In general, the use of predictive models based on sampling is a good way of obtaining a general impression of an area, of being able to tell with a certain degree of precision and reliability what is "out there". As such, they are of value to planners and developers in choosing among possible alternatives and to help minimize damage, delays, and expense (Klesert 1987:231).
Many archaeologists and managers consider predictive modelling to be a practical reality at the methodological level. On the basis of archaeological surface survey or previously recorded data taken from site files, the occurrence of sites correlated with environmental variables, most of which can be derived from topographic or other maps or from such data sources as remote sensing: elevation, slope angle and direction, different types of landform, (assumed to be good defensively, for lookouts, or for shelter), vegetation and soil types, and distance to water sources. The "best" indicators are those that correlate with occurrences of sites that are already found, and these are used to predict where sites will or will not be discovered, at least potentially, in unsurveyed areas (Ebert 1988a:3)..
"Surely there was more to prehistoric human behaviour than this implies," said one colleague, himself a Native American trained as an archaeologist, and geographic information system researcher. "This is what we do to map fox or squirrel habitats: look for water and shelter and food and then draw polygons and isopleths around them. Do these archaeologists think they know all about how complex past peoples' seasonal rounds were, why they went where they did?" We as archaeologists do not know all about the complex systemic behaviour that must be the basis of archaeological predictive modelling (Ebert/Kohler 1988:98).
It seems clear that the use of predictive models has basic limitations: models are nothing more nor less than simplified abstractions of a variable and complex reality. Archaeologists have always employed predictive models, in all situations, whether they have been explicit about that use or not. over the years archaeologists have become more sophisticated in their approach towards sampling and modelling, but it has been a matter of progressive refinements rather than anything approaching a paradigm shift or conceptual revolution. With the introduction of sampling as a replacement for intensive surveys, it is essential that the sampler demonstrate that the model has been adequately tested prior to its implementation, and that the sample does supply the researcher with what is required: either a representative picture of the whole, or labour-saving approximation of an entire population (Klesert 1987:234).
Since it is not the primary objective of the predictive modelling effort now underway to achieve any reduction in the amount of survey required, the way surveys are currently performed will likely not be affected at all. That is assuming that surveys currently being conducted are appropriate, technically sound, and in conformance with existing standards and guidelines. In cases where this is not true, the increased control required of surveys providing data for model development should improve survey techniques and increase the value of the data produced (DeBloois 1985:10).
Archaeologists with management responsibilities fear that the current/suggested potential of predictive modelling is too limiting. They are looking for practical methods to provide better information about cultural resources in order to make realistic recommendations to management. Archaeologists without management responsibilities fear that the technology, if allowed to go unchecked, would be applied by land management in an irresponsible manner.
Archaeologists are concerned about the explanation of past human behaviour, and there is general agreement that they should not be satisfied with only the demonstration of correlations, but that they must also provide explanations for those correlations.
Management objectives are sometimes thought to be limited to a narrow concern over "how many sites are where", and indeed, models can suggest what types of sites are in a specific area and where in that area they might occur. Some models can also be used to generate population estimates and statements concerning the probability of site occurrence in a particular location. These classes of information are important in management decisions about possible surface-disturbing actions. But the more research-oriented objectives of modelling are also important because such models can help to indicate data gaps and highlight research issues needing additional work. Such models can assist in understanding existing data and, in some cases, can expedite and streamline inventory programs (Kincaid 1988:549).
Archaeological sites in a region are not only non-randomly distributed, but often exhibit pronounced patterns with respect to landform, soils, vegetation, hydrology, and other features. Explanations of archaeological distributions or prehistoric landscape uses have centered around these environmental characteristics. It has been variously argued, depending op cultural context and geographical region, that in placing their activities prehistoric peoples selected such environmental features as level ground, proximity to water sources, good views, good soil conditions, south-facing slopes, and a host of other factors. Many regional analyses that have attempted to demonstrate these tendencies have not been very convincing or successful, however, at least until the advent of GIS Kvamme 1989:168).
There is good reason to believe that site locations cannot, in general, be fully predicted from environmental variables alone. In land management applications, therefore, models of natural phenomena and models of cultural phenomena should not be considered equivalent. Managers need to have a realistic understanding of what models can and cannot do in order to use them effectively (Kincaid 1988:551).
From a management perspective, the most important issue facing land managers is not whether to invest in predictive models but whether the modelling process should be an integral part of the overall cultural resource management program. It can be argued that managers should utilize models and the modelling process because it is in their best interest to do so. In the short run, the first few predictive models will probably not be very powerful. They will not be substitutes for inventory surveys, and perhaps they will not even be very good planning tools. Moreover, a commitment to the model-building process may require the restructuring of the cultural resource management program to ensure that projects are designed to meet specific objectives and their results are cumulative. Standardization will have to give way to flexibility in research design, and the agencies may have to be prepared for larger rather than smaller sampling fractions. In the long run, however, a commitment to modelling may be the land managing agencies' best hope for the creation of useful tools to guide future development and management of this country's cultural resources (Altschul 1988:88).
Predictive modelling can be a worthwhile component of cultural resource management, if for no other reason than that it injects rigor into the management process and serves to integrate management with archaeological research. The process of modelling and the preparation and development of models are extremely valuable assets to management, regardless of their ultimate "success" of the models (Judge/Martin 1988:579).
However, the Archaeology Branch feels at present that archaeologists do not know as much as they should about how to do predictive modelling. Through the proper combination of rigor and research, they can probably learn to do such modelling in the near future, but at "this stage in our understanding of the modelling process, it would be premature to produce a guidebook or manual" (Ebert/Kohler 1988:97).
Modelling is a cyclical process of ongoing refinement, rather than a one-time event, and thus models cannot be developed by archaeologists and then simply "turned over" to land managers for "application". Predictive modelling is potentially the most cost-effective way to combine sound management practices with valuable research programs (Judge/Martin 1988:580).
Undoubtedly the most important criterion to consider in evaluating a model is whether or not it has been tested. Without testing or evaluation, a model is little more than a guess. Another reason for model testing is to determine the nature and strength of relationships that may have been discovered (Kincaid 1988:565).
Are logistic models with a significant "fit" sufficient to allow interpretations of site probability for areas as yet unsurveyed? If predictive models are to contribute to a greater understanding of human settlement patterns and to the land-use decision-making process, it is imperative that a validation methodology be developed to allow for more rigorous model testing. Such procedures should allow the incorporation of further field tests (Parker 1985:189).