Exploring Soil-Landscape Relationships through Classification and Regression Tree Analysis.
Sheila Slevin and Jay Noller. Oregon State Univ, 3017 Als, 3017 Als, Corvallis, OR 97331, United States of America
The distribution of soils across a landscape is a function of complex, spatially-variable environmental factors. Conceptual models of soil-landscape relationships combined with field observations regarding site-specific environmental factors allow experienced soil scientists to predict soil map unit boundaries. The environmental factors and processes controlling soil development in any given landscape may be intuited by field scientists, but not explicitly described as part of the mapping process. Using classification and regression tree and data mining software, we have analyzed data collected by field scientists preparing an Ecological Unit Inventory of the WinemaNational Forest in South Central Oregon. These techniques provide a mechanism to explicitly identify the critical environmental factors controlling soil development and vegetation distribution across the landscape, and for preparing predictive maps based on geospatial data.