Sensitivity and first-order/Monte Carlo uncertainty analysis of the WEPP hillslope erosion model.

摘要

Performing a comprehensive sensitivity/uncertainty analysis is a valuable step in understanding and using a predictive hydrologic/water quality (H/WQ) model. This article applies one-factor-at-a-time (OAT) sensitivity analysis (SA) and first-order error analysis (FOEA)/Monte Carlo simulation with Latin hypercube sampling (LHS) uncertainty analysis techniques for evaluation of a complex, process-based water erosion prediction tool, the USDA Water Erosion Prediction Project (WEPP) model (version 2010.1). Assessment of the WEPP hillslope profile model on a Midwestern U.S. Miami silt loam soil for three cropping/management scenarios and three erosion process cases (as defined by topography) is described. WEPP model runoff, soil loss, and corn (Zea mays L.) yield output responses in the form of expected values and error variances were determined to illustrate model prediction uncertainty. The OAT SA showed that WEPP runoff and soil loss output responses were most sensitive to changes in the baseline effective hydraulic conductivity (K_b) and sand content. WEPP model corn yield output response was most sensitive to crop input parameters affecting the simulation of biomass development. The FOEA showed that the largest contributions to runoff, soil loss, and corn yield total error variance came from K_b and sand/clay content, K_b and baseline soil erodibility factors, and the biomass energy ratio of a crop and harvest index, respectively. The FOEA total variances presented in this study for runoff and soil loss were considerably larger than the corresponding Monte Carlo LHS simulation total variances. The Monte Carlo LHS total variance results were reasonable, making Monte Carlo LHS appear to be a better alternative for quantifying WEPP output response error variance. The Monte Carlo LHS soil loss output responses were also compared to Universal Soil Loss Equation (USLE) soil loss predictions. The USLE soil loss estimates were within the Monte Carlo LHS 90% prediction intervals for six of the nine cropping/management and erosion process cases. Results of this study illustrate the usefulness of combining SA and Monte Carlo LHS for providing detailed uncertainty analysis information for complex, physically based models such as WEPP.