Analysis of Uncertainty in Evaluation of Watershed Management Practices

Utility of complex distributed-parameter watershed models for evaluation of the effectiveness of non-point source sediment and nutrient abatement scenarios such as Best Management Practices (BMPs) often follows the traditional {calibrate → validate → predict} procedure. Despite its simplicity, this approach is subject to non-uniqueness of the calibrated parameter set. In this study, a computational framework is developed, in which investigation of uncertainty provides complementary quantitative and qualitative information in support of BMP evaluation. The Generalized Likelihood Uncertainty Estimation (GLUE) method is employed to generate a cumulative likelihood for sediment and nutrient outputs of the Soil and Water Assessment Tool (SWAT) for two scenarios representing outputs with and without representation of BMPs. Quantile analysis of the cumulative likelihoods yields expected sediment and nutrient loads as well as their corresponding uncertainty bounds. While comparison of expected values determines the effectiveness of BMPs, uncertainty bounds could be used to obtain a margin of safety (MOS) for such evaluations. The methodology was applied in a small watershed in the Maumee River basin. Maumee River is the longest river in the Great Lake system, where elevated levels of phosphorus have been a major concern. Results indicated that effectiveness of BMPs evaluated through the traditional method fell well between the estimated uncertainty bounds. It was concluded that parameter uncertainty accounted for nearly 15% of the variation in the estimated effectives of BMPs.