Methods and Areas for Predictive Model Testing

Predictive modelling using logistic regression should be tested against "judgemental" type potential mapping. Essentially, a predictive model must be better than an expert examining a 1:50,000 topographic map and predicting site potential. The Dean River example would be an excellent test of modern modelling, since hardcopy 1:50,000 Canada Land Inventory standard resource mapping is available for the study area. The map layers include terrains, soils, forest cover, ungulate habitat, and fisheries values as well as archaeological potential. These could be digitized if they are not presently available in this form. The quadrat data could form the "training model", which could then be tested against the judgementally collected site locations. This same technique, if successful, could be extended to test other probabilistically surveyed areas, in slightly different environments, and then, as the model is refined, applied to much of the Fraser Plateau.

Another good area to test predictive modelling is the southern Strait of Georgia, using the OSRIS GIS of the Environmental Emergencies Service Branch. This system contains about 1400 archaeological sites as one map layer, together with environmental data for over 50 other variables. An archaeological predictive model would be relatively easy to produce as the quality of data is generally very good. A early example of a potential map is available for a small part of the area (Cassidy 1979). The Emergencies Service Branch is willing to work with the Archaeology Branch to conduct a modelling experiment, subject to system availability and personnel replacement (Howes 1992, personal communication).

If a GIS useful for archaeological resource management is developed, it would be very advantageous to include SPOT and LANDSAT satellite images with the system. Demonstrations of the OSRIS, which includes these images, have shown them to be extremely valuable in assessing current land uses, development boundaries, and unmapped local topographical and cultural features. Although the cost of a 60 x 60 km block is $1,800, it would be worthwhile to obtain coverage of areas of rapid development. In fact, for most internal Archaeology Branch uses, the satellite images by themselves combined with an overlay of known archaeological sites would be much more useful than a TRIM map base, which would cost at least $2,400 for the same area (16 maps @ $150). However, modelling or exploratory data analysis could not be done on the satellite imagery by itself.