UNCERTAINTY ANALYSIS OF THE WEPP SOIL EROSION MODEL

Abstract

ABSTRACT Predictions from hydrologic and erosion models contain a large degree of uncertainty. The Modified Point Estimate Method (Harr, 1989) used in conjunction with a response surface exploration technique (Brooks, 1958) provides a simple, computationally efficient, and powerful tool for evaluating uncertainty of predictions by natural-resource models. The method allows analysis of models with a large number of input parameters which may be correlated and for which the exact input parameter distribution is unknown. The method was applied to the Water Erosion Prediction Project single rainfall-event erosion model. Sixty treatment combinations were selected to determine WEPP output uncertainties for a wide range of soil, crop, management, topographic, and storm conditions. The levels of the treatment combinations were randomly selected to span the entire factorial space of the 28 WEPP inputs, but with a finite number of treatment combinations. Five WEPP outputs were studied: peak runoff rate, average soil loss, average deposition, sediment yield, and sediment specific surface enrichment ratio. Maximum and average output uncertainties, given by the coefficient of variation, were determined for each output of the 60 treatments. Maximum coefficients of variation for peak runoff rate, soil loss, sediment yield, and sediment enrichment ratio were 196, 267, 323, and 47%, respectively. Average coefficients of variation for the same set of variables were 65, 99, 106, and 13%, respectively. Coefficient of variation was less for larger runoff and erosion events, which account for a large percentage of the total soil loss at a location over extended time periods. Significant, positive correlations existed between the coefficients of variation of peak runoff average soil loss, and average soil loss and sediment yield, indicating that the uncertainty in average soil loss and in sediment yield may be directly related to the uncertainty in peak runoff rate.