Uncertainty Analysis and Calibration of SWMM Using a Formal, Bayesian Methodology

Importance of uncertainty analysis (UA) to estimate the degree of reliability associated with model predictions is being understood. Consequently, literature that describes various Bayesian methods for the assessment of parameter and model predictive uncertainty has been steadily rising. Applications dealing with urban stormwater management are, however, very limited. This study demonstrates successful application of a formal Bayesian methodology for UA of the U.S. EPA Stormwater Management Model (SWMM), a widely used urban stormwater management model, and illustrates the methodology using a highly urbanized watershed in southern California. DREAM(ZS), a recently developed effective and efficient sampling algorithm, and a generalized, formal likelihood function that addresses the assumptions commonly made regarding error structure including independence, normality and homoscedasticity are used for the UA. Results will include comparison of the simulated error structure with the assumptions made by the likelihood function, histogram of the parameters posteriors, bounds of the 95 percent confidence interval, and the maximum likelihood (ML) predictions. A conventional calibration attempted to compare the ML results derived from the UA with the optimal solutions identified by the single objective calibration will also be presented. Besides illustrating state-ofthe-art in UA, the study will highlight application of the methodology to developing a watershed management model to mitigate stormwater quantity and quality problems associated with urbanization.