Impact of Parameter Uncertainty on Critical SWAT Output Simulations

Abstract

Watershed models are increasingly being utilized to evaluate alternate management scenarios for improving water quality. The concern for using these tools in extensive programs such as the national Total Maximum Daily Load (TMDL) program is that the certainty of model results and efficacy of management scenarios are not often measured and therefore are not well known. In this study, we used the mean value first-order reliability method (MFORM), a computationally efficient uncertainty analysis method, to determine the contribution of parameter uncertainty to total model uncertainty in streamflow, sediment, and nutrient outputs in a small Maryland watershed. Examination of sensitive and uncertain parameters revealed that parameters not considered highly sensitive contributed to model output uncertainty to a large extent. Therefore, highly sensitive parameters should not be the only parameters considered in uncertainty or calibration analyses. Measures of output uncertainty showed that sediment had the largest amount of variance from its mean value (CV = 28%), while nitrate, phosphate, and streamflow had considerably less variance, with annual average CVs of 19%, 17%, and 15%, respectively. The largest amounts of model uncertainty occurred during wet periods. This study concluded that with improved knowledge of the true value and associated uncertainty of input parameters, and improved algorithms to capture the variability of rainfall and associated flow, watershed water quality models will be of much greater use to TMDL studies and studies alike.