Wednesday, November 7, 2007 - 11:30 AM
300-8

Using Decision Tree Analysis for Predictive Soils Mapping in the Great Basin.

Sarah Hash and Jay S. Noller. Oregon State University, Crop & Soil Sciences Dept., 3017 Agricultural and Life Science Bldg., Corvallis, OR 97331

The process of soil survey is often a tedious, time-consuming, and costly endeavor. Well-trained field scientists must cover tremendous expanses of land, dig and describe soil profiles, determine taxonomic classification, and carry out mental interpolation with the collected data to accurately draw soil map unit boundaries. When delineating soils, the soil scientist uses knowledge of the soil-forming factors–spatial and temporal environmental variables whose unique combinations produce unique soils—and how variations in these factors across the landscape dictate where a certain soil type will end and another begin.

Soil scientists may soon be able to take advantage of the plethora of digital data available that actually quantify the environmental variables that affect soil formation. These data can be collected for a previously mapped area analogous to the area of interest, prepared in a GIS, and then analyzed with decision tree classifier software. Generated rule sets are then applied to produce predictive soil maps for the unmapped area. In the near future, rule-based predictive maps that are generated with a high level of confidence could give field scientists a valuable tool for fieldwork planning and project management.

This paper outlines the procedure used in the active initial soil survey of Malheur County, Oregon. Raster datasets are prepared in ArcGIS 9.1 and ERDAS IMAGINE 8.7, and then data are analyzed using the CART sampling tool in IMAGINE and See5 decision tree classifier software.