A framework for uncertainty analysis of complex process-based models
Abstract
Many complex hydrologic water quality models have been developed, but a comprehensive description of expectations in model response and how they may be applied are rarely provided. Performing an uncertainty analysis is a step towards understanding and using a predictive model. This thesis presents a framework for uncertainty analysis of an evaluated, complex, process-based water erosion prediction model, the USDA Water Erosion Prediction Project (WEPP). The uncertainty analysis framework combines sensitivity analysis, first order error analysis (FOA), and Monte Carlo simulation with Latin Hypercube Sampling (LHS). Assessment of the hillslope profile version of the WEPP model on midwestern cropped lands for three separate slope conditions is described. WEPP model responses in the form of probability distributions functions, expected values, variances, and prediction intervals were determined to show model prediction uncertainty. Sensitivity analysis results found the WEPP model output response for runoff to be most sensitive to effective hydraulic conductivity and soil parameters used in the crusting factor adjustment. The soil loss response was most sensitive to erodibility factors and soil and management parameters influencing infiltration. The FOA did not approximate the WEPP model responses for runoff and soil loss well due to the nonlinearity of the model. Error variance determined through Monte Carlo simulation techniques is recommended for WEPP and other natural resource model output responses. The WEPP response is compared to predictions made by the Universal Soil Loss Equation (USLE). WEPP response prediction intervals showed the USLE estimates to be statistically similar for interrill erosion process cases. WEPP and USLE were the least similar in the no-till management situations and in the cases for clay soils. Recommendations for the use of sensitivity analysis in conjunction with Monte Carlo LHS simulation are made for future uncertainty analyses of other complex natural resources models.
Degree
Ph.D.
Advisors
Engel, Purdue University.
Subject Area
Agricultural engineering|Hydrology
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.