Professional Documents
Culture Documents
David Ebert†
driven is hotly debated. While this level trends in data sets or interpolate scat-
may be the highest level of application, tered points to a wider distribution
it seems to be less commonly employed pattern. There are two types of point
than the previous two levels. procedures: density mapping and inter-
polation. Density mapping is the creation
Spatial Data of maps showing the distribution of a
There are two main types of spatial data variable of interest across a surface, such
in archaeology: point data and areal as artifacts in a plowed field or sites in
data. Point data includes such things a region. From this sort of mapping, it
as spot locations for artifacts, features, is possible to analyze locational trends,
and archaeological excavation units. although these approaches are gener-
They are single locations identified by ally crude (Kvamme 1988: 339). Density
their three-point provenience. Areal mapping tends not to appear as an
data includes things such as a surface, individual procedure in the archaeologi-
landscape, site, or region. Each of these cal literature, although it is a mainstay
types of data has specific GIS analytical of most archaeological reports as part
procedures, and these, in turn, offer dif- of the overall presentation. Generally
ferent possibilities for management and speaking, density mapping would fall
analysis of archaeological data. into the visualization level of GIS use.
An example of a simple density map is
Point Procedures shown in Figure 1.
Point procedures focus on point loca- It is also possible to use point data to
tions, and are often used to analyze create continuous surface data through
the use of interpolation, which consists of was created, which made it easier to see
a number of mathematical procedures patterns in certain areas.
to convert point distributions to a con- The archaeological literature is
tinuous surface (see Ebert 1998, 2002 somewhat scant in terms of GIS projects
for extended discussion). Many of these that employ interpolation types of point
interpolation techniques are based on procedures. An early study by Zubrow
theories other than probability theory and Harbaugh (1978) employed kriging
(Altschul 1988: 69), such as gravity or to determine archaeological site loca-
density models. An example of interpola- tions using non-GIS methods. While they
tion is kriging, which is based on the prem- obtained good results from a synthetic
ise that things nearby tend to be more site system, their use of kriging was
alike than those further away (Goodchild misguided. Kriging uses the relation-
1996: 243); when maps are interpolated, ship between a continuously distributed
data closer to the spot being interpolated variable to make predictions of values
have greater influence than data further where that variable is unknown, such as
away. An example of a map created by interpolating soil types, elevation values,
interpolation is shown in Figure 2. This or snowfall. However, Zubrow and Har-
map was created from fieldwalking data baugh argue that archaeological sites
gathered during the Als Archaeological can be considered as being continuously
Project (Ebert 1998). The purpose of this distributed. To achieve a continuous
map was to create a distribution map in distribution, they used a binary pres-
order to aid in interpretation of surface ence/absence system, meaning a value
finds. Patterns are often hard to discern of zero is assigned to those areas where
when the data are presented as single there are no sites. Other kriging meth-
points. However, by interpolating the ods, such as co-kriging, would have been
surface finds, a continuous distribution more appropriate (Ebert 1998).
aspect, and distance to water (Kvamme regression, the most commonly used
1985: 218–219; 1992: 25–27; Kvamme statistical interpolator, on lithic density
and Jochim 1989: 5–6), becoming a data from Stonehenge, Wheatley (1996a:
“usual suspects” list of predictors. When 287) could account for only 25% of
choosing variables, the preference is for the variability within the data. This led
variables that are related to site loca- him to question whether this method is
tions, but not correlated to each other appropriate as an interpolator for pat-
(i.e., relatively independent) (Rose and terned human behaviour.
Altschul 1988: 185). The assumption
made in this type of modeling is that Inductive Modeling Methods
non-cultural aspects of environment The weighted map layer approach
will correlate and predict site location has been the most popular inductive
(Marozas and Zack 1990: 165). However, modeling method and makes use of
the tendency of archaeological sites to categorical or class-based map layers,
recur in favourable environmental set- with each category being assigned a
tings has been the basis of how many weight relative to conditions found at
archaeologists have found sites through archaeological sites (Brandt et al. 1992:
their “archaeological gut instinct” and 271; Dalla Bona 1994a). This allows
professional expertise for a long time specific variables to have more influence
(Kuna and Adelsbergerova 1995; Warren over predicted site locations than other
1990b: 201; Warren and Asch 2000). variables. One of the ways that weights
Inductive modeling takes an essen- may be determined is through the use of
tially cultural-ecological view of human multivariate statistical procedures, such
settlement systems (Kohler 1999: 32; as logistic regression (Parker 1985).
Wheatley 1993: 133). The unit of analysis Kvamme (1990) proposes a method of
is the land parcel, not the site (Warren determining the relationship between
1990a: 94). Because sites are compared the distribution of sites and the environ-
against the physical environment of the ment, using statistical methods to eluci-
study area as a whole, we must have the date this relationship. Using one-sample
ability to be able to analyze environmen- statistical tests, the background environ-
tal units, rather than points representing ment (i.e., all grid cells within a study
sites. The identification of correlations area) is treated as a control, and statis-
between known archaeological sites and tical deviations from the distribution
certain attributes, usually aspects of the of environmental features are sought
physical environment, is the primary (Kvamme 1990). For such continuous
goal (Kamermans and Wansleeben 1999: variables as slope, aspect, or distance
225; Kohler 1999: 37). to water, the Kolmogorov Goodness-of-
Inductive models have a number of Fit test is preferable (Kvamme 1990:
limitations (Ebert 2000: 129–137), the 370). One of the major weaknesses of
most significant of which is that (a) their this method is that by simply changing
success is unexplainable and (b) we do weights, exponentially different results
not know how they work (Sebastian and may occur (Brandt et al. 1992: 271). It is
Judge 1988: 5). Inductive modeling has only possible through the use of GIS to
also been criticized for the methods quantify the background environment
employed, especially the statistical test- for a large study area in order to com-
ing. For example, using linear multiple pute this test (Kvamme 1990: 370).
Figure 3 illustrates a map created ing of the model is paramount. This can
for a project I am currently working on, be done solely through laboratory meth-
examining the role of time and predic- ods, such as red flag modeling (Altschul
tive modeling. The map shows the basic 1990), or through statistical evaluation,
divisions of high, medium, and low such as a gain statistic (Kvamme 1988). In
potential. Although not a particularly red flag modeling, sites with anomalous
good predictive model because there settings are examined for possible pre-
are too many cells classified as high and dictive variables that have been missed
medium potential, having too many cells (Altschul 1990). The gain statistic quan-
high and medium potential would also tifies the predictive power of the model,
make the predictive model a poor man- based on a scale of −1 (reverse predic-
agement tool. For example, if all of the tive power) to 1 (very strong predictive
cells were high and medium potential, power). However, laboratory testing is
a predictive model would correctly find generally considered to be less effective
sites 100% of the time. However, that than field testing predictive models.
would not save archaeologists any time or
effort, as it would not allow for any focus Extending Inductive Predictive Modeling
of a potential archaeological survey. An important question in the analysis of
sites concerns how many axes the data
Inductive Model Testing may be divided into in order to examine
Once a model is created, whether by temporal, functional, and spatial dif-
inductive or deductive techniques, test- ferences (Kincaid 1988: 557; Rose and
Figure 3. Example of a predictive model map (black = low potential, grey = medium potential,
white = high potential). Dark circles are known sites from which predictive model was created.
Altschul 1988), given that different types mappable proxies, with archaeologi-
of sites may be associated with different cal site location. However, the role of
sets of variables (Rose and Altschul 1988: social, ideological, and political factors
205). It is unclear if adding this level of has received little attention in the pre-
detail provides any significant advan- diction of site locations (Weimer 1995:
tages. This question has received little 91). While a large body of ethnographic
attention and is currently unresolved. data is available for use in prediction, it
This type of approach is illustrated by has received limited usage despite the
Robert Hasenstab’s (1996: 230) study of potential it has. For example, Dalla Bona
Iroquoian villages. His classification of and Larcombe (1996) incorporated an
sites was based on three types: 1) func- ethnographic data in reconstructing the
tional (villages versus campsites); 2) tem- seasonal round of boreal forest hunter-
poral (five occupation periods); and gatherers for a predictive land-use model
3) spatial (three physiographic/cultural for northern Ontario (see also Larcombe
zones). These were analyzed on the basis 1994). One reasons they chose to use to
of three classes of environmental data: do this was because the ethnographic
1) those related to hunting territories; sources provided information on the
2) those related to maize horticulture; social and spiritual significance of natu-
and 3) those influencing trade, espe- ral resources (Dalla Bona and Larcombe
cially canoe routes (Hasenstab 1996: 1996: 254).
230). The results of this study were that Stancic and Kvamme (1999) also
the data and methods return only tenta- incorporated what they termed social
tive answers, and that it was difficult to variables into their analysis of Bronze
evaluate the results because confound- Age hillforts on the island of Bra in
ing factors, such as autocorrelation and Croatia. These variables included: 1) dis-
covariance between the variables, may tance between hillforts; 2) intervisibility;
play a role (Hasenstab 1996: 238). 3) distance from the sea; and 4) location
of long barrows (Stancic and Kvamme
Criticisms of Inductive Modeling 1999: 234). While at first glance, distance
A variety of concerns have been raised to the ocean seems no different from an
regarding predictive modeling. These environmental predictive variable, in
include questions about the accuracy of this case it is the reverse because hillforts
site locational data(Dalla Bona 1994a: 29; were located a considerable distance
Duncan and Beckman 2000: 55; Ebert from the coast (Stancic and Kvamme
2000), the accuracy of the environmental 1999: 234).
data set (Duncan and Beckman 2000: 55; Predictive models based solely on
Ebert 2000), and how areas of archaeo- environmental considerations do seem
logical potential are defined—something to predict the settlement patterns of
seldom explicitly stated in modeling hunter-gatherers fairly well (Maschner
reports (Dalla Bona 1994a: 15). 1996b: 176). However when the focus
One of the greatest failings of tradi- is on more “complex” social or political
tional inductive modeling projects has forms, such as the complex hunter-gath-
been the lack of any non-environmental erer populations of the Northwest Coast,
predictor variables. The focus of predic- the predictions do not seem to work as
tive modeling has been the correlation well. This phenomenon is likely due to
of environmental variables, as readily the fact that in complex social systems,
Figure 4. A sample viewshed (white dot = observer location, grey = non-visible areas,
black = visible areas).
plexity can be added by calculating a study area, which is visible from the hill-
multiple viewshed, which is calculated forts (Madry and Rakos 1996: 111), and
from the Boolean union of viewsheds indicates that roads were built specifically
from a number of individual locations to be in view of one of the hillforts, rather
(Kvamme 1999: 178). than randomly. Moreover, 59.85% of the
One way that a multiple views- roads were visible from the two main
hed could be applied would be the hillforts in the region, suggesting that
calculation of all of the areas visible roads were built to be observed from the
from a series of hilltops. Cumulative hillforts (Madry and Rakos 1996: 111).
viewsheds are the sum of those taken In addition to these applications,
from a number of individual locations viewsheds have been used with a predic-
(Kvamme 1999: 178; Wheatley 1995). tive capacity. For the Great Lakes area,
The result of a cumulative viewshed is for example, Krist and Brown (1994:
that each cell in the map holds a value of 1130) employed viewsheds to evaluate
locations from which the point is visible. the degree to which precontact hunters
Visibility analysis has been proposed as situated themselves to observe the pre-
a method of bridging the gap between dictable caribou migrations. A cost sur-
current data and social and cognitive face was thus created by them to simulate
landscapes (Lock 1995: 16). possible caribou migrations paths, with
Viewsheds have been employed to hunters situating themselves in sheltered
understand the social landscape, includ- areas or near look-out points to observe
ing such aspects of it as the relationship migration (Krist and Brown 1994: 1133).
between visual dominance and territo- Although this model provided important
riality (Lock and Harris 1996: 224). For information about Paleoindian and Early
example, one of the goals of Danebury Archaic sites in the region, the authors
project mentioned previously was to concluded that better data regarding
determine whether the hillforts were caribou migration were necessary to
located were they were to provide visual fully evaluate the predictive power of this
dominance that could be exerted over model (Krist and Brown 1994: 1135).
a territory, or for purely defensive pur-
poses. Lock and Harris (1996: 232–233) Methodological Problems with Viewsheds
found that the maximum defensive There are several methodological prob-
protection was waived in order to obtain lems with viewshed analysis. First, there
maximum visibility over the local popu- is the difference between the calculated
lace of lower-order farmsteads. Visibility viewshed and what can actually be seen
was also one of the factors evaluated by by the observer (van Leusen 1999: 218);
Madry and Rakos (1996) in their study what can be seen, according to the
of Celtic roads and hillforts in the Arroux viewshed, might not be perceived by the
River valley in Burgundy, France. One viewer. A related problem to this question
of the hypotheses examined was that is the “tree-problem” (Maschner 1996a:
forts had a view over the roads, perhaps 8; Wheatley 1996b: 97)—viewsheds tend
for sentries to observe the road network to be created as if the landscape was flat,
(Madry and Rakos 1996: 111). They and trees or other vegetation are not
found that 86.31% of all the Celtic roads factored in. This problem can be allevi-
were in view of one of the hillforts. This ated by raising or lowering the height
is greater than the 68.37% of the total of the observer to simulate the average
thus are comfortable with its use and Compounding this problem is that the
understand its capabilities. Forestry departments of the four individuals
companies are accustomed to the use of listed above have either terminal Mas-
GIS models to delin eate wildlife habitat, ters’ programs or no graduate program
for example. As a result, in working with at all.
forestry companies, I have found first- Given the paucity of academics with
hand that having forestry GIS specialists interest in this area, few classes in archae-
and archaeological GIS specialist speak- ological GIS are offered in archaeology
ing a common language enabled them programs across Canada. This apparent
to translate the concerns of the various lack of interest has important implications
stakeholders to their forestry and archae- for the archaeology departments who are
ology colleagues, respectively. training the students, many of whom go
It is unfortunate for Canadian on to careers in consulting archaeology,
archaeologists that so much of the where they may find themselves in a job
CRM-oriented GIS work remains poorly where it is helpful to have experience
known. Failure to publish the results of in GIS. It is certainly possible to obtain
such projects is one reason. In addition, training in GIS through other depart-
since much of it contains proprietary ments, primarily geography, but these
information, access to these reports is programs do not provide information
very limited. However, some consulting about archaeological applications. How
archaeologists have presented their work then do our students get the training in
at conferences (e.g., the CRIMP models something that is highly marketable in
of Western Heritage). the CRM industry? Clearly, more exper-
The situation in academic archae- tise is needed, but that will only come with
ology is also problematic. There, the time as more faculty with an interest in,
literature on archaeological GIS has and training in, archaeological GIS are
been limited to unpublished theses and hired by academic departments.
dissertations. Furthermore, there are
currently few professors at Canadian Archaeological GIS Projects in Canada
universities and colleges that list GIS as To date, there have been only two
a research interest. In my experience as major academic GIS projects in Canada,
a member of the graduate admissions both of which were concerned with
committee at the University of Manitoba predictive modeling and forestry. The
(1998–2002), increasing numbers of CARP project (Dalla Bona 1994a, b;
graduate students expressed interest in Hamilton et al. 1994; Hamilton and Lar-
doing a research project involving GIS. combe 1994; Larcombe 1994), based at
There are, however, few supervisors in Lakehead University, was a pioneering
Canada who are trained in GIS. A review examination of predictive modeling
I conducted of Canadian university web- projects. It remains a widely cited project
sites revealed that only four archaeology in the archaeological GIS literature,
professors in Canada list GIS as an inter- widely used in many other projects as
est: James Conolly (Trent University), a primer in predictive modeling, and
Scott Hamilton (Lakehead University), has influenced how inductive, weighted-
Quentin Mackie (University of Victoria), layer modeling is done.
and myself. A slightly larger number list A second, more recent predictive
“computer applications” as an interest. modeling project is the Manitoba Model
plistic and might create more problems are regionally-based applications of GIS.
that it solves, the addition of a temporal Recently there has been more attention
dimension will revolutionize the use of paid to other scales of analysis, such as the
GIS in archaeology and the many other site itself (e.g., Quesada et al. 1995) or
fields that are concerned with temporal even levels within the site (e.g., Abe et al.
aspects of spatial distributions. Finally, 2002; Marean et al. 2001). While such
with a temporally oriented GIS, archae- scale of sites, level, and artifact analyses
ologists will be able to look at cultural are not new, the addition of archaeologi-
change, culture process, and cultural cal GIS methods to address these holds
evolution, as time is one of the key vari- much promise.
ables of interests to archaeologist, often GIS clearly provides a powerful set
even more so than space. of tools for the analysis and exploration
Similarly, GIS must be made to be of settlement patterns of past peoples. It
truly three-dimensional. GIS currently is reaching greater levels of acceptance
supports primarily a two-dimensional in many fields, especially as hardware
map, to which a third dimension is becomes more powerful and afford-
appendedmaking it what some refer to able, as the software more sophisticated
as 2.5 dimensional. For archaeologists and user-friendly, and as more digital
who work with sites that are three dimen- datasets come online. The future of GIS
sional, this makes the representation in archaeology will be one of growth as
of things like stratigraphy and cultural more academics adopt it for their own
levels very difficult, if not impossible. research, more departments offer train-
Furthermore, the underlying theo- ing, and more CRM companies delve
retical basis of GIS applications in deeper into its applications GIS. While
archaeology needs to be examined. some aspects of archaeological GIS in
Some (e.g., Church et al. 2000) have Canada remain underdeveloped, there
argued that GIS is a method in search remains much reason for optimism for its
of a theory. While this argument may be future here. This is especially true if more
somewhat of a red herring, since other students seek training in GIS techniques,
methodologies (e.g., zooarchaeology) for their knowledge and skills will have
are not attacked for being atheoretical, substantial long-term benefits for the
itis clear that those involved in using GIS practice of archaeology in Canada.
in archaeology must take a step back and
consider both how we are using GIS and Acknowledgements. An early version of this
how we are applying it within a theoreti- paper was given in the Arch 990 Seminar
cal framework with a rigourous research Series for the Graduate Student in the
design. It is likely that as GIS applications Department of Archaeology, University of
mature and as more GIS-trained archae- Saskatchewan. Thanks are due to the review-
ologists work the methods will find their ers of this article for the CJA, whose com-
theoretical place within established gen- ments greatly strengthened the paper and to
eral theories in archaeology. Dr. Ariane Burke.
Even without many of these changes,
GIS has evolved from earlier archaeolog- REFERENCES CITED
ical applications to new and innovative Abe, Y., C. W. Marean, P. J. Nilssen,
uses at a variety of scales. For example, Z. Assefa, and E. C. Stone
most of the examples discussed above 2002 The Analysis of Cutmarks on
tive Modeling Project: Volume 5 Sum- tifying the Present and Predicting the
mary and Recommendations. Center for Past: Theory, Method and Application
Archaeological Resource Prediction, of Archaeological Predictive Modeling,
Lakehead University, Thunder Bay. edited by W. J. Judge and L. Sebastian,
Hamilton, S., and L. Larcombe pp. 549–569. Government Printing
1994 Cultural Heritage Resource Pre- Office, Washington, D.C.
dictive Modeling Project: Volume 1, Kohler, T. A.
Introduction to the Research. Center for 1999 Putting Social Sciences Togeth-
Archaeological Resource Prediction, er Again: An Introduction to the
Lakehead University, Thunder Bay. Volume. In Dynamics in Human and
Hasenstab, R. Primate Societies, edited by T. Kohler
1996 Settlement as Adaptation: Vari- and G. Gumerman, pp. 1–18. Oxford
ability in Iroquois Village Site Selec- University Press, New York.
tion as Inferred through GIS. In New Kohler, T. A., and S. C. Parker
Methods, Old Problems: Geographic Infor- 1986 Predictive Models for Archaeo-
mation Systems in Modern Archaeological logical Resource Location. In Advances
Research, edited by H. D. G. Maschner, in Archaeological Method and Theory Vol.
pp. 223–241. Southern Illinois Univer- 9, edited by M. B. Schiffer, pp. 397–
sity at Carbondale Occasional Paper 452. Academic Press, New York.
No. 23. Centre For Archaeological Krist, F. J., and D. G. Brown
Investigation, Carbondale, Ill. 1994 GIS Modeling of Paleo-Indian
2003 Archaeological Predictive Period Caribou Migrations and View-
Modeling Using GIS: Problems and sheds in Northeastern Lower Michi-
Prospects. Paper presented at the 69th gan. Photogrammetric Engineering and
Annual Meeting, Society for American Remote Sensing 60(9):1129–1137.
Archaeology, Milwaukee, Wisconsin. Kuna, M., and D. Adelsbergerova
Hunt, E. D. 1995 Prehistoric Location Prefer-
1992 Upgrading Site-Catchment Anal- ences: An Application of GIS to the
yses with the Use of GIS: Investigating Vinorsky-Potok Project, Bohemia,
the Settlement Patterns of Horticultur- the Czech Republic. In Archaeology
alists. World Archaeology 24(2): 283–309. and Geographical Information Systems: A
Kamermans, H., and M. Wansleeben European Perspective, edited by G. Lock
1999 Predictive Modeling in Dutch and Z. Stancic, pp. 117–132. Taylor
Archaeology, Joining Forces. In New and Francis, London.
Techniques for Old Times: Computer Kvamme, K. L.
Applications and Quantitative Methods 1985 Determining Empirical Rela-
in Archaeology, edited by J. A. Barcelo, tionships between the Natural Envi-
I. Briz, and A. Vila, pp. 225–229. Bar ronment and Prehistoric Site Loca-
International Series 757. Tempus tions. In For Concordance in Archaeologi-
Reparatum, Barcelona. cal Analysis: Bridging Data Structure,
Kincaid, C. Quantitative Technique and Theory, edit-
1988 Predictive Modeling and Its ed by C. Carr, pp. 208–238. Westport
Relationship to Cultural Resource Publishers Inc, Kansas City, MO.
Management Applications. In Quan- 1988 Development and Testing of