Abstract
Data mining from existing County soil surveys can improve the utility of maps for research, management and decision making. Aggregated soil series information in soil survey map units can be disaggregated by following a possibilistic decision tree approach to provide maps at the soil series level. The “overall map unit composition percentage” available in the soil survey tabular data can be treated as a possibility distribution of the prevalence of different soil series within all occurrences of the map unit in the survey area. A case study was conducted in Monroe County in southeastern Ohio. Three different learning approaches including C4.5 decision trees, nonspecificity based possibilistic decision trees and clustering based possibilistic decision trees were applied to the existent County survey, and the efficiency of predicting soils at the series level was assessed using an independent soil series point data set. The results showed an improvement in prediction accuracy by following the clustered possibilistic decision tree approach. The maps were useful in identifying the locations of component soil series within map unit associations, consociations, and complexes. Data mining from existent soil survey maps using the component information available in the tabular data can serve as a guide for disaggregating soil map units to create soil series maps, identifying misplacement of polygon boundaries, identifying presence of inclusions, and correcting mislabeled polygons, when updating soil surveys.
•Data mining SSURGO maps can enhance survey updates and harmonize adjacent surveys.•Disaggregation of map units intensifies surveys by locating included series.•SSURGO component percentages were useful for mapping at the series level.•Clustered possibilistic decision trees provided most accurate mapping of components.