Predictive Soil Mapping in Southern Arizona.

Predictive soil mapping in southern Arizona

M.R. Levi, C. Rasmussen, and N. Starman

A fundamental knowledge gap to understanding land-atmosphere interaction is an accurate, high resolution representation of soil properties. Large tracts of land in the western U.S. are currently without published soil data, or are in need of updated information.  Remote sensing and geographic information techniques can create quick, reliable pre-maps for soil survey applications, and bridge the gap between site-specific soil properties and landscape variability.  Measurement of soil properties involves intensive field sampling and laboratory analyses that are time and cost prohibitive. Our objectives were to compare methods of predicting soil map unit boundaries for a large tract of land (~160,000 ha) in an arid/semi-arid ecosystem (Graham County, AZ) with remotely sensed reflectance and elevation data.  Surface reflectance data (ASTER and Landsat 7ETM+) and topographic indices derived from digital elevation models (e.g., slope, wetness index) were combined in a principal component analysis to determine layers accounting for the most data variability across the study area.  The layers accounting for the majority of the variance were retained for segmentation using Definiens Developer 7.0 and also for a maximum likelihood classification of isodata clusters.  Segmentation of slope and individual reflectance bands proved to be easier and more effective for predicting map unit boundaries than the unsupervised/supervised classification of principal component layers.  This research can advance the scientific understanding of planet Earth via space borne satellites and remotely sensed data by providing valuable soils information to improve ecosystem models and land management in similar landscapes.