Use of weights of evidence statistics to define inference rules to disaggregate soil survey maps

摘要

Our objective was to produce a digital (raster) soil map for a tributary watershed within the Okanagan Basin of southern British Columba by disaggregating soil series from within polygons of harmonized legacy soil maps. Individual soil series were assigned to grid cells of a 25 m digital elevation model. We used a fuzzy membership inference using the output from weights of evidence calculations to help define inference rule curves. Weights of evidence is a probabilistic calculation based on the expected spatial relation between a predictor (environmental covariate) and a mapped soil class (soil series) where covariate values are placed into classes. We used the contrast values generated by the calculation to quantify the strength of association between covariate classes and the mapped soil series and the Studentized contrast values to identify statistically significant associations. In this way we could identify the most robust predictors for each soil series and define the membership rule curve for each covariate for each soil series within the ArcSIE (soil inference engine) software. A limiting factor function was used to integrate the fuzzy membership values of all covariates to produce a single value for each soil series for each grid cell. The inference was run for each of 23 soil series to produce a final map. Field validation indicated satisfactory (>70%) prediction accuracy for the method although uncertainty at individual grid cells as measured by dispersion of membership (entropy) values was highly variable.