See more from this Session: Spatial Predictions In Soils, Crops and Agro/Forest/Urban/Wetland Ecosystems: III (Includes Graduate Student Competition)
Using a digital soil mapping approach we modeled biological soil crust development classes and ecological site classes across Canyonlands National Park. Soil and site observations were obtained from a recent soil survey update (2007-2010). Digital topographic data were derived from digital elevation models and included slope, various estimates of solar insolation, and hydrologic functions. Spectral data were derived from Landsat imagery and included a normalized difference vegetation index (NDVI), a cyanobacteria-dominated crust index, and customized band ratios to elucidate soil parent material mineralogy.
Data were analyzed using Random Forests, which is a non-parametric quantitative classification algorithm useful for complex data. The environmental variables most important for predicting biological soil crust classes included elevation (a proxy for precipitation regime), diffuse radiation, Landsat normalized band ratio 5/2, and NDVI, indicating the importance of moisture, solar radiation, soil mineralogy and texture, and vegetation in the development of biological soil crusts. The environmental variables most important for predicting ecological site classes were NDVI and elevation. Random forest models are anticipated to be extendable to unmapped areas adjacent to Canyonlands National Park.