Possiblistic Decision Trees to Disaggregate Component Soil Series From An Ohio County Soil Survey Map.

The accuracy of soil maps could be improved by data mining existing County soil survey maps.  Aggregated soil series information in the soil survey map units could be disaggregated by following a possiblistic decision tree approach. Information available in the form of “overall map unit composition in percentage” in the  soil survey tabular data depicting a possibility of occurrence of different soil series for all occurrences of the map unit in the survey area could be utilized to predict soils at the series level. The case study was conducted in Monroe County in southeastern Ohio. We applied three different learning approaches including C4.5 decision tree, non specificity based possiblistic decision tree and clustering-based possiblistic decision tree approach to compare the efficiency of predicting soils at the series level on a surveyed soil series data set. The clustered possiblistic decision tree approach showed an improvement in prediction accuracy of the mapped soils. Data mining from existent soil survey maps by making use of the component information available in the tabular data could serve as a guide in disaggregating soil series, identifying misplacement of polygon boundaries, identifying presence of inclusions and incorrect labeling, when updating soil surveys.