A probabilistic approach for analysis of uncertainty in the evaluation of watershed management practices

A computational framework is presented for analyzing the uncertainty in model estimates of water quality benefits of best management practices (BMPs) in two small (< 10 km(2)) watersheds in Indiana. The analysis specifically recognizes the significance of the difference between the magnitude of uncertainty associated with absolute hydrologic and water quality predictions, and uncertainty in estimated benefits of BMPs. The Soil and Water Assessment Toot (SWAT) is integrated with Monte Carlo-based simulations, aiming at (1) adjusting the suggested range of model. parameters to more realistic site-specific ranges based on observed data, and (2) computing a scaled distribution function to assess the effectiveness of BMPs. A three-step procedure based on the One-factor-At-a-Time (OAT) sensitivity analysis and the Generalized Liketihood Uncertainty Estimation (GLUE) was implemented for the two study watersheds. Results indicate that the suggested range of some SWAT parameters, especially the ones that are used to determine the transport capacity of channel network and initial concentration of nutrients in soils, required site-specific adjustment. It was evident that uncertainties associated with sediment and nutrient outputs of the model were too large, perhaps limiting its application for point estimates of design quantities. However, the estimated effectiveness of BMPs sampled at different points in the parameter space varied by less than 10% for all variables of interest. This suggested that BMP effectiveness could be ascertained with good confidence using models, thus making it suitable for use in watershed management plans such as the EPA's Total Maximum Daily Load (TMDL) program. The potential impact of our analysis on utility of models and model uncertainties in decision-making process is discussed. (c) 2006 Elsevier B.V. All rights reserved.