Uncertainty analysis of a steady-state erosion model
Abstract
Uncertainties inherent to parametric inputs propagate within models, generating uncertainty in model predictions. The first objective of this study was to determine the coefficients of variation for five outputs of the Water Erosion Prediction Project (WEPP) model, Hillslope Profile Version (version 90), for a wide variety of modeling situations. The WEPP model, a physically-based, steady-state erosion model was the deterministic transfer function studied. The error propagation (stochastic) routine used was the Modified Point Estimate Method. Means and coefficients of variation of model outputs were generated for a wide variety of modeling conditions, using experimental and synthetic data as inputs. The average coefficients of variation in all simulations were 65% for peak runoff rate (PEAKRO), 99% for soil loss (AVSLOS), 106% for deposition (AVDEP), 107% for sediment yield (AVLOST), and 13% for enrichment ratio (ENRATO). Yet, the overall average coefficient of variation of the 28 model inputs was 21%. A second objective in the study consisted of emulating the WEPP model with simple polynomials, generated by a neural network system identification algorithm (SONN), given a series of WEPP input and output data, representing a wide range of modeling situations. These polynomials could substitute the computer-intensive WEPP model in extension offices of less developed countries, where computer hardware is scarce. The results were mixed. Good fits were obtained between the data fitted by the Kolmogorov-Gabor polynomials generated by SONN, and the WEPP original data, for the outputs PEAKRO, AVSLOS, and AVLOST (R$\sp2$ values were 0.89. 0.85, and 0.84, respectively). R$\sp2$ values for ENRATO and AVDEP were 0.63 and 0.48, respectively. SONN-generated polynomials also outperformed simple linear regression models by an average R$\sp2$ value of.22, using less than half of the 28 original model inputs.
Degree
Ph.D.
Advisors
Nearing, Purdue University.
Subject Area
Agronomy
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.