Visual Data Mining Tools for Digital Soil Survey Update.
James E. Burt, A-Xing Zhu, and Rongxun Wang. Geography, University of Wisconsin-Madison, 550 North Park Street, Madison, WI 53706
Most soil survey work in the United States is directed at updating existing surveys. These surveys, though typically decades-old, are for the most part available in digital form. Other studies have shown that by combining survey maps on raster data representing environmental covariates, one can recover soil-landscape relationships that are implicit in the maps. Such relationships approximate soil-landscape models developed by the original soil scientist and can be used to jumpstart the revision process. This paper presents a variety of tools developed for visual data mining that allow a user to recover explicit depictions of soil-landscape concepts, identify inconsistencies in application of those concepts, and develop new soil concepts. The tools offer a mix of univariate and bivariate displays to permit examination of single factors or pairs of factors in both parameter space and environmental (i.e., physical) space. The tools also provide for rule construction that can be fed directly into the SoLIM Suite predictive mapping software.