Search

Abstract: Management of agricultural soils, most notably tillage, influences wind, and water erosion, which in turn has implications for non-point source pollution of pesticides, fertilizer, and sediment in agro-ecosystems. No-till (NT) practices improve soil, water, and aquatic ecosystem quality by reducing soil erosion and chemical runoff. The ability of cropland soils to sequester C from the atmosphere might help mitigate global warming. Classification of Landsat ETM+ satellite images has the potential to identify tillage practices and soil disturbance over large areas, enabling efficient monitoring of these agricultural practices. Previous studies predicting tillage management had relatively small study areas (located in a single county), relatively low numbers of fields (6–51), and were temporally focused on non-planted fields to reduce the potential effects of crop canopy interference and/or field patterning. Our objectives were to predict in the presence of crop canopy and over a spatially large, management diverse study area (1) tillage systems (NT versus tilled) and (2) soil disturbance. A farm survey of the study area, north central Montana, was used to as a means to obtain extensive field-level farm management data. We compared logistic regression (LR), traditional classification tree analysis (CTA), and boosted classification tree analysis (BCTA) for identifying NT fields. Logistic regression had an overall accuracy of 94%, BCTA 89%, and CTA 87%, but tillage was not well distinguished. Soil disturbance was estimated using linear regression (LM), regression tree analysis (RTA), and stochastic gradient boosting (SGB), an RTA variant. Classification of soil disturbance was best achieved using RTA (predicted mean soil disturbance not significantly different than known soil disturbance, p-value=0.08). Classification of Landsat ETM+ imagery showed promise for predicting tillage and agricultural soil disturbance over large, heterogeneous areas. [Copyright &y& Elsevier]