Systems evaluation of erosion and erosion control in a tropical watershed

Soil erosion represents a significant, often hidden cost of land conversion to human use. It is particularly acute in sub-Saharan Africa, where long-term dependence on soil functional capacity is profound and intervention resources are scarce. This dissertation explores soil erosion, focusing on densely populated regions of western Kenya, in two complimentary ways. First, emergy evaluation, which allows comparison of ecological and economic flows in common units, quantified erosion severity at three scales (national, district and landuse subsystem). Second, probabilistic erosion risk models were calibrated based on empirical observations across the Awach River basin.; Emergy analysis revealed that over 4% of national emergy use is lost topsoil; soil loss severity increases at the district (2.4 to 14.2%) and landuse subsystem scale (14–76%). Agricultural benefit (agricultural yields given natural capital costs) ranges from a high 7.6 nationally to 2.25 for Nyando District; landuses ranged between 7.4 and 1.32.; Field observation of potential soil erosion predictors was undertaken at 420 sites distributed throughout the study basin. Tree-based regression models allowed inference of soil properties from visual/near infrared reflectance spectra based on an existing library of laboratory-analyzed samples. Properties successfully delineated included degradation status, infiltration class (<60 mm/h) and 13 soil performance measures.; Erosion risk was conceptually divided into three factors—site protection, detachment resistance and hydrologic/terrain risk—to which observed variables were assigned. Graphical models selected variables conditionally associated with degradation, and multivariate logistic regression quantified effect strength and direction. Resulting models correctly classified 73, 74 and 76% of sites, respectively; an integrated risk model increased accuracy to 84%. The most significant predictors of risk were infiltration class, ground cover and soil organic carbon content.; Satellite imagery facilitated spatial inventory of degradation, infiltration and land use. Over 46% of the basin was classified as degraded (27% severely) by a screening model with 86% accuracy. Infiltration class was classified with 82% accuracy; land use (8 classes) with 73% accuracy. Landuse change trajectories were inferred by comparing a 1986 scene with a 2001 scene. Markov models, both spatially indiscriminate and risk-weighted, were developed to compare landuse change scenarios for erosion attenuation and emergy-based agricultural benefit. Results suggest limited protective effect of moderate changes in landuse change patterns; cattle density reduction and reforestation appear most promising. Extreme changes in landuse were observed to restore the basin, but were highly unrealistic. Spatially targeting low risk landuses to high-risk sites provided small but significant improvement.