Vast areas of the earth need new or updated soil survey data. However, traditional methods of soil survey are inefficient, expensive, and often inaccurate. A methodology incorporating geographic information systems (GIS), remote sensing (RS), and modeling to predict and map soil distribution was developed and tested in a pilot project in the Powder River Basin, Wyoming, USA. Topographic data derived from digital elevation models (DEMs) and Landsat RS spectral data were selected to represent soil-forming factors and analyzed using ERDAS Imagine image processing software. Unsupervised and supervised classifications were used to develop representations of soil-landscape patterns and to plan locations for collection of field data. As more was learned about the survey area, a knowledge-based classification model was built based on the concept of a decision tree. Final map quality was checked using traditional qualitative means and a quantitative accuracy assessment (88% overall accuracy for eight map units). |