An Exercise
SHORTCOMINGS
The shortcomings with this exercise can be divided into three sections: theoretical, methodological, and analytical.
Theoretical Shortcomings
In addition, some archaeologists question the very utility of predictive models at all. They argue that the heavy dependence of the these models on environmental factors renders them too environmentally deterministic and ignores the likelihood that human decision-making processes are neither entirely logical, nor uniform.
Methodological Shortcomings
However, even if I was accurate on all my site locations, the locations of all the sites in Ontario (unsurprizing given the authors of the posting) means I have no knowledge of sites across the river in Quebec. This would be especially interesting as the landform differs sharply between the two areas - the Precambrian highlands and the Paleozoic lowlands.
The grid size I chose for my raster layers also created problems. On the advise of the teaching assistant I chose a grid of 500 (by 424) as this was the grid size used in most of our problems. It was also a manageable size analytically - my computer was less apt to whirr and grind and print jobs were reasonably fast. However, this led to a grid cell size of about 135 metres square - not altogether well suited for some of my buffer distances. Also, in the conversion from vector to raster, some of my site locations landed in the water and I had to go back in and re-digitize a number of site locations so they would remain on dry land once through the vector/raster conversion.
Analytical Shortcomings
A knowledge of where sites weren't located (i.e. areas that had been surveyed and sites were not found) would have been helpful. A comparison of the environmental conditions for "nonsite" to "site" locations would have helped in a more statistically significant prediction of site potentials.
Lastly, my MCE values ended up quite low. It is quite possible that I did not come up with the correct weights for the factors.
Some archaeologists question the use of GIS for producing predictive models of site location. They argue that the analysis can be done by simplier means - even map overlay (website). They also question the relevance of the results. One great weakness with predictive modeling in archaeology is that to some extent you are predicting what you already know. It is no coincidence that most sites have been found along waterways, or highways - these are areas of easy access and therefore places where it has been easy to find sites. In many areas, surveys have not covered the entire range of landscapes and it is unlikely that the total site distribution pattern is understood. A tautology thus develops: the 'unknown' sites will not be modeled for in predictive models based on current understandings of site distribution - and not modeling for these sites makes it less likely these areas will be surveyed and the sites discovered. Although some progress is being made in surveying less accessible areas, until information from less accessible areas are included in predictive models, their results should be viewed with some caution.
A glaring problem in this analysis is the degree of uncertainty of the location of the known sites. Although I used 1:50,000 maps, the information on the website and the roads, rivers, and streams layers to locate the sites as well as I could, there is no guarantee that I accurately located any one (or all) of the site. For a couple of factors, this wasn't really a huge problem. In terms of distance from rivers - well, pretty much all the sites were close to the river. Slope did not vary much in the neighbourhood of the site locations, nor did the surface geology and hydrology categories. Of course, creating slope from a 25 metre contour interval is probably introducing enough error in and of itself and no doubt is masking the location of a number of smaller landforms that might have been useful in the prediction of site location. That factor is out of my control. Where the uncertaintly of site location really showed up was in aspect - moving just slightly around the designated site location created vast differences in aspect. Either many of the sites were situated on ridges, with a range of aspect, or the aspect data was just too variable for my coarse approximations of site location. I suspect the latter. It was for that reason that I chose the more general "south facing" criterion for the aspect factor.
Most of my analytical shortcomings can be summarized as follows - lack of familiarity with the area hindered my ability to fully analyze the data. My lack of knowledge of the area's surface and hydrology has already been discussed. I also lacked knowledge as to the nature of the archaeological sites, was a given site a large scale settlement, or an isolated artefact eroding out of some cutbank. Further, with no knowledge of site stratigraphy, I was unable to better correlate site location with the surface geology or hydrological features.