You are on page 1of 10

Landscape and Urban Planning, 2 1 ( 199 1) 8 l-90

Elsevier Science Publishers B.V.. Amsterdam

81

he utility of fractal geometry in la dscape design


Bruce T. Milne
Department qf Biology, Universit_v
of New Me_Gco,Albuquerque. NM 8 713 I, ClS.4
(Accepted 14 September 1990)

ABSTRACT
Milne, B.T., 199 1. The utility of fractal geometry in landscape design. Landscape CrbanPlann., 2 1: 8 l-90.
Natural landscapes possess remarkable regularity in the patterning, sizes, shapes, connectedness. and density of patches.
Landscape ecological studies that explore the fractal geometry of nature have found many examples of consistent variation
in landscape pattern with scale. Fractal landscape patterns affect the distributions and movements of animals, and therefore may be an important characteristic to include in designed landscapes. A computerized design system enables fractals
to be created interactively and automatically overlayed on images or existing maps. The fractal designs are then used in
simulations of foraging animals to determir
how the new design affects the movements, energetics, and locations of
species. The combination of computerized frdctal designs and ecological simulation models may enhance both the ecological relevance and aesthetic value of the designed landscape.

The planner or landscape designer seldom


knows how the designed landscape affects the
movements of animals, the distribution of species, the rates at which resources are depleted,
or distu.rbance and community assembly. Although some quantities such as the number of
bird species in forest islands may be estimated
empirically (e.g. Van Dorp and Opdam, 1987 ),
no simple way exists for determining how
landscape structure (e.g. hedgerow intersections, Forman and Baudry, 1984) modifies
species distributions. Even if such procedures
were available for all landscapes and all species, the design and construction of new landscapes could be enhanced further by simple and
rapid procedures for configuring landscape
elements (e.g. trees, fields, boulder outcrops)
in ways that emulate fundamental characteristics of natural landscapes. Here, the fundamental characteristic of interest is the remarkably consistent way in which patch density,

0 169-2046/9 l/$03.50

forage biomass, patch edge length, and patch


shape vary as a function of the resolution, or
scale, at which measurements are made (Mandelbrot, 1983; Milne, 1988, 199 la; Voss,
1988).
Ecological phenomena such as the spread of
fire, nutrient redistribution, hydrological cycling, pollutant flow, predation, and nest parasitism are affected by landscape geometry
(Risser et al., 1984; Forman and Godron,
1986; Turner, 1987). Often, these ecological
interactions are modified by several factors
that operate simultaneously at different scales.
For example, Gambel oak is common at dry,
southerly latitudes, but paradoxically it reaches
grecttest abundance at higher, wetter elevations
Neilson and Wullstein, 1983 ). Interactions
between m oisture availability and factors affecting oak seedling establishment make water
a poor. predictor of oak abundance. Indeed,
landscape patterns are regulated by many processes that ;*ange from the day-long process of
seedling germination, to successional changes

0 199 1 Elsevier Science Publishers B.V. All rights reserved.

R.T.MILNE

82

lasting decades after a fire, to the variation of


the solar constant stretching over millennia
(Delcourt et al., 1983). Spatially, these and
other processes may impact square centimeters or the entire globe. As a rule, short-lived
processes (e.g. seedling establishment ) OCCUPY
small areas, whereas persistent processes leave
marks over hundreds of square kilometers.
Thus, landscapes are affected by processes that
vary across many temporal and spatial scales.
Animals and plants interact with a given
landscape at a variety of scales (Woodward,
1987 ). Animal species differ in home range
area, density, metabolic rates, ingestion rates,
reproductive rates, and locomotion, all of
which are readily predicted from measurements of body mass (Peters, 1983). Consequently, body mass is a surrogate for the scale
at which mammals, birds, grazing animals, and
reptiles interact with the environment (Milne
et al., 199 I ). Put another way, a given landscape pattern, be it natural or designed, is perceived quite differently by various species, depending on the scale at which they use the
landscape (Morse et al., 1985 ),
The regularity of temporal, spatial, and biotic
scaling relationships makes it possible to assess whether a .andscape and the organisms
contained withi it function in a way that is
compatible with the natural scales that the organisms are adapted to. For example, the Serengeti, with its hundreds of thousands of hectares and multitudes of grazing animals, is one
of the last remaining landscapes in the world
that supports 2 megafauna one would expect in
a large natural g;assland; the Serengeti ecosystem persists because the region is the correct
size for supporting large beasts.
In contrast, a 100 kg animal such as a sabl?
antelope at a typical zoo is granted less than
0.15% of the 3 km* the animal would occupy
in the wild (see I-Iarestad and Bunnell, 1979 ) .
Contrasting a cattle ranch with a dairy farm reveals similar discrepancies of scale. The western ranch, with perhaps 60 ha per animal, is
much closer to the scaling of a natural herbi-

vore system than the dairy. The dairy farm is


successful only because the farmer imports or
grows sufficient feed to pack the animals into
a fraction of their natural home range area. Artificial systems can only be maintained if an
external source of food energy and supplemental resource cycling procedures are built in.
Every departure from the natural scale exacts
a cost in management effort.
Below, a general method for generating naturally scaled fractal landscape patterns is presented. Fractal designs preserve many aspects
of natural landscapes by ensuring that the spatial scaling of the landscape is built in. The designs increase the chance that temporal and
biotic scaling will follow. After constructing the
fractal design, a naturally scaled simulation of
mammalian grazing can be used to explore the
potential biological implications of the design.
Together, the design and simulation provide
the designer with ecologically meaningful
interpretations of the new landscape. This approach to design focuses on the pervasive fractal character of the landscape, and attempts to
maintain a corlsistent change in ecological
phenomena with scale, rather than attempting
to design and control the smallest nuances of
landscape function and structure.
TAL G

LA

Fractal geometry is one of the widest reaching mathematical developments of the 20th
century. Mandelbrot ( 1983) and others have
observed striking regularities in the shapes of
coastlines, the perimeters of rain clouds. the
roughness of terrain, and the shapes of plants
(Burrough, 198 1; ovejoy, 1982; Feder, 1988).
The regularity is readily described by fractals,
which are mathematical representations of nature. Fractals often exhibit statistically regular
patterns that occur when a quantity such as
patch density at one location is predictable
from measu.rements of the same quantity
nearby. For example, the coast of Maine is
characterized by peninsulas, many of which
have smaller peninsulas attached, ad infini-

UTILITY OF FRACTAL GEOMETRY

IN LANDSCAPE DESIGN

83

turn. The fractal geometry of the coastline implies that the presence of a peninsula, 15 km
long, ensures the presence of yet smaller spits
jutting off from the large one (Milne, 199 1a).
More significantly, coastlines and other patterns that exhibit fractal structure obey strict
relationships between the scale of observation
and the quantity measured (Mandelbrot, 1983;
Stanley, 1986 ) . For example, fractal scaling
was found in analyses of remotely sensed imagery obtained for the Sevilleta National Wildlife Refuge in New Mexico. Regions occupied
by 0- 10% grass cover exhibited regular decreases in grass density as the spatial resolution was decreased (Milne, 199 1a). However,
areas trampled heavily by cattle 15 years before the study, and then left ungrazed by cattle,
did not exhibit simple fractal relationships
(Milne, 199 1c). Natural grasslands without
heavy disturbance exhibited archetypal fractal
relationships while disturbed areas did not.
This contrast suggests that natural configurations of plants and resources, similar to those
found in undisturbed landscapes, can be created by using fractal patterns to design
landscapes.
There are many alternative scaling relationships for a given landscape. For example, one
could examine the relatifDnships between patch
area and perimeter with changes in area
(Krummel et al., 1987; ONeill et al., 1988),
the density of convolutions on the coastline, or
the density of pixels on a digital image of a
landscape (Milne, 1988, 1991a). Each of the
quantities (i.e. perimeter to area ratio, convolutions per kilometer, and density) are predicted to vary as a function of the length scale
used, according to the ge era1 fractal scaling
relation
Q(L)=kL

(lb

where Q(L) is the quantity measured using a


length scale L, k is a constant, and D is the fractal dimension of the quantity (Stanley, 1986:,
Voss, 1988 ) . This expression applies in all cases
considered here. If animals and plants are en-

visioned as operating at different scales, due to


differences in home range area or dispersal
distance, then eqn. ( 1) describes how the species perceptions of a resource vary with scale.
For landscape analysis, a versatile fractal
method has been developed that provides both
an estimate of the fractal dimension describing
the concentration of pixels of a given cover type
and a visualization of scale-dependent pixel
density on a digital representation of the landscape (Milne, 199 1a). The method entails
constructing a digital image of a cover type or
class of interest (e.g. forest ). Then a square
window is centered on each pixel of the class,
and the number of pixels of the class within
each window is counted. The procedure yields
the frequencies at which m = 1, 2, ... L2 pixels
of the class are found on the image within windows of length L. The counts are formed for a
series of L values ranging from very small (e.g.
3 ) to large (e.g. 3 1- 10 1) . By transforming the
frequencies to probabilities, the probability
density function

(2)

Pb%L)=l

is formed, which requires that the sum of the


probabilities p@z,L)
of finding m pixels in
windows of length L equals 1.Osummed across
all values of m from 1 to n(L); n(L)SL
(Voss, 1988 ). Next, the quantity to be used in
eqn. ( 1) to estimate the parameters k and D is
the sum of each p( m,L) value multiplied by
the respective m value
n(L)

Q(L)=

tn= I

uzp(mJA

(3)

Thus, Q(L) is the expected number of pixels


(i.e. the first moment of the distribution )
found in a window of length L, and Q(L) increases as a power of L with an exponent equal
to the mass fractal dimension of the mosaic of
pixels (Voss, 1988 ). The mass fractal dimension and the constant k describe how the number of pixels varies with scale.
Spatial information about the scale dependence of pixel density can be represented by

B.T.MlLNE

84

tallying the number of neighboring pixels of a


given class included in each window of a given
length. By displaying the respective tallies for
each pixel on a gray scale image, high numbers
of neighbors are revealed for dense clusters of
pixels. and few neighbors are found for sparsely
clustered pixels. If this analysis is performed
for each of three values of L, and the brightness of each gray-scale image is resealed such
that the maximum total number of visits by
windows of length L is displayed as brightly as
the image display permits, and the three images are overlayed in the red, green, and blue
color planes of the image processor, then the
resulting image reveals the spatial pattern of
scale-dependent density at each scale (Milne,
199 1b). Theoretically, this analysis reveals
where, on a fractal distribution of food resources, animals with home ranges of length L
are likely to go to find food that is at a high
density when perceived at the animals characteristic scale. Interestingly, the image also
shows that animais operating at different scales
are likely to go to different regions to find high
concentrations of resources.
GN

CTA

SCAPES

An interactive method is required to design


fractals. A useful method should facilitate
creativity while maintaining the scaling behavior of naturally occurring patterns. Bamsley
( 1988a, b) developed the mathematics and
software for so-called iterated function systems (IFS) that allow fractals of an infinite
variety to be constructed easily (Fig. 1). By
coupling the IFS method with existing corn-puter-aided design software, geographic information systems, or remote sensing image analysis packages, the IFSs become another tool for
producing designs with aesthetic value and
ecologically meaningful structure.
Iterated function systems work by rotating,
stretching, and translocating points in a coordinate system by means of simple equations,
not at all unlike the equations used to compute

, i

(/

djj
,,

1=
..i
4 ...

_,

(A)

ii,
:(.I

I!I

Fig. 1. Exampies of fractals generated using the iterated function system method. The panels illustrate the continuous dependence of the pattern on small chdnges in the IFS. Each
panel was made by slightly alteringthe functions used to create panel (A).

perspective
projections.
Surprisingly,
the
equations of the IFS .may be applied randomly
to a set of starting coordinates, and after many
iterations, the set of transformed points may
take on a spatial configuration reminiscent of
natural patterns. Each of the k= 1,2, ... n functions of the IFS is of the following form

w(k)

=[,:;:I=[;
;][;]+E]

in which the original coordinates of a point


specified by x and y are transformed by the rotational matrix containing the parameters a, 6,
c, and d, and then translocated (i.e. moved laterally or verticaily ) by the vector containing
parameters e andJ The design of the IFS and
the pattern it creates are controlled by the parameter values. Applying the rules of matrix
multiplication and addition, the new values of
x and y are
x = ax+by+e
and
y'=cx+dy+/-

UTILITY OF FRACTAL GEOMETRY IN LANDSCAPE DESIGN

85

Thus, after one iteration, each of the points in


the original square boundary of Fig. 2 (A) take
on coordinates indicated by the polygon within
the original boundary. The dots on the curved
lines of Fig. 2 (A) are the successive locations
of the vertices of the polygon for several iterations of the transformation.
The rotation,
stretching, and translocation in Fig. 2 (A) occur if the parameters are set such that a= 0.6,
b=O.l, c=O.2, d=0.4, e=O.l, andf=0.3. After
many iterations, the polygon converges to the
so-called fixed poirt shown inside of the polygon. The fixed point is the point whose value

FPI

is not affected by the function, and it may be


thought of as an attractor of all points within
the original boundary; points travel along paths
whose shapes are determined by the lunction.
The function effectively warps the space within
the coordinate frame such that the shortest
distance along which a transformed point
moves towards the fixed pointis not necessarily straight, but may be curved.
An IFS with several functions leads to much
more complex patterns, because the functions
are applied randomly to the current coordinates of a point (Fig. 2 (B ) ). The curved lines

FP2

Fig. 2. Graphical explanation of how iterated function systems create fractal patterns. (A) Transformation of the original boundary by a pair of equations shrinks and rotates the boundary to create a new polygon (transformed boundary). Reiteration of the
transformation projects all points toward the fixed point. (B) The fixed points produced by two functions, showing how trajectories toward fixed points vary between functions. (C) A series of points created by applying functions 1 and 2 in the order 1,2,
I, 2,2, 1. Dashed lines represent the trajectory each point would make if it were subjected to repeated iterations of the respective
functions. Intersections indicate where the procedure switched from one function to the next. (D) The resulting fractal pattern
after 500 iterations.

86

in Fig. 2 (B) indicate that function 1, with fixed


point 1, tends to attract points originating
from within the coordinate frame towards the
fixed point along routes that are shaped quite
differently than the routes leading to fixed
point 2, which is the fixed point of the second
function. Rarely will the fixed points of each
function be equal, and consequently the application of a randomly chosen function forces
points off the fixed points, thus perpetuating
the series of transformations.
A point follows a jagged trajectory if the
transformations are applied in the order 1, 2,
1,2,2, 1 (Fig. 2 (C) ). The point begins at position 0 at the top of the coordinate frame, and
is transformed using function 1, which drives
the point closer to fixed point 1. Then, in a
manner that is analogous to flipping a coin, a

B.T. MILNE

random number is used to choose function 2,


thereby resulting in a new trajectory towards
fixed point 2, but only as far as point 2. Random selection of function 1 drives the point toward the first fixed point, and so on. Once the
transient behavior of the trajectory disappears, the point oscillates back and forth in the
region between the fixed points, and traces out
a pattern (Fig. 2 (D) ). Thus, the parameters
of the functions determine all possible routes
the point takes to create the fractal pattern, and
manipulation of the parameters regulates the
final pattern.
Fractals created by an IFS have four fundamental properties that are of interest to the designer, and a fifth characteristic of ecological
significance. First, a small change in the parameters of the IFS results in small, continu-

Fig. 3. Fractal created using a three-function IFS and then symbolizing each point based on the number of the last functionused
the point. The straight lines connect consecutive points to illustrate thar the image is not created one row or columnat
a time.

to create

UTILITY OF FRACTAL GEOMETRY IN LANDSCAPE DESIGN

87

ous changes in the fractal pattern produced,


thereby
facilitating
animation
(Barnsley,
1988b). A wide range of different patterns can
be produced from simple manipulations of the
IFS (Fig. 1).
Second, each point on the attractor has an
address. For the designer, the addresses provide a means of assigning different qualities to
each point. The address is a list describing the
order in which the various functions w(k) for
k= 1,2, 3, . .. n were chosen randomly to create
the point. For example, the points can be symbolized based on the last function that was applied to create a point. Thus, a point created
by the string of functions in the order 2,2,1,3,1
could be represented by the numeral 1 to indicate that function w( 1) was used last. Substituting different colors, symbols, landscape elements (e.g. tree species, land cover types), or
even other IFS for the numerals provides a texturing scheme that maintains a high degree of
fidelity with the overall pattern (Fig. 3 ). The
hierarchical structure of the fractal can be in-

-200
c

it
0100.
3

A
A
A

- -0

66

15

50

Number

-Ip

2'0
length

Windcw

7
0.0

tm

_100

of

152

2oc)

piwls

Fig. 4. Statistical behavior of the fractal used to make Fig. 3.


showing the distribution of the number of pixels found in 3 X 3,
7 x 7, and 15 x 15 pixel windows. The inset illustrates the scale
dependence of the expected number of pixels (i.e. the first
moment of the frequency distributions)
in a window as a
function of window width; squares represent an analysis of
the pattern in Fig. I (A), triangles show the results for the
fractal used to make Fig. 3.

corporated into the design by using the penultimate function number to color a pixel. This
produces a much more intricate texture, with
each of the n possible symbols nested within
the prominent clusters that would have appeared if the points had been mapped using the
symbol for the last function, rather than the
penultimate function.
Third, the fractal can be magnified, subject
to the numerical accuracy of the computer, to
reveal additional details at high resolution.
Thus, one imagines zooming in on a fractal that
represents different cover types to reveal individual trees within the cover types (Bamsley,
1988b ). A sophisticated rendering of the image would enable oblique views at any scale.
Fourth, the fractal created by an IFS exhibits the same kind of scaling found in natural
landscape mosaics. For example, the pattern in
Fig. 3 exhibited a strong dependence on the
number of image pixels found within sampling
windows of various lengths (Fig. 4). The effect of scale was much greater for the image in
Fig. 3 than it was for Fig. 1 (A), as indicated
by the slope of the curves (Fig. 4, inset ). Ecologically, the scale dependence suggests that
organisms that gather resources with different
home range areas will experience exponentially different concentrations
of resources
(represented bv the pixels) in a given fractal
landscape (M&e et al., 1989). The reason for
this difference relates to the low fractal dimension of 1.1 for Fig. 1 (A ), indicating a very
simple, linear structure of the fractal (recall
that lines are one-dimensional).
In contrast,
the dimension of 1.4 for the fractal used to
construct Fig. 3 indicates the greater tendency
for the fractal to fill the plane, thereby offering
much greater variation in the clustering of pixels at all scales.
Finally, fractals generated in this manner are
necessarily phenomenological. In nature, fractals such as snowflakes and dendritic patterns
are thought to be formed by microscopic mechanical processes (Nittman and Stanley,
1986; Meakin, 1988 ). For example. dendritic

88

can be explained as the steady formation of macroscopic features from microscopic


asymmetries, which occur when a molecule is
added to the exterior of a growing mass of molecules. Any variation in the process that leaves
more molecules in one growing branch than
another biases the future addition of molecules to the larger mass. Thus, the dendritic
feature grows into a shape reflecting the historical events which led to its creation (Nittman
and Stanley, 1985; Meakin, 1988). The IFS
does not have the mechanistic qualities characteristic of natural fractal patterns. Rather, an
IFS produces patterns that fortuitously resemble familiar objects, or objects of the
imagination.
features

ECOLOGICAL CONSEQUENCES:
ALLOMETRIC HERBIVORY
Ultimately, any design should be evaluated
to determine its ecological function relative to
species distributions or the flows of resources
through the landscape. Each landscape design
implies specific interactions with water and
nutrient runoff from one area to the next, as
the wooded vegetation intervening between
farm fields, for example, may filter nitrogen
and phosphorus, thus reducing pollutants in
water courses (Correll, I 983 ) . Likewise, environmental gradients a
e lobes of wooded
peninsulas may alter the distribution of species, and thereby regulate the number of species (Milne and Forman, 1986). Finally, the
economics of foraging may differ between species living in the same landscape, depending on
the scale at which the animals gather resources
(Milne, et al., 1989, 1991).
A model of allometric herbivory was developed to assess the consequences of fractal
landscape patterns on the foraging success of
small mammalian herbivores (Milne et al.,
199 1). Allometric herbivory refers to the way
in which the scale at which animals forage is
predictable from the animals body mass. For
example, small mammals are present at much

B.T. MILNE

greater density than large mammals (Peters,


1983 ), they have smaller home ranges, and
much higher metabolic rates. These, and other
ecological, behavioral, and physiological characteristics were used to regulate animal density, home range area, speed of movement,
ingestion rate, metabolic rate, and the time
scales at which the species foraged in a simulation model.
The animals were placed randomly on the
landscape and allowed to eat within a fractal
pattern. The movements of animals throughout the landscape were determined by a simple
rule. If the animal obtained enough forage
within its home range to offset its metabolic
costs, then it was allowed to remain in place
and feed until the next time period. However,
animals finding insufficient food were allowed
to move in a random direction to another home
range location.
Successful animals tended to cluster together on the landscape in areas with high concentrations of food, while hungry animals
moved about in search of food (Fig. 5). The
fact that 3 kg animals used much smaller home
ranges, lived at higher densities, and had higher
metabolic costs resulted in a rapid depletion of
resources on the landscape (Fig. 5 ). Large an-

CJ 60,
,-E
:

40 L
0

i125000

Time

__~_

(win)

.__l-25ocoo

_-.I

Fig. 5. Variation in the percent qf simulated animals foraging


in a fractal landscape as modeled for 180 days.

UTILITY OF FRACTAL GEOMETRY IN LANDSCAPE DESIGN

imals were more mobile on average, reflecting


the difficulty they experienced in locating high
concentrations of food within their larger home
ranges. Thus, the fractal geometry of the landscape resulted in very different perceived patterns of food distribution for each species, and
consequently altered the foraging success of
animals that operated at different scales.
CONCLUSIONS
In nature, complex landscape patterns are
formed by a multitude of processes, including
the birth and death of plants, soil erosion,
grazing, and disturbance by fire. The interactions between random events and constraints
on growth processes create patterns of seemingly infinite complexity (e.g. Stanley, 1986 )
that exhibit remarkably consistent variation
with scale, e.g. Fig. 4. While foraging or searching for mates and nest sites, organisms are confronted with heterogeneous distributions of resources or habitat; the fractal geometry of the
distributions may alter movement patterns
(Wiens and Milne, 1989) or abundance
(Morse et al., I985 ). Theoretically, the magnitude of the effect on animals is related to the
scale at which
species
perceive
the
environment.
Tools are needed both for the analysis of
landscape designs and for the creation of landscape patterns that have geometrical regularity
similar to the regularity found in natural landscapes. Ecological research is needed to determine the consequences of large changes in resource density with scale vs. small changes with
scale. Until such research is complete, simulation models based on allometric herbivory, or
other properly scaled ecological processes, are
one way in which ecologically sound assumptions about animal activity can be coupled with
computer-aided design packages to evaluate
whether designed landscapes have desirable
ecological characteristics. The generation of
fractal patterns by iterated function systems
should be regarded as one of the best available

89

techniques for creating patterns with the same


statistical properties observed in natural
landscapes.
ACKNOWLEDGMENTS
I thank A.R. Johnson and Y. Marinakis for
helping decipher the mathematics of iterated
function systems. The comments of an anonymous reviewer helped clarify an early manuscript. This work was supported by the National Science Foundation (grant nos. BSR8806435 and BSR-86 1498 1) and the Department of Energy (grant no. DE-FG0488ER607 13 ).
REFERENCES
Barnsley. M.F., 1988a. Fractal modelling of real world images. In: H.-O. Peitgen and D. Saupe (Editors), The Science of Fractal Images. Springer, New York, pp. 2 19-242.
Bamsley, M.F.. 1988b. Fractals Evei ywhere. Academic Press,
New York.
Buirough, P.A., 198 1. Fractal dimensions of landscapes and
other environmental data. Nature, 294: 24 l-243.
Correll, D.L., 1983. N and P in soils and runoff of three coastal
plain land uses. In: R. Lowrance. R. Todd, L. Asmussen
and R. Leonards (Editors). Nutrient Cycling in Agricultural Ecosystems. Univ. Agric. Exp. Stn. Spec. Publ. 23.
pp. 207-224.
Delcourt, H.R.. Delcourt, P.A. and Webb, III, T.. 1983. Dynamic plant ecology: the spectrum of vegetational change
in space and time. Quat. Sci. Rev., 1: 53-175.
Feder, J.. 1988. Fractals. Plenum Press, New York.
Forman, R.T.T. and Baudry. J., 1984. Hedgerows and hedgerow networks in landscape ecology. Environ. Manage., 8:
495-5 10.
Forman, R.T.T. and Godron. M., 1986. Landscape Ecology.
Wiley, New York.
Harestad, A.S. and Bunnell. F.L., 1979. Home range and body
weight-A reevaluation. Ecology, 60: 389-402.
Krummel, J.R., Gardner, R.H., Sugihara, G.. ONeill, R.V.
and Coleman. P.R., 1987. Landscape pattern in a disturbed environment. Oikos, 48: 32 l-324.
Lovejoy, S., 1982. Area-perimeter relation for rain and cloud
areas. Science, 2 16: 185- 187.
Mandelbrot. B., 1983. The Fractal Geometry of Nature. W.H.
Freeman, New York.
Meakin, P., 1988. The growth of fractal aggregates and their
fractal measures. In: C. Domb and J.L. Lebowitz (Editors), Phase Transitions and Critical Phenomena. Academic Press, New York, pp. 335-589.
Milne, B.T., 1988. Measuring the fractal geometry of landscapes. Appl. Math. Comput., 27: 67-79.

H I h-lII.Nt

41)

I Y&ii. I hc
Cambridge
Kisser,

ixolqycd

University
Ksrr.

P.G.,

J.R. and

scrlpc ecology: Directions


hlllnc,

B.T. and Forman.

R.T.T..

Wood! plrlnt dlversltj.


terns. Ecolog!.
Xlllne.

1986. Peninsulas in Maine:

distance.

and environmental

pat-

B.T.. Sohns:ln. K. and Formap.

ofwildlife

deC,endcnt proxtmitb

M.G..

I~)?I. lntc,actlons

19X9. Scale-

IO I - 119.

Ecol.. 2:

the fractal geometry

hcrbivcl),.

Theor.

Popul.

.A.R..

of landBlot.. In

press.
Illorsc.
hq.H

D.K..

Lawton.

J.H..

Dodson.

. 19S5.Fractnl dlmcnslon
of

arthropod

Xmerican

namics. J. Biogeogr..
J. and Stanley,

ir.terfacial

M.M.

;nd WIlliamson.

ot\egeta~lon

body

Icngths.

behavior.

L.H.,

and the d~sNature,

3 14:

1983. Biogeography

of two

oaks in relation

anisotropy.

growth patterns

Nature,

32

I : 663-668.

R.T.T.,

1984.

Land-

111. Nat.

Hist.

IL.

An introduction

to self-similarity
and N. Ostrowskl

( Edi:ors ). t )n Growth and Form: Fractal and Non-Frac-

iurner.

in Physics. Martipus

Nijhoff,

Boston, pp. 2 1-

( Editor ). L987. Landscape Heterogeneity

M.G.

Disturbance.
Van Dorp.

Ecological Studies,

11. and Opdam,

nltles. Landscape
VOU.

R.F..

dy-

without
arising

P.F.M.,

I9)xX

to simulation.

19X7. Efiects of patch silt.


on ;i)rcst bird commu-

I : 59-73.

Ecol.,

Frxtal5

and

64. Sprlngr%r. New York.

and regIor,al aoundzncc


In nature:

In: H.-O.

Fron

characteriration

Peitgcn and D. Saupe (Editors),


Images.

Springer,

New York, pp.

2 t-70.
Wiens, J.A. and Milne,
landscape

1986. Tip splitting

tension and dendritic

from molecular

to atmospheric

10: 275-297.
H.H..

Forman,

In: I1.E. Stanley

The Science of Fractal

Neilson. R.P. and Wullstein.


sollthwest

1386. Form:

tal Patterns

Isolation

trlbution
731-734.

Nittman.

H.E..

SILL.

53.

Wrens. J.,-.. and Johnson,

between

scapes and allometric

R.T.T..

habitat in ;1 spatiall:-neu-

tra1 Ba> cslan model. Landscape


%lllne. B.T.. Turner.

Stanlq,

un3od~

and approaches.

Surv. Spec. Publ. No. 2. (hampaign.


and fractal

67: 967-973.

Illlpllcallol~~

Press. NLW York.

perspective.
Woodward,

B.T..

I%c). Scaling of landscapes ir

ecology. or, landscape


F.I.,

Landscape

1987. Climate

bridge University

ecology from a beeties

Ecol., 3: 87-96.
and Plant Dis;ribution.

Press. New York.

Cam-

You might also like