailsheim 2004)
with the empirical data reported by Seeley (1992), as it is shown in Figure 7.
1,2
Probability to dance
to dance
1,0
P(response)
0,8
P (danc e )
0,6 P (abandon)
Probability
P (no- danc e )
0,4
0,2
0,0
0 40 80 120
Figure 7. Probabilities to perform either a waggle dance or a tremble dance. A: Empiric observations in
honeybees (Seeley 1992). B: Probabilities that were modeled according to equation 4 and 5. We used
values of Θ=10, n=2.5 (for p waggle) and Θ=40, n=5 (for p tremble) in our multi-agent models.
worker groups by sharing a social water stomach was predicted also based on individual-
based proximate mechanisms.
Also in ants, food is handled often from one individual to another. Reyes and Fernández-
Haegar (1999) reported a special case of TP in the seed-harvester ant Messor barbarus:
Larger ants take over the food from smaller ants on the foraging trail. Similar observations
were also reported from other ant species, e.g., grass-cutter ants. As loaded ants always walk
towards the nest and empty ants predominantly walk in the outside direction, this behavior
finally leads to an interesting pattern, which emerges from parallel execution of these simple
behavioral rules: The ants end up sorted on the trail, having the larger ants closer to the nest
than the smaller ones. To analyze and interpret this phenomenon, Anderson et al. (2002)
constructed an individual-based model of this system and showed that simple rules suffice to
generate the observed pattern: (1) Ants move only along a line (of discrete patches) between
the nest and a seed pile. (2) Unloaded ants move towards the seed pile where they pick up a
seed and become loaded ants. Loaded ants move towards the nest side where they drop the
seed and become unloaded ants. (3) Ants are of differing size, drawn either from a continuous
distribution or from a discrete set of sizes. (4) When a loaded smaller ant meets an unloaded
larger one, the food is handled over. This set of rules led to a correctly size-sorted ant trail
within the time period a single ant would need to perform one whole foraging trip in the
system. The process described above reminds us on a sort of 'bubble-sort' algorithm, as it is
known from computer science, performed by ants. Anderson et al. (2002) discuss in their
paper whether or not this should be interpreted as being an 'artifact' that arises from the fact
that the larger ants just grab food items if they can. They mention many arguments pro and
contra the adaptive value of this system: The most prominent dis-advantage is the additional
time the ants spent for handling over the item multiple times during one round trip. This is a
typical 'con' when discussing possible ultimate causations of any TP system. The major
advantage of this TP could be that ants frequently run on the similar part of the foraging trail,
thus get familiar with it and probably increase their movement speed. And, if large ant run
faster, they will maximize their contribution to the total foraging process, thus they will
increase the average total foraging tempo.
Not only food transmission from one individual to another is proposed in literature to
regulate DOL and TP. There is also frequent transmission of chemicals (and tactile stimuli)
involved in the social behavior of ants, wasps, bees and termites. Such transmissions are also
assumed to have a regulatory effect on DOL and TP. Huang & Robinson (1992) suggested an
'activator-inhibitor' model, which was later extended by Naug & Gadagkar (1999), by adding
variable brood-to-adult ratios, which change the frequencies of adult-to-adult interactions. In
these models, all agents produce an activator at an age-dependent rate, which progresses their
'physiological age'. In addition, they produce an inhibitor (intrinsic inhibitor), which is
transferred to their partners on social interactions. These received inhibitor substances in turn
slow down the progress of physiological age progression in the receiving workers. This
model predicted the emergence of AP, similar to observed dynamics in the eusocial wasp
Ropalidia marginata. The activator and the inhibitor could either be considered as exchanged
pheromones, or any other kind of exchanged signals that alter the physiological status (e.g.,
hormonal status) of the receiving worker. For the case of honeybees, Watmough (1997)
produced a model to analyze the basic properties of the spread of queen-produced
pheromones within the colony. This transmission works by direct contact (tactile, licking) and
via the wax of the comb. Watmough (1997) used a set of stochastic differential equations,
Modeling Population Dynamics, Division of Labor … 241
which describes the rate of change of pheromones in worker bees, in the queen and in/on the
wax of the comb. Their model shows that high congestion in the hive impedes transmission of
queen-derived pheromone, because the pheromone is 'diluted' by more workers and, due to
slower movement, the workers less frequently meet 'new' bees to transfer the pheromone to.
This crowding effect is discussed to be one of the major factors affecting the timing of colony
reproduction in honey bees (swarming). It will be interesting to produce an individual-based
model concerning this 'crowding issue' and to merge this model with a model derived from
the Naug & Gadagkar (1999) model.
It was found that the role of juvenile hormone, which was the first candidate hormone to
act as an 'activator' in such an 'activator-inhibitor' model, is maybe not as prominent for
physiological progression as it was primarily thought. Beshers et al. (2001) formulated a
revised version, called 'social-inhibition' model. In this model, the variable Xi,t represents the
physiological state (physiological age) of individual i in time t. This variable is increased at a
given rate, as the individual gets older. Social contact with other workers slows down this
progression. Several patterns of DOL, like for example the reallocation of workers after a
sudden removal of a worker age-group were predicted with this model in a way that is
comparable to empiric observations.
In an insect colony, not only pheromones and information are exchanged. Also parasites
and diseases find a rich substrate to spread. Naug & Camazine (2002) produced a spatially
discrete (cellular) individual-based model which depicts worker-to-worker interactions by
allowing agents in neighboring cells to potentially interact. This way, also a pathogen spreads
from one agent to another. The model does not contain birth and death explicitly. Infected
agents stay ill for some time, in which they infect other nearby agents. After some recovery
time, they get susceptible again (or are replaced by new susceptible workers). The model
predicts that DOL alone does not affect pathogen transmission significantly. But after
implementing also heterogeneity in the interaction network and age demography of agents,
the authors found pathogen transmission slowed down. This modeling study shows a different
approach to find ultimate causations for DOL and TP, besides the usual economic
considerations of energy, time and risk.
Using individual based models, many options for proximate mechanisms have been
investigated that were proposed to regulate DOL in many social insect species. Such models
demonstrate that the limited amount of information available for individual workers suffice to
affect task selection of workers in a way that colony level behavior is regulated. Thus, such
models have manifold advantages: Many models demonstrate swarm intelligence of
distributed task selection, and show how this kind of intelligence arises as emergent
phenomenon of self-organization. Such models often nicely demonstrate how worker-to-
worker interaction and the exploitation of side-effects of other workers' behaviors ('cues')
support colony-wide homoeostasis and collective decision making.
However, such models frequently do not deal with the question of ultimate reasoning,
which is important to interpret how these proximate mechanisms emerged through natural
selection. In the upcoming section, we describe the individual based models
‗HoFoReSim‘,‗FlowerSim‘ and ‗TaskSelSim‘' in detail, which are aimed to answer both
important questions concerning DOL in honeybees: The 'how?' and the 'why?' question. To
answer the ‗how?‘ question, one has to model the hypothetical mechanisms that regulate
DOL, which was done by various models in history. To answer the ‗why?‘ question, worker
physiology, nutrient balance and energy economics have to be modeled. Our set of models
242 Thomas Schmickl and Karl Crailsheim
described in the upcoming sections belong to the rare species of models that target both
objectives simultaneously, allowing to generate predictions and hypotheses that cannot be
produced by any specialized model that just targets one of these two questions alone.
, (6)
where gain refers to the energetic gain of the foraging trip, that is the amount of energy
contained in the nectar load that is finally brought back to the hive (in Joule). The variable
cost (also measured in Joule) refers to the energetic cost that was invested by the forager.
Clearly, gain increases with the sugar concentration of the offered nectar and cost increases
with increasing distance of the food source to the hive. Such colony-level foraging was
investigated with an ODE model (Seeley et al. 1991) and with multi-agent simulation (de
Vries & Biesmeijer 1998, 2002). In our research, we aimed for understanding the colony-
level implications of foraging decisions. Thus we aimed for modeling the energetic costs
which dynamic foraging environments impose on the colony. We hypothesize that every
foraging decision causes characteristic costs in terms of energy which is spent by the colony
during making this decision. To allow such an analysis, we created the first model HoFoSim
(Honeybee Forgaing Simulation), which is a multi-agent implementation of the honeybee
foraging system (Schmickl & Crailsheim 2004, Schmickl et al. 2005). We implemented each
forager as an autonomous agent, having its own crop volume for foraging, having its own
load-dependent weight and its own metabolism which consumes energy. During every
foraging trip, and also during the period a bee agent stays in the hive, energy is spent for
every time step. After a successful foraging trip, foragers deposit new amounts of energy
(sugar) to the colony's global nectar stores. In a later version of the model, we implemented
also nectar receivers with their physiological parameters, thus we could also investigate the
colony-level costs of mismatches between the group of foragers and the group of receivers
(Thenius et al. 2006, Schmickl et al. 2010). This version of the model is called HoFoReSim
(Honeybee Forager and Receiver Simulation). The probabilities of performing either a waggle
dance or a tremble dance were modeled according to Figure 7. The correlation of dance
duration with food source profitability was modeled according to Seeley (1994). In the
models HoFoSim and HoFoReSim, transitions from one behavioral state to another, as they
are depicted in Figure 3b, are triggered either by events that happen to agents, by fixed
delays, or by stimulus-response functions, as they are described in equation 1.
Modeling Population Dynamics, Division of Labor … 243
Figure 8a shows a screen-shot of the simulator and Figure 8b shows a schematic drawing
of the finite state automaton that we used to implement the agents' behaviors. Based on these
behavioral states (e.g. flying or not flying) and based on their current weight (e.g. caused by
their nectar load), the energetic expenditures of agents were calculated.
Using HoFoSim and HoFoReSim, we were able to investigate how foraging decisions
affect the global energy economics of the colony. We found that foraging decisions have a
significant impact on the nectar-collection efficiency of differently sized colonies (Thenius et
al. 2006). Figure 9a,b show a comparison of collective foraging decision making, as it was
observed in real honeybees and as it is predicted by HoFoReSim. Figure 9c shows that our
model predicts that such sudden environmental fluctuations, which cause such foraging
decision, impose a significant cost to the colony. We established several colony conditions
(ratio of forager cohort to receiver cohort) and found that these intrinsic properties of the
colony affect foraging decisions and nectar economics of the colony significantly, especially
if the environment is dynamic (Schmickl et al. 2010).
Figure 8. (A) Screen shot of our multi-agent model HoFoReSim at runtime. a: Hive. Grey dots in the
hive are bee agents. Darker dots indicate dancing bees, advertising their food source. b: Entrance to the
hive. c,d: Nectar feeders. e,f: Forager bees flying to and from these feeders. g: Scout bees, flying
randomly to search new sources. (B) Finite state automaton that we used in HoFoReSim to model the
agents' behaviors.
244 Thomas Schmickl and Karl Crailsheim
In Figure 9d,e,f, we show that fluctuations in the environment are predicted to allow a
bigger colony to react much more precisely in our model compared to a smaller colony: To
investigate the ability of our virtual colony to adapt to sudden changes in the environment, we
spontaneously doubled the length of the foraging flight cycle as proposed by Anderson &
Ratnieks (1999a) and switched it back to initial duration after a period of 2 hours of simulated
time. We measured the number of forager bees recruited to the source. Using HoFoReSim,
we found, that bigger colonies react to changes of the foraging flight cycle length withab a
stronger and more precise recruitment of forager bees than smaller colonies (see figure 9d,e).
Colonies with 200 bees show little adaptation, during the phase of elongated foraging flight
cycle, which lasted from hour 2 to hour 4 of the experiment. Colonies with higher numbers of
sm
[ ]
l
bees (1000 bees) show a strong recruitment of forager bees within the first hour of the phase
ys
toe
r
of elongated foraging flight cycle.
o
co
l y
nhn
oe
(a) 4
5
8
9
0
1
2
3
4
5
60
0
8:0
50
00w
f
lu
f
it
f
c
o
hl
t
uo
u
r1
u
c
a
a
t
0
tg
i
u
o:0
e
n
a0
r
t
io
bn
e
:e
2
s.
t
,
5
1im
M
2
2:
5
e
o
000
l
/Lr
:
e1
c.
0eM
i
v1
oe
4
l
:
/
rL
0b
0e
es1
6:00
empiric data model results 500 forager bees, 1500 receiver bees
fluctuation: 2.5Mol/L:1.0Mol/L
620
Source A with a
[ml] [ml]
fluctuation
bees on source / 30min
600
Source A 100 Source B
bees on source / 30min
without
fluctuation
stores
580
honey stores
Source B 560 b
colonyhoney
50 540
520
colony
500
0 480
08:00 12:00 16:00 08:00 12:00 16:00 08:00 10:00 12:00 14:00 16:00
(a) time of day (b) time of day (b)
(c) time
Small colony (200 forager bees) Bigger colony (1000 forager bees) Economics of different colony sizes
Figure 9. Results obtained from the model HoFoReSim. (a) Collective behavioral foraging pattern
emerging in honeybees as a result of a sudden switch in the qualities of two nectar sources at t=12:00.
Source: Seeley et al. (1991). (b) Corresponding results from the model. Source: Schmickl & Crailsheim
(2004). (c) Such fluctuations lead to a significant decrease in the nectar accumulation by the colony.
Source: Schmickl et al. (2010). Number of recruited forager bees in a colony of 200 bees (d) and in a
colony of 1000 bees (e): During the phase of elongated foraging flight cycle (hour 2-4) there is hardly
any reaction visible in the number of recruited forager bees in the small colony. The bigger colony
reacts to the phase of elongated foraging flight cycles with massive forager recruitment. When the
foraging cycle length drops back to initial level, the number of recruited foragers drops back to normal
level. Bold lines: Mean values (left y-axis); Thin lines indicate standard deviations; Gray line: relative
flight cycle duration (right y-axis); n = 6. (f) The nectar economic efficiency of honeybee colonies is
predicted to be dependent on colony size. (d-f).
Modeling Population Dynamics, Division of Labor … 245
In addition to that, the model predicted a significant decrease of working efficiency per
forager bee for small colonies, showing that the more precise reaction to environmental
fluctuations of larger colonies are predicted to result also in a colony-level fitness
enhancement.
One interesting aspect of honeybee foraging that we wanted to investigate with our model
HoFoReSim was the proximate-reasoning of the observed variance of dance responses in
honeybee foragers. Seeley (1994) showed that some bees very well adapt the number of
dancing-rounds they spent in waggle dancing to the observed quality of the food source. But
some foraging bees do not react well to different source qualities and dance almost the same
amount of time (rounds) for all tested sources.
Our hypothesis was that bees with steep dance-response curves are necessary for quick
food source selection, as they impose positive feedback on the honeybee foraging by re-
directing other foragers quickly to the best food sources. In contrast to that, we hypothesized
that other, not-so-responsive bees are necessary to prevent the foraging cohorts from falling to
quickly into the traps posed by ‗local optima‘. Thus, the empirically observed variance in
dance response curves was hypothesized to represent a near-optimal balance of ‗exploitation‘
(steep curves) to ‗exploration‘ (flat curves). Such a balance is important for all kinds of
optimization or search algorithms. Another hypothesis was that this balance between
exploitation and exploration is of higher importance in fluctuating environments than in stable
environments. To test these hypotheses with our model HoFoReSim, we generated colonies
(by parameterization) which reflect the empirically observed data (called ‗homogeneous
parameterization‘) and other colonies, which just contain the greedy bees with steep dance-
response curves. Both colonies were then simulated in environments with stable conditions
and in environments with fluctuations. In such fluctuations, nectar sources that initially were
of high quality became suddenly worse and nectar sources which initially were of bad quality
improved suddenly. As all food sources were placed in identical distance to the hive in our
simulation, quality of food sources was varied by changing the sugar concentration of those
food sources, as it was done empirically by Seeley et al. (1991). We found that in all
environments, regardless of the stability of food sources‘ qualities, the heterogeneous (nature-
near) parameterization yielded the highest collective nectar collection rate (see figure 10).
The ‗very sugar-selective‘ homogenous set of dance response curves was predicted to
perform well neither in the often changing environmental conditions, where quick re-
decisions of the colony were expected to be optimal, nor in the stable environment, where the
first decision never has to be reverted (local optimum = global optimum). We found that
excessive dancing for food sources was a waste of time in the homogenous parameterization
as the bees spend more time and energy in dancing was necessary to recruit all flying
foragers. These findings gave new input to the discussion why not all honeybee foragers
dance intensively for profitable food sources (Thenius et al. 2005).
Our model predicts that in some environmental conditions, it is worth to pay the price of
changing the focus of massive forager recruitment, while in other conditions such new
recruitments are too costly for the colony (Schmickl and Crailsheim, 2004). To analyze this,
we conducted the following experiment. We performed a simulation run of 8 hours of
simulated time in an environment containing one high quality nectar source (2.5 Mol/L) and a
one poor nectar source (1.0 Mol/L). At time point t=12:00 we saved the status of all agents
and of the environment, afterwards we continued the simulation until t=16:00 without any
246 Thomas Schmickl and Karl Crailsheim
disturbance of the system. Such a ‗time-machine‘ procedure can only be done in a simulation,
never in real world experiments.
heterogenous set of dance response curves homogenous set of dance response curves
50 50 50
N = 20 N = 20
45
40 40
dance rounds done
bee #1
* quality-index
80
*
60 * * *
40
20
0
0 1 3 7 17 31
c) number of changes
Source: Thenius et al. (2005).
Figure 10. Empiric data, model parameterization and predictions of the model HoFoReSim. (a)
Empirically observed dance-response curves of individually marked bees. Seouce: (Seeley 1994). (b)
Two different parameterizations that we used in our model HoFoReSim for a comparative study. Left:
Parameter set ‗heterogeneous‘, which reflects empiric data observed in honeybees. Right: Parameter set
‗homogenous‘, which makes all bees show maximum responsiveness to food source quality. In this
setting all bees discriminate very well between food sources of different qualities and adjust their
dancing accordingly. (c) Model predictions: HoFoReSim predicts that in fluctuating environments
(number of food quality changes during runtime on x-axis) the heterogeneous (nature-like) parameter
setting leads to colonies that collect more honey over time, as they stay responsive enough but waste,
on average, less time for dancing.
The final nectar gain of the colony in this simulation run was memorized as variable A.
Then we loaded again the configuration of t=12:00 from hard disk and manipulated the food
sources according to the empiric experiment described in (Seeley et al. 1991): The former
good food source decreased its quality to 0.75 Mol/L of sugar concentration and the former
bad source increased its quality to 2.50 Mol/L. As already depicted in figure 9a and 9b, our
model predicts plausible patterns of forager recruitment and re-recruitment for this
experimental procedure. We continued this run then until t=16:00 and memorized the final
honey gain now as variable B. Then we reloaded the configuration of t=12:00 for a third time,
again performed the same nectar source fluctuation. But this time, we did not allow the bees
in our model to switch nectar sources in this 4 hour of the afternoon. We simulated non-
functional honeybees concerning foraging target decisions this time. The final honey gain at
t=16:00 was recorded as variable C. Based on these data we were now able to calculate the
benefit of the foraging decision as
Modeling Population Dynamics, Division of Labor … 247
, (7)
as it can also be read also from figure 11a. The benefit of the foraging decision is the amount
of nectar that can be additionally accumulated by the colony compared to the run where we
impaired the bees from switching their foraging target after t=12:00. Also the costs of the
environmental fluctuation can be predicted by comparing the undisturbed honey gain A with
the final nectar, described by
. (8)
Additionally, we investigated how the frequency of fluctuations affects the costs that are
imposed by this environmental instability to the colony nectar economics. As is shown in
figure 11b, the step from no fluctuation to one fluctuation caused the highest honey deficits in
our simulated colonies. Additionally increasing the frequency of fluctuations did increase
these costs but with decreasing significance. This could be explained that we simulated
environments with only two nectar sources that switched their quality frequently from ―good‖
to ―poor ‖. The collective decision making of honeybees takes some time thus in simulations
with high frequency of environmental changes the foragers distributed 50:50 over both nectar
sources and encountered there a good source for 50% of all foraging trips and a bad source
for the other 50% of instances.
We also altered the amplitude of environmental fluctuations in the one-fluctuation
environments to see how this amplitude is predicted by our model to affect costs. We defined
‗amplitude of fluctuations‘ as follows: After 12:00 the quality of the ‗worsened‘ nectar source
was varied. While in Seeley et al. (1991) the nectar source that had poor quality in the
afternoon was set to 0.75 Mol/L, we tested several sugar concentrations on that source in
repeated runs. The smaller the difference of quality of this source to the good source in the
afternoon (always 2.50 Mol/L) was, the higher the amplitude of the fluctuation was. As is
shown in figure 11c,d the highest colony level cost were caused by medium-sized amplitudes.
With low amplitudes, the colony simply doesn‘t switch it‘s major foraging target, thus
minimizing costs. With high amplitudes, the source selection process happens fast, also
minimizing costs.
Additional runs of these experiments were performed with HoFoReSim in (Schmickl et
al. 2010), where we investigated also the energetic costs caused by storer bees in such
scenarios.
In the experiments described so far the environmental fluctuations happened suddenly,
that is from one time step to another. However, in nature the process in which plants increase
and decrease their nectar flow is slower. As these flows lead to ‗turning on‘ and ‗turning off‘
of nectar sources of different distances to the hive and of different sugar concentration, we
assume that also the ‗speed of change‘ in environmental fluctuations affects colony level
costs of decision making. To test this in Schmickl et al. (2005) we performed again
experiments as they are described in Seeley et al. (1991), but the changes of quality of nectar
sources around t=12:00 take a certain period of time, which was expressed by the variable h.
Figure 12 shows that, the faster fluctuations happen, the lower is the global nectar cost
imposed on the colony, as fast fluctuations cause fast recruitment patterns and fast
reallocation of foragers to better food sources.
248 Thomas Schmickl and Karl Crailsheim
(a) (b)
time time
Figure 11. Analyzing the colony-level gains and costs of foraging in a fluctuating environment. (a)
Graphical analysis of costs and benefits of decision making in a single-fluctuation environment. (b)
Analysis of costs and gains of decision making in colonies confronted with environments that change
the quality of nectar sources with different frequencies. (c) Analyzing costs and gains of decision
making in single-fluctuation environments in which the fluctuations happen with different amplitudes.
(d) Dependency of the final honey gain on the amplitude of the fluctuation. The higher the sugar
concentration on source A was in the afternoon (x-axis) the lower was the amplitude of the
environmental fluctuation.
In addition to the studies mentioned above, the costs imposed by the interplay of forager
bees and storer bees were predicted in additional studies performed with HoFoReSim
(Schmickl et al. 2010). In these studies, also several empiric studies were mimicked by the
model (‗equilibrium dancing‘, ‗cross-inhibition‘, ‗choices‘) and the resulting nectar
economics that are predicted by our model for these special environmental conditions were
predicted.
Modeling Population Dynamics, Division of Labor … 249
75
70
65
60
55
50
h=.01 h=.25 h=.5 h=1 h=2 h=4 h=6 h=8
single fluctuation scenario
Figure 12. Analyzing the colony-level costs of foraging in a fluctuating environment in which
fluctuations happen with different length of the transitional period. The smaller the parameter h is, the
faster (the more sudden) the quality changes in the environment happened in our simulation.
soon as a receiver bee is sufficiently filled, it disappears from the interaction platform for
some time to add the nectar to the global nectar stores of the colony. Figure 13 shows a
screenshot of this simulation at runtime. The agents (foragers and receivers) are driven by a
finite state automaton similar to the one used in HoFoReSim (see Figure 8b). We analyzed the
emergence of partial crop loads in foraging bees, which were found to correspond well with
empirical measurements in honeybees (Thenius et al. 2008).
In contrast to other works on this field (Bartholdi et al., 1993; Camazine & Sneyd, 1991)
FlowerSim models the focal honeybee colony not using differential equations. Instead, it
models individual honeybees to predict the emergent colony behavior in reaction to individual
experienced information and clues inside of the colony and outside in the environment. This
is especially important as individual differences between bees lead to colony-level
predictions, that are not predicted by model that assume a homogeneous colony build from
identical actors. A special focus in FlowerSim is made on varying flows of nectar per
blossom, individual honey-stomach sizes of foragers and receivers, workload balancing
mechanisms inside the colony and possible informational clues that can be experienced by a
single forager.
Our model is aimed to make predictions about how individually experienced multiple
unloading might serve as a source of information for the workers, as changes in the nectar
flows offered by blossoms changes the filling-status of the workers that return to the hive. As
is described below, our model FlowerSim shows that individual foragers have not only
information available that informs them about the current work-load balancing (foragers vs.
storer bees) inside the colony, but also about changes in flow of nectar on other sources,
exploited by a different forager cohort, regardless of sugar-concentration of the nectar.
Empiric work regarding the partial nectar loads and their possible feedback onto colony
behavior was done by Dornhaus et al.(2006), by Huang & Seeley (2003) as well as by Hart &
Ratnieks (2001). In contrast to other multi agent simulations in this field (e.g. Dornhaus &
Klügl, 2006) the interplay of foragers and receivers on a individual level was modeled in
FlowerSim incorporating many literature-derived morphological and physiological details.
This allowed us to investigate the effects of mismatches between nectar-stomach fillings of
foragers and storers based on noise and changes in the environmental nectar flow conditions.
Figure 13. Screenshot of the simulation FlowerSim at runtime. The spotted grey areas in the upper part
represent flower meadows or trees filled with blossoms. The lower narrow longitudinal area represents
the transfer zone between foragers and receivers. The white circles indicate honeybee agents, grey
circles indicate storer bees.
Modeling Population Dynamics, Division of Labor … 251
Using the model, we performed the following experiment: After an initial period in which
both flower fields offered the same nectar flow, we increased the nectar flow in one flower
field for some time and then the original nectar-flow was restored. The model predicts that
during the period of high nectar flow on one field, the foragers that return from the other
flower field -- which was never altered -- experience a significant increase in forced multiple
unloading acts (see Figure 14a), as the receivers were filled to a higher level by the foragers
returning from the other flower field. In addition, also the queuing delays increased
significantly (see Figure 14b). Thus, the model predicts that there might exist a
communication channel between these two groups of foragers via the commonly shared and
limited resource of receiver bees. Such a communication channel was not described so far and
is yet subject to be investigated by empiric experimentation.
Figure 14. (a) Even though the nectar source B did not change at all, the foragers returning from that
source encounter an increase of multiple unloading as long as nectar source A improved concerning
nectar flow. (b) In addition to multiple unloading, also the experienced queuing delays encountered by
foragers returning from source B increased significantly. Both results indicate a potential unknown
channel of communication between distinct forager groups that communicate via the shared group of
receiver bees. Bold line: median; dashed line: 1st and 3rd quartiles. (c) Allowing the foragers to adapt
their dancing behavior to this additional source of information significantly increased the nectar
accumulation efficiency of the colony.
One hypothesis how honeybees could react individually to changes of unvisited flower
patches in a way that is efficient for the collective can be described as follows: We assume
that foragers might change their waggle dance behavior in response to an encounter of sudden
high queuing delays without a quality enhancement of the own visited nectar source. If a bee
just ceases their dancing behavior in such situations, more recruitment will happen to other
flower patches, which just have increased their nectar flow per time unit and/or per blossom.
Such a reaction would be beneficial for the colony which has recently discovered another
flower patch which is better than the one the focal bee is currently foraging on. Figure 14c
shows, that such a hypothetical way of 'social learning' is predicted by our model to
significantly enhance the colony's global nectar accumulation potential. Thus, such an
individual-based reaction of the 'unloading cue' is assumed to enhance the colony's overall
efficiency.
252 Thomas Schmickl and Karl Crailsheim
Figure 15. Screen shot of our multi-agent model TaskSelSim at runtime of a simulation. a: Overview of
the modeled honeybee hive. b: Forager bees (blue) are leaving the hive, fly to a nectar source, fill their
crops with nectar and return to the hive. c: A forager bee performs a waggle dance (blue patches) to
recruit other foragers. d: A forager bee performs a tremble dance (pink patches) to recruit storing bees.
e: Larvae are nursed by nurse bees (red). There are also unemployed (yellow) bees around. The slight
brown shading depicts the local hunger signal intensity, which is emitted by larvae. f: A returning
foragers emits the 'unload me' signal (green patches). g: Unemployed bees. h: Unemployed bees and
storing bees in the honey storage area. Yellow patches represent nectar/honey cells.
Modeling Population Dynamics, Division of Labor … 253
Again, this model inherited the sub model of physiological (energetic) expenditures of
individual workers from the previously described models HoFoReSim and FlowerSim: Adult
bees consume a certain amount of nectar, depending on their behavioral status and depending
on their weight, which in turn depends on their cropload. Flying bees consume more nectar
per time step than walking bees. In addition, also larvae consume nectar per time step. All
agents have implemented a 'hunger threshold'. If the crop contents of agents fall below this
threshold, they choose actions aiming to refill their nectar load: Adult bees head towards the
honey comb area for refilling themselves and larvae emit a chemical hunger stimulus that
attracts nurse bees which are located nearby who then feed the larvae. All agents, that run out
of nectar, (adults and brood) die and are removed from the model. The behavioral programs
of the agents was again modeled as finite state automatons, one for each worker group
'unemployed', 'forager', 'storer', and 'nurse'. Transitions between these jobs were modeled
again by stimulus-response functions, according to equation 1 (fixed thresholds setting) or
according to equations 2-3 (adaptable threshold setting).
The inclusion of nectar consumption and physiological costs allows us to predict the
nectar economics of mechanisms that regulate DOL by simply observing the colony's net
nectar gains or losses over time. By testing several proposed proximate mechanisms, we are
able to assess their effects on system variables that allow ultimate reasoning: Nectar gains and
brood survival rate directly affect the colony fitness, because high colony fitness is achieved
only by high colony-level survival and with a high colony-level reproduction rate. For colony
survival, a good honey/nectar accumulation is important and for successful reproduction of
colonies, a high population is required for swarming. This is supported by low brood
mortality. In TaskSelSim, adult bee agents can engage in one of the following tasks every
time step:
The stimulus to engage in foraging is the local reception of a waggle dance. The stimulus
to engage in the storing task is local reception of the tactile unloading stimulus. Recruitment
to nursing is induced by the hunger stimulus emitted by larvae. All bees quit their job with a
254 Thomas Schmickl and Karl Crailsheim
fixed probability per time step. In a basic parameter setting, we allowed the thresholds
(Θforaging, Θstoring, Θnursing) to adapt over time by using the threshold reinforcement mechanism
described by equation 3.
In Schmickl & Crailsheim (2008a), we showed that varying the length of the foraging
flight cycle has significant influence on the predicted nectar economics of the modeled
colony: The longer the foragers have to fly the lower is the net nectar gain of the colony per
time. The results showed also that with distant nectar sources, the colony needs a significant
time period to achieve any positive nectar balance. The period of 8000 time steps depicted in
figure 16a reflects a whole foraging day of a honeybee colony. Figure 16b shows that the
flight cycle duration has also a significant effect on DOL, that is how many workers are
recruited to the modeled tasks.
100 100
0
50
-100
-200 0
0 2000 4000 6000 8000 foraging bees storer bees nursing bees
Figure 16. Results of simulation runs of the model TaskSelSim with different distances from the hive to
the nectar source. A: Dynamics of nectar stores. B: Size of task-related worker groups.
The model predicted that the flight cycle duration between the nectar source and the hive
has also a significant effect on DOL, as is expressed by the number of workers are recruited
to the modeled tasks. These results demonstrate that the model TaskSelSim allows to measure
colony-level fitness variables, which result from the emergent DOL, which in turn results
from the modeled proximate mechanisms and from the simulated environmental conditions.
Thus, the model allows to link proximate mechanisms to ultimate reasoning. In addition to
the results shown here, we demonstrated in Schmickl & Crailsheim (2008a) that the modeled
colony is able to adapt to changes of the flight cycle length within one simulation run and that
the system stays adaptive even with significant changes of the values of ξand θ, what is
important as these parameters are hard to measure empirically.
To further investigate the system of honeybee colony integration, we confronted the
model colony with several specific disturbances: We suddenly removed brood, added brood
or removed adult bees at half-time of simulation runs. Thus, the colony first achieved colony
homoeostasis and equilibrium DOL and was then suddenly disturbed. After this disturbance,
the colony had to re-organize its workforce and to adapt to the new circumstances. Such
disturbances were empirically demonstrated with honeybees by Robinson et al. (1992). We
hypothesized that this adaptation is associated with significant costs (Schmickl & Crailsheim
2008b).
Modeling Population Dynamics, Division of Labor … 255
Figure 17. Resulting nectar stores predicted in simulation runs of the model TaskSelSim with sudden
disturbances of the colony's population structure at t=10000 steps. The arrow indicates the point in time
the population structure was experimentally disturbed.
Figures 17 and 18 show that each of these disturbances indeed has significant impact on
the nectar economics of the modeled colony as well as on brood survival and worker
distribution among task groups.
In addition, our results showed that the threshold reinforcement mechanism leads to a
significant degree of specialization of nurse bees, to medium specialization of storing bees but
to almost no specialization of forager bees (Schmickl & Crailsheim 2008b). This suggests
that the threshold-reinforcement works better with long-lasting, locally almost omnipresent
stimuli in the environment, as it is the case with chemical signals. These signals are used by
hungry bee larvae and also in many ant species, where such threshold-reinforcement models
were frequently applied. With short-termed and not-persistent stimuli this model predicts less
specialization, although the two-fold adaptation of Θstoring achieved at least some degree of
specialization in the modeled workers (Schmickl & Crailsheim 2008b).
These results made us confident to use our model TaskSelSim to predict nectar stores of
experimentally disturbed honeybee colonies by investigating different proximate mechanisms
(or threshold distributions). In addition to nectar stores, we investigated also the predicted
brood survival of the colonies, which is an important colony-level fitness variable (Schmickl
& Crailsheim 2011): We tested the colony in three distinct environmental conditions: no
disturbance, sudden removal of 50% of adult bees, and sudden increase of the flight cycle
length of forager bees. The following four distinct settings of proximate mechanisms
(threshold distribution regimes) were tested:
Random: All values of Θ were distributed between 0.0 and 1.0 (uniform random
distribution) and were fixed during runtime.
1Caste: All values of Θ were drawn randomly (normal distribution) around a value of
0.5. These values were fixed during runtime.
3Castes: The population was sub-divided into 3 task groups. Within each task group,
the Θ value associated with the preferred task was distributed (normal random
distribution) around 0.25, the other two values of Θ were distributed (normal random
distribution) around 0.75. All values of Θ were fixed during runtime.
Dynamic: All Θ values were initially distributed (uniform distribution) randomly
between 0.0 and 1.0. These values were adapted during runtime according to
equation 3. We used values of θand ξ which were shown in Schmickl & Crailsheim
(2008a) to be reasonable, as the model reacts insensitive in this parameter region.
256 Thomas Schmickl and Karl Crailsheim
0% 0% 0%
0% 20% 40% 60% 80% 100% 0% 20% 40% 60% 80% 100% 0% 20% 40% 60% 80% 100%
a brood removed b brood added c adults removed
0% 0% 0%
0% 20% 40% 60% 80% 100% 0% 20% 40% 60% 80% 100% 0% 20% 40% 60% 80% 100%
d brood removed e brood added f adults removed
0% 0% 0%
0% 20% 40% 60% 80% 100% 0% 20% 40% 60% 80% 100% 0% 20% 40% 60% 80% 100%
g brood removed h brood added i adults removed
Figure 18. Emerging changes in DOL and in nectar economics after sudden disturbances of colony
structure: Brood removal, brood addition and worker removal at t=10000 steps. Figures depict the final
colony status.
Plowright (1988), and by Theraulaz et al. (1991, 1998). As Figure 19 shows, the tested
proximate mechanisms showed significant differences in the colony's predicted final
performance: In all tested situations, the threshold-reinforcement mechanism led to
significant costs concerning nectar accumulation and brood survival. These results suggest
that not only environmental fluctuations cause costs for the honeybee super organism, but
also the mechanisms of decentralized task selection impose additional costs concerning
colony-level fitness. Our model suggests that in honeybees a rather rigid AP schedule offers
better economics than dynamic adaptation by threshold reinforcement.
Source: Schmickl & Crailsheim (2011). Letters indicate significant differences of medians (ANOVA;
Wilcoxon test, p<0.05, N=16 per setting) between groups.
Figure 19. Resulting nectar stores and brood survival predicted in simulation runs of the model
TaskSelSim with sudden disturbances of the colony's population structure and with different settings of
proximate mechanisms of task selection.
258 Thomas Schmickl and Karl Crailsheim
1) Top-down models that predict whole years should be enriched by selective runs of
bottom-up models. Foraging decisions can be predicted by a model like HoFoReSim,
while the resulting nutrient allocation and population dynamics are predicted by a
model like HoPoMo (see figure 20b).
2) As the specific multi-agent simulations focus on specific areas of the bees‘
environment, it is possible to fuse such models in a layered approach, allowing
information exchange between them. This way, a colony can be modeled in its full
richness of feedback loops. We call this approach 'BeeCube', see figure 20a.
260 Thomas Schmickl and Karl Crailsheim
Figure 20. Our vision of future research that builds on the experiences we made with the models
HoPoMo, HoFoReSim, FlowerSim and TaskSelSim. (a) These models are fused together in layers.
Additional layers will be needed to model self-organized comb construction, brood cannibalism and
pollen foraging. Several of these 'BeeCubes' will then be joined to a community (ecology), having their
foragers (small black circles) competing for limited resources and escaping from predators (birds,
depicted as triangles). Parameters of the sub-models will be inheritable and mutating traits, thus
initiating a process of artificial evolution. (b) Multi-timescale model that is able to join HoFoReSim and
TaskSelSim with HoPoMo. Arrows indicate how (and when) these submodels influence each other.
In conclusion, we can state that both modeling approaches of DOL in honeybees (top-
down and bottom-up) gave answers to relevant scientific questions, as we mentioned them in
the introduction. We identified potential new communication pathways and were able to
investigate the gains and costs of DOL in terms of energy, which was not done before. By
comparing colony-level fitness parameters resulting from several kinds of proximate
mechanisms within the same modeling framework, we could generate new hypotheses about
the ultimate causation for a rather strict regime of AP in honeybees, at least for the tasks of
nectar foraging, nectar handling and brood care.
Modeling Population Dynamics, Division of Labor … 261
ACKNOWLEDGEMENTS
This work was supported by: EU-IST FET project ‘I-SWARM‘, no. 507006; EU-IST-
FET project ‘SYMBRION‘, no. 216342; EU-ICT project ‖REPLICATOR‖, no. 216240; EU-
ICT project ―CoCoRo‖, no. 270382.Austrian Science Fund (FWF) research grants: P15961-
B06 and P19478-B16.
REFERENCES
Anderson, C. 1998. Simulation of the feedbacks and regulation of recruitment dancing in
honey bees. Advances in Complex Systems, 1, 267-282.
Anderson, C. & Ratnieks, F. L. W. 1999a. Task partitioning in insect societies. I. Effect of
colony size on queueing delay and colony ergonomic efficiency. The American
Naturalist, 154, 522-535.
Anderson, C. & Ratnieks, F. L. W. 1999b. Task partitioning in insect societies. II. Use of
queueing delay information in recruitment. The American Naturalist, 154, 536-548.
Anderson, C., Boomsma, J.J. & Bartholdi, J.J. III. 2002. Task partitioning in insect societies:
bucket brigades. Insectes Sociaux, 49, 171-180.
Bartholdi, J.J., T.D. Seeley, C.A. Tovey, & Vande Vate, J.H. 1993. The pattern and
effectiveness of forager allocation among flower patches in honey bee colonies. Journal
of Theoretical Biology 160:23-40.
Beshers, N.B. & Fewell, J.H. 2001. Models of division of labour in social insects. Annual
Reviews of Entomology, 46, 413-440.
Beshers, S.N., Huang, Z.Y., Oono, Y. & Robinson, G.E. 2001. Social inhibition and the
regulation of temporal polyethism in honey bees. Journal of Theoretical Biology, 213,
461-479.
Bonabeau, E., Theraulaz, G. & Deneubourg, J.-L. 1996. Quantitative study of the fixed
threshold model for the regulation of division of labour in insect societies. Proceedings of
the Royal Society of London B: Biological Sciences, 263, 1565-1569.
Bonabeau, E., Theraulaz, G. & Deneubourg, J.-L. 1998. Fixed response thresholds and the
regulation of division of labor in insect societies. Bulletin of Mathematical Biology, 60,
753-807.
Britton, N.F., Partridge, L.W., & Franks, N.R. 1996. A mathematical model for the
population dynamics of army ants. Bulletin of Mathematical Biology, 58, 471-492.
Camazine, S. & Sneyd, J. 1991. A model of collective nectar source selection by honeybees:
Self-organization through simple rules. Journal of Theoretical Biology 149, 547-571
Camazine, S., Deneubourg, J.-L., Franks, N.R., Sneyd, J., Theraulaz, G. & Bonabeau, E.
2001. Self-Organization in Biological Systems. pp. 538, Princeton University Press.
DeGrandi-Hoffman, G., Roth, S.A., Loper, G.L. & Erickson jr., E.H. 1989. BEEPOP: A
honeybee population dynamics simulation model. Ecological Modelling, 45, 133-150.
DeGrandi-Hoffman, G. & Hagler, J. 2000. The flow of incoming nectar through a honey bee
(Apis mellifera L) colony as revealed by a protein marker. Insectes Sociaux, 47, 302-306.
Deneubourg, J.-L., Goss, S., Pasteels, J.M., Fresneau, D. & Lachaud, J.-P. 1987. Self-
orgaization mechanisms in ant societies (II): learning in foraging and division of labor.
262 Thomas Schmickl and Karl Crailsheim
In: Pasteels, J. & Deneubourg, J.-L. (eds). From Individual to Collective Behavior in
Social Insects, Experientia Supplementum, 54, 177-196.
De Vries, H., Biesmeijer, J.C. 1998. Modelling collective foraging by means of individual
behaviour rules in honey-bees. Behavioral Ecology and Sociobiology, 44, 109–124.
De Vries, H., Biesmeijer, J.C. 2002. Self-organization in collective honeybee foraging:
emergence of symmetry breaking, cross inhibition and equal harvest-rate distribution.
Behavioral Ecology and Sociobiology, 51, 557–569.
Dornhaus A., Collins, E. J., Dechaume-Moncharmont, F.-X., Houston, A. I., Franks, N. R. &
McNamara, J. M. 2006. Paying for information: partial loads in central place foragers.
Behavioral Ecology and Sociobiology 61:151–161
Dornhaus, A., Klügl, F., Oechslein, C., Puppe, F. & Chittka, L. 2006. Benefits of recruitment
in honey bees: effects of ecology and colony size in an individual-based simulation
model. Behavioral Ecology 17(3): 336-344
Franks, N.R. & Tofts, C. (1994). Foraging for work: How tasks allocate workers. Animal
Behaviour, 48, 470-472.
von Frisch , K. 1967. The dance language and orientation of bees. Harvard University Press,
Cambridge, MS.
Gautrais, J., Théraulaz, G., Deneubourg, J.-L., & Anderson, C. 2002. Emergent polyethism as
a consequence of increased colony size in insect societies. Journal of Theoretical
Biology, 215, 363-373.
Gregson, A.M., Hart, A.G., Holcombe, M. & Ratnieks, F.L.W. 2003. Partial nectar loads as a
cause of multiple nectar transfer in the honey bee (Apis mellifera): a simulation model.
Journal of Theoretical Biology, 222, 1-8.
Hart, A.G., Anderson, C. & Ratnieks, F.L.W. 2002. Task partitioning in leaf-cutting ants.
Acta Ethologica, 5, 1–11.
Hart, A.G. & Ratnieks, F. 2001. Why do honey-bee (Apis mellifera) foragers transfer nectar
to several receivers? Information improvement through multiple sampling in a biological
system. Behavioral Ecology and Sociobiology, 49, 244–250.
Hölldobler, B. & Wilson E.O. 2008. The Superorganism: The Beauty, Elegance, and
Strangeness of Insect Societies. W. W. Norton & Company, pp 544.
Huang, Z. & Robinson, G.E. 1992. Honeybee colony integration: worker-worker interactions
mediate hormonally regulated plasticity in division of labor. PNAS, 89, 11726-11729.
Huang, M., & Seeley, T.A. 2003. Multiple Unloadings by Nectar Foragers in Honey Bees: a
Matter of Information Improvement or Crop Fullness? Insectes Sociaux, 50, 1–10.
Jeanne, R.L. 1986a. The evolution of the organization of work in social insects. Monitore
Zoologico Italiano, 20, 119-133.
Jeanne, R. L. 1986b. The organization of work in Polybia occidentalis: costs and benefits of
specialization in a social wasp. Behavioral Ecology Sociobiology, 19, 333-341.
Jeanson, R., Fewell, J.H., Gorelick, R. & Bertram, S.M. 2007. Emergence of increased
division of labor as a function of group size. Behavioral Ecology and Sociobiology, 62,
289-298.
Karsai, I. & Wenzel, J. W. 1998. Productivity, individual-level and colony-level flexibility,
and organization of work as consequences of colony size. PNAS, 95, 8665-8669.
Karsai, I. & Wenzel, J. W. 2000. Organization and regulation of nest construction behavior in
Metapolybia wasps. Journal of Insect Behavior, 13, 111-140.
Modeling Population Dynamics, Division of Labor … 263
Karsai, I. & Balazsi, G. 2002. Organization of work via a natural substance: regulation of nest
construction in social wasps. Journal of Theoretical Biology, 218, 549-565.
Karsai, I. & Runciman, A. 2008. The effectiveness of the 'common stomach' in the regulation
of behavior of the swarm. In: Troch, I. & Breitenecker, F. (eds). Proceedings
MATHMOD 2009, Vienna.
Lindauer M. 1952. Ein Beitrag zur Frage der Arbeitsteilung im Bienenstaat. Zeitschrift für
vergleichende Physiologie, 34, 299-345.
Lloyd J.E. 1983. Bioluminiscence and communication in insects. Annual Reviews of
Entomology, 28, 131-160.
Lorenz, K. 1978. Vergleichende Verhaltensforschung. Springer-Verlag, Wien, NewYork.
Makela, M.E., Rowell, G.A., Sames, W.J. & Wilson, L.T. 1993. An object-oriented
intracolonial and population level model of honey bees based on behaviors of European
and africanized subspecies. Ecological Modelling, 67, 259-284.
Mayr, E. 1961. Cause and effect in biology. Science, 134, 1501-1506.
Merkle, D. & Middendorf, M. 2004. Dynamic Polyethism and Competition for Tasks in
Threshold Reinforcement Models of Social Insects. Adaptive Behavior, 12, 251-262.
Moritz, R.F.A. & Fuchs, S. 1998. Organization of honeybee colonies: characteristics and
consequences of a superorganism concept. Apidologie, 29, 7 – 21.
Naug, D. & Gadagkar, R. 1999. Flexible division of labor mediated by social interactions in
an insect colony – a simulation model. Journal of Theoretical Biology, 197, 123-133.
Naug, D. & Camazine, S. 2002. The role of colony organization on pathogen transmission in
social insects. Journal of Theoretical Biology, 215, 427-439.
Omholt, S.W. 1986. A model for intracolonial population dynamics of the honeybee in
temperate zones. Journal of Apicultural Research, 25, 9-21.
Omholt, S.W. 1988. Relationship between worker longevity and the intracolonial population
dynamics of the honeybee. Journal of Theoretical Biology, 130, 275-284.
Omholt, S.W. 1992. The heuristic value of mathematical modelling for elucidation of the
honey production dynamics of Apis melliferacolonies. Norwegian Journal of Agricultural
Sciences, 6, 99-110.
Plowright, R.C. & Plowright. C.M.S. 1988. Elitism in social insects: a positive feedback
model. In: Jeanne, R.L. (eds). Individual behavioral variability in social insects. pp. 419-
431. Westview Press, Boulder, CO.
Ratnieks, F.L.W. & Anderson, C. 1999. Task partitioning in insect societies. Insectes sociaux,
46, 95-108.
Reyes, J.L. & Fernández-Haegar, F. 1999. Sequential co-operative load transport in the seed-
harvesting ant Messor barbarus. Insectes Sociaux, 46, 199-125.
Robinson, G.E., Page, R.E., Strambi, C. & Strambi, A. 1992. Colony integration in
honeybees: Mechanisms of behavioral reversion. Ethology 90, 336-348.
Robinson, G.E. 1992. Regulation of division of labor in insect societies. Annual Reviews of
Entomology, 37, 637-665.
Schmickl, T. & Crailsheim, K. 2001. Schmickl, T., Crailsheim, K.: Cannibalism and early
capping: strategy of honeybee colonies in times of experimental pollen shortages. Journal
of Comparative Physiology A, 187(7) (2001) 541–547
Schmickl, T. & Crailsheim, K. 2004. Costs of environmental fluctuations and benefits of
dynamic decentralized foraging decisions in honey bees. Adaptive Behavior, 12, 263-277.
264 Thomas Schmickl and Karl Crailsheim
Chapter 11
Manuel Gámez
Department of Statistics and Applied Mathematics
Almería University, Spain.
ABSTRACT
The paper is a review of a research line initiated two decades ago. At the beginning
the research was concentrated on basic qualitative properties of ecological and
population-genetic models, such as observability and controllability. For population
system, observability means that, e.g. from partial observation of the system (observing
only certain indicator species), in principle the whole state process can be recovered.
Recently, for different ecosystems, the so-called observer system (or state estimators)
have been constructed that enables us to effectively estimate the whole state process from
the observation. The methodology of observer design can be also applied to estimate
unknown changes in ecological parameters of the system. Clearly, both observation (i.e.
monitoring) and control are important issues in conservation ecology. For an ecological
system, in an appropriate setting, controllability implies that a disturbed ecosystem can be
steered beck to an equilibrium state by an abiotic human intervention. Recent research
concern the effective calculation of such control functions. While the considered
ecological models are density-dependent, observability and controllability problems also
naturally arise in frequency-dependent models of population genetics. As for the
frequency-dependent case, observation systems typically occur in case of phenotypic
observation of genetic processes; control systems can be used to model e.g. artificial
selection. In this survey, in addition to the basic methodology and its applications, the
recent developments of the field are also reported.
1. INTRODUCTION
The mathematical analysis of control and observation systems has been developed from
engineering practice. In fact, for linear models, the theory of controllability and observability
has been built up in the framework of Mathematical Systems Theory by Kalman et al. (1969).
Control (or input) usually means a human intervention on a given object, observation (or
output) is a transform of the state process available to us instead of the state itself. The basic
theory of nonlinear systems has been laid down in Lee and Markus (1971). A recent reference
on Mathematical Systems Theory is Zak (2003). Systems-theoretical approach in biology is
applicable both at infra-individual and supra-individual levels. The present survey deals with
control and observation problems in supra-individual systems, in particular with population
ecology and population genetics. The considered population systems are nonlinear, where a
basic approach of nonlinear systems theory, namely the linearization of the system around an
equilibrium is an efficient approach, see Varga (2008). The case of frequency-dependent
models required the development of the theory of controllability and observability of
nonlinear systems with invariant manifold published in Varga (1989) and Varga (1992).
In Section 2, first a brief summary of necessary concepts and theorems concerning
nonlinear observation systems is given, formulating a sufficient condition for observability
and recalling the basics of observer design, and then the application of this methodology is
illustrated on a Lotka-Volterra type model. It is also presented how the methodology of
observer design can be also applied to estimate unknown changes in ecological parameters of
the system. Section 3 is devoted to control problems in density-dependent populations
models. Both finite-time open-loop equilibrium control and asymptotic closed-loop
equilibrium control of population systems are presented and illustrated with applications. In
Section 4, based on the concept of observability, the state monitoring in frequency-dependent
population systems is considered, namely, a sufficient condition for phenotypic observability
of genetic processes is presented. Observer system for an evolutionary game model is also
constructed.
Furthermore, in Section 5, control in a frequency-dependent populations system as a
model of articicial selection is shortly recalled. Finally, in section 6, we also give an outlook
to further applications of the surveyed methodology.
Let m, n N , f C1 (R n , R n ) , h C1 (R n , R m ) , x R n , f ( x ) 0 ,
h( x ) 0 ; and consider observation system
Observation and Control in Density … 269
x f (x) (1)
y h(x) . (2)
where (2) defines the transform, observed instead of the state itself.
Definition 1. System (1)-(2) is called locally observable at x on [0, T ] , if for any
solutions x1 , x 2 of (1) defined on [0, T ] , initially close enough to x ,
h x1 h x 2 x1 x 2 .
Linearizing (1)-(2) around x , we get
A : f ( x ) , C := h ( x * ) .
Example 1. We consider now a Lotka-Volterra type model with two preys and one
predator, determined by the following differential system (Gámez et al. 2008):
with ai , bij 0 for all i, j 1, 2, 3. We suppose that we observe the densities of both prey
populations, i.e., the observation equation is
* *
y h( x) : ( x1 x1 , x3 x3 ).
2 T
It is easy to check that rank[C | CA | CA ] 3. Thus, by Theorem 1 the system is
locally observable near the equilibrium, and the whole system state, in principle, can be
recovered observing only the prey populations.
270 Manuel Gámez
Observer System
Recently, for different ecosystems, the so-called observer system (or state estimators)
have been constructed that enables us to effectively estimate the whole state process from the
observation. Below it will be shown that the methodology of observer design can be also
applied to estimate unknown changes in ecological parameters of the system. Clearly, both
observation (or monitoring) and control are important issues in conservation ecology.
Consider again observation system (1)-(2)
x f (x)
y h(x) .
z G( x, y) (4)
is called a local (exponential) observer for system (1)-(2) at x , if for the composite
system (1)-(2),(4) we have
Theorem 2. (Sundarapandian, 2002). Suppose equilibrium x of system (3) is Lyapunov
stable, and there exists nxn matrix K such that A KC is stable. Then system
z f ( z) K[ y h( z)] (5)
x F ( x, u ( w))
w s( w) (6)
y h( x),
Observation and Control in Density … 271
Example 2. To apply this construction to our population system (3), we remind that in
López et al. (2007b) it was proved that system (3) is asymptotically stable for an equilibrium
* * *
x * ( x1 , x 2 , x3 ) . we consider the following two-prey one-predator model with the
presence of an unknown environmental disturbance w :
*
System (8) has an equilibrium: x (1.4608, 3.9307, 0.6506, 0) , and with
0 0 0 1
T
K :
0 0 0 0.1
x f ( x, u u) , (10)
corresponding to a control u U [0, T ] , starting from any point close enough to x , is
defined on the whole interval [0, T].
Definition 3. System (10) is said to be locally controllable to x on [0, T ] , if R
a G ( x ) , the -neighbourhood of x , u U [0, T ] such that a
u
x .
rang B AB ... An1B n system (10) is locally controllable to x on [0, T ] .
If a nearly natural ecological system is deviated from its equilibrium, an important task of
conservation ecology may be to control it back into equilibrium. In this subsection, based on
Gámez et al. 2010a, we illustrate the application of the above Theorem 4, combined with a
software solving optimal control problems.
To this end we consider a relatively simple food web: A typical terrestrial trophic chain
consists of the following components: resource, the 0th trophic level (solar energy or
inorganic nutrient), which is incorporated by a plant population, the 1st trophic level
(producer), which transfers it to a herbivorous animal population, the 2nd trophic level
(primary consumer). Let us denote by x 0 the time-varying quantity of ―f ree‖ resource
present in the system, x1 and x 2 , in function of time, the biomass (or density) of the
producer (species 1) and the primary consumer (species 2), respectively. Let Q be the
274 Manuel Gámez
resource supply considered constant in the model. Let 0 x 0 be the velocity at which a unit
biomass of species 1 consumes the resource, and it is assumed that this consumption increases
the biomass of this species at rate k1 . A unit biomass of species 2 consumes the biomass of
species 1 at velocity 1 x1 , converting it into its own biomass at rate k 2 . Both the plant and
the animal populations are supposed to decrease exponentially in the absence of the resource
and the other species, with respective rates of decrease (Malthus parameters) m1 and m2 .
Then with model parameters
the dynamic model for the trophic chain can be set up as follows:
x 0 Q 0 x0 x1 1m1 x1 2 m2 x 2 (12)
In Shamandy (2005), a necessary and sufficient condition has been obtained for the
coexistence of the population system. The latter means that there exists a non-trivial
ecological equilibrium x of dynamic system (12)-(14), where all components are present:
system (12)-(14) has a unique equilibrium x ( x0* , x1* , x2* ) 0 if and only if the resource
supply is high enough, i.e.
m1m2 1m1m2
Q Q2 : .
1k1k 2 1k 2
In the following, we will illustrate how an optimal control software can be applied to
effectively control the trophic chain into equilibrium, intervening at a given trophic level.
Let us suppose first that the resource supply is controlled in function of time in the form
Q u(t ) , considering control functions u defined on a fixed interval [0,T]. Then our model
(12)-(14) takes the form
x 0 Q u (t ) 0 x 0 x1 1m1 x1 2 m2 x 2 (15)
Observation and Control in Density … 275
x f ( x, u u (t )) (18)
Since
2 2 2
det[ B | AB | A2 B] 0 1k1 k 2 x1 x2 0 ,
2
we get rank[ B | AB | A B ] 3 , and applying Theorem 4 we obtain the local
controllability of system (15)-(17) into x on interval [0, T ] .
The obtained local controllability means that from nearby states, the system can be
steered into the equilibrium applying an appropriate small control u U 0 [0, T ] . Now we
proceed to the determination of such a control.
Fix an initial state x 0 from a neighbourhood of local controllability of system (18), and
for each control function u small enough (i.e. u U 0 [0, T ] ), let x be the solution of (18),
defined on [0, T ] and corresponding to the initial value x 0 . Then a control u u will steer
initial state x 0 into equilibrium x , if and only if it minimizes the functional
276 Manuel Gámez
2
(u ) : x (T ) x .
Theorem 5. For any parameter values (11), system (15)-(17) is locally controllable into
equilibrium x on interval [0, T ] , and an initial state x 0 will be steered into x by a control
u U 0 [0, T ] if and only if the latter is a solution of the following optimal control problem:
2
(u ) : x (T ) x min , (19)
x F ( x, u u (t )) . (21)
Example 3. Let us consider system (15)-(17) with parameters (see Gámez et. a. 2010a):
Q : 10; 0 : 0.3; 1 : 0.1; 1 : 0.2; 2 : 0.3; m1 : 0.1; m2 : 0.4; k1 : 0.5; k 2 : 0.5.
0
Taking as initial condition x : ( 4 , 7 , 5) and time interval T:=5, we apply the
MatLab toolbox mentioned above. Figure 2(A) shows the solution u of the optimal control
problem, the corresponding solution x , ending at equilibrium x ( 4.52 , 8 , 5.78) can be
seen in Figure 2(B). On the other hand, we note that, for the uncontrolled system, x is
asymptotically stable, and arrives at the equilibrium only in ―i
nfinite time‖, see Figure 2(C).
(A) (B)
Observation and Control in Density … 277
(C)
Figure 2. (A) Control function of system (15)-(17) for T=5, with initial value x(0)=(4,7,5), (B)
Solutions of system (15)-(17) for T=5, with initial value x(0)=(4,7,5) and Solution of system (12)-(14)
for T=5, with initial value x(0)=(4,7,5).
If we want to control the system into a required new equilibrium state, and maintain this
new equilibrium by a constant control, a technique of theory of optimal control can be applied
to construct a feedback control system.
n n n r
For n, r N , L R , BR , and continuously differentiable function
g : R R , consider the control system
n n
x Lx g ( x) BU (22)
Lx * g ( x* ) Bu * 0 . (23)
* *
Then, from (22) and (23), for the new variables y x x ; u U u we have
y Ly q( y) Bu with q( y ) : g ( y x* ) g ( x* ). (24)
A feedback control will be given below which asymptotically steers system (24) into the
zero equilibrium. In Rafikov et. al. 2008 we have,
278 Manuel Gámez
n n
Theorem 6. If there exist matrices P, Q, R R ; P positive definite and Q symmetric,
such that the function l ( y ) : y Qy q ( y ) Py y Pq ( y ) ( y R n ) is positive definite,
T T T
u ( y ) : R 1BT Py ( y R n ) (26)
Corollary 1. Using the notation of the previous theorem, let us suppose that function l is
locally positive definite. Then there exists a neighbourhood V of zero in Rn such that for all
x(0) V , for the solution x of system (22) we have lim x x* .
Now we illustrate the application of above Corollary to the following population system.
Example 4. We consider a dynamic model (see Gámez et. al. 2009), describing the
interaction between the populations of healthy and irradiated cells of an organ under the effect
of a constant radiation. Let x1(t) be the number of healthy cells, and x2(t) the number of
irradiated cells at time t. For this situation, from Freedman and Pinho (2008) we recall the
following model:
x
x1 r x1 1 1 x1 px2
(27)
K0
x2 x1 px2 x2
The mentioned authors proved that the system leaves the nonnegative orthant invariant.
We apply Corollary 1 to system (27) utilizing the following control system, corresponding to
system (23),
x
x1 r x1 1 1 x1 px2
K0 (28)
x2 x1 px2 x2 U
Now we want to obtain a feedback control that steers the population of healthy cells to a
desired level x1* x1d . The corresponding value x2* x2 d and u* can be calculated solving
the following system of linear equations (23).
If we choose for the (25) and (26),
Observation and Control in Density … 279
r p r 2
( y1 2 y1 x1* )
L : and h ( y ) : K ( y R2 )
( p ) 0
It is easy to prove that the condition of Theorem 6 are satisfied and the feedback control
can be obtained by (26).
For the concrete parameters: r := 1; Δ :=5; K := 100; p := 1, and δ := 0:1, it is easy to
prove that system (22) has an asymptotically stable equilibrium, where x1 = 54. Now, we
suppose that the objective is to increase the number of healthy cells, for example to a level x1d
= 80. To this end, we calculate x2d = 344 and u* = 21.6.
For matrices L and B we have
4 1 0 1 0
L : ; B : and coosing Q : ; R : 1.
5 1.1 1 0 1
we calculate the matrix Riccati equation (21), and using the function LQR of
MATLABTM v 7.0, we obtain a solution of P. Therefore, applying (26), we obtain the
required feedback control, and from inequalities x = x* + y and U = u* + u, we can calculate
the closed loop control system. In Figure 3 shows how the first coordinate of the solution of
the controlled system asymptotically reaches the desired value x1d = 80.
Figure 3. Solutions of (27) and (28) with the same initial value x(0) = (100; 0).
Following Cressman et al. (1996), we shall consider a large panmictic diploid Mendelian
population with alleles A1,…,An at a single autosomal locus. By assumption, zygotes are
produced according to the Hardy-Weinberg proportions. We consider N possible phenotypes
or behavior strategies identified with the vertices of the standard simplex of the phenotype
vectors,
N
N : {( s1 ,..., sN ) R N | sk 0 (k 1, N ), sk 1}.
k 1
We also suppose that the phenotype of a zygote is uniquely determined by its genotype:
for each genotype AiAj , let be Sij the phenotype of an AiAj zygote. For each allelic state
N
N
S ij AS S
k ,l 1
ij , k a kl S l ( p ) (29)
where stands for the scalar product of vectors Sij;k and Sl(p) are the k-th and the l-th
components of the vectors, Sij and S(p); respectively. Let now p(t) be the allelic frequency
vector in the adult population at time t. Then the standard continuous-time dynamics reads as
follows:
n n
p i pi pi Sij pk pl Skl A pk pl Skl (30)
j l k ,l l k ,l
n
, S ( p ) : pi Sij for the above dynamics we obtain the following compact form
i
j l
p i pi S i ( p) S ( p) AS ( p) (31)
For the frequency-dependent systems (31), the general sufficient conditions for local
observability of nonlinear systems with invariant manifold, is proved in Varga (1992).
y h(x) , (33)
Observation and Control in Density … 281
Tx ( M ) KerQ {0} (32)-(33) is locally observable in M near x on
t [0, T .
Following Gámez et al. (2003), we obtain a sufficient condition for de local observability
of model (31) for a phenotypic observation:
o
Example 5. We suppose that S* is a polymorphic ESS ( S
*
N ), and p* is a
o
polymorphic allelic state ( S * n ) realizing this ESS: S(p*) = S*: It is easy to prove that p*
is an equilibrium of dynamics (31), see Cressman et al. (1996).
For the effective phenotypes at the dynamic equilibrium, we introduce the notation S*i:=
Si(p*) (i 1, n). and we consider de observation defined by,
h : Rn R N , h( p ) S ( p ) S * (34)
h( p) : S ( p) S ( p* ) . (35)
282 Manuel Gámez
z f ( z) K[ y S ( z)
we shall construct will not only approximate the solution of the original system, but the
solution of the observer is also a substitute of the latter in the sense that the interior of the
simplex is locally long-term invariant for the observer system, that is, for any solution p of
system (31), that is initially close enough to p , p(t) remains in int n for t large enough, see
López et al. 2008.
Consider now dynamics (31) with the following payoff matrix: fix an R and define
1 1
A( ) : 1 1 . (37)
1 1
Let us fix : 0.5 . We know that this system has a positive equilibrium
p : (1 / 3, 1 / 3, 1 / 3) see López et. al. (2008). It is easy to check that for this game, the
phenotypic image of equilibrium p , i.e., the mixed phenotype
Observation and Control in Density … 283
h( p) : S ( p) S ( p* ) . (39)
S *1 S 1 ( p * ) (2 / 3,0,1 / 3),
S *2 S 2 ( p * ) (1 / 3,2 / 3,0),
S *3 S 3 ( p * ) (0,1 / 3,2 / 3),
4 / 3 2 / 3 0
C 0 4 / 3 2 / 3.
2 / 3 0 4 / 3
For an effective (although approximate) calculation of this state process we apply the
observer design method provided by Theorem 2.
1 0 0
For K 0 1 0 we obtain that matrix L KC has eigenvalues with negative real
0 0 1
parts and by Theorem 2, we can construct a local exponential observer for observation system
(38)-(39):
zi zi S i ( z) S ( z) AS ( z) K[ S ( p) S ( z)] (i 1,3) (40)
By the particular choice of the gain matrix K , we have that int n is locally long-term
invariant for the observer system (40). .
284 Manuel Gámez
(A) (B)
(C)
Figure 4. (A) 1st coordinates of p, z and w (B) 2nd coordinates of p, z and w and (C) 3rd coordinates of
p, z and w.
Now we illustrate how the observer approximately provides the required solution from
the observed function. Suppose that the initial condition is p0 (0,3, 0,4, 0,3) for the
original system (38) and z0 (0,35, 0,35, 0,3) for the observer system (40). In the next
three figures the coordinates zi (i 1,3) of the solution z of the observer system practically
end up in the corresponding coordinates pi (i 1,3) of the solution p of the original system
(38). Since p is asymptotically stable, if a solution p is initially close enough to p , the
‖unknown‖ solution p could also be asymptotically approximated by another solution w of
system (38), with the same initial value as z . The coordinates wi (i 1,3) of this solution
are also represented in the Figures 4(A)-(C), showing that the observer performs much better.
5. CONTROL IN FREQUENCY-DEPENDENT
POPULATIONS SYSTEMS
For the frequency-dependent systems (31), the general sufficient conditions for local
controllability of nonlinear systems with invariant manifold, is proved in Varga (1987).
Observation and Control in Density … 285
Using notation and conditions of Section 1, fix k 1, n 1 , and suppose that a k-
1
dimensional regular C -submanifold M R n is locally positively invariant for the control
system
x f ( x, u u) (41)
Definition 5 System (14) is said to be locally controllable to x on [0, T ] in M , if
]0, 0 [ a G ( x ) M u U [0, T ] such that a
u
x .
rang B AB ... An1B k system (41) is locally controllable to x on [0, T ] in
M
The observation of food chains at different trophic levels, including observer design was
studied in López et al. (2007a). In Varga et al. (2003) and López et al. (2007b) monitoring of
Lotka-Volterra population systems, under different environmental conditions was
investigated.
286 Manuel Gámez
REFERENCES
Banga, J.R., Balsa-Canto, E., Moles, C.G., Alonso, A.A., 2005. Dynamic optimization of
bioprocesses: efficient and robust numerical strategies. Journal of Biotechnology 117,
407–419.
Cressman, R., Hofbauer, J., Hines, W.G.S., 1996. Evolutionary stability in strategic models of
single-locus frequency-dependent viability selection. J. Math. Biol. 34, 707–733.
Farkas, Gy., 1998a. Local controllability of reactions. J. Math. Chem. 24, 1-14.
Farkas, Gy., 1998b. On local observability of chemical systems. J. Math. Chem. 24, 15-22.
Freedman, H.I., Pinho, S.T.R., 2008. Persistence and extinction in a mathematical model of
cell populations affected by radiation. Period. Math. Hungarica 56 (1), 25–35.
Gámez, M., Carreño, R., Kósa, A., Varga, Z., 2003. Observability in strategic models of
selection. BioSystems 71, No 3, 249-255.
Gámez, M., López, I. and Molnár, S., 2008. Monitoring environmental change in an
ecosystem. BioSystems 93, 211-217.
Gámez, M., López, I., Garay, J. and Varga, Z., 2009. Observation and control in a model of a
cell population affected by radiation. BioSystems 96, 172-177.
Gámez, M., López, I. and Shamandy, S., 2010a. Open-and closed-loop equilibrium control of
trophic chains. Ecological Modelling, 221 , 1839-1846.
Gámez, M., López, I., Szabó, I. and Varga, Z., 2010b, Verticum-type systems applied to
ecological monitoring. Applied Mathematics and Computation, 215 3230-3238.
Hirmajer, T., Balsa-Canto, E., Banga, J.R., 2009. DOTcvpSB, a software toolbox for dynamic
optimization in systems biology. BMC Bioinformatics 10, 199.
Hofbauer, J. and K. Sigmund, (1988). The Theory of Evolution and Dynamical Systems.
Cambridge University Press, Cambridge
Kalman, R. E., Falb, P. L., Arbib, M. A., 1969. Topics in Mathematical System Theory.
McGraw-Hill, New York.
Lee, E. B., Markus, L., 1971. Foundations of Optial Control Theory. Wiley, New York -
London - Sydney.
Observation and Control in Density … 287
Chapter 12
ABSTRACT
We analyse the effects of environmental noise in three different biological systems:
(i) mating behavior of individuals of Nezara viridula (L.) (Heteroptera Pentatomidae); (ii)
polymer translocation in crowded solution; (iii) an ecosystem described by a Verhulst
model with a multiplicative Lévy noise. Specifically, we report on experiments on the
behavioral response of N. viridula individuals to sub-threshold deterministic signals in
the presence of noise. We analyze the insect response by directionality tests performed on
a group of male individuals at different noise intensities. The percentage of insects which
*
E-mal address: bernardo.spagnolo@unipa.it
°
E-mal address: davide.valenti@unipa.it
290 B. Spagnolo, D. Valenti, S. Spezia et al.
1. INTRODUCTION
During last decades noise-induced effects have been experimentally observed and
theoretically studied in different physical and biological contexts (Agudov and Spagnolo,
2001; Fiasconaro and Spagnolo, 2009; Spagnolo et al., 2003; Valenti et al., 2004; Spagnolo et
al., 2002; Zimmer, 1999; Bjørnstad and Grenfell, 2001; Grenfell et al., 1998; Chichigina,
2008; Giuffrida et al., 2009; Korobkova et al., 2004; Pizzolato et al., 2009), such as neuronal
cells, excitable systems and threshold physical systems (Braun, 1994; Moss et al., 1994;
Gingl et al., 1995; Pei et al., 1995; Pikovsky, and Kurths, 1997; Nozaki and Yamamoto,
1998; Longtin and Chialvo, 1998; Stocks, 2001; Wiesenfeld et al., 1994; Gammaitoni, 1995;
Wannamaker et al., 2000; Lindner et al., 2004).
In particular, stochastic resonance, resonant activation and noise enhanced stability
phenomena in neuronal activation have been recently discussed (Lindner et al., 2004; Duarte
et al., 2008; Prankratova et al., 2005).
Nature consists of open systems characterized by intrinsically non-linear interactions and
subject to environmental noise (Spagnolo et al., 2004). The presence of random fluctuations,
that are an uneliminable component of natural ecosystems, makes difficult detection and
transmission of signals and can modify the information transported.
Environmental Noise and Nonlinear Relaxation in Biological Systems 291
However, in the presence of some specific non-linearity of the system and for suitable
intensity of noise, counterintuitive phenomena, such as stochastic resonance (SR), can be
observed. This indicates that noise can play a constructive role, improving the conditions for
signal detection.
SR phenomenon, initially observed in the temperature cycles of the Earth (Benzi et al.,
1981), can be found in many physical and biological non-linear systems (Gammaitoni et al.,
1998; Mantegna and Spagnolo, 1994; Mantegna et al., 2001; Agudov et al., 2010). SR can be
modeled by a bistable potential subject to periodical driving force in the presence of external
additive noise. The signature of SR is a nonmonotonic behavior, characterized by a
maximum, of the signal-to-noise (SNR) ratio as a function of the noise intensity. This
indicates that the noise can enhance the amplitude of deterministic signals, improving the
response of the system through a resonance-like phenomenon (Moss et al., 1994; Gingl et al.,
1995; Pei et al., 1995; Nozaki and Yamamoto, 1998; Longtin and Chialvo, 1998; Stocks,
2001; Wiesenfeld et al., 1994; Gammaitoni, 1995; Wannamaker et al., 2000; Lindner et al.,
2004; Gammaitoni et al., 1998; Mantegna and Spagnolo, 1994; Mantegna et al., 2001; Vilar
et al., 1998; Longtin et al., 1991; Bulsara et al., 1991; Chialvo and Apkarian, 1993; Neiman
and Russell, 2002; Bahar et al., 2002; Douglass et al., 1993; Russell et al., 1999; Freund et al.,
2002; Greenwood et al., 2000; Gailey et al., 1997). However, SR does not occur only in
bistable systems, but also in monostable, excitable, and non-dynamical systems. In these
situations we name this effect non-dynamical (or threshold) stochastic resonance, because the
phenomenon is connected with the crossing of a threshold and can occur also in the absence
of an external potential (Moss et al., 1994; Gingl et al., 1995; Vilar et al., 1998). Sensory
neurons, that are threshold systems characterized by intrinsic noise, are an ideal workbench to
observe non-dynamical SR (Longtin et al., 1991; Bulsara et al., 1991; Chialvo and Apkarian,
1993; Neiman and Russell, 2002; Bahar et al., 2002). Historical experiments revealed the
presence of non-dynamical SR in the neural response of mechanoreceptor cells of crayfish
(Douglass et al., 1993), and the improvement of sensorial activity of paddlefish in the
detection of electric signals produced by preys (Russell et al., 1999; Freund et al., 2002;
Greenwood et al., 2000). Such sensory neurons are ideally suited to exhibit SR as they are
intrinsically noisy and operate as threshold systems (Longtin et al., 1991; Bulsara et al., 1991;
Chialvo and Apkarian, 1993; Neiman and Russell, 2002; Bahar et al., 2002).
In this contribution, we study the effects of external noise in three different biological
systems. We start analyzing the mating behavior of individuals of N. viridula (L.)
(Heteroptera Pentatomidae). In particular, we investigate the role of noise in the response of
male insects to mechanical vibrations emitted by female individuals and transmitted in the
substrate (Čokl et al., 1999; Čokl et al., 2003; Čokl et al., 2007). N. viridula, the southern
green stink bug, is a pentatomid insect highly polyphagous and quite harmful for agriculture
(Todd, 1989; Pannizzi, 2000). N. viridula has up to five generations per year (Borges et al.,
1987; Kiritani, 1964; Tremblay, 1981; Fucarino, 2003).
The mating behavior of N. viridula can be divided into long-range location and short-
range courtship. The first one includes those components of the behavior that lead to the
arrival of females in the vicinity of males. The long range attraction mediated by male
attractant pheromone enables both sexes to reach the same plant.
Here, we analyze the mating behavior of insects during the short-range courtship, when
bugs of both sexes are very close and the acoustic stimuli (improperly called songs) can be an
important element in the sexual communication (Čokl et al., 1999).
292 B. Spagnolo, D. Valenti, S. Spezia et al.
The sound is produced by the tymbal, an organ sited in the back and present in adult
individuals (Čokl et al., 2003). The vibrations, produced by a bug at the frequency of about
100 Hz, propagate through the legs into the plant stem and can be detected by the vibro-
receptors placed in the legs of another insect (Tremblay, 1981; Bagwell et al., 2008). Many
experimental studies have been performed on this acoustic communication, analyzing the
different signals characteristic of populations of N. viridula from Slovenia, Florida, Japan and
Australia (Čokl et al., 2000).
The fundamental role of the vibratory signals suggests that a better knowledge of the
mechanism of acoustic communication during the short-range courtship can help to point out
more efficient strategy to control N. viridula populations, devising "biologic" traps whose
working principle is the emission of acoustic signals. In natural conditions, N. viridula
populations interact strongly with environment, and therefore the presence of surrounding
noise becomes an essential component of the acoustic communication.
In the second part of this contribution, we consider transport phenomena of polymers in
crowded solutions. In fact, the translocation of DNA and RNA across nuclear membranes as
well as the crossing of potential barriers by many proteins represents a fundamental process in
cellular biology. The study of the transport of macromolecules across nanometer size
channels is important for both medical research in anticancer targeted therapy (Higgins, 2007;
Halwachs et al., 2009) and technological applications (Mannion et al., 2006; Sundaresan et
al., 2008).
First experiments on the passage of DNA molecules across an -hemolysin (-HL)
protein channel revealed a linear relationship of the most probable crossing time p with the
molecule length (Kasianowicz et al., 1996). Moreover, p scales as the inverse square of the
temperature and the dynamics of biopolymer translocation across an -HL channel is found
to be governed by pore-molecule interactions (Akeson et al., 1999; Meller et al., 2000; Deng
et al., 2003). More recent experimental studies have shown that the application of an AC
voltage to drive the translocation process of DNA molecules through a nanopore plays a
significant role in the DNA-nanopore interaction, and provides new insights into the DNA
conformations (Deng et al., 2003; Vernier et al., 2004; Sigalov et al., 2008; Lathrop et al.,
2009; Nikolaev and Gracheva, 2009).
The complex scenario of the translocation dynamics coming from experiments has been
enriched by several theoretical and simulative studies (Lubensky and Nelson, 1999; Storm et
al., 2005; Forrey and M. Muthukumar, 2007; Luo et al., 2008; Gracheva, and J. P. Leburton,
2008; Pizzolato et al., 2008; Panja, and G. T. Barkema, 2008; Pizzolato et al., 2009; Pizzolato
et al., 2010). The mean first passage time of a Brownian particle to cross a potential barrier in
the presence of thermal fluctuations and a periodic forcing field has been theoretically and
experimentally investigated as a function of the driving frequency in Refs. (Doering and
Gadoua, 1992; Bier and Astumian, 1993; Boguna et al., 1998; Mantegna and Spagnolo, 2000;
Dubkov et al., 2004; Spagnolo et al., 2007). The translocation time of chain polymers has
been theoretically studied in the presence of a dichotomically fluctuating chemical potential
only as a function of its amplitude in Ref. (Park and Sung, 1998).
In particular we investigate the role of an external oscillating forcing field on the transport
dynamics of short polymers surmounting a barrier, in the presence of a metastable state. We
find a minimum of the mean first translocation time (MFTT) of the molecule center of mass
as a function of the frequency of the forcing field. This nonlinear behavior represents the
Environmental Noise and Nonlinear Relaxation in Biological Systems 293
resonant activation (RA) phenomenon in polymer translocation. We find that a suitable tuned
oscillating field can speed up or slow down the mean time of the translocation process of a
molecule crossing a barrier, using the frequency as a control parameter. This effect can be of
fundamental importance for all those experiments on cell metabolism, DNA-RNA sequencing
and drug delivery mechanism in anti-cancer therapy.
In the third part of this chapter we investigate the transient dynamics of the Verhulst
model perturbed by arbitrary non-Gaussian white noise. The nonlinear stochastic systems
with noise excitation have attracted extensive attention and the concept of noise-induced
transitions has got a wide variety of applications in physics, chemistry, and biology
(Horsthemke and Lefever, 1984). Noise-induced transitions are conventionally defined in
terms of changes in the number of extrema in the probability distribution of a system variable
and may depend both quantitatively and qualitatively on the character of the noise, i.e. on the
properties of stochastic process which describes the noise excitation. The Verhulst model,
which is a cornerstone of empirical and theoretical ecology, is one of the classic examples of
self-organization in many natural and artificial systems (Eigen and Schuster, 1979). This
model, also known as the logistic model, is relevant to a wide range of situations including
population dynamics (Horsthemke and Lefever, 1984; Morita, 1982; Ciuchi et al., 1993;
Mathis and Kiffe, 1984), self-replication of macromolecules (Eigen, 1971), spread of viral
epidemics (Acedo, 2006), cancer cell population (Ai et al., 2003), biological and biochemical
systems (Derise and Adam, 1990; Ciuchi et al., 1996), population of photons in a single mode
laser (McNeil and Walls, 1974; Ogata, 1983), autocatalytic chemical reactions (Schlögl,
1972; Chaturvedi et al., 1976; Gardiner and Chaturvedi, 1977; Bouché, 1982; Leung, 1987),
freezing of supercooled liquids (Das, 1983), and social sciences (Herman and Montroll,
1972; Montroll, 1978).
By considering the season fluctuations and the random availability of resources we
analyze the stochastic Verhulst equation in the presence of a non-Gaussian stochastic process.
By investigating the transient dynamics of this model we obtain exact results for the mean
value of the population density and its non-stationary probability distribution for different
types of white non-Gaussian noise sources. Noise-induced transitions for the probability
distribution of the population density and a nonmonotonic behavior of the nonlinear
relaxation time as a function of the Cauchy noise intensity are found.
The chapter is organized as follows. In section 2.1 we report on experimental setup and
methods used in the investigation of behavioral response in N. viridula. In section 2.2, we
present our experimental results of directionality tests on the behavior of male individuals of
N. viridula. In section 2.3 we discuss the experimental findings and compare them with
theoretical results obtained by a hard threshold model.
In Sect. 3 we present our polymer chain model and give the details of the molecular
dynamics simulation process. Results are reported in Sect. 3.1. In the next sections 4. - 6. we
present our Verhulst stochastic model with Lévy noise excitation together with all the
theoretical results obtained. Finally conclusions are drawn in Sect. 7.
294 B. Spagnolo, D. Valenti, S. Spezia et al.
are two identical wooden branches, 25 cm long, and 0.4 cm thick, as shown in Fig. 1.d. The
angle between two branches is 30° - 50°.
In our experiment a signal is sent along one branch of the Y-shaped substrate and the
behavior of single male individuals, initially placed at the center of the vertical stem, is
observed [39] (see Fig. 1.d). The source of vibratory signals (i.e. the cone used as an electro-
acoustic transducer) is in contact with the right apex of the Y-shaped dummy plant. Vertical
stem and lateral branches are not in direct contact, albeit in close (0.5 cm) proximity.
Figure 1. (a) Oscillogram (b) Power spectrum density and (c) Sonogram of the non pulsed type of
Nezara viridula female calling song; (d) The block diagram of the experimental setup.
We consider a trial valid if the insect, before choosing one direction in the Y-shaped
structure, has checked the two possible directions of signal origin, touching the lower
extremity of both branches. By following these criteria, we made several observations for
different intensities of female calling songs, recording the choices (left or right) of each male
296 B. Spagnolo, D. Valenti, S. Spezia et al.
individual used in our trials, and obtaining a set of statistical data that allows us to determine
the intensity threshold value at which the bugs start to "hear" the calling song.
The presence of an "oriented" behavior, that is the tendency of the insects to choose the
branch with the signal source, is revealed by performing directionality tests on a group of
male individuals. When we observe a percentage of insects higher than 65% going towards
the acoustic source, source-direction movement (SDM), we consider that the signal has been
revealed by the insects. In Fig. 2.a we plot the relative frequency of SDMs, that is the number
of SDMs divided by the total trials, at different signal intensities. The exact number of trials,
performed for each intensity, is reported beside the corresponding point in the graph. For
small values (lower than 0.0010 V) of the signal power approximately 50% of the insects
choose one direction and the remaining 50% the other.
Figure 2. Plots of the Source-Direction Movement (SDM) Ratio as a function of: (a) the female calling
song Root Mean Square (RMS) amplitude (purely deterministic signal); (b) the noise intensity D. In
each experimental value is reported the error bar and beside the corresponding number of the performed
trials.
Conversely, for values greater than 0.0020 V, the insects show a preferential behavior,
choosing the direction from which the signal originates in the 80% of the trials.
Consequently, we have chosen the value 0.0015 V of the signal power as the threshold level
for signal detection.
Then, by using a sub-threshold signal plus a Gaussian white noise signal we have
investigated the response of the test insect for different levels of noise intensity D. In Fig. 2.b
we report the percentage of SDMs as a function of D, finding the optimal noise intensity that
maximizes the recognition between individuals of opposite sex. The graph shows a maximum
−5 2 -5 2
for D≈ 1. 30⋅ 10 V . For values of D both lower and higher than 1.30 10 V , the
-5 2
response of insects is not significant. In particular, for D > 1.30 10 V the percentage of
individuals going towards the acoustic source decreases below 0.5 reaching 0.2 for
−5 2
D≈ 3. 75⋅ 10 V . The other values of the SDM ratio close to 50%, indicate that
individuals of N. viridula randomly choose the direction of their motion, that is no oriented
behavior occurs. The nonmonotonic behavior of SDM, with a maximum at
Environmental Noise and Nonlinear Relaxation in Biological Systems 297
−5
D≈ 1. 30⋅ 10 V2, indicates that in the presence of a sub-threshold deterministic signal, the
environmental noise can play a constructive role, amplifying the weak input signal and
contributing to improve the communication among individuals of N. viridula. The occurrence
−5
of a minimum in the SDM behavior at D≈ 3. 75⋅ 10 V2, will be subject of further
investigations. A possible conjectural explanation is the following: when the noise intensity is
so large that the signal received from the vibro-receptors is significantly modified, the male
insects are not able to recognize the female calling song, and they exchange it for the song of
some rivals.
A further increase of the noise intensity causes the spectrum of the received signal to
become indistinguishable from a pure environmental noise and therefore the insect is unable
to recognize any signal of N. viridula individuals. This implies that no significant response is
observed in terms of percentage of source-direction movements (SDMs ~ 50%).
Figure 3. Evolution driven by a sinusoidal function plus noise. (a) Time series generated by consecutive
pulses (dashed line: threshold level, solid line: mean value of the periodic signal); (b) Temporal
sequence of threshold crossing events.
298 B. Spagnolo, D. Valenti, S. Spezia et al.
Stochastic resonance (SR), initially observed in the temperature cycles of the Earth (Benzi
et al., 1981), is a counterintuitive phenomenon occurring in a large variety of non-linear
systems, whereby the addition of noise to a weak periodic signal causes it to become
detectable or enhances the amount of transmitted information through the system (Moss et al.,
1994; Gingl et al., 1995; Pei et al., 1995; Pikovsky and Kurths, 1997; Longtin and Chialvo,
1998; Stocks, 2001; Wiesenfeld et al., 1994; Gammaitoni, 1995; Wannamaker et al., 2000;
Lindner et al., 2004; Gammaitoni et al., 1998; Mantegna and Spagnolo, 1994; Mantegna et
al., 2001; Vilar et al., 1998; Longtin et al., 1991; Bulsara et al., 1991; Chialvo and Apkarian,
1993; Neiman and Russell, 2002; Bahar et al., 2002; Douglass et al., 1993; Russell et al.,
1999; Freund et al., 2002; Greenwood et al., 2000). When SR occurs, the response of the
system undergoes resonance-like behavior as a function of the noise level. In spite of the fact
that initially this phenomenon was restricted to bistable systems, it is well known that SR
appears in monostable, excitable, and non-dynamical systems.
Figure 4. Temporal evolution of the simulated calling signal over the threshold for noise intensity
−6 −5 −3
D =2. 6⋅ 10 V2 a), D =1. 30⋅ 10 V2 (b) and D =1. 0⋅ 10 V2 (c). The corresponding
power spectral densities are shown in panels b, d, f. The threshold level is sth = 0.045 V and RMS
amplitude of the subthreshold signal is 0.031 V2. In the figures (a), (c) and (e) we have rescaled the
values of the signal amplitude in such a way that the zero value corresponds to the threshold value.
and characterized by: (i) an energetic activation barrier (threshold); (ii) a weak coherent input
such as a periodic signal (sub-threshold signal); (iii) a source of noise which is inherent to the
system, or is added externally to the deterministic input (Moss et al., 1994; Gingl et al., 1995;
Pei et al., 1995). Since the three ingredients are often present in nature and the idea of the
existence of a threshold is quite intuitive, TSR has migrated into many different fields, so that
during the last decades a considerable amount of literature on this subject has appeared in
several areas of science and engineering.
We have simulated a system with a threshold 0.045 V and a subthreshold signal of RMS
amplitude 0.031 V (a. u.), obtained by the recorded female calling song. In Fig. 4 we show
the output signal, and the corresponding PSD, for three different levels of noise added to the
subthreshold deterministic signal (calling song). In the Figs. 4.a, 4.c, 4.e we have rescaled the
values of the signal amplitude in such a way that the zero value corresponds to the threshold
value. For low noise intensities the signal crosses the threshold (dashed line in Fig. 4.a) very
rarely, and in the corresponding PSD (Fig. 4.b) no frequency shows any significant power
enhancement. By increasing the noise level the threshold crossings become more frequent
(Fig. 4.c) and the PSD appears to take a larger value for f = 102.5 Hz (Fig. 4.d), that is the
dominant frequency contained in the input signal. A further increase of noise intensity
produces a degradation of the signal, a loss of coherence in the temporal sequence (Fig. 4.e)
and a reduction of the main peak (f = 102.5 Hz) in the PSD (Fig. 4.f). The signal-to-noise
ratio (SNR) at f = 102.5 Hz is reported in Fig. 5. For each value of the noise intensity we have
performed 5000 numerical realizations. The noise intensity for which the SNR is maximum is
−5
D≈ 1 . 17⋅ 10 , which is very near the value of the noise intensity that maximizes the
percentage of SDMs (see Fig. 2.b).
Figure 5. Signal to noise ratio versus variance noise D of the output signal model when the input female
calling song is subthreshold, at the dominant frequency f = 102.5 Hz. All the other parameters are the
same of Fig. 2.
The results obtained from our model suggest that in the biological system analyzed,
stochastic resonance plays a key role, since it permits information to be extracted from a weak
deterministic signal, thanks to the constructive action of environmental noise. In other words
there is a suitable noise intensity which maximizes the activating behavior of the green bugs
300 B. Spagnolo, D. Valenti, S. Spezia et al.
and this effect can be described by the simplest threshold model which shows stochastic
resonance. In Fig. 5 we report also the best fitting curve of the simulations (cross points in the
figure) obtained by the formula of the SNR for a single frequency coherent signal (Moss et
al., 1994)
a b
SNR=c log[ exp(− ) ], (1)
2 D
D
−5 −6
where a =6. 6⋅ 10 , b =1. 70⋅ 10 , and c = 2.18.
(2)
(3)
where Kr is the elastic constant, rij the distance between particles i and j, d the equilibrium
distance between adjacent monomers, K the bending modulus, LJ the LJ energy depth and
the monomer diameter. The effect of temperature fluctuations on the dynamics of a chain
polymer escaping from a metastable state is studied in a two-dimensional environment. The
polymer translocation is modeled as a stochastic process of diffusion in the presence of a
potential barrier having the form
−3 −4
with parameters a =3⋅ 10 and b =2⋅ 10 , as already adopted in Ref. (Pizzolato et al.,
2009). A three-dimensional view of UExt is plotted in Fig. 6.
The drift of the ith monomer of the chain molecule is described by the following
overdamped Langevin equations
Environmental Noise and Nonlinear Relaxation in Biological Systems 301
dx i ∂ U ij ∂ U Ext
=− − +√
D ξ x +Acos(ωt+θ ) (6)
dt ∂x ∂x
dy i ∂ U ij
=− +√Dξ y (7)
dt ∂y
where Uij is the interaction potential between the ith and jth beads, x and y are white
Gaussian noise modeling the temperature fluctuations, with the usual statistical properties,
namely <x (t) > = 0 and <k (t)l (t+) > = kl ()0 for (k,l = x,y). A and are respectively
the amplitude and the angular frequency of the forcing field and a randomly chosen initial
phase. In our simulations, the time t is scaled with the friction parameter as t = tr/, where tr
is the real time of the process.
Figure 6. 3D-view of the potential energy UExt ,which is included in our system to simulate the presence
of a barrier to be surmounted by the polymer. A sketch of the translocating chain molecule is shown.
2 1/2
The standard Lennard-Jones time scale is η LJ = ( mζ /ε LJ ) , where m is the mass of
the monomer. A bead of a single-stranded DNA is formed approximately by three nucleotide
bases and then ζ ≈ 1 . 5 nm and m≈ 936 amu (Luo et al., 2008). Orders of magnitude of
the quantities involved in the process are nanometers for the characteristic lengths of the
3
system (polymer and barrier extension) and microseconds for the time domain. A set of 10
numerical simulations has been performed for different values of the frequency of the forcing
field and two values of the noise intensity D, namely D = 1.0, 4.0. The values of the potential
energy parameters are: Kr = K = 10, LJ = 0.1, = 3 and d = 5, in arbitrary units (AU). The
−2
amplitude of the electric forcing field is A =5⋅ 10 in AU, because it is scaled with . The
number of monomers N is 20. The initial spatial distribution of the polymer is with all
monomers at the same coordinate x0 = 0, corresponding to the local minimum of the potential
energy of the barrier. Every simulation stops when the x coordinate of the center of mass of
the chain reaches the final position at xf = 15.
302 B. Spagnolo, D. Valenti, S. Spezia et al.
Figure 7. MFTT vs. frequency of the forcing field for two different values of the noise intensity D. The
values of the potential energy parameters are: Kr = K= 10, LJ = 0.1, 3 and d = 5, in arbitrary units
−2
(AU). The amplitude of the electric forcing field is A =5⋅ 10 (AU). The number of monomers N is
20.
Figure 8. Probability density function (PDF) of the first translocation time (FTT). Each panel shows
two PDFs, each one characterized by a specific value of the noise intensity. The three panels differs for
the frequency of the forcing field. The panel (a) shows the time distribution in the low frequency
region. The long tails indicate that the polymer crosses the potential barrier with a longer mean time. In
the panel (b) the FTTs are distributed towards shorter values, because of the lowest time scale
characterizing the translocation process in the resonant activation regime. The panel (c) shows the
probability distribution for the high frequency domain, where the time scale is the same that is
characterized by the presence of a static potential.
Environmental Noise and Nonlinear Relaxation in Biological Systems 303
The MFTT shows three different translocation regimes as a function of the frequency
(Fig. 7). In the low frequency domain (-3), the period of the forcing field oscillations is
very long with respect to the typical values of the mean crossing time of the chain molecule.
In this regime the MFTT is equal to the average of the crossing times over upper and lower
configurations of the barrier, and the slowest process determines the value of the mean
crossing time. As a consequence, the MFTT increases and we observe long tails in the
probability density function (PDF) shown in Fig. 8.a. In the high frequency domain
(-1), a saturation of the translocation time is obtained. In this case, very rapid
oscillations act on the polymer motions as the mean potential, i. e. the static field, and
therefore the MFTT becomes equal to that obtained without any additional periodic driving.
In other words, the polymer chain feels the average potential barrier. For intermediate
frequencies (10-2 < -3), the crossing event is strongly correlated with the potential
oscillations and the MFTT vs. exhibits a minimum at a resonant oscillation rate. This
frequency region corresponds to periods of oscillations which are of the same order of
magnitude of the mean time the polymer takes to cross a static barrier with its shape
corresponding to the lowest configuration of the oscillating potential. In other words, the
potential remains around its lowest configuration for enough time to allow the polymer to exit
and, even in the case of an initially high or intermediate value of the height of the barrier, the
potential feature turns into the lowest configuration within a sufficiently short time lag to
facilitate the translocation process. The polymer, driven by a periodic field oscillating at a
period comparable with a characteristic time of the crossing dynamics, reaches a resonant
regime that accelerates the translocation process. For each of the frequency values, the
thermal noise intensity D is able to speed up or slow down the crossing process, as described
by the three frequency regions (Fig. 7) and the corresponding translocation dynamics
(Pizzolato et al., 2010). The probability density function of the first translocation time (FTT)
is shown in Fig. 8 for three frequency values characterizing the different dynamical domains.
Each panel shows two PDFs, each one characterized by a specific value of the noise intensity.
In the resonant activation regime (Fig. 8.b) the PDFs do not present the long tail at higher
crossing times, observed in Fig. 8.a. Consequently, the MFTT reduces its value. The PDF
assumes an interesting two-peaks structure that suggests the presence of two characteristic
times of translocation. This feature, being present both at low and high noise intensity, can be
ascribed to two different translocation dynamics of the polymer chain surmounting the
barrier. In the high frequency domain (Fig. 8.c) the PDFs show the characteristic feature of
the static potential case.
dx x
dt ( )
=rx 1−
Ω
(8)
where there is the Malthus term with the rate constant r and a saturation term with the
factor, which is the upper limit for the population growth due to the availability of the
resources.
Really the parameters r and are not constant. In fact the parameter r changes randomly
due to season fluctuations, and the parameter fluctuates due to the environmental
interaction which causes the random availability of resources. As a consequence we have the
following stochastic Verhulst equation
dx
dt [
=r (t )x 1−
x
Ω (t )]. (9)
In the context of macromolecular self-replication, the model equation (9), with constant
and a white Gaussian noise in r(t), was numerically studied in Ref. (Leung, 1988) and the
critical slowing down, i.e. a divergence of the relaxation time at some noise intensity, was
found. Later Jackson and co-authors (Jackson et al., 1989) investigated the same model, by
analog experiment and digital simulations. They analyzed specifically in detail the nonlinear
relaxation time defined as (Binder, 1973)
(10)
and did not observe the critical slowing down. They explained this discrepancy by the
incorrect approximate truncation of the asymptotic power series for T used in Ref. (Leung,
1988). The stability conditions were derived in Ref. (Golec and Sathananthan, 2003). Similar
investigations for colored Gaussian noise r(t) were performed in Ref. (Mannella et al., 1990),
where a monotonic dependence of the relaxation time and the correlation time on the noise
intensity was found. Some exact results for Eq. (9) with constant r and Markovian
dichotomous noise excitation (t) = r/ (t) was obtained in Ref. (Zygadło, 2008).
The generalization of Eq. (9), to study a Bernoulli-Malthus-Verhulst model driven by a
multiplicative white and colored Gaussian noise, was analyzed in Refs. (Calisto and Bologna,
2007; Suzuki et al., 1982; Brenig and N. Banai, 1982; Makino and Morita, 1985; Morita and
Makino, 1986). In Refs. (Makino and Morita, 1985; Morita and Makino, 1986) the authors,
using perturbation technique, obtained the exact expansion in power series on noise intensity
of all the moments and found the long-time decay of t-1/2 (see also Ref. [Ciuchi et al., 1993]).
In the present chapter, using the previously obtained results for a generalized Langevin
equation with a Lévy noise source (Dubkov and Spagnolo, 2005; Dubkov et al., 2008), we
investigate the transient dynamics of the stochastic Verhulst model with a fluctuating growth
rate and a constant value for the saturation population density , that is . The exact
results for the mean value of the population density and its non-stationary probability
distribution for different types of white non-Gaussian excitation r(t) are obtained. We find the
Environmental Noise and Nonlinear Relaxation in Biological Systems 305
interesting noise-induced transitions for the probability distribution of the population density
and the relaxation dynamics of its mean value for Cauchy stable noise. Finally we obtain a
nonmonotonic behavior of the nonlinear relaxation time as a function of the Cauchy noise
intensity.
dx
=r (t )x (1− x ). (11)
dt
(12)
where x0 = x(0). Now by substituting in Eq. (12) the following expression for the random rate
r(t)
r (t )=r+ξ (t ), (13)
Where r > 0 and (t) is an arbitrary white non-Gaussian noise with zero mean, we can rewrite
the solution (12) as
1− x 0 − rt− L (t ) − 1
(
x (t )= 1+
x0
e
) . (14)
Here L(t) denotes the so-called Lévy random process with L(0) = 0, and ξ (t )= L̇ (t ) . As it
was shown in Refs. (Dubkov and Spagnolo, 2005; Dubkov et al., 2008; Feller, 1971), Lévy
process having stationary and statistically independent increments on non-overlapping time
intervals belongs to the class of stochastic processes with infinitely divisible distributions. As
a consequence, the characteristic function of L(t) can be represented in the following form
(see Eq. (6) in [Dubkov and Spagnolo, 2005])
306 B. Spagnolo, D. Valenti, S. Spezia et al.
(15)
where (z) is some non-negative kernel function. The case (z) = 2D(z) corresponds to a
white Gaussian noise excitation (t), while for a symmetric Lévy stable noise (t) with index
1− α
we have a power-law kernel ρ (z )=Q∣z∣ , with 0<.
In the model under consideration the stationary probability distribution has; (i) a
singularity at the stable point x = 1 for white Gaussian noise; and (ii) two singularities at both
stable points x = 0and x = 1 for Lévy noise. To analyze the time behavior of the probability
distribution in the transient dynamics it is better not to use the Kolmogorov equation for the
probability density P(x,t), but rather the exact solution (14). Using the standard theorem of the
probability theory regarding a nonlinear transformation of a random variable, we find from
Eq. (14)
(1− x 0 )x
P (x,t )=
1
x (1− x ) ([
P L ln
] )
x 0 (1− x )
−rt,t , (16)
where PL(z,t) is the probability density corresponding to the characteristic function (15). For a
white Gaussian noise (t), this distribution reads
1 z2
P L (z,t )=
2√πDt
exp −
4 Dt { }
. (17)
The time evolution of the probability distribution P(x,t) for D = 0.3, r = 2, and x0 = 0.1 is
plotted in Fig. 9.
Figure 9. Time evolution of the probability distribution of the population density for white Gaussian
noise excitation with intensity D . The values of the parameters are: x 0 =0 .1 , r =2 , D =0. 3 .
Environmental Noise and Nonlinear Relaxation in Biological Systems 307
As it is easily seen, the maximum of the unimodal distribution with initial position at x0 =
0.1 shifts with time towards the stable point at x = 0.1. At the same time, as it follows from
Eqs. (16) and (17), for all t > 0 we have
(18)
The same picture is observed for another kernel function ρ (z )=Kz / (2sinh z ) (K >0 ) ,
corresponding to a Lévy process L(t) with finite moments and the following probability
density of increments
2 Kt− 1
P L (z,t ) = 2
π Γ (Kt )
Γ (
Kt iz
+
2 π
Γ
2 π)(
Kt iz
− , ) (19)
where (x) is the Gamma function. The corresponding time evolution of the probability
distribution P(x,t) for K = 0.2, r = 2, and x0 = 0.1 is shown in Fig. 10.
Figure 10. Time evolution of the probability distribution of the population density in the case of Lévy
noise with distribution (19). The values of the parameters are: x0 = 0.1, r = 2 , K = 0.2.
A different situation we have for a Cauchy stable noise (t) with constant kernel
ρ (z )=Q (α =1 ) . After evaluation of the integral in Eq. (15), the probability density of the
Lévy process increments takes the form of the well-known Cauchy distribution (Feller, 1971)
D1 t
P L (z,t )= , (20)
π z 2 + (D1 t )2
[ ]
where D1 = Q is the noise intensity parameter. In such a case from Eqs. (16) and (20) for all
t > 0 we find
308 B. Spagnolo, D. Valenti, S. Spezia et al.
(21)
As a result, from an initial delta function we immediately obtain a trimodal distribution for t
> 0 and then after some transition time tc a bimodal one with two singularities at the stable
points x = 0 and x = 1 (see Figs. 11 - 13).
Figure 11. Time evolution of the probability distribution of the population density in the case of white
Cauchy noise excitation. The values of the parameters are: x0 = 0.1, r = 2, D1 = 0.7.
We should note that the transition from trimodal to bimodal distribution is a general
feature of the model in the presence of a Cauchy stable noise, and it is not limited to some
range of parameters. In fact, from Eq. (21) and a delta function initial distribution inside the
interval (0,1), this transition always takes place.
In the following Figs. 12 and 13 we show the time evolution of the probability
distribution of the population density for two other values of the noise intensity, namely D1 =
1.2 and D1 = 1.7. As the noise intensity increases the probability distribution shows two
singularities near x = 0 and x = 1 with different amplitude.
Figure 12. Time eolution of the probability distribution of the population density in the case of white
Cauchy noise. The values of the parameters are : x0 = 0.1, r = 2, D1 = 1.2.
Environmental Noise and Nonlinear Relaxation in Biological Systems 309
Figure 13. Time evolution of the probability distribution of the population density in the case of white
Cauchy noise. The values of the parameters are x0 = 0.1, r = 2, D1 = 1.7.
This transition in the shape of the probability distribution of the population density is due
to both the multiplicative noise and the Lévy noise source. Using Eqs. (16) and (20) and
equating to zero the derivative of P(x,t) with respect to x, we obtain the following condition
for the extrema in the range 0 < x < 1, and particularly for a minimum in the same interval
z ( x,t ) 1
=x− , (22)
2
z ( x,t ) + ( D1 t ) 2 2
with
(1− x 0 )x
z ( x,t )= ln
[ ]
x 0 (1− x )
−rt . (23)
This condition can be solved graphically by finding the intersection between the functions
2 2
y1 =z ( x,t )/ ( z ( x,t ) + ( D1 t ) ) and y2 = x – 1/2. This is done in the following Figs. 14 -
16, where the function y1 is plotted for three different values of time and noise intensity. In
each figure the black blue curve (color on line) corresponds to the critical value of time tc for
which we have a noise induced transition of the probability distribution of the population
density from trimodal to bimodal, that is from two minima and one maximum to one
minimum inside the interval 0 < x < 1. The appearance of one minimum in the probability
distribution is the signature of this transition.
The three values of the critical time tc corresponding to the three values of the Lévy
noise intensity investigated are: (D1)1 = 0.7, tc = 1.75; (D1)2 = 1.2, tc = 1.3; (D1)1 = 1.7, tc =
0.95. One rough evaluation of the critical time tc is obtained by putting equal to 1 the scale
parameter of the Cauchy distribution of Eq. (20), that is 1/D1. The critical time tc is the time at
which the maximum and one minimum of the probability distribution (see Figs. 11 - 13)
coalesce in one inflection point and in this point x the function y2 = x – 1/2 becomes tangent
at the function y1 (see Figs. 14 - 16). It is interesting to note that the critical time tc decreases
310 B. Spagnolo, D. Valenti, S. Spezia et al.
with the noise intensity D1. This is because by increasing the noise intensity, more quickly the
population density reaches the two points near the boundaries x = 0 and x = 0.
Figure 14. Plots of both sides of Eq. (22) (white Cauchy noise): function y1 (solid curves), function y2
(dashed curve), for three values of time, namely: t = 0.8, 1.75, 2.5. The critical time is tc = 1.75 (black
blue curve). The values of the other parameters are: x0 = 0.1, r = 2, D1 = 0.7.
Figure 15. Plots of both sides of Eq. (22) (white Cauchy noise): function y1 (solid curves), function y2
(dashed curve), for three values of time, namely: t = 0.9, 1.3, 3. The critical time is tc = 1.3 (black blue
curve). The values of the other parameters are: x0 = 0.1, r = 2, D1 = 1.2.
Environmental Noise and Nonlinear Relaxation in Biological Systems 311
Figure 16. Plots of both sides of Eq. (22) (white Cauchy noise): function y1 (solid curves), function y2
(dashed curve), for three values of time, namely: t = 0.4,0.95,1.5 . The critical time is tc = 0.95 (black
blue curve). The values of the other parameters are: x0 = 0.1, r = 2, D1 = 1.7.
where
−1
1− x 0
(
f (q )= 1+
x0
q
) . (25)
Then, by expanding the smooth function (25) in a standard Taylor power series in q around
the point q = 0 we have
(26)
(27)
(28)
For white Gaussian noise (t) with kernel (z) = 2 D (z) we obtain from Eq. (28) the
following asymptotic series
(29)
By considering a finite number of terms in this expansion leads to a wrong conclusion about
the critical slowing down phenomenon in such a system, as found in Ref. (Leung, 1988). The
exact result is obtained, of course, by summing all the terms in Eq. (29). Moreover, for most
of the kernels (z) the integral in Eq. (28) diverges. Thus, this approach is inappropriate for
our purposes, and it is better to use the direct averaging in Eq. (14). Therefore, using this
second approach we have
(30)
Figure 17. Nonlinear relaxation of the mean population density in the case of white shot noise
excitation, for three values of the mean frequency , namely 0.3, 3. The values of the other
parameters are: x0 = 0.1, r = 2, a0 = 1.
Environmental Noise and Nonlinear Relaxation in Biological Systems 313
ai δ (t − t i ) (31)
− νt (1− cosa 0 u )
〈e iuL(t )〉=e . (32)
By making the reverse Fourier transform in Eq. (32) we find the probability distribution of
the corresponding Lévy process
(33)
where In(x) is the n-order modified Bessel function of the first kind. The relaxation of the
mean population density <x(t)> is shown in Fig. 17. According to the Eqs. (30) and (33) the
stationary value of the population density in such a case is <x(t)>st = 1, but the relaxation
time (10) increases with increasing the mean frequency of pulses.
Figure 18. Nonlinear relaxation of the mean population density in the case of Lévy noise with
distribution (19), for three values of the parameter K, namely K = 0.4, 2. The values of the other
parameters are: x0 = 0.1, r = 2.
For white non-Gaussian noise with the kernel ρ ( z )=Kz / (2sinh z ) we observe a
similar transient dynamics, which is shown in Fig. 16. We have the same stationary value
<x(t)>st, and the relaxation time T increases with increasing the parameter K, which is
314 B. Spagnolo, D. Valenti, S. Spezia et al.
proportional to the noise intensity. Finally, in the case of white Cauchy noise (t) we obtain
interesting exact analytical results. First of all, substituting Eq. (20) in (30) and changing the
variable z = D1ty under the integral, we obtain
(34)
For the stationary mean value <x(t)>st we find from Eq. (34)
(35)
where 1(x) is the step function. After evaluation of the integral in Eq. (35) we obtain finally
1 1 r
〈 x 〉st = + arctan . (36)
2 π D1
As it is seen from Fig. 19 and Eq. (36), for small noise intensity D1, with respect to the value
of the rate parameter r = 2, the stationary mean value of the population density is
approximately 1, as for the other white non-Gaussian noise excitations considered. But for
large values of D1, this asymptotic value, which is independent from the initial value of
population density x0, tends to 0.5.
Figure 19. Nonlinear relaxation of the mean population density in the case of white Cauchy noise, for
three values of the noise intensity D1, namely D1 = 0.2,2,7. The values of the other parameters are: x0 =
0.1, r = 2.
Environmental Noise and Nonlinear Relaxation in Biological Systems 315
It is interesting also to analyze, for this case of white Cauchy noise, the dependence of the
relaxation time T from the noise intensity D1. Substituting Eq. (34) in (10) and changing the
order of integration, for initial condition x0 = 0.5, we are able to calculate analytically the
double integral in t and in y obtaining the final result
π ln2
T= . (37)
r(1+D 12 /r 2
)arccot( D1 /r )
We find a nonmonotonic behavior of the relaxation time T versus the noise intensity D1 with a
maximum at the noise intensity D1 = 0.75, as shown in Fig. 20.
Figure 20. Nonmonotonic behavior of the nonlinear relaxation time T as a function of the white Cauchy
noise intensity D1. The values of the other parameters are: : x0 = 0.5, r = 2.
This nonmonotonic behavior is also visible for another initial position: x0 = 0.1 in Fig.
19. Here the relaxation time to reach the stationary value of mean population density <x(t)>st,
increases from very low noise intensity (D1 = 0.2) to moderate low intensity (D1 = 2), while
decreases for higher noise intensities (D1 = 7). This is also due to the dependence of <x(t)>st
from the noise intensity D1 (see Eq. (36)). We note that this nonmonotonic behavior of the
relaxation time T is related to the peculiarities of the transient dynamics of the mean
population density and it will be object of further investigations.
CONCLUSIONS
In this contribution we have studied the effects of random fluctuations, i.e. noise, in the
vibrational communications occurring during the mating of N. viridula, i.e. the green bug. In
our experimental work we analyzed the behavioral response of different individuals of N.
viridula to a deterministic signal (calling song), measuring for these individuals the threshold
of the neural activation. Afterwards, we analyzed the green bug response when a sub-
316 B. Spagnolo, D. Valenti, S. Spezia et al.
threshold deterministic signal is added with an external noise source. By using the Source-
Direction Movement ratio as indicator of positive response to the external signal, we observe
that the behavioral activation of the insects is characterized by a nonmonotonic behavior as a
5 2
function of the noise intensity D, with a maximum at D = Dopt 1.30 10 V . The
value Dopt maximizes the efficiency of the sexual communication among individuals of
Nezara viridula (L.), and therefore represents the optimal noise intensity during the mating
behavior of these insects. The nonmonotonic behavior observed in the insect response as a
function of the noise intensity is the signature of the threshold stochastic resonance (TSR)
[117]. By using a threshold model we obtained numerical results for the threshold crossing,
which corresponds in our model to the behavioral activation, finding a theoretical value for
the optimal noise intensity. Experimental and numerical results are compared, finding a good
agreement between the values of the optimal noise intensity obtained by the experimental
work and model (see Figs. 2.2(b) and 3).
We analyzed the influence of an external oscillating driving field on the translocation
dynamics of short polymers embedded in a noisy environment. We simulate the translocation
process by letting the polymer to cross a potential barrier starting from a metastable state, in
the presence of thermal fluctuations. The mean translocation time as a function of the
frequency of the driving force shows a nonmonotonic behavior, with the noise intensity acting
as a scaling factor of the values of the crossing times. The forcing periodic electric field
jointly with the temperature of the system can be able to speed up or slow down the polymer
translocation. In this view, the oscillating electric field constitutes a tuning mechanism to
select a suitable translocation time of the polymer. This feature may have important biological
effects on the cell metabolism, for example, during a cancer targeted therapy.
Finally, the transient dynamics of the Verhulst model, perturbed by arbitrary non-
Gaussian white noise, has been investigated. This well-known equation is an appropriate
ecological and biological model to describe closed-population dynamics, self-replication of
macromolecules under constraint, cancer growth, spread of viral epidemics, etc... By using
the properties of the infinitely divisible distribution of the generalized Wiener process, we
analyzed the effect of different non-Gaussian white sources on the nonlinear relaxation of the
mean population density and on the time evolution of the probability distribution of the
population density. We obtain exact results for the non-stationary probability distribution in
all cases investigated and for the Cauchy stable noise we derive the exact analytical
expression of the nonlinear relaxation time. Due to the presence of a Lévy multiplicative
noise, the probability distribution of the population density exhibits a transition from a
trimodal to a bimodal distribution in asymptotics. This transition, characterized by the
appearance of a minimum, happens at a critical time tc, which can be roughly evaluated as
1/D1 (where D1 is the noise intensity) and exactly evaluated from the condition (22). Finally a
nonmonotonic behavior of the nonlinear relaxation time of the population density as a
function of the Cauchy noise intensity was found.
ACKNOWLEDGMENTS
Authors acknowledge the financial support by MIUR.
Environmental Noise and Nonlinear Relaxation in Biological Systems 317
REFERENCES
Acedo L. 2006. A second-order phase transition in the complete graph stochastic epidemic
model. Physica A 370:613-624.
Agudov NV, Dubkov AA, Spagnolo B. 2003. Escape from a metastable state with fluctuating
barrier. Physica A 325:144-151.
Agudov NV, Krichigin AV, Valenti D, Spagnolo B. 2010. Stochastic resonance in a trapping
overdamped monostable system. Phys. Rev. E 81:051123 (8).
Agudov NV, Spagnolo B. 2001. Noise-enhanced stability of periodically driven metastable
states. Phys. Rev. E 64:035102(R)(4).
Ai BQ, Wang XJ, Liu GT, Liu LG. 2003. Correlated noise in a logistic growth model. Phys.
Rev. E 67:022903(3).
Akeson M, Branton D, Kasianowicz JJ, Brandin E, Deamer DW. 1999. Microsecond Time-
scale Discrimination Among Polycytidylic Acid, Polyadenylic Acid, and Polyuridylic
Acid as Homopolymers or as Segments Within Single RNA Molecules. Biophys. J.
77:3227-3233.
Bagwell GJ, Čokl A, Millar JG. 2008. Characterization and Comparison of Substrate-Borne
Vibrational Signals of Chlorochroa uhleri, Chlorochroa ligata, and Chlorochroa sayi.
Ann. Entomol. Soc. Am. 101(1):235-246.
Bahar S, Neiman A, Wilkens L, Moss F. 2002. Phase synchronization and stochastic
resonance effects in the crayfish caudal photoreceptor. Phys. Rev. E, 65:050901(R)(4).
Benzi R, Parisi G, Sutera A, Vulpiani A. 1982. Stochastic resonance in climatic change.
Tellus 34:10-16.
Benzi R, Sutera A, Vulpiani A. 1981. The mechanism of stochastic resonance. J. Phys.: Math
Gen. 14:L453-L457.
Bier M, Astumian RD. 1993. Matching a diffusive and a kinetic approach for escape over a
fluctuating barrier. Phys. Rev. Lett. 71:1649-1652.
Binder K. 1973. Time-Dependent Ginzburg-Landau Theory of Nonequilibrium Relaxation.
Bjørnstad ON, Grenfell BT. 2001. Noisy Clockwork: Time Series Analysis of Population
Fluctuations in Animals. Science 293:638-643.
Boguna M, Porra JM, Masoliver J, Lindenberg K. 1998. Properties of resonant activation
phenomena. Phys. Rev. E 57:3990-4002.
Borges M, Jepson PC, Howse PE. 1987. Long-range mate location and close-range courtship
behaviour of the green stink bug, Nezara viridula and its mediation by sex pheromones.
Entomol. Exp. Appl. 44:205-212.
Bouché V. 1982. A new instability phenomenon in the Malthus-Verhulst model. J. Phys. A:
Math. Gen. 15:1841-1848.
Braun HA, Wissing H, Schäfer K, Hirsch MC. 1994. Oscillation and noise determine signal
transduction in shark multimodal sensory cells. Nature 367:270-273.
Brenig L, Banai N. 1982. Non-linear dynamics of systems coupled with external noise: Some
exact results. Physica D 5:208-226.
Brumer Y, Michor F, Shakhnovich EI. 2006. Genetic instability and the quasispecies model.
J. Theor. Biol. 241:216-222.
Bulsara A, Jacobs EW, Zhou EW, Moss F, Kiss L. 1991. Stochastic resonance in a single
neuron model: Theory and analog simulation. J. Theor. Biol., 152:531-555.
318 B. Spagnolo, D. Valenti, S. Spezia et al.
Mantegna RN, Spagnolo B. 1994. Stochastic resonance in a tunnel diode. Phys. Rev. E
49:R1792-R1795.
Mantegna RN, Spagnolo B. 2000. Experimental Investigation of Resonant Activation. Phys.
Rev. Lett. 84:3025-3028.
Mathis JH, Kiffe TR. 1984. Stochastic Population Models: A Compartmental Perspective,
(Springer--Verlag, Berlin).
McNeil KJ, Walls DF. 1974. Nonequilibrium phase transitions in chemical reactions. J. Stat.
Phys. 10:439-448.
Meller A, Branton D. 2002. Single molecule measurements of DNA transport through a
nanopore. Electrophoresis 23:2583-2591.
Meller A, Nivon L, Brandin E, Golovchenko JA, Branton D. 2000. Rapid Nanopore
Discrimination Between Single Polynucleotide Molecules. Proc. Natl. Acad. Sci. USA
97:1079-1084.
Michor F, Iwasa Y, Nowak M. 2006. The age incidence of chronic myeloid leukemia can be
explained by a one-mutation model. Proc. Natl. Acad. Sci. USA 103:14931-14934.
Montroll EW. 1978. Social dynamics and the quantifying of social forces. Proc. Natl. Acad.
Sci. USA 75:4633-4637.
Morita A, Makino J. 1986. Simple analytical solution for multiplicative nonlinear stochastic
differential equations by a perturbation technique. Phys. Rev. A 34:1595--1598 (1986).
Morita A. 1982. An exact expression for the average value of the population governed by the
stochastic Verhulst equation. J. Chem. Phys. 76:4191-4194.
Moss F, Pierson D, O'Gorman D. 1994. Stochastic resonance: tutorial and update. Int. J.
Bifurcat. Chaos 4:1383-1397.
Neiman A, Russell D. 2002. Synchronization of Noise-Induced Bursts in Noncoupled
Sensory Neurons. Phys. Rev. Lett. 88:138103(4).
Nikolaev A, Gracheva M. 2009. Controlled DNA translocation through a nanopore membrane
with different electrostatic landscapes. Biophys. J. 96:649a.
Nozaki D, Yamamoto Y. 1998. Enhancement of stochastic resonance in a FitzHugh-Nagumo
neuronal model driven by colored noise. Phys. Lett. A 243:281-287.
Ogata H. 1983. Logistic equations in nonlinear systems. Phys. Rev. A 28:2296-2299.
Panja D, Barkema GT. 2008. Passage Times for Polymer Translocation Pulled through a
Narrow Pore. Biophys. J. 94:1630-1637.
Pankratova EV, Polovinkin AV, Spagnolo B. 2005. Suppression of noise in FitzHugh–
Nagumo model driven by a strong periodic signal. Phys. Lett. A 344:43-50.
Pannizzi AR. 2000. Suboptimal nutrition and feeding behavior of hemipterans on less
preferred plant food sources. An. Soc. Entomol. Bras. 29:1-12.
Park PJ, Sung W. 1998. A stochastic model of polymer translocation dynamics through
biomembranes. Int. J. Bifurcat. Chaos 8:927-931.
Pei X, Bachmann K, Moss F. 1995. The detection threshold, noise and stochastic resonance in
the Fitzhugh-Nagumo neuron model. Phys. Lett. A 206:61-65.
Pikovsky AS, Kurths J. 1997. Coherence Resonance in a Noise-Driven Excitable System.
Phys. Rev. Lett. 78:775-778.
Pizzolato N, Fiasconaro A, Persano Adorno D, Spagnolo B. 2010. Resonant activation in
polymer translocation: new insights into the escape dynamics of molecules driven by an
oscillating field. Physical Biology, 7:034001(5).
322 B. Spagnolo, D. Valenti, S. Spezia et al.
Vilar JM, Gomila G, Rub JM. 1998. Stochastic Resonance in Noisy Nondynamical Systems.
Phys. Rev. Lett. 81:14-17.
Wannamaker RA, Lipshitz SP, Vanderkooy J. 2000. Stochastic resonance as dithering. Phys.
Rev. E 61:233-236.
Wiesenfeld K, Pierson D, Pantazelou E, Dames C, Moss F. 1994. Stochastic resonance on a
circle. Phys. Rev. Lett. 72:2125-2129.
Zhdanov VP. Quantitative modeling of selective lysosomal targeting for drug design. 2008.
Eur. Biophys. J. 37:1329-1334.
Zimmer C. 1999. Life After Chaos. Science 284:83-86.
Zygadło R. 1996. Kinetics of a Verhulst-type system with nonlinearly coupled noise. Phys.
Rev. E 54:5964-5969; Zygadło R. 2008. Power-law distribution as a result of
asynchronous random switching between Malthus and Verhulst kinetics. Phys. Rev. E
77:021130(4).
In: Ecological Modeling ISBN: 978-1-61324-567-5
Editor: WenJun Zhang, pp. 325-334 © 2012 Nova Science Publishers, Inc.
Chapter 13
Alessandro Ferrarini
Department of Evolutionary and Functional Biology,
University of Parma,
Via G. Saragat 4, I-43100 Parma, Italy.
ABSTRACT
Landscape modelling is founded on the idea that the patterning of landscape
elements strongly influences ecological characteristics, thus the ability to quantify
landscape structure is a prerequisite to the study of landscape function and change over
time as well. For this reason, much emphasis has been placed until now on developing
methods to quantify landscape structure.
Unfortunately, on one side landscape (i.e., landcover or landuse) and vegetation
maps are very complex mosaics of thousands of patches, and this makes the interpretation
of their structure very challenging. On the other side, methods developed so far to
quantify landscape structure just return numerical results, that are not linked to
cartographic outputs. Last, landscape pattern indices are numerous, and the need for a
synthetic representation is more and more impelling.
I provide here the description and application of a novel approach to landscape
structural modelling based on the combined use of GIS (Geographical Information
Systems) and multivariate statistics. First, landscape structure of the study area (Ceno
valley, Italy) is analyzed through 5 patch-based, non-redundant indicators (area, isolation,
compactness, shape complexity, interspersion) with indirect link to functional aspects.
Second, PCA (principal component analysis) is used in order to synthesize structural
indicators, and cartographic output is given. Third, KCA (k-means cluster analysis) is
applied in order to group landscape patches into homogeneous clusters, and again GIS
output is supplied. Last, LDA (linear discriminant analysis) is employed to provide
evidence for the differences among clusters.
This modelling approach provides the chance for a deep and cost-effective exegesis
of landscape structure, with promising consequences on conjecture formulation about
functional aspects as well.
326 Alessandro Ferrarini
1. INTRODUCTION
Landscape (i.e., vegetation or landcover-landuse matrix) is defined as a ―het erogeneous
land area composed of a cluster of interacting ecosystems that is repeated in similar form
throughout‖ (Forman and Godron, 1986). Landscape structure is characterized by both its
composition and configuration. Landscape composition refers to features associated with the
presence and amount of each patch type (i.e. vegetation type or landcover-landuse class) but
without being spatially explicit. Instead, landscape configuration refers to the spatial
distribution of patches across the landscape.
A well-documented assumption in landscape studies is that the spatial arrangement of
landscape elements bears a strong influence on landscape processes and functions (Ferrari and
Ferrarini, 2008; Forman, 1995). For example, many vertebrate species require specific habitat
types, and the total amount of suitable habitat likely influences the occurrence and abundance
of these vertebrate species. Adjacency effects have been documented for habitats, animals,
plants and abiotic resources. An organism‘s ability to utilize resources in adjacent patches
depends on the nature of boundaries between patches, since edges can act as differentially
permeable membrane that eases some ecological flows but obstructs others. The analysis and
exegesis of landscape structure is also a prerequisite for the success of animal and plant
species conservation efforts within protected areas (Ferrarini et al., 2008; Parolo et al., 2008;
Parolo et al., 2009; Rossi et al., 2009).
To this reason, landscape modelling is founded on the idea that the patterning of
landscape elements strongly influences ecological characteristics, thus the ability to quantify
landscape structure is also a prerequisite for the study of landscape function and change over
time. Hence, so far much emphasis has been placed on developing methods to quantify
landscape structure (e.g., Ferrarini et al., 2005; Ferrarini and Tomaselli, 2010).
Unfortunately, on one side landscape and vegetation maps are very complex mosaics of
thousands of patches, and this makes the interpretation of their structure very thought-
provoking. On the other side, methods developed so far to quantify landscape structure just
return numerical results, that are not linked to cartographic outputs. Last, landscape pattern
indices are numerous, and the need for a synthetic representation is more and more impelling.
I provide here the description and application of a novel approach to landscape structural
modelling based on the combined use of GIS (Geographical Information Systems) and
multivariate statistics. It provides the chance for landscape anamnesis and prognosis, with
promising consequences on conjecture formulation on functional aspects as well.
Landscape Structural Modeling 327
The Ceno valley is a 35,038 ha wide valley situated in the Province of Parma, Northern
Italy. It has been mapped at 1:25,000 scale (1846 GIS polygons; Ferrarini, 2005; Ferrarini et
al., 2010; Tomaselli, 2003) using the CORINE Biotopes classification system (Table 1;
C.E.C., 1991). It is characterized by a substantial human use in the form of crops (9779 ha,
27.98% of the study area), and wooded landcover given by Ostrya carpinifolia (8212 ha,
23.45%) and Quercus cerris (4562 ha, 13.02%) woods. Wetlands account for about 1.8% of
the study area. The Ceno valley can be divided into three belts. The lower supra-
mediterranean belt is dominated by Quercus pubescens woods which are prevalent on South
and S-E expositions. The higher supra-mediterranean belt goes up 1000 m a.s.l. and is
characterized by Ostrya carpinifolia woods and by the scattered occurrence of Quercus cerris
woods. The montane belt, starting at about 1000 m a.s.l., is ruled by Fagus sylvatica woods.
The cartographic model has been applied to 1448 out of 1846 GIS polygons of the study
area, because 398 patches correspond to anthropogenic polygons (residential areas, villages,
industrial sites) which have not been analyzed since the proposed approach is referred to
natural and semi-natural patches.
In this study, multivariate stats (Legendre and Legendre, 1998) are joined to GIS
capabilities in order to achieve a deep model of the inner structure of the study area.
Scores from PCA have been joined to the 1448 GIS polygons of the study area. Resulting
PCC is showed in Figures 1, 2 and 3. White areas correspond to anthropogenic polygons.
Figure 4. Cartography of the 3 resulting clusters. White areas correspond to anthropogenic polygons.
Cluster k1 (317 GIS polygons; 7418 hectares) can be conceived as the group of ―n atural
patches‖ having wide extension, high levels of isolation, complex shapes, low compactness
and high interspersion (Table 4). It‘s expected that this portion of the study area responds to
natural dynamics, and it is the result of natural driving forces (height a.s.l., acclivity, slope
aspect, soil, geomorphology). It‘s also awaited that future dynamics will be almost
exclusively driven by natural causes (climate changes and ecological successions). Due to the
low compactness, biodiversity (plants and animals) lying inside these patches is also expected
to be very sensitive to episodic human disturbance.
Cluster k3 (459 GIS polygons; 2721 hectares) is the opposite, i.e. patches largely
influenced by contiguous human patches (croplands and developed areas) determining
fragmentation (small extension), straight borders and high compactness. Hence, cluster k3 can
be reified as ―pat ches largely influenced by humans‖. It‘s awaited that this portion of the
study area responds almost exclusively to anthropogenic dynamics (expansion of developed
areas and croplands, fragmentation due to roads), which determine both the cause and future
destination of these patches.
Cluster k2 (672 GIS polygons; 14631 hectares) is halfway between previous clusters,
hence can be conceived as ―s emi-natural patches with low or average levels of human
influence‖. Ecological anamnesis and prognosis of these patches are much more difficult to
depict, they respond to a complex interplay between human and natural driving forces. A
deep understanding of this cluster of patches requires more effort, for instance a cost-effective
monitoring through satellite images in order to capture its spatial and temporal properties.
actual group
k1 k2 k3
k1 310 1 0
predicted
k2 4 665 6
group
k3 3 6 453
overall accuracy = 98.6%
3. CONCLUDING REMARKS
Landscape and vegetation mosaics are very complex mosaics of thousands of patches.
There‘s an urgent need for synthetic, cartographic models of its structural attributes having a
well-documented join with functional properties (e.g., naturality or human influence). In this
chapter, I have illustrated an ad hoc methodology, named here ―m ultivariate cartography‖ and
based on the conjoint use of GIS and multivariate statistics, which can be applied to any
landcover-landuse or vegetation map. Although I applied it to 5 structural indicators, the
proposed modeling approach has not a superior limit; hence a larger set of indicators might be
used in case they are thought meaningful for different study areas.
In recent years, remote sensing techniques have undergone considerable development in
applications, particularly with regard to territorial monitoring of both natural and developed
areas. If applied to remotely sensed data, multivariate cartography supplies the opportunity of
accurate and cost-effective surveys for capturing the inner properties of the landscape.
ACKNOWLEDGMENTS
I wish to thank Centro Studi Val Ceno ―
Cardinale Antonio Samorè‖ (Bardi, Italy) for
kindly making the CORINE map of Ceno valley available for this study.
334 Alessandro Ferrarini
REFERENCES
C.E.C. (Commission of European Community) 1991. CORINE BiotopesManual, habitats of
the European Community. A method to identify and describe consistently sites of major
importance for nature conservation. EUR 12587/3, Bruxelles.
Ferrari I, Ferrarini A. 2008. From ecosystem to landscape ecology: a progression calling for a
well-founded research and appropriate disillusions. Landscape Journal online, 6:1-12.
Ferrarini A. 2005. Analisi e valutazioni spazio-temporale mediante GIS e Telerilevamento del
grado di Pressione Antropica attuale e potenziale gravante sul mosaico degli habitat di
alcune aree italiane. Ipotesi di pianificazione. Ph.D. Thesis, Università degli Studi di
Parma, Parma, Italy.
Ferrarini A, Bollini A, Sammut E. 2010. Digital Strategies and Solutions for the Remote
Rural Areas Development. Acts of Annual MeCCSA Conference 2010, London School of
Economics, 6-8 January 2010.
Ferrarini A, Rossi P, Rossi O. 2005 Ascribing ecological meaning to habitat shape through a
piecewise regression approach to fractal domains. Landscape Ecology, 20:799-809.
Ferrarini A, Rossi G, Parolo G, Ferloni M. 2008. Planning low-impact tourist paths within a
Site of Community Importance through the optimisation of biological and logistic
criteria. Biological Conservation, 141:1067-1077.
Ferrarini A, Tomaselli M. 2010. A new approach to the analysis of adjacencies. Potentials for
landscape insights. Ecological Modelling, 221:1889-1896.
Forman RTT, Godron M. 1986. Landscape ecology. New York: John Wiley & Sons. 619 p.
Forman, R.T. 1995. Land Mosaics, the Ecology of Landscapes and Regions. Cambridge
University Press, Cambridge, UK.
Legendre P, Legendre L. 1998. Numerical Ecology. ed. Elsevier.
McGarigal K, Cushman SA, Neel MC, Ene E. 2002. FRAGSTATS: Spatial Pattern Analysis
Program for Categorical Maps. Manual available on line at
http://www.umass.edu/landeco/research
Mladenoff D, DeZonia B. 2004. APACK 2.23 Analysis software. Manual available on line at
http://flel.forest.wisc.edu
Parolo G, Rossi G, Ferrarini A. 2008. Toward improved species niche modelling. Arnica
montana in the Alps as a case study. Journal of Applied Ecology, 45: 1410-1418.
Parolo G, Ferrarini A, Rossi G. 2009. Optimization of tourism impacts within protected areas
by means of genetic algorithms. Ecological Modelling, 220: 1138-1147.
Rossi G, Parolo G, Ferrarini A. 2009. A rapid and cost-effective tool for managing habitats of
the European Natura 2000 network. Biodiversity and Conservation, 18:1375-1388.
Tomaselli M. 2003. CORINE map of the Ceno Valley. Technical Report, Centro Studi Val
Ceno ―Cardi nale Antonio Samorè‖, Bardi, Italy.
In: Ecological Modeling ISBN: 978-1-61324-567-5
Editor: WenJun Zhang, pp. 335-390 © 2012 Nova Science Publishers, Inc.
Chapter 14
ABSTRACT
Our days, the climatic change, manifested by strong and brutal precipitation, violent
wind and long drought, has as direct consequence to damage the plant canopies (forests,
sylviculture, oasis, pastoral lands and agricultural fields) so menacing the human feeding
either from plants or animals (caprine, ovine, bovine, cameline..), exhausting the water
resources, increasing the need for energy in buildings used for all activities (industrial,
agricultural and services). Which solution the ecological modelling is capable to
participate with, at short and long dated, in order to buffer the climatic change effect and
to assume the need of food and clean energy for human? In this chapter we will present
the basic concepts to model the plant architecture (species, densities, positions and
orientation) the most adaptable to the sudden calamity, the energy use efficiency in
building (material of construction, isolation system, organisation of accessories and
apparatus), and the produce of clean energy from the wind velocity (founding wind
sources and evaluating regional wind potential offshore and inshore, conceptualising
wind turbine and testing their efficiencies)
1. INTRODUCTION
The solar radiation is behind all the phenomenon of transport in the environment for all
scales: electron, atomic, molecule, particle, element, substance, material, part, body. In fact it
is needed for the accomplishment of the atmospheric water cycle by exciting the
evapotranspiration of water from soil, lakes, water tables and plant canopies, accelerating the
creation of depression zone in the atmosphere which induce the motion of air mass and after
the wind generating, finally cloud displacement and precipitation happening. So it is on
behalf of the solar radiation that the vegetation canopies of all kinds (oasis, forests, sylvi-
culture, pastoral lands, agricultural fields, …) on the glob participate actively on the
atmospheric water cycle, found their needs of water (precipitation) and it‘s on behalf of solar
radiation that they take their requirement of photosynthetic active radiation for their biomass
development. In fact there is a mutual reaction between high plant canopies, CO2 storing,
wind formation and rain creation As results, animals (caprine, ovine, bovine, camels) would
get their source of life and after the human wellbeing and alimentary security were assumed.
But this later and because of the diversity of his activity,(industrial, agriculture, services) he
needs the solar radiation in only a desired quantities depending on his request for comfort that
permits to him realising all his potencies. Also he must optimise the quantity of energy
needed in his different sectors of actions in order to avoid the waste of energy, to limit the
green house gas emission, to assume the environment equilibrium by minimising the causes
of the climatic change and to use a clean renewable energy in a controlled manner. How to
quantify this mutual interaction between solar radiation, wind intensity, energy use efficiency,
precipitation, plant development, biodiversity conservation (animal, vegetation),
environmental balance, human wellbeing and comfort? Which ecological indicators we can
use? Which equations we can apply? Which models we can develop? in the purpose to give
solutions, at short and long dates, that buffer the climatic change effect (strong and brutal
precipitation, violent wind and long drought )and assuming the need of food (plants and
animals) and clean energy for human. In this chapter we will try to respond those questions
by putting in the basic concepts to model the plant architecture for high developed canopies
like oasis, forests, sylvopastoral lands (species, densities, positions and orientation) the most
adaptable to the sudden calamity, the energy use efficiency in buildings for all human
interests specifying industrial, agricultural and services activities (material of construction,
isolation system, organisation of accessories, machines and apparatus), and the clean energy
offering from wind velocity by founding wind sources, evaluating regional wind potential
offshore and inshore, conceptualising wind turbine and testing their efficiencies.
We will begin by presenting the theoretical background common for the three
compartment of this modelling approach: biodiversity conservation, building energy
efficiency and wind potential (energy producing). After we will present specific equations
and models for biodiversity conservation in wet lands like oasis, forest and sylvopastoral
spaces with results from some case studies. Third., specifications for rationalising the use of
energy inside building with the corresponding equations will be detailed. Finally, our focus
will be about modelling for wind energy.
Basic Concepts for Modelling in Different and Complementary Ecological Fields 337
2.THEORETICAL BACKGROUND
For all systems, the phenomenon of transport at every scale of resolution (electron,
atomic, molecular, particle, element, substance, material, part, body…) and time steps
(second, minute, hours, day..) concerns generally the displacement, gain and loss of flux of
mass or energy under an external excitation effect. Either, those two parameters, flux of mass
and flux of energy, are linked because of the strong interaction between them so determining
both of them or only one and deducing the other from it is the same (Jannot 2009;
Beniston.1998; Comolet 1963). In fact, in the case of plants, as explained above, under the
effect of climatic demand(solar radiation, heat, pressure difference, temperature difference,
humidity…) plant takes water from soil by its roots and absorbs photosynthetic active
radiation by its leaves, forms the sap flow, develops its biomass and transpires water in order
to humidify the surrounding environment. The amount of water transpired constitutes a mass
of humid air that will absorbs and emits energy, losses or gains material, moves up, and
participates with the action of earth rotation (coriolis force) and atmospheric vapour pressure
difference in the formation of wind, cloud displacement and giving out precipitation. Solar
radiation and precipitation, govern the microclimatic conditions and the meteorological
parameters of the milieu surrounding buildings and influence the human need of energy for
his activities inside. So monitoring the evolution in a system (single plant or plant canopy
development‘s, wind velocity and direction, efficiency use of energy in a building) and
quantifying its inputs and outputs consist in fact to model the heat and mass transfer in the
system and to estimate the corresponding flux at every moment, every position and any scale
of resolution. Any system, at general conditions, is submissive to the following fundamental
dynamic equations (Lamoureux.and Fortuné 2001; Spinnler 1997; Comolet 1963):
Ftra Ffric Fvol Fsur Fcor m dV
dt
V : speed of training or displacement
m: Masse of the system or element of the system
Fcor : Coriolis force
Ffric : Friction force
Fsur : Forces of Surface (pressure force, centripetal force)
Fvol : Forces of volume ( electric field, magnetic field, gravity field)
By expressing each term after considering the particularity of any system we obtain the
following general set of equations (Jannot 2009; Beniston.1998; Comolet 1963):
Equation of mass conservation :
(V)0
t
d , s,t
K s,t , s,t s,t , s,t Jes,t 1 s,t P( ) s,t d
, ,
ds 4 , ,
, 4
(c) T (c) f VT (eT)
t
It‘s on behalf of this reasoning and those equations that we will develop our modelling
approach for the three compartments: Biodiversity conservation, thermal building efficiency
and wind energy producing.
at time in order to give the opportunity to the specialists to intervene at the exact moment
before their spreads, to provide the biomass development and to quantify the fodders
harvests, crops yields and fruit production, to organise the actions of grazing by fixing the
ideal number of the flock (sheeps , goats, calfs, cows, caprine, ovine, bovine, camels) to
forage as function of nutritional values that the vegetable canopies can offer, to plan the acts
of deforestation in concordance with those of renewing in order to assume the canopy
durability.
For our modelling approach here we will try to give to our model a global organisation
chart formed by a set of equations organised in a general manner in the purpose that they can
be applied to any species and to many vegetations canopies structures. So our focus will be
about plant canopies in which we can found a diversity of species, with different heights,
planted in multi-layers architecture, used for divers interests (environmental equilibrium,
animal feeding, human utilities and socio economic stability). The suitable vegetables
canopies helping us to develop this kind of global model, the most accommodates to do the
precedent cited obligations and the conveyable ones having those common architectural
characteristics are the following kinds of wet lands: forests, sylvo-pastoral zones and
traditional oasis (Eden P.2011; K.A.I.G. 20011; Sellami October 2010; Sellami 2010, 2008;
Aidoud A. 2006; Ouelmouhoub S. 2005; Le Houerou 2005/2006, 2002, 2001/2002,
1995/1996, 1975; Ionesco T. 1980) . In fact, in those type of canopies we can found at least
three production layers, with many plant sorts in each layer and very important densities
adaptable to different climatic conditions (arid and hot climate in the south for the oasis,
humid and cold one in the North for forest and sylvopastoral zones)( Farooqi 2010, 2008;
W.T.M.A. 2010; Butler R.A. 2010, 2006; Goosem S. and Tucker N. 1995). For the
traditional oasis there are a storey for medicinal and ornamentals species, market gardening,
cereals and forages species (different sorts: Geranium, Vineyards, Chillis, Proteas, Poppies,
Lupins, Phoenix Roebelini, Butia Capitata, Adenium variety, Green Anthuriums, Roses,
Gerbera jamesonii, Portlandia Grandiflora, Bougainvillea, Salvia, ,Cestrum nocturnum,
Lannea grandis, Heliotropium indicum, Citrus grandis, Syzygium cumini, fodder crops, cereal
growers, corn flour, wheat, flour, oats, barley, rice, sunflower, sorghum, maize, lucerne,
alfalfa, tomato, pepper, potato, bean, bersly, salad, cucumber, onion… ), a storey for fruit
trees(many kinds for all seasons of the year: olive, almond, fig, apricot, pomegranate, orange,
vine, peaches, apples, grapes, citrus…), and that of palm dates(divers types, about 100
varieties for Algeria and 50 in Tunisia)(Sellami 2010, 2008; ). For forests and sylvopastorals
stands we found at least three layers: a short one (herb, graze, food plants, medical plants:
Lady Fern, The Tawny Milkcap Mushroom, Saprophytes, Orchid, Passion flower, Ficus
hispida, Streblus asper, Curcuma longa, Typhonium giganteum, Terminalia belerica,
Parthenocissus quinquefolia, Terminalia chebula, ivy, soybean, …), a middle one (shrubs of
different sorts: Guelder Rose, Orchids, Cacti, Aroids, Lichens, White Birch, White Oak,
Strangler fig tree, ponderosa pine, American elm, California laurel, various oaks …) and a tall
one (hardwood trees and deciduous trees of many kinds: Pecan, Conifers trees, American
Beech, Linden tree, Shagbark Hickory, Cheyenne, Sioux, Pine maritime, Wichita, Pawnee,
Buttress Roots, Lianas) (Eden P. 2011; CRPF 2011,2007; Guérin et al 2010; Hadjadi Aoul et
al S.. 2009; Sghaier . and Garchi . 2009 ; Benabadji N., et al. 2009 ; Hammi,S. 2007 ; Cheikh
AL Bassatneh 2006 ; El Euche F. 2006; Nsibi et al 2006 ; Le Houerou 2005/2006;
340 Mohamed Habib Sellami
Ouelmouhoub . 2005 ; Le Houerou . 2002 ; E.P.A. 2000 ; Carrière 1995; Joffre et al. 1991 ;
Le Houerou 1975).
So a general model established for those kind of canopies can be applied easily for other
simple agriculture fields planted in different climatic regime and if validated experimentally
in one of them we can say that the results can be generalised to all of them. In this tendency a
case study to validate this model in the traditional oasis of Tozeur by an experimental
protocol will be detailed.
In spite the fact that they are fragile and delicates, the wet land ecosystems, like forests,
sylvopastoral lands, traditional oasis are considered by experts and environmentalists as wet
lands of great interest for the environment harmony and balance because they avoid the
erosion of soil, humidify the atmosphere and clean it, eliminate the greenhouse effect by
storing the CO2, they assume the nutrition of animals by offering durable pastoral zones,
shelter migrant birds, excite wind formation and precipitation descending, protect the human
infrastructure from landslide and inundation (habitation, road, barrage, bridge, cultivation
fields…), conserve the human alimentary security from either animals product(milk and meat
), nutritional plants(agriculture fields, fruit trees), medical plants, endemic species, wood and
biomass, enhance human wellbeing and ameliorate his socio-economic situation.
Those wetlands occupy large surface and rally the interest of an important community,
either the direct profiteers (verener population, farmers, investors, consumers), scientific
researchers, and decision makers at International, Mediterranean and North African standards
(Sellami 2010, 2008; Butler R.A. 2010, 2006). In fact vegetation canopies characterised by
high densities formed from plants of different heights and species are often called rainforests
and considered as the glue that holds the climate of our planet together (Eden P. 2011; Butler
R.A. 2010, 2006). They control our climate by absorbing CO2, acting as a store for carbon,
and by making huge white clouds. White reflects heat, keeping the Earth cool. They are called
rainforests because they make rain. In fact, it is estimated that 12% to 20% of all current
carbon emissions come from deforestation and so, as said by many experts, if we can help the
forests to survive, they will help ensure our future survival (Eden P. 2011; W.T.M.A. 2010;
Butler R.A. 2010, 2006; Goosem S. and Tucker N. 1995). Knowing that the Amazon forest
releases 20 billion tonnes of moisture every day, helping water crops thousands of miles away
(Eden P. 2011; W.T.M.A. 2010; Butler R.A. 2010, 2006). The tropical forests store more than
half of all carbon found in terrestrial vegetation worldwide and contain at least two thirds of
the world‘s terrestrial biodiversity, making reducing emissions from deforestation and
degradation a critical component in the global fight against climate change (W.T.M.A. 2010;
Butler R.A. 2010, 2006). Approximately 60 million indigenous people rely on rainforests for
their way of life(Eden P. 2011). That the correspondent biodiversity has great medicinal and
economic value (Farooqi 2010, 2008). According to the US National Cancer Institute, more
than 70% of plants with anti-cancer properties are found here. Agricultural scientists use wild
strains of rainforest crops to increase yields and resistance to pests and diseases in cultivated
varieties (Eden P. 2011; Farooqi 2010, 2008; W.T.M.A. 2010; Butler R.A. 2010, 2006;
Goosem S. and Tucker N. 1995).
Basic Concepts for Modelling in Different and Complementary Ecological Fields 341
3.3. Modelling Solar Radiative Transfer and Heat and Mass Exchange for
Wetlands Canopies
Model the energetic transfer inside wetlands like forest, or sylvopastoral zone or
traditional oasis means determining the amount of solar radiation intercepted and diffused by
every species inside and calculating the flux of mass and heat exchanged between layers. This
estimation permits later to deduce the water really needed by every plants, to calculate the
biomass production in each crop storey and to monitor the eventual plant diseases before their
spreads. In fact the net radiation, latent and sensible heat flux are directly linked to the evapo-
transpiration and after the amount of water really needed by plant to respond to the climatic
and biological demands, the photosynthetic active radiation absorbed quantify the biomass
product, the optical properties of every specie permit de detect the stress situation and
diseases if they change comparing to their normal values.
rs(f)rs,0 exp,hs f
rd (f)rd,0 exp ',h f
rs f : Direct solar radiation at depth f inside the vegetation
rd f : Diffuse solar radiation at depth f inside the vegetation
f : Leaf Area Index held from the top of the canopy
342 Mohamed Habib Sellami
' (α, h) is the mean extinction coefficient for the diffuse flux density.
,hs extinction coefficient for the direct solar radiation
h: elevation of the solid angle containing the vault of heaven
hs: Solar elevation
α : co-latitude for the distribution of plant orientation
rs, is the direct solar radiation received above the canopy
rd,0 is the diffuse solar radiation received on a horizontal plane above the vegetation
Next penetrating the vegetation, the incident beam will suffered a multiple rediffusion
under the effect of leaves, trunks and branches of every specie. The beam will be scattered in
upward and downward directions. In order to simulate the scattered radiations, in a tall
canopies like those we are studying in this chapter (forests, sylvopastoral zone and traditional
oasis), we propose dividing up all the production stories into thin layers and to study the
radiative balance for every layer inside the vegetation. Mathematically, the vegetation layer is
bounded by the level just above indicated by (f) and the level just below appointed by f + df.
Inside every layer there is a multitude of species that participate individually in the exchange
acts.
Every specie inside a layer is identified by its physiological parameters and optical
properties which depend on the internal molecular components. The canopy architecture
characteristics (species densities and orientations) are considered in the extinction coefficients
and in the zenith-azimuth distribution. The biomass development, crop yields and fruit
production intervene in the definition of the leaf area index which in the case of tall canopies
where we can found trees, saplings and shrubs introduce the leaves, branches, trunks and
fruits. We must make morpho-physiologic measurements for each species to establish
mathematical relationships calculating the intervention rate of every component( Niklas and
Enquist 2002; Varlet Grancher C et al 1993).
When passing through a vegetation layer of thickness df, the downward and upward flux
undergo a diminution that expresses the parts of radiation trapped at the designated level
inside the canopy. The radiation balance of the re-scattered flux at level f can be expressed by
a 2nd order linear differential equations for which a general analytical solution is (Sellami and
Sifaoui 1999):
r f X1 exp f X2 exp f Y1 rs,0 exp f Y2 rd ,0 exp ' f
r f X3 exp f X4 exp f Y3 rs,0 exp f Y4 rd ,0 exp f
'
r ,int df rs f rd f r f df r f rs f df rd f df r f r f df
The total sum of every solar radiation component (direct radiation, diffuse, descendant
and ascendant flux and that intercepted) for each production storey or for all the canopy is
done as integration over the leaf aria index differentiation (df) which represents the thickness
of a vegetation layer. The number of partition layer to adopt depends on the vertical
heterogeneity of the canopy, the precision degree asked and the input parameters we dispose.
We can write this general form:
r,com Gcom(f)df
r,com : The radiation flux component‘s (direct, diffuse, upward, downward, intercepted)
Gi,com(f) takes the expression of the designed radiation component given above.
3.3.2. Formulating the Mass and Heat Flux Transfer Inside Multi-Storeys Canopy
As said above, the total radiation intercepted, will be in part used by plants for
photosynthetic activity and in other part for evapotranspiation in order to respond to the
climatic demand and to regulate its internal temperature and metabolism. In term of energy,
the evapotranspiration in a vegetation layer is determined from the balance between the net
radiation trapped, the sensible heat flux released and the latent heat flux disengaged by all the
layers or by each plant in the interior. Those flux are determined by applying the equations of
energy conservation, mass balance and motion given above to the system studied. In our case
the system is a vegetation layer inside a wetland canopy (Sellami October 2010; Sellami
2010, 2008).
The canopy stand either for the forests, sylvopastoral zone or traditional oasis was
visualized as an electrical analogue with a succession of circuits for the sensible and latent
heat exchanges. For every layer inside it corresponds a circuit. The net radiation intercepted
by the specie (j) within each layer (i) of the canopy represents its current source (electricity
generator) . Sensible and latent heat flux supplied by the kind (j) in the layer ( i ) inside the
canopy are considered as intensity of the electric current. Temperature and air vapour
pressure are the tensions. Resistances were of three kinds:
To elaborate the energy balance equations, we have neglected the heat storage term. For
every layer, net radiation absorbed can be considered to be equal to the sum of sensible and
latent heat flux densities. These energy balance equations together with the relationships
between flux densities and relevant driving forces, using the analogy of electrical circuit
theory (Ohm‘s law), constitute a closed set of equations for all unknown variables.
We can formulate the latent and sensible heat flux exchanged inside a layer ( i ) with the
following equations (sellami et sifaoui 2008):
j
j
c veg ,i vjveg,i rjnet,i
j j
T j T j
cjveg,i cp l,i j a,i
rc,i
cjveg,i and vjveg ,i are respectively the sensible heat flux and the latent heat flux
corresponding to the vegetation specie j in the layer i
rj net,i flux of net radiation intercepted by the specie j in the layer i
ρcp is the volumetric heat capacity of air,
γ is the psychrometric constant,
Tl,ij is the temperature of the leaves corresponding to the plant sort j in the canopy layer i
Ta,ji is the temperature of the air corresponding to the plant sort j in the canopy layer i
eTl,ij and eTa,ji are water vapours pressures of the air for the specie j at the canopy
layer i for respectively Tl,ij and Ta,ji .
rjc,i : Resistance to sensible heat flux displacement for the specie j in the layer i
rjv,i.: Resistance to latent heat flux displacement for the specie j in the layer i
If we consider m species in every canopy layer and n layers in the canopy we obtain the
following system:
Basic Concepts for Modelling in Different and Complementary Ecological Fields 345
n m cp(Ta,i 1Ta,i)
j
c veg ,i
i 1 j 1 rtt ver,i
To found an explicit analytical solutions to the precedent set of equations we will make
the following variables change:
Φjent,i=Φjc-veg,i+Φjv-veg,i
Φjsat,i=Φjc-veg,i- Φjv-veg,i
Φjent,i : Enthalpy heat flux density for the specie (j) in layer ( i ),
Φjsat,i : Saturation heat flux density for the specie (j) in layer ( i )
This reasoning leads to the following solutions:
Φjent,i=Φjr-net,i
c p
( )Di
rjnet,i
sat
j
,i
)rs,ji 1( ) rc,ji
j
rc,ji (
rs,i
Because net radiation intercepted by every specie within each layer is a known variable
simulated by the first part of the model, Φjent,i and Φjsat,i can be obtained immediately. By
considering (m) species and (n) layers inside the wet land, the new cumulative flux densities
for all the stories of the canopy are denoted as follows:
n m
sat sat,i
cum j
i 1 j 1
n m
ent ent,i
cum j
i 1 j 1
346 Mohamed Habib Sellami
Stomata conductance greatly affects the energy partitioning into latent and sensible
forms, and is very sensitive to the radiation received at leaf surfaces. The openness of the
stomata controls both the inward flow of carbon dioxide used in photosynthesis and the
outward flow of water vapour (Sinoquet et al. 2001). Thus, modelling sensible heat flux and
latent heat flux provide us the basis for quantifying biomass production (Sellami 2010, 2008;
Sellami et Sifaoui 2008; Cowling and Field 2003; Kull 2002; Eastman et al ., 2001; Farquhar
2001).
3.3.3. Formulating the Biomass Production: Forage Unity, Crops, Fruit Quantities
For plants, many researchers consider that the relationships among metabolic production,
growth, and biomass partitioning are pivotal to ecological theory from the level of
understanding the individual organism to that of predicting complex macro-ecological
dynamics (Austin 2007; Schizas and Stamou. 2007 ; Varlet Grancher et al 1993). Also they
suggest that plant organs growth in any season is dependent on metabolic productivity in the
previous growth season, which is dependent on the standing biomass and growth conditions
of the previous season. This is due to the fact that plant organs photosynthetic activity is
controlled by both environmental and physiological factors and involves stomatal and
nonstomatal mechanism. Following partitioning roles, the available carbon pool is allocated
into fruits, leaves, stems, roots and reproductive organs and it is generally characterised by
the quantum yield efficiency which represents the ratio of the moles of CO2 reduced to the
moles of photons captured. Hence a quantum yield of about 0.066 corresponds to a fixation of
0.0135 mg of CO2 per Joule of photosynthetic active radiation absorbed (Daudet and
Tchamitchian 1993). In fact, the capture of solar energy as fixed carbon in biomass via
photosynthesis, during which carbon dioxide (CO2) is converted to organic compounds, is the
key initial step in the biomass production (Ewert. 2004; Delucia et al 2002; Joseph et al
2001;).
So to convert from photosynthetic to structural material (Rosati A. et al 2004; 37, 38, 40,
41], we must estimate the photosynthetic rate for every part of a plant. The gross daily
photosynthetic production of the specie j in a layer i inside the whole canopy is given by
(Sellami et Sifaoui 2008; De-Xing and Coughenour 1994):
pj ,i Cv psl
24
j
,i f sl,i psh,i f sh,i
j j j
1
Cv : a conversion constant
psl,
j
i : average leaf photosynthesis rates for sunlit leaves of specie j in layer i
psh,
j
i : average leaf photosynthesis rates for shaded leaves of specie j in layer i
In order to calculate biomass product, many authors used quantum lighting for
photosynthetic active radiation (Abdul Awal M. et al 2005, Bréda. 2003; Varlet Grancher et
al 1993) defined as energy rate or photon flux density and expressed in µmol.m-2 s-1 . It
represents the number of photons reaching a plant unit area during 1 s. They suggest that 1
Watt of global solar radiation is equal to 2.02 µmol.m-2 s-1 and that the ratio of the
photosynthetic active radiation to global solar radiation is about 0.48 (Bonhomme 1993).
After by considering the mass of each molecular composition of every specie or every part in
the specie(fruit, leave, stem, trunk, roots…) inside the wetland and by calculating the
correspondent radiative climate we can deduce the detailed biomass product either for human
use or for animal feeding at different scales of time and space. For our study here we will give
later an estimation of the hourly evolution of the global solar radiation flux and the
photosynthetic active radiation inside a traditional oasis measured and calculated by the
model. The results can be used to evaluate the biomass product for any wetland canopy
because of the commons characteristics between them as detailed above and to plan later the
exploiting actions( grazing, deforestation, cultivation).
For the wetlands like forests, sylvopastoral stands and traditional oasis, the vegetation is
the keyword around it orbit the actions of soil conserving, climate balancing, infrastructures
protecting, animal sheltering and feeding, human alimentary security and welfare assuming,
socio-economic situation stabilising. The most important human activities menacing wetlands
durability are overgrazing and deforestation.
quantification of the aliments energetic values (Gerard et al 2005; CingMars 2003; INRA
1988):
After adopting a scientific sampling procedure, a field and experimental studies by specie
for all the vegetable canopy of the wetland in the objective to estimate their nutritious values
as forage unity, total digestible nutrient, congestion unity, nitrate unity, non nitrogenous
extractive digestible unity, ethereal extractive digestible unity, mineral unity, calcium unity,
phosphorus unity, brute cellulose unity, metabolic protein unity with the real need, from those
components, by animal category for all its physiologic phase are very encouraged. After we
can, using the precedent formulas, calculate the amount of every nutritional rate for each
biomass quantity and type produced inside the wetland, dividing by the real need from those
alimentary value for each animal to feed we can deduce the optimal number to intervene in
the grazing cycles. We can propose the following general relationship to estimate the optimal
number of the flock either for the ovine, caprine, bovine or cameline by considering their real
need from the above cited alimentary value existing in plants (Sellami October 2010):
i
k j
(Qtotj i Qcons
j
i )
cat
Nani j
Bcatk j
i
N cat
ani : Optimal number of animal category to enter in the grazing cycle
ik j : Nutritious unity of the component k restrained in the vegetal specie j of the layer i
inside the wetland canopy
Qtotj i : Total biomass quantity calculated by the model for the specie j of the layer i
inside the wetland
i : Biomass quantity for the specie j of the layer i to keep standing to conserve the
j
Qcons
environmental role‘s of wetland
Bcatk j : Real need of an animal category (by development physiologic phase) from
i
nutritional unity of a component k contained in the vegetal specie j of the layer i inside the
canopy
k : Indicator of the nutritional component contained in a vegetal specie (calcium,
phosphorus, nitrate matter, cellulose brute, non nitrate extractive, forage unity …)
Basic Concepts for Modelling in Different and Complementary Ecological Fields 349
For the ovine specie, the requirement from calcium and phosphorus for the physiologic
growth phase can be expressed as function of the alive weight as follow (Meschy 2002;
Kessler 1991):
grCa 9.5Wgr
grP 5.5Wgr
Wgr : Alive weight for the physiologic growth phase of the specie in kg
: Calcium requirement for the growth phase in g per kg of weight gain
: Phosphorus requirement for the growth phase in g per kg of weight gain
350 Mohamed Habib Sellami
For the ovine and caprine and for their growth phases, the relationships between their
alive weight and their needs from magnesium (Mg), sodium(Na) and potassium (K) as
nutritional elements are (Meschy 2002; Kessler 1991):
grMg 0.4Wgr
grNa 1.6Wgr
grK 2.4Wgr
Wgr : Alive weight for the physiologic growth phase (kg)
grMg : Magnesium requirement for the growth phase in g per kg of weight gain
grNa : Sodium requirement for the growth phase in g per kg of weight gain
grK : Potassium requirement for the growth phase in g per kg of weight gain
For the physiologic upkeep phase of the ovine and caprine species and for their
nutritional requirement from the magnesium, sodium and potassium we can adopt the
following relations ships (Meschy 2002; Kessler 1991):
The modelling approach and the set of equations presented above permit the estimation
of the biomass evolution and the conceptualisation of a varied and dense vegetal architecture
for the wetland in order to have a productiveness biodiversity over all the years. So we can
assume a durable and normal grazing rhythm (nomads, conventional transhumance, economic
transhumance conjectural...) and a complimentary between its pasture lands.
primordial roles of the vegetation in buffering the climatic change effect( Sellami October
2010; Sellami 2010, 2008; Sellami and Trabelsi 2009; Jeffrey et al 2005; Zhang and Sellers
1996).
We can determine the number of consumers to satisfy with the wetlands under products
without exhausting them by this mathematical relationship (Sellami October 2010; Zhang H.
and Sellers A.H. 1996):
,i Qbiom,i Qbiom,i
tot, j anim, j prot, j
,i
Qbiom
N pop ren
j
per, j
Bbiom ,i
Npop : Population number to satisfy for all their needs from the wetlands biomass products
Qbiom,itot,j : Total quantity of biomass produced from specie j in layer i
Qbiom,ianim,j : Biomass quantity produced needed for animal feeding from specie j in layer i
Qbiom,iprot,j : Minimal biomass quantity from specie j in layer i to let stand
Bbiom,iper,j : Real biomass need by person from specie j in layer i for his divers actions
τren,ij : Rate of plant renewing inside the wetland canopy for the specie j in layer i.
The total quantity of biomass produced from specie j in layer i could be calculated by the
first part of the model (radiative exchange and heat and mass transfer inside wetland). The
biomass need by person can be determined from socio-economic studies, statistical studies,
the censuses studies and analysis of the management plans for durable development. The rate
of plants renewing and the minimal quantity of biomass to keep stand for protecting resources
from the water streaming, wind effect and assuming a minimal climatic demand inside
wetland are determined from soil characterisation and topography, plant density and
architecture quantification and the flowing energy estimation (Sellami 2010, 2008; Sellami
October 2010; Sellami and Trabelsi 2009; Sellers and Mintz 1996 ; Jadhav., and Buchberger.
1995; ).
In order to validate this model for a very complicate canopies like forest, sylvopastoral
zones and traditional oasis, we must first detect the repartition of the different micro-
heterogeneities for sorts of soil (texture and structure) and micro-topography because the
correspondent flux emitted it depends, nature of species (optical properties, physiological
characteristics), architectural consideration (density, plant orientation, zenith and azimuth
angles, gap frequencies), micro-climatic parameters and their horizontal and vertical profiles
(temperature, humidity, wind speed, solar radiation) (Sellami 2010, 2008; Jeffrey et al, 2005;
De-Xing C. and Coughenour 1994; Varlet Grancher et al 1993; Gash et al 1989). After we
pass to make a micro-zoning and networking and for every knit or zone detected we must
make the measurement of all parameters characterising the purl which will be used as either
inputs for the model or to compare its outputs. So we must install the necessary apparatus for
micro-climatic measurement, taking patterns of soils and plants for conductive and optical
properties measurements, sampling representative trees and plants to be equipped by
physiological probes (sap flow meter, temperature, xylem diameters…) and for which we will
352 Mohamed Habib Sellami
estimate other morphological measurements like the leaf area index, zimuth and azimuth
angle distribution, plant heights, number of leaves, proportion of trunks, branches, fruits,
forage unities…. Those measurement must be done for every knit identified in order to
consider the horizontal heterogeneity inside the field . For the vertical heterogeneity, we must
do the same measurements for every vegetation layers. So we have to install many masts
equal to the numbers of mails signalled and repartitioned vertically in levels equalizing the
number of layers for the canopy ( Sellami 2010, 2008; Tournebize and Sinoquet 1995; Adama
1990). In each level of the mast we must install the necessaries probes and apparatus to
measure all micro-climatic and physiologic parameters (diffuse radiation, direct radiation, net
radiation, photosynthetic active radiation, temperature, vapour pressure, humidity, wind
speed, optical properties, sap flow, angles, surfaces, gaps,…). This is in the case when we
choice a fix installation. In other case we can install a limited number of masts in a mobile
system which can sweep a representative surface inside the wetland so we can obtain an
average representation of all parameters.
As said above there are many common and similar characteristics between forest,
sylvopastoral zones and traditional oasis. In fact we found in the same field a multitude of
species, with different highs and important density, stratified in many production layers, and
conceptualised in natural, hazardous and stochastic architecture (Sellami October 2010;
Sellami, 2010, 2008) . So if the model is validated for one canopy there is a great probability
that it can be applied to the others and there is an important chance, that the empirical
relationships established for one and the variables estimated for it, could be extrapolated for
the others. In this context a first application of the model and experimental validation have
been done for a traditional oasis situated in Tunisia which is a Mediterranean country and the
Mediterranean shares its climate with at least 4 other regions around the world, South
Africa, South West Australia, Central Chile, and California (Sellami 2010, 2008). So the
high degree of results applicability‘s. The experimental protocol consisted at installing a mast
inside a representative parcel. To consider the three production stories inside the traditional
oasis (date palms, fruit trees and market gardening) the mast was partioned in three levels. In
each level we have installed the needed probes for microclimatic and physiologic
measurements (Sellami et Sifaoui 2003, 1998). The relevant results are in the following
paragraphs.
3.6.1.Model Validation for Solar Radiative Transfer inside a Traditional Oasis Canopy
For validating the radiative part of the model we have compared the time course of solar
radiation calculated to that measured inside the oasis for the date palm storey, fruit trees
storey and market gardening storey. Results have been detailed in previous papers (Sellami
2010, 2008; Sellami et Sifaoui 1999, 1998). We will remind the most important.
We have noticed that values simulated are close to those measured with a weak
overestimation at noon and a light underestimation at the beginning and at the end of the day.
The discard between measurement and calculation is more important for the radiation
received by the market gardening, sustained by that received by the fruit trees and finally the
Basic Concepts for Modelling in Different and Complementary Ecological Fields 353
date palms. The model shows a good simulation of the distribution and the sharing of solar
radiation between the stories. In fact, we have noticed a clear discard between the radiation
received by the date palms, fruit trees and the market gardening: between 11 and 14 h, the
overation is about 35w/m² for the radiation transmitted across the palmgrove, 60 w/m² for that
transmitted through the fruit trees and 85 w/m² for that reaching the ground. At the beginning
and at the end of the day, the underrate doesn‘t pass through 30 w/m². We deduced that the
same walk is visible at the different levels and that the market gardening intercept 35% of the
incident global radiation, the fruit trees intercept 23% while the date palms intercept only
18%. The average of daily global radiation intercepted by each canopy story was about 640
w/m² for date palm storey, 853w/m² for fruit trees storey, 1380w/m² for the market gardening.
The gap between measurement and calculus does not overrun through 15% we may conclude
that the agreement between estimated and observed values is good.
date palm storey represents 42 % from the net radiation above the oasis, for the fruit trees
storey it represents 34 % while the market gardening is only about 7%.
3.6.3. Results from Simulation of Plant Diseases Monitoring and Biomass Product
To validate the biomass product part of the model we have compared the photosynthetic
active radiation (µmol m-2 s-1), simulated by the model to that measured inside the traditional
oasis (Sellami et Sifaoui 2008; 2003, 1998). We have noted an agreement between the values.
A little difference between modeling and measurement is advisable more for the market
gardening about 171.7 µmol m-2 s-1 which represents a rate of about 5.5 %, sustained by the
fruit tees about 121.2 µmol m-2 s-1 so 3.1% and the date palm 70.7 µmol m-2 s-1 equivalent to
1.6%. The total calculated efficiency for the photosynthetic active radiation intercepted inside
the oasis canopy is about 65 % while that measured is 67.16 %. For the average
photosynthetic active radiation intercepted by the date palm layer we have about 621.51µmol
m-2 s-1 which is equivalent to 14.11 % from that recorded above the oasis, for the fruit trees
layer we have 824.52 µmol m-2 s-1 representing a rate of 18.73 % from the photosynthetic
radiation over all the canopy, finally the market gardening can intercept 1338.048 µmol m-2 s-
1
equivalent to a rate of 30.4%. Those values were given with a global error of about 3.5 %.
To explain the repartition rate of photosynthetic active radiation between the three stories
we can say that the rate of PAR intercepted depends on the associated structure. For that,
particular disposition of species involves a heterogeneity of the transmitted radiation. The
intercropping system ensures the best occupation of space permitted to the dominant space, to
intercept increasing radiation (Sellami 2010, 2008). This is owed to the spacing between the
trees, which is especially important for the date palm, while the market gardening occupied
all the ground. For that, the lower storeys receive either incident radiation that arrived directly
through gaps in the foliage, those scattered by the upper storeys (leaves, trunks and branches),
or complementary radiation reflected by the soil and then re-scattered.
Human in his modern life and for all his activities needs energy. Generally those
activities belong to one of the following sector: industrial, agriculture, services and domestic,
which are mostly accomplished inside buildings and they necessitate a suitable level of
comfort to be accomplished in success. The use of that energy in an excessive manner without
optimisation leads to the exhaust of natural resources (biomass, fuel) and an emission of a
great quantity of greenhouse gas from buildings. Which can contribute in the acceleration of
the climatic changes and in the worsening of their effects to allocate later, directly or
indirectly, the plants survive, wind and cloud creation and all the atmospheric water cycle.
In fact there is an international focus and anxiety about the important part with which the
building sector contributes actually in the aggravation of the world environmental situation.
Knowing that in European Union , the energy consumed in the building sector represents 40%
from the total consumption and its behind 36% of the total CO2 emission. While in the USA
the building sector consumes 49% from the total energy produced, uses 77% from the total
electricity supplied and disengages 46.9% from the total CO2 emission (U.S. Energy
Information Administration 2010). The results of a foresee study about the evolution of the
energy consumed by many sectors up to 2030 show clearly that the building sector is and will
356 Mohamed Habib Sellami
rest the first consumer (U.S. Energy Information Administration 2010). That‘s why there is an
urgent international need to ameliorate the thermal building efficiency so the importance, the
complementary sense, and the link between the three compartments studied in this chapter.
Which equations we propose to conceptualise the positioning and the orientation of
buildings in order to profit from external climatic conditions so we minimise the need to air-
conditioning either in warm or cold period? How to model the internal architecture in order
to make the optimal partitions in the purpose to profit from the transfer of energy between
them, to choose the material of construction for the beneficial insulation, to organise the
emplacement of tools, apparatus, machines, movables, desks, computers…, to make the
calendar of their displacement and the timing of their functioning in order to optimise the
consumption of energy? Which practical method we can develop that it promises to define a
program of certification for energy saving more realistic responding to the international
inquiry concerns about the emission of gases with greenhouse effect, the approach of
petroleum peak, and the increase of the price/cost of combustibles having a fossil origin‘s?.
That way we will present a model allowing the analyse of building thermal efficiency for all
usages (domestic, industrial and services). The model utilises the basic equation of radiation
transfer, energetic transfer, mass transfer and motion. It is backed on the thermal ohm law and
it considers the contribution of either the external factors namely climatological /
meteorological parameters and surrounding environment conditions and the internal factors
knowing walls and roofs, isolation system, accessories, machines and human activities.
Like inside every system, thermal transfer inner buildings is governed by radiation
exchange, convection exchange and conduction exchange. Those later are ruled by the
general equations of energy transfer, mass transfer and motions presented above. The
mathematical representation of these equations depends on the scale of resolution to follow:
mean values, fluctuant values, macro, micro, molecular, atomic… and permit us to estimate
the exchanges between the building and its surrounding by considering the local meteor-
climatic conditions added to the exchanges between the internal components by considering
human activities at the interior, tools, movables, machines, and the molecular composition of
the construction and isolation materials.
The coefficient a and b were respectively 0.52 and 0.065 with a correlation coefficient of
about 0.97.
The radiation stemmed from clouds in the overcast sky is expressed as follow:
ter Ts4
σ: Coefficient de Stéphan-Boltzman
The net radiation arriving at the building is the balance between the flux of radiation
exchanged. We can write:
rnet rg1alb t atm Ts
4
alb: Albedo of ground surface
: Surface emissivity
tatm : Total incoming atmospheric flux
During daylights, the net radiation is positive and the surface receiving radiation lost
energy. After desk, during night, net radiation is affected by a negative sign and the surface
gains energy.
The water vapor pressure can be deduced from the following formula:
ea(T)610.8exp( 17.27T )
T 237.3
T: Temperature
358 Mohamed Habib Sellami
The diurnal course of solar radiation for different time steps (second, minute, hours…)
depends on the sun elevation which is expressed as function of the altitude, latitude, and
declination. It gives an indication about the hours in the day, the season, the years…The
elevation of the source can be expressed by the following function:
s hbul cp Ks h T
z
s hbul : Sensible heat flux for the building
T : Temperature
z : altitude
ρ : Volume mass
Cp : Specific heat coefficient
Ks-h : Exchange coefficient for the sensible heat
q : Humidity or vapour air pressure
For the latent heat flux:
q
L hbul LK L h
z
Lhbul : Latent heat flux for the building
KL-h: Exchange coefficient for the latent heat
For the heat flux by conduction
Basic Concepts for Modelling in Different and Complementary Ecological Fields 359
chbul Kch T
x
c h bul : Conduction heat flux for the building
Kc-h: Exchange coefficient for the conduction heat
T: Temperature
x: Distance from the source
When considering the total exchanges for the building, the balance is written (Jannot
2009, Pons 2008, Wright 2007a, Strub 2005) :
dQi
( (
j i
rnet,i c h bul,i Lhbul,i s hbul,i ph,i)) (
j i dt
)
rnet,i : Flux of net radiation exchanged for the component ( i ) inside the layer ( j )
ph ,i: Heat flux disengaged by the biologic system (i) depending on the activity inside
the building
Qi : Calorific energy
i: Indicator of the component (i) for an identified zone in the building(depending on the
scale of resolution)
j: Indicator of the zone in the building (insulation layer on the wall, wall, roof, part of a
room, room, partition, apartment, floor, all the building…
Resolving the precedent equations permits to us, when considering the local boundary
conditions, to determine the energy stocked or warehoused by every element of the building
and by all the components existing inside for use. It can be manifested as the rate of local
volumetric absorbed energy for each component:
2 2
s
Ei x,t K r,t d d sin ,,t *exp K s ,t ds
1 1 s sR
Ei : Local volumetric absorbed energy by an element (i) inside the building
360 Mohamed Habib Sellami
Depending on the scale of reasoning, the rate of volumetric absorbed energy expresses
the energy absorbed at molecular level of every building element or/and every façade, and/or
every component inside and/or all the building (Rivière 2009; Abou Khalil 2008;
Haralambopoulos and Paparsenos 1998). The volumetric absorption coefficient Kν is a linear
function of the identification parameter for the absorbent element for every scale. We write:
Kν,i = αν,i Ci
αν,i Molar absorptivity of the element i inside the building
Ci Identification parameter for absorbant element (density, concentration, volume,
dimension, size…)
For all the building components (i= 1, 2, ….,m), the total absorption coefficient can be
expressed by:
i m i m
K
i 1
K,i
i 1
, i
Ci
4.3. Formulation of Exchanged Flux Inside and between Storeys for All the
Kind of Building’s Usage: Multi-Layers Model
2009, 2008; Franco 2007; Matteis 2007; Jannaot. and Acem . 2007; Zho et al 2007; Jannot et
al 2006; Haralambopoulos and Paparsenos 1998; Hladik 1990). The model elaborated is said
multi-layers and is based on a thermal analogy, equivalent to the ohm law in electricity,
corresponding to the process of heat propagation from the ground to the roof of the most high
storey in the building. The air temperature and air vapour pressures represent the tension in
the ohm law; latent heat flow and sensible heat flow schematise the intensity. Herewith a
schematic figure for the model functioning:
Figure 1. A functioning scheme of the multi-layers model: A layer is notified by (i) which varies from
0(soil) to n (highest floor), Φs-h-bull,,j, ΦL-h-bul,j, Φrnet,j are respectively the sensible heat flux,
latent heat flux and the net radiation for the layer (j), rt-ver-bul,j, rs-h-j, et rL-h-j are respectively the
resistance to the vertical turbulent transfer, sensible heat flux transfer and latent heat flux transfer for
the layer (i).
The purpose of this model is to express the global heat flux under a combined equation
describing the heat exchange through all the building height with conserving to the usage type
(domestic, service, industrial) a physics signification. The flux of energy are homolog to the
intensity, temperature and humidity are analogue to the tension.
362 Mohamed Habib Sellami
For an ( n ) layers building, the flux density (intensity of courant) are composed of the
given up flux by the accessories inside the layers and between them. Between the soil, layer
number zero, and the high ranking layer number n, the sums of sensible and latent heat flux
are written:
n m
cp(Ta, j 1Ta, j)
( j shbul,i)
j 0 i 1 rt ver, j
j 0 i 1 rt ver, j
j shbul,i : Sensible heat flux for the component ( i ) inside the layer ( j ) in the building
j Lhbul,i : Latent heat flux for the component ( i ) inside the layer ( j ) in the building
ea,j-1and ea,j : Air vapour water pressure in respectively the layer j - 1 and the layer j
rt-ver,j: Resistance to the vertical transfer of heat flux in the layer ( j )
Ta,i-1 , Ta,i : Air Temperature for respectively the layer j – 1 and the layer j
γ : Psychrometric constant
We have also presented the basic equations and the running principle of the multi-layers
model for estimating the different flux exchanged inside and with the environment for an (n)
stories building by considering the components that exist in the interior and which depend on
the activity realised in the building. The equations and the modelling approach presented can
be applied for different scales by starting from a macro resolution (shape, dimensions, global
flux) tell a micro level(molecular and atomic composition, electric field, magnetic field)
(Sellami December 2010; Fiabitat 2009; Rivière et al 2009; Aoun 2008; Abou Khalil B.
2008).
To assume the well-being of the building users for all sectors (domestic, industrial and
service) we must fix the optimal levels of heat we need to accomplish our activities. When
those levels are known and after the calculus of heat flux coming from the exterior and those
exchanged inside by the precedent formulas we must pass to the conception of the isolation
system permitting to reassure the difference. The choose of isolation material depends on its
heat exchange capacity, thermal and electrical properties and optical characteristics. So the
questions that we must ask are: If the thermal isolation can protect us from the strong
difference of heat? If yes is it due to the important density of insolating material? What is the
de-phasing effect in thermal insulation?
By utilising the precedent equations we can delimitate the insulation system parameters
and estimate the importance of their contribution in the phenomenon of thermal transfer
inside buildings. The insulation technique utilising a thin reflecting material that must be
Basic Concepts for Modelling in Different and Complementary Ecological Fields 363
embedded in the building envelope joins the thermal insulation performance of the material
by air cavities or pores to the reduction of the radiative heat transfer by a barrier of low
thermal emissivity (Chami N. 2009; EMPA 2009; Fiabitat 2009; Haralambopoulos and
Paparsenos 1998). The characterisation of the insulation material and the determination of
their efficiencies necessitate the knowledge of the physics phenomenon induced in this
insulation. Modelling the transfer of heat by radiation, convection and conduction in the
walls, partitions and roofs having insulating products with thin reflecting layer for different
dimensions and inclination help us to calculate the rate of heat transmission under different
climatic conditions and their thermal performance by considering the degree off ventilation of
the air and the permeability of the roofing, the outer cover of soils and walls. Our model is
full of promises in this axe.
The results of many studies done in this sense were focalised around the following rates
(Chami 2009; EMPA 2009; Fiabitat 2009; Millet 2006 ; Th-CE 2005; C.E.R.T.U. 2003):
nocturnal sub-ventilation estimated impact 1 to 4.5°c, solar protecting estimated impact 1 to
3.5°C, internal charges(heat flux from equipments and human activities) impact estimated 1
to 3°C, building inertia (total mass) impact estimated from 1 to 2.5°C, nature of the isolating
estimated impact from 0 to 1°C, duration of heating period intervenes by 1 to 2°C, walls and
partitions covering interfere by 0 to 1.5°C.
30
ext cont ( sol ref neig )Sequi recNhday
j 1
The equivalent receiving surface for the building can be estimated by:
Sequi-rec = Sbay Fumbrage Fsolar
Sequi-rec : Equivalent receiving Surface
Sbaie : Surface of the bay
364 Mohamed Habib Sellami
Generally, on the base of the building usage, we have tendency to give a global value of
the energy emitted by all accessories, apparatus and human motion as energetic flux emitted
by m² of building surface or room surface. Said also Building energetic indice‘s. Yet we can
write the following relationship (Th-CE 2005):
For example, the calculus of heating installation inside buildings constructed for domestic
usage we admit a global value for the heat internal contribution by considering human
activities and apparatus closed to 5 watt per m² habitable. This value is proposed by the 2000
thermal regulating in the laws Th-C (Millet 2006 ; Th-CE 2005). So we can propose this
formula:
Basic Concepts for Modelling in Different and Complementary Ecological Fields 365
intcontr 5ShabNhday
intcontr : Heat flux from internal contribution during a day
Nhday : Number of functioning hours in a day (generally we make the calculus for 24
hours)
Shab: Habitable surface for every floor inside the building
The model presented above permits the estimation of the contribution of each component
inside the building by introducing the network of resistance to heat transfer.
Analytically many researchers in the energy field for buildings have agree about the
following values for the emitted flux by m² of occupied surface for different usages. Here
with a recapitulating table (Millet 2006 ; Th-CE 2005):
Table 1. Values for emitted flux and disengaged humidity by unity of surface for
different building usage
Those values were estimated without considering the energy needed for the sanitary
warm water circuit, the humidity produced in the bathrooms and kitchens. The question to ask
here if these values represented exactly the thermal exchange inside a building?, what is the
degree of their precision?, if they can be adapted our days for building thermal certification
and if we can make their bringing up to date?. The modelling approach proposed here can
contribute.
If we want to consider the heat flux needed for the sanitary warm water inside any
building, we can propose the following formula (Millet 2006 ; Th-CE 2005):
h sww wCc wVw wn(Tww Tcw)sww
h sww : Heat flux needed for the sanitary warm water
ρw: Water volumic mass
Cc-w: Water calorific capacity
Vw-w-n: Volume of warm water needed
Tww: Temperature of warm water
366 Mohamed Habib Sellami
4.5.3. Estimating the Energetic Contribution for the Recovery of the Lost Heat Flux
As said above every body or installation(apparatus, machine, frigorific system, boiler,
pipes for fluid transport and distribution, pipes for sanitary water, reservoirs, movables,
computers…) inside the building participates in the heat climate by emitting a flux either by
convection, conduction or radiation. If the flux disengaged by that body or installation is
inside a room that must be heated for a desired action so we can recover the flux and benefit
from it in the calculus of the room or floor warming need‘s. Also when the desired body or
installation is in a room or a partition under, above or next to a room or a floor that must be
warmed so the flux emitted is said a recoverable one and we can profit from it after
considering the inertia effect of walls, partitions and ceilings. In the reverse case when that
body, apparatus or installation emitting a heat flux is in a zone under, above or next to a room
that have to be cooled so the emitted flux is a parasite one and if not noted it will affect badly
our calculus. It is said a non recoverable flux. Yet the important questions to ask are how to
architect and arrange the apparatus or installation positioning inside the building and how to
organise relatively the usage and the different activities in every room?. Our model permits to
take those kinds of decisions. Empirically the formulas permitting to evaluate the recoverable
heat flux are deduced from those general presented above. If we focus our interest about the
system of pipes and cables for flux transport, adduction or distribution (flux of heat water,
flux of cold water, flux of energy or heat...) that we can found in any tools or installation(
sanitary heat water, frigorific machine and system, production machine, production line, air-
conditioning system…) we propose the following empirical formula (Millet 2006 ; Th-CE
2005):
In the case when there is a re-circulation system we must consider the energy emitted by
the circulator or the driving ,generally a motor-pomp, the most used empirical formula is
(Millet 2006 ; Th-CE 2005):
The system of storage either of heat, warm fluid or water, cold fluid or water…can emit
through the reservoir or balloons inner sides a non negligible amount of heat that we have to
consider. It can be estimated by the following empirical formula:
To consider the loss of heat by the recoverable and non recoverable effect, the
researchers introduce a coefficient reducing the heat loss‘s named emplacement coefficient.
This later contemplates if the system is installed in a position inside the building that must be
heated or not and it deems the passage from a heated zone to a cold one. There is many
abacus that fix the values of this coefficient for different situations (Millet 2006 ; Th-CE
2005). So the global recoverable heat flux between the transport and the storage systems can
be expressed by:
For the energy loss‘s through the sides of the system of boilers we can present the
following expression(Millet 2006 ; Th-CE 2005):
one and it depends on the global volume of the building and its different components(walls,
roofs, rooms, apartments, floors…), the thermal conductivity of the construction material
used, the isolation system installed and their molecular formations (Rivière et al 2009; Abou
Khalil B. 2008; Franco 2007; Jannot et al. 2006; Strub et al. 2005). In our modelling approach
we have considered the inertia effect in the volumetric absorption term which is presented as
a linear relation between the molecular absorption of the material and an identification
parameter for the receiving surface (density, concentration, volume, dimension, size…) and
also in the network of resistance to the flux transfer that we have introduced. So the scale of
resolution can vary between a macro one (all the building or all the absorbing surface) to a
molecular level. Actually for energetic audit studies, the specialists have classified the
building inertias into five classes: very light inertia, light inertia, average inertia, heavy
inertia, very heavy inertia (Millet 2006 ; Th-CE 2005; C.E.R.T.U. 2003). So with this
thermal inertia scale the time dephasing to diffuse the heat received by the different
component of a building increases gradually from the very light inertia to the very heavy one.
The experts have characterised the thermal inertia of building by a parameter called the daily
thermal capacity by unity of habitable surface. The following table gives the different values
(Millet 2006 ; Th-CE 2005; C.E.R.T.U. 2003):
4.5.5. Estimating the Season Duration for Heating or Cooling the Building
The heating or cooling season depends on the building usage and for the same utilisation
it is function of the action to undertake. For an industrial utility that period is subordinated by
the production line, type of product, kind of machines, management method, number of
technical and administrative staff… For a service usage of the building the season is fixed by
the service nature, working timetable, time of vacations, number of the staff, tasks
realised…For a domestic usage it is dominated by the habitant number by apartment, their
ages, their professions, if it is a primary residence, where they pass their
weekends…Generally we assume that the cooling or heating period becomes a necessity
when the internal and externals influx can not offer an ambient temperature of about Tambient
Basic Concepts for Modelling in Different and Complementary Ecological Fields 369
+ 1°C (Millet 2006 ; Th-CE 2005; C.E.R.T.U. 2003). The external temperature from which
the heating of the building becomes obligatory can be estimated by (Th-CE 2005):
The total flux contributing in the heating of the building is the sum of the external
contribution, the internal contribution and the recovered contribution. We write:
In the objective to estimate with great precision the cooling or heating periods and their
exact starting time for all activities inside a building, the specialists of the domain introduced
the notion of degree hour and degree day (Millet 2006 ; Th-CE 2005; C.E.R.T.U. 2003). So
they can consider the intermittent effect either for the work or for the use of the different
electrical consumption sources also they contemplate the fact that the temperature inside the
building differs from room to room, from a partition to the other, from a floor to that
successive… and inside the same space (zone in the room, room, partition, apartment,
floor…) it varies from position to the other because it depends on the nature of the apparatus,
machines, movable… and the management of their emplacement. The calculus must be done
for every usage period of the building (work timing, week-end, vacancies, night, seasons…)
and for every particular zone inside. If many rooms, because of the activity inside, have the
same temperature, their representative surfaces can be cumulated. The different hours degrees
calculated will be cumulated to estimate the days degrees for every particular zone inside the
building and finally the month degrees (Th-CE 2005).
The formula to use for the hour degree is:
The research of methods and options to limit the CO2 emission from buildings and
ameliorating their thermal performances could not be separated from using renewable energy
370 Mohamed Habib Sellami
in particular the wind energy for the combined production of heat, energy and electricity at
large and micro levels. We will next present the basic equations to found the potential wind
sites and to conceptualise propellers and wind parks.
Globally, the wind is defined as gases motions at the surface of any planet enclosed
inside the atmosphere. The motion of an air mass as a gas is initiated by warming the
terrestrial surface by solar energy and the earth rotation. All points and positions in the
terrestrial planet have the same rotation velocity but the degree of heating differs. It depends
on the soil nature and topography, surface of lakes, sea and oceans, the surface occupied by
vegetation, the sorts of species and the type of plant canopies (agriculture field, forests,
sylvopastoral, oasis…). That‘s why the speed of wind is variable in space and in time either
inshore or offshore (Beniston, 1998; Rakipova and Efimova 1974; Guterman. 1974; World
Meteorological Organisation 1974). In fact, the wind considered by many specialists as an
indirect form of the solar radiation, is the result of the move of an air mass as water vapour
under the effect of a pressure difference. The mass of air has tendency to displace vertically
from position of high pressure to an another characterised by a weak pressure value. The
effect of earth rotation (Coriolis force) will generate the displacement of that mass in different
directions. The direction of motion of the air mass represents the wind direction and the speed
of its displacement is exactly the wind velocity.
To reaffirm the link between the three compartment studied in this chapter we can say
that a vegetation stand well developed, formed of divers species, conserved and well kept
participates actively by transpiration in the atmospheric water cycle and in generating the
wind. An enormous amount of CO2 emitted from buildings accelerates the climatic changes
and enhance the environmental imbalance. So we assist to long drought, sudden violent
averse and calamity, inundation, vanishing of vegetation canopies, absence of clouds and
precipitation and as results the disappearance of normal and positive wind (Lamb 1974;
Landsberg 1974; Konstantinov and Sakaly 1974). Knowing the fact that the earth receives in
30 minutes as solar radiation the equivalent of the humanity energy consuming in a year. 1 to
2% of that energy in converted into wind energy. Whish is 50 to 100 time more than the solar
energy converted into biomass by photosynthesis (Wikipédi A, 2011).
The winds, behind all the meteor-climatic phenomenon, are classified according to their
spatial extent, speeds, geographical localisation, generating forces, their effects and impacts.
How to localise the zones with high wind potential? How to test the wind turbine efficiency ?
are questions asked frequently by either farms generating electricity from wind energy or
particulars who want to produce it for individual uses. Actually those wind customers to
begin their actions utilize global and general data supplied by Meteorological services
Basic Concepts for Modelling in Different and Complementary Ecological Fields 371
(meteorological station, satellite photos) established for large strides of space and time and
market information furnished by turbine sellers and manufacturers.
We will in this compartment give the basic equations that permit to predict wind velocity
locally at short step and to choose the wind turbine the most coherent for our activities.
Climatic change, CO2 emission, Kyoto protocol, End of fossil resources, Worldwide
environmental funds, Clean technologies funds, Clean energy, Mediterranean solar plan,
Tunisian solar plan, ELMED project, MENA renewable energy projects, Ecological village
project, plate-form to export the renewable energy, Project 20-20-20 goal for 2020,
DESERTEC Project…are new designations and projects at international level that focus and
encourage the electricity regeneration from renewable energy and to use it in urban or rural
zones for all activities (agricultural, industrial and domestic) with the possibility to sell the
amount beyond our need or to stock it for future use. Added to the fact that actually the petrol
outtake is done at the rate 2% from the worldwide fossil reserve per year, if this rate is
conserved the resources will be finished in 2060, if the outtake rate reaches 4% the date of
end will be 2035 but if it reaches 7% per year the resources will finish in 2020 (Abdallah
2008). That‘s why the concern at international level is how to produce 50 % to 80% of the
worldwide electricity need from renewable energy at the year 2050 (Wikipédi A, 2010).
Nevertheless, by considering the climatic and environmental circumstances (increase of sea
level, increase of the earth temperature, long drought, sudden averse, inundation,
calamities…) a great objective fixed by specialists to buffer and alleviate those effects
consisting to realise a climate change mitigation of a about 2°C at the horizon 2030. . This
could not be done without acting on three factors: Conserving the plants biodiversities
(forests, sylvopastoral canopies, oasis…) to play the role of atmospheric cleaner by
transpiration of water vapour and absorption of CO2 , minimising the release of greenhouse
gases in the atmosphere by ameliorating the thermal building efficiency for all activities
(agriculture, industrial and domestic) and using clean and sophisticate technology, increasing
the percentage of renewable energy usage in electricity producing for every country and in
particular the wind energy. In fact, at international level, the Aeolian channel represents the
third renewable energy source used in electricity producing at a rate of about 3.5% after the
hydraulic channel (89 % ) and the biomass (5.7%) (U.S. Department of Energy, 2010;
Wikipédi A, 2011). The worldwide installed wind power is about 157131 MW tell the end of
2009. It was only about 6115 MW in 1996, about 74117 MW in 2006 and about 121 266 MW
in 2008. Shared between the Europe (53.54%), USA (22.31%) and Asia (24.75%) (Charlotte
at al 2008). Also a propeller of 2 MW functioning ¼ year produces 4 to 5 millions kwh
which suffuse as domestic electricity without heating to 4000 persons. A wind turbine
installed offshore for a capacity of about 5MW produces 15 GWh per year, whish is sufficient
to supply 10000 standards electric cars (type Renault Megane 100% electric) for a distance of
10000 km/year(U.S. Department of Energy, 2010; Wikipédi A, 2011; Charlotte et al 2008;
Ryan and Mark 2008). Wind could meet 12% of global power demand by 2020, up to 22%
by 2030, and could play a key role in satisfying the world‘s increasing power demand, while
at the same time achieving major greenhouse gas emissions reductions according to a study
372 Mohamed Habib Sellami
published by the Global Wind Energy Council and Greenpeace International (Angelika et al..
2010) . The 1000 GW of wind power capacity projected to be installed by 2020 would save as
much as 1.5 billion tonnes of CO2 every year. These reductions would represent 50-75% of
the cumulative emissions reductions that industrialised countries committed to in their 2020
‗Copenhagen pledges‘. By 2030, a total of 34 billion tons of CO2; would be saved by 2300
GW of wind power capacity (G.W.EC 2010; Angelika et al.. 2010 ; Christian et al.. 2009).
The purchase cost of the electricity generated from wind energy was fixed to 8.38 centimes of
Euro for the kwh during the past 10 years (2000 to 2010). For the owing 5 years the buying
cost will vary between 2.8 and 8.2 centimes of Euro for the kwh (G.W.EC 2010; Christian et
al.. 2009).
An operation of that large-scale (climate change mitigation of a about 2°C at the horizon
2030) has an estimated cost of about 139-175 US$/year and it is in that wide-ranging that
belongs the European project 20-20-20 goal for 2020 (decrease of green house gases emission
at 20%, conserving 20% from energy use efficiencies, 20% from electricity generating must
be from renewable energy). So we signal an other time the importance and the link between
the three compartments studied here. In the purpose to diminish costs and attract investors
which represent the important requests for renewable energy in general and wind energy in
particular at both International and National levels, the primaries technical solutions are the
research of sites ―onshore‖ or ―of f shore‖ with an important wind potential permitting to
offer the amount needed and how to test the efficiency of wind turbine to use. In this work we
will present a model to foresee the zones with high wind intensity, to establish a wind atlas
and to analyse the wind turbines capacities.
5.3.1. Forces Affecting Air Motion, Wind Directions and Wind Intensities
Characterising wind sites means to determine the direction of air motions and the speed
of their displacement. Which are directly linked to the water vapour pressure differences
created in many points in the atmosphere under the warming effect of solar radiations in their
trajectories from the sun to the earth.. Then the air mass is guided by a resultant force for
which the dominant direction correspond to wind direction and the intensity signals the
importance of wind speed. At any point on the environment surrounding the earth an air mass
suffered the effects of the following forces:
Air pressure difference force‘s resulting from the solar warming and water vapor in
p
the air, because it is the force engendering the air displacement ( ; p: pressure;
ρ: air density)
Coriolois force for earth rotation that deviates the air in different directions
perpendicular to an observer on the ground and proportional to the speed of the air
motion ( fV ; f is the Coriolis factor proportional to earth rotation)
Frictional forces for soil roughness, morphology and geomorphology effect,
vegetation canopies, lakes and sea
Basic Concepts for Modelling in Different and Complementary Ecological Fields 373
V2
Centripetal forces for rotation of the air mass around earth ( ; r: radius of the
r
flow curvature)
Gravity force for the air weight and gravity acceleration ( g ; g: gravity
acceleration)
Motive force (m dV ; m: air mass and V : Wind speed)
dt
The mode of representation of those forces depends on the coordinate system to use
(polar, cylindrical, spherical, Cartesian). In general and by considering the pressure as vertical
coordinate and the Cartesian plane (x, y) as horizontals coordinates we can write the balance
between the Coriolis force, the friction force, the centripetal force, the pressure difference
force, the gravity force for the air mass in atmosphere (onshore or offshore) in the form of the
fundamental dynamic equation as presented above in paragraph 2.
To evaluate the wind importance with high precision we must consider and estimate all
the presented forces at different points in the atmosphere. Actually researchers in the domain
of meteorology and oceanography to calculate these forces use numerical models based on an
atmospheric resolution of about 100 km. But between points in that category of knit there will
be many phenomenon not well represented and for which the effect is not neglected. To have
a real representation of the atmosphere they say that we must consider a discard between
particular points less than 10 m, which is, in accordance to many researchers, impossible in
the view of calculators capacities (Edwards 2010; Lynch 2008, 2006; Stensrud 2007; Kalnay
Eugenia 2003; Pielke, Roger 1984; Thompson 1961).Developing analytical model, based on
the equations we will present later, for local zone of high wind potential considering local
conditions, situations and parameters (topography, plant canopies, microclimates) is a
solution.How to consider the effect of high plant canopies (forests, sylvopastoral zones,
traditional oasis), knowing the fact that they participate actively in the atmospheric water
cycle, the pressure difference appearance, wind generation and precipitation descend? How to
validate experimentally that kind of model?
found winds blowing in the direction North East in the north hemisphere and in the direction
South-East in the south hemisphere.
The second, in the middle latitude, characterised by transient low-pressure systems in the
west direction.
Finally, polar cellule in the North and south of the parallel number 60 with an East
surface circulation. At large scale and for the north hemisphere, the wind turns in the hourly
sense around anticyclone and anti hourly around a depression. Inversely, for the south
hemisphere, winds turn in the hourly sense around a depression and anti-hourly around
anticyclone.
The meteorologists have agree on the following global directions of wind:
North (N), South (S), East (E), West (W), North West (NW), North East (NE), South
West (SW), South East (SE). In the field those directions could be found by installing
weathercocks on masts at different highs. Mathematically they are considered in the
expressions of the wind speed coordinates and the correspondent intensities as we will show
later. Many researchers to formulate the effect of wind directions offshore or inshore on the
intensity consider the scale and shape parameters and use the Weibull probability density
function for the total wind distribution over all directions. This later can be expressed by the
following formula (Lavagnini et al 2006; Jaramillo and Borjazool, 2004):
Bshape 1
FW p d (V)( )( V ) shape exp( V ) shape
B B
Based on this equation many models have been established in order to predict and draw
maps for the potential wind energy sites specially the offshore ones. The most widely used
tool for direction wind assessment based on Weibull probability is the Geo WAsP one of Riso
National Laboratory (Roskilde, Denmark) (Lavagnini et al 2006; Rise National Laboratory,
2006). To obtain the offshore wind climatology over the whole Mediterranean region and by
applying this model the European Centre for Medium-Range Weather Forecasts retrieve 6
hourly wind data over a grid resolution of 0.5° * 0.5° for a period of 24 years (Lavagnini et al
2006). For each grid point, in the Mediterranean area, mean wind speeds and Weibull
probability density function parameters have been computed for 12 directions of 30° each.
Maps of mean wind speed, scale parameter and shape parameter established show that the
areas most suitable for wind energy application in the Mediterranean basin are located close
to the Gibraltar strait, in the Golfe du Lion, western Sardinia, south of Sicily, the Greek
Island, the southwestern part of Turkish coast, and most of the North Africa coast (Lavagnini
et al 2006).
The following table presents values of V, Bscale and Bshape for different region in the
Mediterranean basin (Lavagnini et al 2006):
Basic Concepts for Modelling in Different and Complementary Ecological Fields 375
Iwind = mV dV
dt
Iwind: Intensity of the wind
: Vector representing the Wind velocity
m : Air mass in the limited airspace volume‘s studied
t : Time
The air mass is determined after limiting virtually the volume of airspace in which we
will install our propellers. So considering the air density (volume mass) we can deduce the air
mass and the correspondent weight. Wind speed will be modelled on behalf of Navier-
Stockes equations as we will present in the next paragraph.
5.3.4. Air Motion Equations to Estimate Wind Speed Onshore and Offshore
The general equations formulating the displacement of any body, in particular, the air
mass are the Navier Stockes equations. They are deduced from applying the fundamental
dynamic relation to the system studied and by considering all the forces exerted and the
boundary conditions relatives to the system. In an explicit manner these equations are the
combination of the continuity equation, energy equation and motion equation presented above
(paragraph 2.).
For atmospheric circulation, by considering, hypothesis related to air motion in the
atmosphere, the Navier Stockes equations have been simplified to obtain the atmospheric
primitive equations(Edward 2010; Graber, 1978; Comolet 1963). They are applied after
376 Mohamed Habib Sellami
considering the following hypothesis: Fluid in motion over the surface of a sphere, the
vertical component of the motion is negligible in front of its horizontal component, the deep
of fluid layer is very feeble compared to the radius of the sphere Those hypothesis
correspond to the flows motion at large scale as the synoptic scale for the earth atmosphere.
They are well used in meteorology and oceanography for numerical models of time
forecasting and when simulating the future behaviour of the atmosphere (Edward 2010;
World Meteorological Organisation, 1974; Richardson‘s 1922).
From this general equation, and for the gestrophic conditions, we can obtain the
following equations of air motion either onshore or offshore (Sellami November 2010;
Lamoureux et Fortuné 2001 ; Beniston,.1998; Spinnler 1997; Comolet 1963; Thompson
1962; Richardson‘s 1922):
du fv
dt x
dv fu
dt y
RT
p p
u v w 0
x y p
T u T v T w(T RT ) H f
t x y p pcp c p
cp : Specific heat
Hf: Heat flux by unity of time and mass
T : Temperature
u, v, w : Coordinates of the wind speed, respectively zonal, meridional and vertical
Φ : geo-potential
Ψ: Water content of the air
R: Perfect gaz constant
By considering the stereographic polar projection and the sigma coordinate we can obtain
the Following expressions:
T u T v T wT
t x y z
- For the wind speed coordinate u:
(u 2 x)
u v c p p Exn z u 2
t x x x
- For the wind speed coordinate v:
(v2 y)
v u cp p Exn z v 2
t v y y y
- For the air water content:
u(x )v(y ) z(z )
t x y z
Ψx,y,z : Air waters content in the direction x, y et z
θp : Potential temperature
Exn : Exner function
A general analytical solution of the primitive equations that consider the latitude and the
altitude is:
With these equations we can then estimate the wind speed, the temperature, the pressure
at any points inshore or offshore to evaluate later if we can install our turbines or not. We can
also, on behalf of this resolution, elaborate new wind atlas and wind maps, to make there
bringing up to date and to actualise those existing.
P = Ut I
P: Power in watt
Ut: Tension in volt
I: Intensity in ampere
378 Mohamed Habib Sellami
Lamps for lighting, household appliances, ovens, air conditioner, exhauster, aspirator,
probes, producing machines inside an industrial chain, agriculture product transforming
installations, water pumping, street illuminating, power tools, devices for building and
administration functioning, computer and calculus centrals running…all must be considered
and energy needed for their functioning have to be estimated. The model of wind turbine to
install and the number to fix depend on the amount of energy needed hourly.
Statistical analysis and mathematical models, census, general plan for durable
development by region are tools to evaluate the need of energy and express it by person at
many scales for time (second, hour, day, month, year, decade, …), or by space (house,
apartment, floor, building, village, town, country…), or by usage (lighting, air conditioning,
cleaning, industrial, agricultural…) So we can write:
P
Pcons need
l
l
After fixing the population need from energy in order to accomplish all their activities,
compared to the wind power supplied by the identified site which we can calculate by the
above equations we can judge the feasibility of the wind site, if it is sufficient and if we can
implant in it our wind farm.
Yet we calculate the force of air pressure from the dynamic fundamental equation, the
atmospheric primitive equations and the expression of wind speed presented above, after we
deduce the force of resistance of the blade:
Fbladeres Fpres
Also, at the same time, we can express these force of resistance by considering the blade
fabrication material density and its geometric form as follow:
Basic Concepts for Modelling in Different and Complementary Ecological Fields 379
Fbladeres = b Sb eb 1V
t
ρb: Density or volume mass of the blade material of fabrication
Sb: Surface of the blade
eb: Depth of the blade
t: Time of functioning depending on the wind blowing duration‘s
V : vector of the wind speed
For a circular form of the blade we can write the following simplified form.
Fbladeres = b
Db2
V2
4
Db: Blade diameter
From the equality between the two expressions of the blade resistance force‘s we can
calculate the density or volume mass of the adequate material to fabricate the blade, its
specific diameter and its optimal depth. The number of blades by propeller could be fixed
according to the desired wind power we want to produce with the idea that we must minimise
the lost of wind between the blades.
If we suppose a propeller of (N) blades and in the case when we went to transform all the
wind blowing and the correspondent air mass touching the blades into electricity and not only
the energy needed by the consumers to satisfy we can write the following relationship:
From the precedent equations we can calculate the wind speed and the wind power so we
can deduce the optimal number of blades by a propeller, their diameters and depths. So they
can resist to the wind effect and they can transform all the wind touching them into electric
energy.
We give here some values for blades sizes as function of wind forces: A wind speed of 1
m/s exerts a pressure of 0.14 kg/m² on the blade, a wind speed of 7 m/s exert a pressure of
about 6.64 kg/m², that of 10 m/s has as pressure 13.54 kg/m², at 25 to 30 m/s the pressure
reaches 160 kg/m². For a wind of 10 m/s, and by supposing that the propeller utilize all the
380 Mohamed Habib Sellami
wind touching it, a blade of 2 m diameter sweeps a circle of 3.14 m² can‘t resist to the wind
force, while that of 5 m sweeps a circle of 19.6 m² and resists to a pressure of 263 kg.
The blade can be spiral, fringed or conical and its number varies between 2 to 20 by a
propeller. It depends on wind direction, its speed and if we went to exploit all the power
source with minimising the lost between the blades. The calculus is easy by considering some
geometrical rules, and the wind speed estimated by the precedent equations (Stankovic and al
2009; Roland1981).
Yw-tur = Pwtur
Pwblade
Yw-tur: Yield of a wind turbine
Pw-tur: Energy to be transformed by a wind turbine in unity of power
Pw-blade: Energy of wind flux in the site touching the blades of a propeller
Added to the precedent formulas for the intensities expressed in unity of power that we
have detailed in the above paragraphs, the energy that could be transformed by a wind turbine
can be formulated by (Roland ,1981):
Pw-tur = 0.12 D2 V3
D: diameter of a blade depend on wind speed pressure
V: Wind speed
The energy of wind flux in the site touching the blades of a propeller can be expressed in
more explicit form as follow (Comolet 1963):
Pw-blade = ρ g Qair-flow D
ρ : Air density
g: gravity acceleration
Qair-flow Incident air flow touching the blade of a propeller in the site (depends on wind
speed and the airspace. Surface‘s traversed by the blades).
Determining the capacity of wind farms means calculating its load to change all the wind
energy of the sites into electrical energy. We can then write (Sellami 2010; Stankovic and all
2009 ; Pierre P. 2006; Comolet 1963):
Ptrfor farm
Lw-farm =
Psite
Lw-farm: Load of the wind farms
Ptrfor-farm: Energy to be transformed expressed in unity of power
Psite: Energy existing in the potential sites expressed in unity of power
The energy existing in the potential site could be deduced according to the precedent
equations, that to be transformed can be calculated by the following explicit formula
(Stankovic and all 2009; Spinnler 1997; Roland1981; Comolet 1963):
1 4Vspace farm
Ptrfor-farm = ρ g Qtot
Ywtur NwtD2
ρ : Air density
g: gravity acceleration
Qtot: Airflow touching the propellers
Vspace-farm: Volume of airspace occupied by all the farm (all the wind turbines) in the site
D: Diameter of a wind turbine
Nw-t : Number of wind turbines used
Yw-tur: Yield of a wind turbine
Also, in order to offer the exact need of consumer and to conceptualise correctly the wind
park, we have to consider the eventual lost in the transport system (cables, concentrators,
batteries..) and the position of every propeller on behalf of the penetrating points of the air
flux inside the limited airspace volume‘s containing all the wind turbines.
CONCLUSION
With global environmental challenges, ( greenhouse gaz emission, intensive inundation,
long drought, scarcity of water resources, exhausting of fossil energy resources), specialists in
many scientific fields, experts at International level and decision makers are extremely
worried about proposing solutions and tools to monitor the climatic change and to plug its
effects. Particular attention, is given for models to develop, the necessaries steps to follow
and programs to apply in the purpose to assume the absorption of CO2 existing in the
atmosphere by conserving the plant biodiversity, to limit the CO2 emission from buildings
and to produce energy from the renewable ones in particular from the wind. It is in this
context that belongs our contribution.
For the plant biodiversity conservation, fragile ecosystem management and suggesting
appropriate tools helping in decision-making for the environment harmony and agricultural
planning, we have developed a model to be applied for three kinds of wetlands: forests,
382 Mohamed Habib Sellami
sylvopastoral zones and traditional oasis which are occupied by diversity of species with
different heights. These species have multiple usages (human feeding, animal grazing,
endemic and medical plants, wood biomass, rain canopies, soil stabilisers and infrastructure
protector) and are considered as atmospheric cleaner (CO2 reservoir). The equations
introduced permit to determine the radiative climate and the transfer of heat and mass inside
the wetland after considering the vegetation canopy as a succession of layers and inside every
layer the specie is identified by its optical and dielectric properties, morphologic and
physiologic characteristics. Also we have applied the thermal electrical analogy and admitted
a network of resistances to the transfer of latent and sensible heats expressed as function of
plant physiologic parameters, air temperature and water vapour pressure So we can estimate
the biomass production inside the wetland for every sort of plant, monitor their phytosanitary
situation, propose the plant architecture (specie, density, orientation…) the most adaptable to
climate change when renewing them, organise the actions of grazing and deforestation
without risks of overexploitation. A first validation of the model was done. The results are
satisfying.
For the building sector, we have proposed the basic equations and concepts to model the
thermal exchange inside and with its surrounding environment for all the building usage‘s:
domestic, service and industrial. This model is backed on the equation of heat and mass
transfer, equation of radiative transfer, equation of motion and the Ohm‘s law. So we can
choose the building orientation and the construction material for walls, ceilings and roofs the
most adaptable to the local climatic conditions. Also, we can make the decision about
conceptualising the repartition of offices, rooms, apartments, floors for every activity, internal
arrangement of apparatus, machines, movables, installations…, their positioning inside the
building, the timing of their functioning, the frequency of their moving… for an optimal use
of energy. Either we can, on behalf of the model presented, propose the type of insulation
system (molecular formation, porosity…) to install the most adaptable to both the external
climatic condition and the internal thermal contribution by considering recoverable and non
recoverable heat flux. Finally, the equations established are very promising in defining a
program of certification for energy saving in buildings responding to the international inquiry
concerns about the emission of gases with greenhouse effect and assuming an acceptable
level of building thermal efficiency. The model could, also, be useful for making analytical
analysis to verify the degree of precision of the energetic audit method‘s and the formulas
used actually for building thermal studies and to make their bringing up to date.
For the use of renewable energy, our focus was about the wind energy. We have
presented a set of equations, based on the resolution of the Navier Stockes‘ equations for the
air mass circulation at atmospheric level. We have expressed the wind speed for all the
directions either inshore or offshore. So we could be able to determine zones with high wind
potential, elaborate new wind atlas and wind maps, and to actualise those existing. Also after
estimating the wind speed, we can deduce the wind power existing on the site, to
conceptualise wind turbine and wind farm to install and to test their efficiencies and loads.
The necessary equations were detailed. Currently, a good number of research projects are
underway on international level in the fields of short term forecasting of wind power,
onshore/offshore. The purpose of that action is to co-ordinate the activities in these related
fields, to spread the knowledge gained from these projects among the partners, and to start the
work on some roadmaps for the future. Our work here could be very fruitful.
Basic Concepts for Modelling in Different and Complementary Ecological Fields 383
REFERENCES
Abdallah H. 2008 « Dangers menacing golf petroleum‘s » International Politics Journal ; Al-
Ahram ; Issue n°171; pp.34-37.
Abdul Awal M., Wan Ishak Wan Ismail, Mohd. Haniff Harun and Johari Endan, 2005.
―M ethodology And Measurement of Radiation Interception by Quantum Sensor of the
Oil Palm Plantation. Songklanakarin‖ J. Sci. Technol., Vol. 27 N°. 5, Pp. 1083-1093.
Abou Khalil B. 2008 ―A method for energy and exergy analyses of product transformation
processes in industry‖ ENSMP PhD. Thesis, Paris, 2008.
Adama Diawara 1990« Echanges d‘energie et de masse à l‘intérieur et au-dessus d‘une forêt
de pins des Landes » Thèse, Université Blaisse Pascal, Clermont-Ferrand.
AFRC. 1997 « The nutrition of goats‖ Technical Committee on Responses to Nutrients.
Report 10. Ag. Food Res. Council. Nutr. Abstr. Rev. (Series B) 67; pp. 806-815.
AFRC. 1991 « A reappraisal of the calcium and phosphorus requirements of sheep and cattle‖
Technical Committee on Responses to Nutrients. Report 6. Ag. Food Res. Council. Nutr.
Abstr. Rev. (Series B) 61; pp. 573-612
Aidoud A. 2006 « Les steppes arides du Nord de l‘Afrique » Sécheresse Vol. 17, N° :1-2.
Angelika P., Sven T., and Alexandra D. 2010 ― Wind power to provide a fifth of world‘s
electricity by 2030‖ Press release October 12, 2010, Greenpeece International and
G.W.E.C, http://www.greenpeace.org/ and http://www.windfair.us/press/8220.html
Aoun Bernard 2008 « Micro-cogénération pour les bâtiments résidentiels fonctionnant avec
des énergies renouvelables », Thèse, Ecole Nationale Supérieurs des mines de paris. Pp
186 pages.
Aregheore E.M. 2007 ― Voluntary intake, nutrient digestibility and nutritive value of foliage
of fluted pumpkin (Talfairia occidentialis) – haylage mixtures by goats‖ Live Stock
Research for Rural Development Vol.19 N°.4
Armstrong R and Milne J.A. 1993 ―Nutritive value of pastures and rangelands‖ ICEL. AGR.
SCI Vol.7 pp..37-43.
Auque A.S., Fantoni S. and Lepage N. 2008 «Ambiances thermiques : travail à la chaleur et
au froid » Centre de conseil et de documentation ; I.S.T.N.F. France. p. 28 pages.
Austin, M. 2007. Species Distribution Models And Ecological Theory: A Critical Assessment
and Some Possible New Approaches. Ecological Modelling, Volume 200, Issue 1-2, Pp.
1 – 19.
Bamikole M.A. and Babayemi O.J. 2009 ―M acro-mineral utilisation in West African dwarf
goats fed combination of fertilized and unfertilized guinea grass (Panicum Maximum, cv.
Ntchisi), Verano stylo (Stylosanthes hamata cv. Verano) and a concentrate mix‖. Live
Stock Research for Rural Development Vol.21 N°.9.
Benabadji N., Aboura R. and Benchouk F.Z. 2009 ―The regression of the Mediterranean
steppes.: The case of Lypeum Spartum L.Facies in Oran region (Algeria)‖ International
Journal of Mediterranean Ecology. Vol. 35. pp. 75-90
Beniston, Martin.1998 ―F rom Turbulence to Climate: Numerical Investigations of the
Atmosphere with a Hierarchy of Models‖. Springer, Berlin, ISBN: 3540634959
Bonhomme R. 1993 ― The solar radiation: characterisation and distribution in the canopy‖ in
―CropStructure and Light Microclimate‖ Edition INRA France. pp. 17-28.
384 Mohamed Habib Sellami
Bréda Njj. 2003. Ground-Based Measurements Of Leaf Area Index: A Review Of Methods,
Instruments And Current Controversies. Journal Of Experimental Botany, 54, Pp. 2403–
2417.
Butler R.A. 2010 ― Types Of Rainforests‖ From Http://Rainforest.Mongabay.Com/0103.Html.
Consulted 07/01/2011.
Butler R.A. 2006 ― Diversities Of Images- Rainforest Biodiversity‖ From
Http://Rainforest.Mongabay.Com/0305.Html. Consulted 07/01/2011.
Carrière M. 1995 « Impact des systèmes d‘elevage pastoraux sur l‘environnement en Afrique
et en Asie tropicale et sub tropicale, aride et sub aride » Scientific Environmental
Monitoring Group. Universität des SaarlandesInstitut Fär Biogeographie.
SAARBRÜCKEN. Allemagne. p. 70 pages.
C.E.R.T.U. 2003 « Le confort thermique » Mémento Technique du Bâtiment, France, p. pages
22 pages.
Chami N. 2009, « Evaluation de l‘impact des isolants minces réfléchissants dans les
bâtiments », Thèse, Ecole Nationale Supérieure des Mines de paris, p.155 pages.
Chang T.J., Wu Y.T., Hsu H.Y., Chu C.R and Liao C.M. 2003 ― Assessment of wind
characteristics and wind turbine characteristics in Taiwa‖ Renewable Energy, N° 28, pp.
851-871.
Charlotte B.H; Alfredo P., Merete B.C., Poul A., Morten N., Frank M., Donald T., and
Nielsen P. 2008 ― Remote sensing observation used in offshore wind energy‖ Applied
earth Observations and Remote Sensing, Vol. 1 N° 1 pp. 67-79.
Cheikh AL Bassatneh M. 2006―Facteur du milieu, gestion sylvicole et organisation de la
biodiversité : Les systèmes forestiers de la montagne de Lure (Alpes de Hautes-
Provence, France) Thèse de Doctorat, Université Paul Cezanne Aix Marseille III. Pp. 239
pages.
Christian K.., Bruce D., Raffaella B., and Elke Z.. 2009 ―Wind Energy. The Facts. Executive
Summary‖ European Wind Energy Association, p. 32 pages.
Cing Mars D., 2003 « Récents développments concernant les apports nutritionnels
recommandés en calcium et phosphore chez les caprins » MAPAQ, Direction de
l‘innovation scientifque et technologique.
http://www.agrireseau.qc.ca/caprins/Documents/Recents%20developpements.pdf
Consulted 18/01/2011. p. 5pages
Comolet R. 1963 « Mécanique expérimentale des fluides. Dynamique des fluides réels,
turbomachines » Masson et CLE editeurs, France, p. 442 pages
Cowling S.A., Field C.B. 2003. ― Environmental Control Of Leaf Area Production:
Implications For Vegetation and Land-Surface Modeling.‖ Global Biogeochemical
Cycles, 17, Pp. 1–14.
CRPF 2007 « Fiche Identité Restinclières. : Fiches Essences : Programme Intégré de
Recherche en Agroforestière à Restinclières »Hérault- Languedoc-Roussillon. Pp. 6
pages. Source Internet
http://www.agroof.net/agroof_dev/documents/pirat/protocoles_restinclieres.pdf
consulted 01/01/2011
CRPF 2011« Guides des stations forestières » Centres Régionales des Propriétés Forestières,.
Source Inetrnet : http://www.crpf-lr.com/telecharg.htm#guides. Consulted 15/01/2011.
Daudet F.A. and Tchamitchian M. 1993 ―Radiative exchange and photosynthesis‖ in ―Cro p
structure and light microclimate‖ INRA France. Pp. 401-417.
Basic Concepts for Modelling in Different and Complementary Ecological Fields 385
Delucia Evan H., Kate George And Jason G. Hamilton, 2002. Radiation Use Efficiency of a
Forest Exposed To Elevated Concentrations Of Atmospheric Carbon Dioxide. Tree
Physiology 22, Pp. 1003-1010.
De-Xing C. And Coughenour M.B. Gemtm.,1994: A General Model For Energy And Mass
Transfer of Land Surfaces And Its Application At The Fife Sites. Agricultural And Forest
Meteorology, 68, 145-171.
Dixon S.L., Eng B.. 1998 ― Fluid Mechanics, Thermodynamics of Turbo machinery‖ Fourth
Edition in SI/METRIC UNITS, OXFORD OX2 8DP; ISRN 0 7506 7059 2; 353 pages.
Eastman J.L., Michael B. Coughenour And Roger A. Pielke, Sr., 2001. ― The Regional Effects
Of Co2 And Landscape Change Using A Coupled Plant And Meteorological Model‖.
Global Change Biology, Vol. 7, Issue 7 Page 797.
Eden.P 2011 ―Eden Project‖ Source Internet: Http://Www.Edenproject.Com/Whats-It-All-
About/Index.Php; Consulted 15/01/2011.
Edwards Paul 2010 ― A vast Machine: Computer Models, Climate Data, and the Politics of
Global Warming‖ cambridge, MA: MIT Press, 2010. ISBN-13: 978-0262013925, p. 528
pages
El Euche F. 2006 « Gestion durable des aménagements sylvopastoraux en Tunisie »Direction
Générales des Forêts. Ministère de l‘Agriculture. Tunisie p.30 pages.
E.M.P.A. 2009 « Protection thermique estivale des pièces sous les combles. Analyse des
facteurs d‘influence sur le climat intérieur » Rapport n° 444'383. Pp. 24 pages.
E.P.A. 2000 ―America‘s wetlands: our vital link between land and water‖ US Environmental
Protection Agency. p. 16 pages.
Ewert. F. 2004. Modelling Plant Responses To Elevated Co2: How Important Is Leaf Area
Index? Annals Of Botany, Vol. 93, Pp. 619-627.
Farooqi M.I.H 2010 ―Plants Of The Quran‖ 7th Edition: Sidrah Publisherz, pp. 224 pages.
Farooqi M.I.H 2008 ―Medicinal Plants In The Traditions Of Prophet‖ 4th Edition. Sidrah
Publisherz, Pp. 248 Pages.
http://Openlibrary.Org/Authors/Ol204792a/M._I._H._Farooqi. Consulted. 14/01/2011.
Farquhar Graham D.,* Susanne Von Caemmerer, And Joseph A. Berry. 2001. Models of
Photosynthesis. Plant Physiol., Vol. 125, Pp. 42-45.
Fiabitat 2009 « Concept rapport : construction et ingénierie écologique » Pp 22 pages,
www.fiabitat.com.
Flachowsky G.; Chesson A.; Aulrich K. 2005 ―A nimal nutrition with feeds from genetically
modified plants‖ Archives of Animal Nutrition Vol. 59 N°. 1.pp. 1 – 40.
Franco A 2007 « An apparatus for the routine measurement of thermal conductivity of
materials for building application based on a transient hot wire method » Applied
Thermal Engineering Vol. 27, pp..2495-2504.
Gash J.H.C.and al 1989 « Micrometeorological measurements in Les-Landes forest during
Hapex-Mobilhy », Agr.For.Meteo. n° 46 pp 131-147, 1989.
Graber D. 1978 ―M ethods for estimating wind speed frequency distributions‖ Journal Applied
Meteorology N°17, pp.350-353.
Gerard C., Marcot S., F.Barrabe and A.Guyonvarch, 2005, « Variabilité des valeurs
nutritionnelles de l‘herbe dans le grand Ouest de la France », Renc.Rech.Ruminants. Vol.
12.
386 Mohamed Habib Sellami
Goosem S. and Tucker N. 1995 ―Repairing the rain forest: Theory and paractice of rainforest
re-establishment in North Queens land‘s wet tropics‖ Cassowarry Publication; Wet
Tropics Management Authority. Pp. 70 pages.
Guérin G., Aubron C., Boutonnet J.P., Gallton B. Et Sajdak G 2010 « Construire une
cohérence productive entre sylviculture et pâturage dans les espaces boisés de la zone
meditterranéenne Française. L‘innovation sylvopastorale en question » ISDA 2010,
Montepellier 28-30 juin 2010. pp. 14 pages.
Guterman I.G. 1974 ―Macroturbulence of horizontal motion over the U.S.S.R.‖ in ―Phy sical
and dynamic climatology‖. WMO, Leningrad, pp..214-225
G.W.EC 2010 « Global wind Energy outlook 2010 » Report 2010 for Global Wind Energy
Council
http://www.gwec.net/fileadmin/documents/Publications/GWEO%202010%20final.pdf,
conuslted 15/01/2011
Hadjadi Aoul S., Chouied M., and Loisel R. 2009 « Effect of environmental factor on natural
regeneration of Tetraclinis articulata (Vahl Master) in Oran region Algeria» International
Journal of Mediterranean Ecology. Vol. 35. pp. 19-30.
Hand M.M., Simms D.A., Fingersh S., and Larwood S.M. 2001 ―Unsteady aerodynamics
experiment phase VI: wind tunnel test configurations and available data campaigns‖
NREL/ TP-500-29955. Colorado. US. p. 310 pages.
Haralambopoulos D.A. and Paparsenos G.F. 1998 ―Assessing the thermal insulation of old
buildings. The need for in situ spot measurements of thermal resistance and planar
infrared thermography‖ Energy Corners. Mgmt Vol.39. N°1/2, pp.65-79.
Hladik J. 1990 ―Métrologie des propriétés thermophysiques des matériaux‖ Masson
INRA 1988 ―Alimentation des bovines, ovins et caprins » Jarrige Edition. Institue National
de la Recherche Agronomique, Paris. p. 471 pages.
Ionesco T. 1980 ―Suggestions pour une stratégie de développement des zones pastorales
arides et désertiques de Tunisie‖ CIHEAM- Options Méditerranéennes N° 28. pp. 82-88.
Jadhav Ranjit S., and Buchberger Steven G. 1995 ―Eff ects of vegetation on flow through
surface wetlands‖ Ecological Engineering Vol.5 pp. 481-496.
Jannot Y. 2009 « Transfert Thermique » Ecole des Mines Nancy, pp.160
Jannot Y. 2008 « Théorie et pratique de la métrologie thermique » Séminaire PER AUF
―M étrologie thermique‖ LEA Dakar, 12-18/11/08, pp.1-78.
Jannaot Y. and Acem Z. 2007 ―A quadrupolar complete model of the hotdisc‖ Measurement
science and Technologie Vol.18 pp.1229-1234.
Jannot Y., Acem Z., and Kanmogne A. 2006 ―Transient hot plate method with two
temperature measurements for thermal characterisation of metals‖ Measurement Science
and Technology. Vol.17, pp.69-74.
Jaramillo O.A. and Borjazool M.A.2004‖ Bimodal versus Weibull wind speed distributions:
an analysis of wind energy potential in La Venta, Mexico‖ Wind Engineering Volume 28
n°2, pp. 225-234.
Jeffrey T. Edwards, Larry C. Purcell and Douglas E. Karcher, 2005. ― Soybean Yield and
Biomass Responses to Increasing Plant Population among Diverse Maturity Groups. II.
Light Interception and Utilization » Crop Science, vol. 45, pp. 1778-1785.
Joffre R., Hubert B. et Meuret M. 1991 « Les système agro-
sylvopastoraux Méditerranéens : Enjeux et réflexions pour une gestion raisonnée »
UNESCO. Printed in France. p. 97 pages.
Basic Concepts for Modelling in Different and Complementary Ecological Fields 387
K.A.I.G. 2011 ―King Abdullah International Garden: a giant oasis in the desert‖ Source
Internet: http://www.kaig.net/01_About/1A_01.asp. Consulted13/01/2011.
Kalnay Eugenia 2003 ―A tmospheric Modeling, Data Assimilation and Predictability‖.
Cambridge university Press.
Kessler, J.1991:‖ Mineral nutrition of goats‖. In, Goat Nutrition. Edited by P. Morand Fehr.
Wageningen, Netherlands, Pudoc, 1991, EAAP Publ. No. 46, pp. 104- 119.
Konstantinov A.R. and Sakaly L.I. 1974 ―Thesignificance of the components of heat and
water balances in the formation of Micor and local climate‖ in ―Phy sical and dynamic
climatology‖. WMO, Leningrad, pp..99-105
Kull O. 2002. Acclimation Of Photosynthesis In Canopies: Models And Limitations.
Oecologia, Vol. 133, Pp. 267–279.
Lamb H.H.1974 ―Climates and circulation regimes developed over the Northen hemisphere
during and since the last ice age‖ in ― Physical and dynamic climatology‖. WMO,
Leningrad, pp..233-261.
Lamoureux L. et Fortuné D. 2001« Mécanique générale. Aide mémoire »Dunod, Paris, ISBN
2 10 005389 2, p.181 pages.
Landsberg H.E. 1974 ―M an-made climatic changes‖ in ― Physical and dynamic climatology‖.
WMO, Leningrad, pp..262-303.
Lavagnini A.; Sempreviva A.M; Transerici C. ; Accadia C.; Casaioli M.; Mariani S.and
Sperenza A.; 2006 « Offshore wind climatology over Mediterranean Basin‖ Wind
Energy, N°.9; pp.. 251- 266.
Leduc R. and Gervais R. 1985 ―Co nnaître la météorologie‖ Presses de l‘Université de
Québec; Sillery, Québec, ISBN 27605-0365-8, p. 271 pages
Le Houerou H.N. 1975 « The pastoral situation in the North Africa : progress achieved in
data and research » Options Méditerranéennes N°: 26
Le Houerou H.N. 1995/1996 « Considértions biogéographiques sur les steppes arides du nord
de l‘Afrique » Sécheresse, Vol. 6 N°. 2
Le Hourerou H.N. 2001/2002 « La végétation potentielle de la Tunisie aride et désertique
(notice détaillée de la carte au 1/1000 000) » Revue des Régions Arides, N°. 12.
Le Houerou H.N. 2002 ―M ultipurpose germplasm of fodder shrubs and trees for the
rehabilitation of arid and semi-arid land in the Mediterranean isoclimatic zone: A
photographic catalogue‖ Zaragoza [Espagne] : CIHEAM-IAMZ –Options
Méditerranéennes: Série B. Etudes et Recherches. ISBN.ISSN : 978-2-85352-240-3.
p.118 pages.
Le Houerou H.N. 2005/2006 « Problèmes écologiques du développement de l'élevage en
région sèche » Sécheresse, vol. 16, n. 2
Lewis R.I, 2001 ―Development of vortex dynamics for simulation of turbo machine cascades
and blade rows‖ Journal of Computational and Applied Mechanics, Vol. 2, n° 1, pp.73-
85.
Lynch, Peter 2008 ― The origins of computer weather prediction and climate modeling‖
Journal of computational Physics N° 227 Vol. 7, pp.3431-3444.
Lynch, Peter 2006 "Weather Prediction by Numerical Process". The Emergence of Numerical
Weather Prediction. Cambridge University Press. ISBN 9780521857291. pp.1-27.
Matteis P. Campagnoli E., Firrao D., and Ruscia G. 2007 « Thermal diffusivity
measurements of metastable austenite during continuous cooling » International Journal
of Thermal Sciences
388 Mohamed Habib Sellami
Meschy F. 2002 « Eléments minéraux majeurs : Données récents chez les caprins » INRA
Paris. Prod. Anim. N°15. pp. 267-271.
Millet J.R 2006 « Les nouvelles règles Th-CE » Les Rendez vous du CSTB. France. Source
Internet :
http://www.cantal.equipement.gouv.fr/IMG/pdf/Analyse_du_CSTB_cle2cde1a.pdf.
Consulted 17/01/2011. p. 36 pages
Niklas K.J. and Enquist B.J. 2002 ― On the vegetative biomass partitioning of seed plant
leaves, stems, and roots‖ The American Naturalist Vol. 159 N°.5. pp. 482-497.
NRC. 1981 « Nutrient requirements of goats‖ National Research Council. National Academy
Press. Washington D.C. p. 381 pages.
Nsibi R., Souayah N., Khouja L.M. ; Khaldi A. et Bouzid S 2006 « Impacts des facteurs
biotiques et abiotiques sur la dégradation de suberaie Tunisienne » Geo. Eco. Trop. Vol.
30 N°. 1.pp..25-34.
Ouelmouhoub S. 2005 ―Gestion multi-usage et conservation du patrmoine forestier: cas des
subéraies du Parc National d‘EL Kala (Algérie)‖ Série « Master of Sciences » N°78.
CIHEAM-IAMM Montpellier. ISBN: 2-85352-329-2. pp.. 129 pages.
Pierre P. 2006 ―Estimation of the uncertainty in wind power forecasting‖ Thesis Ecole des
Mines de Paris, p. 266 pages.
Pons Michel. 2008 « Méthode pour analyses exergétiques robustes d‘installations solaires et
de bâtiments » I.B.P.S.A.-France. Ed. J.J. Roux and G. Krauss, Article#NSO3.pdf. p.
8pages.
Rakipova L.R. and Efimova L.K. 1974 ―The ory of circulation and thermal regime of the
mesosphere‖ in ― Physical and dynamic climatology‖. WMO, Leningrad, pp..226-232.
Richardson‘s Lewis Fry 1922.‖ Weather prediction by numerical process‖Cambridge
University Press pp.236.
Rise National Laboratory, 2006―WAsP utility programs, version 3, contents, installation and
support‖ Copyright 1993-2006, by Rise National Laboratory, RosKilde, Denmark.
www.wasp.dk/Downloads/; Accessed 14/01/2011;p. 3pages.
Rivière A., AbouKhalil B., and Berthou M. 2009 ―An energy efficiency assessment tool
based on process energy and exergy analysis method‖ ECEEE 2009 Summer Study,
Pnel5: Energy Efficiency Industry,pp.1109-1114.
Roland Roger 1981―Produire son énergie avec le vent‖ Editions de la Lanterne, Paris, IBSN
2 86588-007-9, p. 114 pages.
Rosati A., Metcalf S.G., and Lampinen B.D. 2004. A simple method to estimate
photosynthetic radiation use efficiency of canopies. Annals of Botany, vol. 93, pp. 567-
574
Ryan W. and Mark B. 2008 ―Annual Report on U.S. Wind Power,Cost, and Performance
Trends: 2007‖ US department of Energy. Energy Efficiency and Renewable Energy;
DOE/GO-102008-2590; p.32 pages.
Sanae Hammi, Vincent Simonneaux, Mohamed Alifriqui, Laurent Auclair, Nicolas Montes
2007 « Évolution des recouvrements forestiers et de l‘occupation des sols entre 1964
et 2002 dans la haute vallée des Ait Bouguemez (Haut Atlas central, Maroc) » Science et
Changements planétaires. Sécheresse. Vol. 18 N°.4. pp. 1-7
Schizas D. and Stamou. G. 2007. What Ecosystems Really Are—Physicochemical Or
Biological Entities?. Ecological Modelling, Volume 200, Issues 1-2, Pp.. 178-182.
Basic Concepts for Modelling in Different and Complementary Ecological Fields 389
Sellami M.H. and Sifaoui M.S. 1998 ―M easurements of microclimatic factors inside the
oasis: interception and sharing of solar radiation‖ International Journal Renewable
Energy, Vol. 13 N°. 1 pp.67-76.
Sellami M.H.and Sifaoui M.S. 1999« Modelling solar radiatif transfer inside the oasis:
Experimental validation » Journal of Quantitative Spectroscopie and Radiative Transfer,
63/1 pp 85-96.
Sellami M.H. and Sifaoui M.S. 2003 ― Estimating transpirationin an intercropping system:
measuring sap flow inside the oasis‖ Agricultural Water Management Vol.59. pp.191-
204.
Sellami M.H. and Sifaoui M.S. 2008 ―M odelling of heat and mass transfer inside a traditional
oasis. Experimental validation‖, International Journal of Ecological Modelling; 210, pp.
144-154.
Sellami M.H. 2008 « A scientific guide for agricultural water management and biodiversity
conservation inside the North African oasis » Chapter In: Agricultural Water
Management Research Trends. ISBN 978-1-60456-159-3. Editor: Magnus L. Sorensen ©
2008 Nova Science Publishers, Inc. pp..171-212.
Sellami M.H. And Trabelsi 2009 « A Modelling Approach Using The Geographic
Information System For Decision Making In The Water Sector » Journal Geographia
Technica, Issue 2/2009.
Sellami M.H. 2010 ―M odelling Approaches For Plant Biodiversity Conservation. Case
Studies In The North African Oasis‖ A chapter in the book ― Biodiversity Hotspots‖
Editors: Vittore Rescigno and Savario Maletta, Nova Science publisher, 2010..
Sellami M.H. October 2010 « Approche de modélisation pour un aménagement hydro-
agricole durable et efficace des ressources sylvo-pastorales : aléas climatiques, bilan
hydrique et biodiversité » 1er Colloque International sur les Ressources Sylvo-Pastorales
et Développement Durable en Méditerranée. Tabarka, Tunisie les 19, 20 et 21 Octobre
2010
Sellami M.H. November 2010 ― A modelling approach for evaluating regional wind potential,
conceptualising wind turbines and testing their efficiencies‖ International Renewable
Energy Congress, IREC2010(www.irec.cmerp.net), Sousse, Tunisia, 5-7 November
2010.
Sellami M.H. December 2010 «Modélisation de l‘efficacité thermique des bâtiments »
Premier Séminaire International sur le Génie Climatique et l‘Energétique.
www.sigcle2010.com. Université Mentouri Constantine, Algérien 06-07 Décembre 2010.
Sellers P.J. and Mintz Y., 1996, ― A simple biosphere model (SiB) for ue within general
circulation models‖ Journal of the Atmospheric Sciences, Vol 43 n°6, pp 505 –531.
Sghaier T. and Garchi S. 2009 « Dominant height growth modelling and fertility of Aleppo
pine (Pinus halepensis Mill) stands in Tunisia » International Journal of Mediterranean
Ecology. Vol. 35. pp. 49-64.
Sinisa Stankovic, Neil Campbell and Alan Harrie‘s 2009 ‖Urban wnd energy‖ Punlished by
Earth Scan UK. and USA, ISBN 978-1-84407-282-8, p. 187 pages.
Sinoquet H., Roux L. E., Adam B., Ameglio T. and Daudet F. A., 2001. RATP: a model for
simulating the spatial distribution of radiation absorption, transpiration and
photosynthesis within canopies: application to an isolated tree crown. Plant, Cell and
Environment, vol. 24, pp. 395-406.
390 Mohamed Habib Sellami
Smith M.C. and David M. 1994 ―Goat Medicine‖ Edited Lippincott Williams and Wilkins,
USA, p. 620 pages.
Spinnler G. 1997 « Conception des machines. Principes et applications » Presses
Polytechniques et Universitaires Romandes, Suisse, ISBN 2-88074-302-8, 513 pages
Stensrud, David J. 2007.Parameterization schemes: keys to understanding numerical weather
prediction models. Cambridge University Press.. ISBN 9780521 865401. p. 137 pages
Strub F., Castaing L.J., Strub M., Pons M., and Monchoux F. 2005 ―Sec ond law analysis of
periodic heat conduction through a wall‖ Int. J. Thermal. Sci. Vol.44, N°.12, pp.1154-
1160.
Thompson, Philip D.1962. ―Numerical Weather Analysis and Prediction” Mathematics of
computation Vol.16 N° 80. pp.503-505
Th-CE 2005 ― Calcul réglementaire des consommations d‘énergie‖ Doc. V7.3.France. Source
Internet : http://www.rt2000.net/documents/news/News25_2.pdf Consulted 17/01/2011.
p. 159 pages.
Tournebize Régis et Hervé Sinoquet 1995 « Light interception and partitioning in a
shrub/grass mixture » Agriculture and Forest Meteorology n°72 pp 277-294.
UN-REDD; 2010 ― UN-REDD Programme applauds the COP16 agreement on Reducing
Emissions from Deforestation and Forest Degradation reached in Cancum‖ Source
Internet: www.un-redd.org Consulted 12/01/2011.
U.S. Department of Energy, 2010 ―Wi nd and water power program: Building a new energy
future‖ National Renewable Energy Laboratory, DOE/GO-102010-3065, p. 2pages.
U.S. Energy Information Administration 2010 ―International Energy Outlook 2010‖ U.S.
Department of Energy. Washington. DOE/EIA 0484(2010).
www.eia.gov/oiaf/ieo/index/.html. p. 338 pages.
Varlet Grancher C., Bonhomme R. and Sinoquet H., 1993, « Crop structure and light
microclimate. Characterisation and applications », INRA Editions, 518 pages.
Vasilescu E. E., Boussehain R., Feidt M., and Dobrovicescu A. 2008 « Analyse exergetique
comparative des systèmes frigorifiques a adsorption a un ou deux étages à simple ou
double effet » Termothenica Issue 2, pp.11-17
Wikipédi A, « le vent »,http://fr.wikipedia.org/wiki/Vent , consulted 13/01/2011
Wikpédi A ―ener gie éolienne »
http://fr.wikipedia.org/wiki/%C3%89nergie_%C3%A9olienne; consulted 15/01/2011.
World Meteorological Organisation‖ Physical and dynamic climatology‖, WNO N° 347,
Leningrad, 1974, 582 pages .
Wright S.E. 2007a « Comparative analysis of the entropy of radiative heat transfer and heat
conduction‖ Int. J. Thermodynamics Vol.10 N° 1.pp.27-35.
W.T.M.A. 2010 ―Plants and animals: Plants- Not all forests are the same‖ Wet Tropics
Management Autority. Source Internet: www.wettropics.gov.au. Consulted 01/01/2011.
Zhang H. and Sellers A.H. 1996 « Impacts of tropical deforestation. Part I: Process analysis
of local climatic change‖ American Meteorological Society. Vol.9 pp..1497-1516.
Zho W., Qi S., An Q., Zhao H. and Liu N.2007 ―Thermal conductivity of boron nitride
reinforced polyethylene composites‖ Materials Research Bulletin. Vol. 42 pp.1863-1873
INDEX
automata, 12, 84
avoidance, 159
C
awareness, 84, 206
cabbage, 14
cables, 366, 381
B calcium, 347, 348, 349, 383, 384
calculus, 349, 353, 362, 363, 364, 365, 366, 369,
Baars, 221 378, 380
background information, 38 calibration, 18, 24, 37, 119, 126, 133, 166
bacteria, 43 campaigns, 386
bacterium, 320 cancer, 293, 316, 320
Balkans, 203 carbon, x, 17, 86, 88, 89, 90, 115, 116, 118, 128,
banks, 328 142, 143, 152, 155, 158, 159, 163, 165, 166, 170,
barriers, 178, 179, 292 187, 202, 210, 214, 340, 341, 346
base, 100, 141, 185, 364 carbon dioxide, 17, 346
basic research, 100 carbon emissions, 340
batteries, 381 carotenoids, 28
beef, 101 case studies, 62, 89, 336
behavioral models, 258 case study, xi, 14, 62, 63, 113, 196, 197, 203, 205,
behaviors, xi, 46, 223, 224, 226, 241, 243, 259, 263 207, 219, 334, 340
Beijing, 12, 14 casting, 92
bending, xii, 5, 290, 300 catchments, 46, 50, 51, 54, 60, 61
benefits, 43, 91, 116, 168, 226, 248, 262, 263 category b, 349
bias, 4, 10, 50, 125, 132, 135, 139, 171, 198, 201 cattle, 383
biochemical processes, 86 Cauchy stable noise, xii, 290, 305, 307, 308, 316
biodiversity, x, 173, 185, 186, 193, 200, 208, 328, causal relationship, 228, 229, 230, 232, 238
332, 336, 338, 340, 350, 381, 389 causation, 225, 259, 260
biogeography, 186, 187, 193, 194, 199, 201, 203 cell metabolism, 293, 316
biological control, 218 cell size, 219
biological media, 215 cellular automaton, 258
biological processes, xi, 87, 99, 205, 206, 216 cellulose, 347, 348
biological systems, xii, 185, 289, 291 certification, 356, 365, 382
biopolymer, 292 challenges, 194, 381
biosphere, 389 changing environment, 88, 245, 288
biotic, 100, 113, 120, 175, 178, 179, 180, 181, 182, chaos, 73, 80
188, 190, 191, 192 chemical, xi, 17, 50, 51, 86, 89, 205, 206, 208, 210,
biotic factor, 100, 175, 180, 182 214, 216, 253, 255, 286, 292, 293, 319, 321
birds, 2, 187, 260, 340 chemical properties, 17
birth rate, 67, 69, 72, 77, 79, 81, 227 chemical reactions, 293, 321
blowfly species, ix, 97, 101, 109, 111 chemicals, 48, 50, 220, 240
body size, 84, 99, 113, 188, 197, 202 chemotherapy, 319
boilers, 367 Chicago, 200
bonds, 300 Chile, 208, 220, 352
boreal forest, 171 China, 1, 11, 12, 14, 39, 111
Brazil, 35, 101, 111, 318 chlorophyll, 21, 23, 24, 28, 35, 88, 89, 91, 214, 215,
breakdown, 44, 61 217, 221
breeding, 12, 183, 195 circulation, 91, 193, 210, 212, 218, 366, 373, 374,
Britain, 37 375, 382, 387, 388, 389
Brownian particle, 292 City, 169, 171, 172
building blocks, 224 classes, 20, 134, 157, 159, 165, 166, 229, 230, 233,
Bulgaria, 171 368
classification, 3, 12, 70, 72, 176, 177, 184, 187, 191,
192, 202, 210
clean energy, xiii, 335, 336
Index 393
ecology, vii, ix, x, xi, 11, 12, 13, 80, 83, 84, 86, 90, equilibrium, xi, 164, 183, 187, 189, 248, 254, 267,
92, 93, 97, 98, 99, 101, 102, 107, 108, 109, 112, 268, 269, 270, 271, 272, 273, 274, 275, 276, 277,
113, 116, 117, 164, 166, 168, 193, 196, 198, 199, 279, 281, 282, 283, 286, 300, 336, 339
203, 219, 220, 221, 258, 260, 262, 267, 268, 270, equipment, 157, 160
272, 273, 279, 293, 334 erosion, 340, 349
economic efficiency, x, 116, 118, 119, 155, 163, 244 ethology, 265
economic values, 136, 159 EU, 43, 46, 47, 64, 261
economics, 225, 237, 241, 243, 247, 248, 249, 252, Europe, viii, 63, 65, 117, 188, 189, 194, 195, 371
253, 254, 255, 256, 257 European Community, 47, 334
ecosystem, x, xi, xii, 37, 42, 69, 73, 87, 89, 91, 100, European Social Fund, 194
116, 202, 205, 206, 207, 208, 209, 210, 214, 215, European Union, 355
216, 217, 218, 219, 220, 222, 267, 270, 286, 289, evapotranspiration, 336, 343, 349
318, 334, 381 evidence, xiii, 92, 109, 208, 325
editors, 110, 194 evolution, xi, 33, 87, 88, 110, 204, 213, 223, 231,
effluent, 2 258, 260, 262, 265, 285, 298, 306, 307, 308, 309,
effluents, 42 311, 316, 319, 337, 347, 350, 355
egg, 101, 231, 232, 318 excitation, 293, 304, 306, 308, 312, 337
elaboration, 220, 227, 238 exclusion, 124, 184, 194, 208
electric current, 343 execution, 240
electric field, 316, 322, 337, 362 expenditures, 243, 253, 259
electrical conductivity, 64 experimental design, 108
electrical properties, 362 expertise, 109
electricity, 343, 355, 361, 370, 371, 372, 375, 379, exploitation, 77, 241, 245
383 exposure, 45, 46, 48, 117, 235
electromagnetic, vii, 15, 16, 17, 20, 31 extinction, 68, 76, 191, 195, 286, 322, 342, 343
electromagnetic waves, 17 extraction, ix, 2, 11, 36, 98, 106, 109, 116
electron, 336, 337
elucidation, 263
emission, 193, 292, 318, 336, 355, 356, 364, 369, F
371, 372, 381, 382
fabrication, 360, 364, 378, 379
employment, 109, 118, 128
factor analysis, xi, 80, 177, 198, 205, 206
endangered species, 329
farmers, 46, 340
endothermic, 194
farms, 370
endotherms, 187
fatty acids, 43
energy efficiency, 336, 388
fermentation, 319
energy parameters, 301, 302
fertility, 389
energy transfer, 356
filtration, 349
engineering, 100, 111, 268, 299
financial, 109, 171, 316, 341
England, 167
financial support, 109, 316
entropy, 177, 200, 201, 390
Finland, 34, 41, 64, 116
entropy model, 200
fish, 13, 80, 90, 111, 207, 208, 209, 219, 322
environmental change, 174, 238, 247, 286, 338
fisheries, 91, 220
environmental conditions, 87, 90, 174, 177, 192,
fitness, 225, 227, 245, 253, 254, 255, 257, 260, 285,
224, 231, 245, 248, 254, 255, 285
287
environmental degradation, 42
fixation, 346
environmental factors, 11, 185, 187, 188, 189, 190
fixed costs, 157, 160
Environmental Protection Agency, 63, 385
fixed rate, 234
environmental resources, 89, 93, 116, 206
flexibility, 62, 235, 262
environmental variables, 89, 177, 179, 180, 182, 187,
flight, 244, 249, 254, 255
188, 190, 217
flights, 237, 249, 318
EPA, 44, 45, 63
floods, 46
epidemic, 317
flour, 339
equality, 379
flow field, 87
396 Index
lying, 332
K
metazoa, 224 natural black spruce (Picea mariana (Mill) BSP.), ix,
meteor, 356, 370 115
meter, 351 natural habitats, 47
methodology, viii, xi, 32, 41, 43, 46, 60, 62, 84, 85, natural resource management, 158
188, 190, 267, 268, 270, 271, 333 natural resources, 355
Mexico, 386 natural science, 84
mice, 194 natural sciences, 84
microclimate, 384, 390 natural selection, 226, 241, 259
microorganisms, 2, 98 nature conservation, 334
migration, 212 Nd, 358
military, 16, 64 negative outcomes, 116
minors, 226, 228, 234, 235 Netherlands, 194, 219, 387
mission, 37 networking, 351
mixing, xi, 32, 87, 88, 89, 206, 210, 211, 212, 214, neural connection, 104
215, 216 neural network, vii, ix, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11,
model specification, 124, 126, 128, 129 12, 13, 14, 97, 99, 100, 101, 102, 103, 104, 105,
Moderate Resolution Imaging Spectroradiometer, 27 106, 107, 108, 109, 110, 111, 112, 113, 114, 177,
modifications, 69, 229 192
MODIS, 27, 36, 37 neural networks, vii, ix, 2, 10, 11, 12, 13, 14, 97, 99,
modules, x, xi, 115, 118, 129, 142, 223 100, 101, 102, 103, 104, 107, 108, 109, 110, 111,
modulus, 300 112, 113, 177, 192
moisture, 16, 21, 22, 26, 35, 116, 165, 294, 340 neuronal cells, 290
moisture content, 16, 21, 22, 26 neurons, 4, 10, 11, 100, 103, 104, 105, 291, 320
mole, 265 New Zealand, 12, 117
molecular dynamics, xii, 290, 293 next generation, 101, 104, 105
molecules, 17, 21, 292, 320, 321 Nigeria, 188
momentum, 105 NIR, 18, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32
monomers, xii, 290, 300, 301, 302 NIR spectra, 22, 29
Montana, 202 nitrogen, 2, 11, 13, 31, 36, 39, 89, 170, 214
Morocco, 186, 187, 190, 201 nitrogen dioxide, 2, 13
morphology, 187, 188, 372 nodes, 100
morphometric, 13 non-linear equations, 87
mortality, 91, 100, 116, 117, 120, 122, 128, 141, nonlinear systems, 267, 268, 280, 284, 288, 320, 321
159, 161, 163, 164, 165, 166, 227, 228, 230, 253 normal curve, 183
mortality rate, 161, 227, 228, 230 normal distribution, 255
Moscow, 80, 81 North Africa, 189, 198, 340, 374, 387, 389
Multi Criteria Analysis (MCA), viii, 42, 43 North America, 117, 171, 199, 207, 208, 294
multicellular organisms, xi, 223, 224 Norway, 170
multidimensional, 177, 329 Norway spruce, 170
multiple regression, xi, 3, 107, 130, 200, 205, 206 nuclear membrane, 292
multiple regression analysis, 130 nurses, 253
multivariate analysis, 203 nursing, 230, 231, 234, 252, 253
multivariate statistics, xiii, 325, 326, 333 nutrient, viii, 15, 86, 88, 89, 207, 208, 209, 210, 213,
museum collections, 178, 180 214, 215, 218, 219, 231, 233, 241, 259, 273, 347,
mutation, 285, 286, 287, 321 348, 383
mutation rate, 285 nutrients, 43, 64, 86, 116, 165, 174, 206, 209, 213,
214, 215, 217, 221, 224, 225, 232, 239
nutrition, 321, 338, 340, 383, 385, 387
N
nanometer, 292 O
nanometers, 301
National Research Council, 388 observed behavior, 225, 226
native species, 193 obstacles, 363, 373
400 Index
population size, 66, 67, 68, 69, 72, 73, 76, 78, 79,
105, 229
Q
population structure, 68, 78, 227, 228, 255, 257
quantification, 35, 348, 351
porosity, 382
quantization, 12
Portugal, 173, 186, 190, 191, 192, 194, 195, 196,
quantum mechanics, 185
199, 202
positive correlation, 24
positive feedback, 236, 238, 245, 263 R
positive relationship, 208
potassium, 42, 350 radial distance, 126
potato, 339 Radiation, 31, 35, 335, 338, 356, 357, 359, 383, 385
precedent, 339, 345, 348, 359, 362, 379, 380, 381 radiometric vegetation indices, vii, viii, 15, 16, 22,
precipitation, xiii, 44, 47, 184, 187, 188, 192, 335, 32, 33
336, 337, 340, 370, 373 radius, 373, 376
predation, 90, 92, 175, 178, 188, 215, 220, 328 rain forest, 386
predators, 67, 70, 90, 92, 220, 260 rainfall, 184
predictability, 3 rainforest, 340, 386
prediction models, 169, 390 random walk, 88, 92
predictor variables, ix, 97, 103, 133, 134, 183 rate of return, 156
preparation, 49, 50 reactions, 78, 225, 286
preservation, 327, 328 real time, 72, 301, 354
pressure gradient, 212 reality, 75, 77, 180
prevention, 43 reasoning, 226, 227, 230, 241, 245, 253, 254, 258,
principal component analysis, xiii, 181, 325 259, 276, 338, 345, 354, 360
principles, viii, 15, 16, 32, 84, 99, 228, 265 recall, 268, 270, 272, 278
probability, vii, xii, 1, 2, 44, 45, 47, 48, 49, 51, 53, recalling, 268
54, 55, 56, 57, 66, 75, 76, 93, 123, 124, 129, 174, reception, 239, 253, 367
179, 194, 198, 229, 230, 234, 235, 238, 239, 242, receptors, 44, 45, 46, 47, 48, 52, 53, 58, 59, 60, 61,
254, 290, 293, 302, 303, 304, 306, 307, 308, 309, 292, 297
313, 316, 318, 352, 374 recognition, 14, 296
probability density function, 123, 129, 303, 374 reconciliation, 199
probability distribution, vii, xii, 1, 2, 290, 293, 302, reconstruction, xi, 93, 206, 216, 218, 286
304, 306, 307, 308, 309, 313, 316, 318 recovery, x, 115, 117, 136, 137, 138, 142, 160, 169,
probability distribution functions, vii, 1, 2 170, 172, 241
probability theory, 2, 306 redevelopment, 64
problem-solving, 84 redundancy, 10
producers, xi, 205, 209, 216 reflectance spectra, 20
production costs, 100 reflectivity, 17, 21, 342, 354
profit, 356, 366 refractive index, 21
profitability, 242 regeneration, 119, 121, 157, 160, 161, 371, 386
prognosis, 326, 332 regionalization, 202
programming, 104, 105, 107, 110 regression analysis, 124, 125, 126, 127, 128, 129,
project, 11, 63, 182, 183, 186, 189, 193, 203, 261, 131, 132, 134, 135, 138, 139
371, 372 regression method, 102, 107
propagation, 19, 338, 361 regression model, ix, 14, 97, 98, 102, 103, 106, 107,
proportionality, 164 109
proposition, 165 rehabilitation, 387
protected areas, 46, 49, 52, 55, 57, 59, 60, 61, 191, reinforcement, 235, 236, 239, 254, 255, 256, 265
192, 193, 326, 334 relatives, 375
protection, 43, 185, 191 relaxation, xii, 290, 293, 304, 305, 312, 313, 314,
proteins, 292 315, 316, 318
pruning, 13 relevance, 182
pulp, 151, 169, 232, 239 reliability, 226
402 Index
relief, 24
remote sensing, vii, viii, 15, 16, 17, 18, 19, 20, 22,
S
31, 32, 33, 34, 36, 37, 202, 333
salts, 42
renewable energy, 336, 341, 369, 371, 372, 382
saturation, 28, 30, 31, 303, 304, 305
repair, 88
sawdust, 136
replication, 119, 286, 293, 304, 316
scaling, 50, 51, 62, 128, 164, 168, 172, 316
reproduction, 91, 197, 224, 231, 233, 241, 253
scaling law, 164, 172
reproductive organs, 346
scaling relations, 128
reptile, 181, 187, 191, 193
scaling relationships, 128
reptile species, 181, 187, 193
scarcity, 381
requirements, 10, 67, 90, 99, 112, 116, 151, 159,
scattering, 17, 21, 27, 32
188, 190, 380, 383, 388
science, vii, x, 116, 164, 166, 240, 299, 386
research institutions, 63
scope, 84, 214
researchers, 340, 346, 349, 365, 367, 373, 374
Scots pine, 117, 171
residual error, 125
sea level, 371
residuals, 67, 73, 75, 77, 78, 103, 107, 124
seasonal changes, 86, 231, 259
resilience, 220
seasonality, 89
resistance, 30, 320, 340, 343, 344, 345, 361, 365,
security, 336, 340, 347
368, 378, 379
sediment, xi, 2, 63, 206, 208, 215, 216
resolution, ix, 35, 97, 99, 178, 180, 192, 193, 195,
sedimentation, 44, 47
228, 337, 356, 359, 362, 368, 373, 377, 382
sediments, 47, 206, 210, 215
resource allocation, 260
seed, 240, 263, 388
resource management, 166, 258, 264
segregation, 190
resources, ix, 39, 85, 86, 89, 93, 97, 98, 99, 113, 138,
self-organization, xi, 223, 224, 241, 264, 293
164, 165, 166, 232, 239, 258, 260, 293, 304, 326,
self-regulation, 224, 225
338, 351, 371, 381
sellers, 371
response surface model, vii, 1, 2, 5, 8, 9, 10
semiconductor, 319
responsiveness, 246
sensing, vii, viii, 15, 16, 19, 20, 32, 33, 38, 39, 384
restoration, 64, 219
sensitivity, xi, 11, 12, 13, 24, 28, 29, 30, 92, 184,
restrictions, 178, 259
205, 207, 209, 232, 235, 294
rhythm, 350
sensitization, 235
rice field, 100
sensors, viii, 15, 16, 31, 33
risk, viii, 13, 41, 42, 43, 44, 45, 46, 49, 50, 52, 53,
sequencing, 293
54, 56, 60, 61, 62, 63, 64, 92, 101, 113, 159, 169,
services, x, xiii, 116, 335, 336, 355, 356, 360, 365,
182, 183, 191, 195, 207, 220, 241
370
risk assessment, viii, 41, 42, 43, 44, 46, 49, 61, 62,
sewage, 64
63, 64, 220
sex, 68, 91, 294, 296, 317
risk management, viii, 42, 44, 64
shade, 87
risks, 63, 217, 347, 382
shallow lakes, 210
river basins, 63
shape, xiii, 6, 19, 35, 102, 120, 123, 127, 129, 130,
RNA, 292, 293, 317
142, 150, 158, 303, 309, 325, 327, 328, 329, 330,
room temperature, 101
334, 362, 374, 375
root, 85, 103, 105, 137
shear, 211, 213
roots, 337, 346, 347, 388
sheep, 113, 319, 383
Roses, 339
shelter, 340
roughness, 18, 372, 373
showing, 87, 93, 182, 239, 245, 284
routes, 188
shrimp, 14
Royal Society, 37, 111, 113, 261, 265
shrubs, 339, 342, 387
rules, ix, 85, 92, 93, 98, 104, 106, 108, 109, 184,
shyness, 165
203, 237, 240, 258, 261, 262, 380
signal transduction, 317
runoff, 49
signals, xi, xii, 25, 27, 223, 237, 240, 255, 289, 290,
Russia, 65, 289
291, 292, 295, 298, 318, 372
signal-to-noise ratio, xii, 31, 290, 299
Index 403
target, 10, 16, 19, 20, 22, 24, 50, 92, 160, 178, 242, translocation, xii, 289, 290, 292, 300, 302, 303, 316,
246, 247 319, 320, 321, 322
task allocation, 239, 265 transmission, 18, 239, 240, 241, 263, 265, 290, 320,
task performance, 227, 234, 239 322, 363, 369
taxa, x, 86, 90, 174, 187, 191, 192 transparency, 18, 88, 215
techniques, ix, xi, 11, 13, 35, 43, 85, 97, 99, 100, transpiration, 38, 341, 370, 371, 389
120, 124, 151, 170, 177, 200, 202, 205, 206, 207, transport, 37, 50, 87, 160, 211, 213, 215, 263, 292,
209, 259, 333 319, 321, 336, 337, 366, 367, 380, 381
technologies, 371 transportation, 12, 157, 159, 160, 163
technology, 33, 371 treatment, 64, 116, 141, 149, 158, 159, 160, 161,
temperature, 2, 11, 89, 90, 91, 92, 100, 107, 109, 162, 166, 216
110, 111, 176, 184, 187, 188, 192, 207, 209, 214, trial, 104, 127, 295, 331
215, 237, 291, 292, 298, 300, 301, 316, 337, 343, triggers, 234, 253
344, 345, 351, 357, 361, 364, 367, 368, 369, 371, tropical forests, 340
373, 376, 377, 386 turbulence, 88, 92, 210, 211, 214, 215, 218, 220
tempo, 240 turbulent mixing, 87, 88, 89, 90, 211
tension, 361 turtle, 197
tensions, 343 two-dimensional space, 6
territorial, 333
territory, 176, 184
testing, xiii, 13, 105, 113, 253, 335, 336, 389 U
texture, 351
UK, 205, 209, 219, 221, 334, 341, 389
therapy, 292, 293, 316, 320
Ukraine, 205
thermal resistance, 360, 386
UN, 341, 385, 390
thinning, x, 115, 116, 117, 119, 121, 122, 141, 149,
underlying mechanisms, 209
155, 156, 157, 158, 159, 160, 161, 162, 163, 164,
UNESCO, 341, 386
165, 168, 169, 170, 171, 172
uniform, 93, 180, 182, 255
threats, 202
United, 110, 113, 168, 173, 341
three-dimensional model, 215, 221
United Kingdom, 173
threshold level, 296, 297, 298
United States, 110, 113, 168
tides, 212, 214, 221
urban, 12, 13, 111, 371
timber production, 168
urban population, 111
time lags, 78
USA, 12, 63, 80, 117, 167, 168, 169, 170, 264, 294,
time periods, 90, 259
320, 321, 355, 371, 389, 390
time series, viii, 2, 65, 66, 67, 68, 69, 70, 71, 72, 73,
USSR, 169, 205
74, 75, 76, 77, 78, 79
topology, 179
total energy, 355 V
total product, 136, 157
tourism, 334 vacancies, 369
tracks, 141, 258 Valencia, 201
trade, 226 validation, vii, 1, 6, 8, 9, 10, 104, 105, 184, 197, 352,
trade-off, 226 382, 389
training, xi, 10, 11, 13, 64, 103, 104, 105, 206, 216, variable costs, 157, 159, 160
337 variations, 21, 24, 26, 27, 86, 87, 89, 99, 187, 188,
traits, 91, 98, 111, 187, 258, 260 190, 196
trajectory, ix, 65, 68, 69, 72, 75, 76, 77, 138, 141, varieties, 35, 339, 340
157 vector, 12, 67, 69, 149, 150, 154, 230, 271, 280, 338,
transducer, 294, 295 379
transformation, 30, 31, 125, 132, 134, 135, 138, 139, vegetables, 339
306, 383 vegetation, vii, xii, 12, 15, 16, 20, 21, 22, 23, 24, 25,
transformation processes, 383 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38,
39, 157, 169, 325, 326, 328, 333, 336, 338, 340,
Index 405
341, 342, 343, 344, 347, 350, 352, 354, 370, 372, wide diameter, 150
382, 386 wildfire, 119, 121
vegetation cover fraction, viii, 15 wildlife, 116, 195, 196, 203
vegetation monitoring, vii, 15 wind farm, 378, 381, 382
vehicles, 86 wind maps, 377, 382
velocity, xiii, 47, 49, 210, 211, 213, 218, 274, 335, wind speeds, 374
336, 337, 349, 370, 371, 375 wind turbines, 372, 381, 389
ventilation, 363, 369 windows, 17, 18
vertebrates, 200, 201 Wisconsin, 35
Viking, 37 wood, x, 115, 117, 119, 138, 139, 142, 143, 154,
viscosity, 211 158, 163, 169, 294, 340, 382
vision, 260 wood density, 119, 138, 139, 142, 143, 154, 158,
volumetric yields, x, 115, 143, 158, 163 163, 169
vulnerability, 191, 192 woodland, 219
workers, xi, 223, 224, 225, 226, 227, 228, 229, 230,
231, 232, 233, 234, 235, 236, 237, 238, 240, 241,
W 250, 253, 254, 255, 258, 262
workforce, 224, 225, 227, 229, 231, 253, 254
Wales, 205
working hours, 365
walking, 253
workload, 227, 229, 234, 235, 236, 250
Washington, 38, 168, 169, 221, 264, 388, 390
worldwide, 340, 371
waste, viii, 42, 44, 47, 48, 51, 226, 245, 246, 336
wastewater, 42, 50, 63
water absorption, 21, 30 X
water chemistry, 218, 219
Water Framework Directive, viii, 41, 43, 64 xylem, 351
water quality, 43, 44, 50, 51, 52, 212
water resources, xiii, 43, 335, 381
water vapor, 17, 357, 372 Y
watershed, viii, 37, 42, 43, 46, 50, 60, 61, 62
wave number, 377 yield, viii, x, 7, 10, 15, 23, 24, 31, 35, 37, 99, 115,
wavelengths, 17, 18, 20, 21, 22 117, 118, 122, 124, 129, 138, 160, 162, 164, 168,
WD, 199 169, 170, 179, 346
wealth, 192
web, 208 Z
weight gain, 349, 350
welfare, 347 zooplankton, ix, 83, 86, 90, 92, 93, 207, 208, 209,
well-being, 362 214, 215, 219
West Africa, 383 zygote, 280
wetlands, 13, 340, 341, 347, 350, 351, 381, 385, 386