Weights of Evidence Analysis for Cultural Resource Site Prediction and Risk Assessment

Funded by PWR GIS 2002, PMIS #79905
Mike Kaberline and Cathy Schwemm, Channel Islands National Park

Background
Santa Rosa Island within Channel Islands National Park is 55,000 acres in size and was purchased by the National Park Service in 1986. The island exhibits diverse topography and land cover. From sea level several peaks rise to over 500 meters and are separated by steep canyons and perennial streams. Vegetation varies dramatically across the island from homogenous patches of exotic grasslands to endemic coastal scrub and woody species stands. Historic cattle ranching, which ended in 1998, had significant impacts on vegetation and soils. Non-native deer and elk are still allowed on the island for an annual hunt conducted under a special use permit.

Santa Rosa Island supported an extremely vibrant Native American culture for most of the last 10,000 years. Over 600 archaeological sites have been documented on the island, and human remains found on Santa Rosa are among the oldest known in North America. Paleontological resources found on the island are also significant, and date from approximately 12,000 to 40,000 yrs. B.P.

Natural processes, combined with the land use history of the island, create an environment that puts in-ground archaeological and paleontological resources at risk. A century of grazing and ranching on Santa Rosa have resulted in many alterations to the soil structure and vegetative cover of the island. Winter storms bring significant periodic rainfall events also leading to loss of soil. Physical damage to the soil layer by ungulate trampling, permanent loss of topsoil due to erosional processes, erosion of coastal sites due to wave action, and heavy impacts by non-native grazers to riparian resources all contribute to threaten in-ground cultural and natural resources.

The goal of this project was to combine existing spatial data describing topography, vegetation, and anthropogenic impacts with expert knowledge of the island to create a predictive model of areas where known and currently unexposed sites are most at risk from erosional processes and other surface disturbances.

Project Description
Several tasks were necessary to develop the model. First, all existing spatial data layers for Santa Rosa Island were evaluated to determine which datasets contained information relevant to analysis of erosion potential. Vegetation, geology and elevation data were determined to be most valuable, while other data, including roads, perennial streams, pasture delineations, and fire history were examined but in general were of less use in evaluating site risk. A soils map is being created for the park, but was unavailable for this project. (The lack of soils information was ultimately determined to be the largest data gap in the project.)

Second, resource experts were consulted for their opinions on the most significant threats to site stability. Spatial datasets that included information on threat factors were attributed and weighted as necessary to represent perceived impacts, and converted to raster format in ArcView Spatial Analyst. Model Builder in ArcView was initially tasked for creation of the model, however the absence of a soils layer and the general lack of quantitative information describing specific impacts hindered a deterministic model. We decided that a weights of evidence (wofe) probabilistic analysis using existing information regarding site locations, and field data collection of ‘training points’ would provide the better alternative.

The ‘Arc-WofE’ extension created for ArcView (http://ntserv.gis.nrcan.gc.ca/wofe/index_e.htm) employs a Bayesian approach to spatial pattern recognition, and in this case would offer more utility than intuitively assigning weights to evidential themes. To support the wofe analysis, random data collection points, or training points, were generated for the island, and data describing ground condition and erosion potential collected. Weather and accessibility prevented data collection on much of the island, so the analysis was constrained to the area from which data were available.

As the risk analysis proceeded, we decided also to create a predictive model for potential prehistoric site occurrence. In this case the known locations of in-ground sites were used as training points. The project then produced response theme maps from the models, one displaying erosion potential and one displaying predicted site locations. Conditional independence, a measure of the independence of the evidential themes used in the model, was 0.99 (possible 1.0) for the erosion potential model and 0.97 for the site prediction model. Several conclusions were drawn from this project:

1. The absence of a soils layer prevented confidence in any a priori models;
2. An observance of 70% or more bare ground at the training points was selected as the determination of site erosion potential. Alternative determinations of erosion risk would have provided different results;
3. The availability of additional ground data would have allowed the expansion of the risk assessment model to a larger area;
4. The site predictive model was useful, but more so when used in concert with the erosion potential map. Analysis showed that no predicted sites were located in areas of high erosion potential;
5. Tobler’s First Law of Geography holds true: ‘Everything is related to everything else, but nearer things are more related to each other than are distant things.’


References
Agterberg, F.P., G.F. Bonham-Carter, Q. Cheng, and D.F. Wright. 1993. Weights of Evidence Modeling and Weighted Logistic Regression for Mineral Potential Mapping. From Davis, J.C. and U.C. Herzfeld, eds. Computers in Geology, 25 Years of Progress. Oxford University Press, Oxford.

Hansen, D.T. 2000. Describing GIS Applications: Spatial Statistics and Weights of Evidence Extension to ArcView in the Analysis of the Distribution of Archaeology Sites in the Landscape. Presented at the 2000 ESRI User Conference, San Diego.

Raines, G.L., G.F. Bonham-Carter, and L. Kemp. 2000. Predictive Probabilistic Modeling Using ArcView GIS. ArcUser, April-June 2000 (http://www.esri.com/news/arcuser/0400/files/wofe.pdf).