Article III. Global sensitivity analysis of rusle illustrates importance of cover management across environments in predicting soil erosion rates
Estrada Carmona, Natalia
Fremier, Alexander K.
DeClerck, Fabrice A. J.
MetadatosMostrar el registro completo del ítem
Soil loss remains a critical issue for sustained agricultural production and reduction of downstream environmental impacts. Modeling soil loss at watershed scales helps researchers and decision makers quantify the impact of policy and land use decisions. The Revised Universal Soil Loss Equation (RUSLE) is a common empirical model used for quantifying soil loss. This model is widely applied across spatial extents and environmental conditions despite a lack of site-specific data for many regions. To better understand the consequences of the broad applications of RUSLE and to provide recommendations for prioritization of site specific data collection, we performed a global sensitivity analysis (GSA) on three dissimilar factor estimate datasets, covering varying scales (plot and watershed) and environmental conditions (temperate and tropical). The GSA technique allowed us to rank factor importance in estimating erosion rates and identify important factor interactions controlling soil loss across environmental conditions. We also compared the robustness of both global and local sensitivity analyses in assessing factor contributions to model uncertainty. Using a non-parametric approach (Random Forest and Classification and Regression Trees), we found that the greatest soil loss comes from small proportions of the watersheds and is largely determined by the interaction of cover management with slope steepness in steep areas, and with soil erodibility in level areas. Results highlight the importance of cover management in soil loss predictions regardless of environmental condition and model parameterization. Our findings reinforce that conservation practices should be targeted at specific locations of high erosion by adjusting cover management, specifically root density and surface cover. In addition, we argue that a global sensitivity approach is more robust than the local sensitivity analysis because higher order interactions among factors are quantitatively considered.