Stormwater Control Measure Modeling and Uncertainty Analysis for Total Maximum Daily Load Compliance

摘要

Cities, counties and other stormwater management agencies throughout the United States face billions of dollars of urban stormwater improvements each year to meet total maximum daily load (TMDL) regulations. In many cases, they will accomplish this by implementing stormwater control measures (SCMs) that are designed to capture urban stormwater and remove pollutants before the stormwater is discharged back to receiving waters. A wide variety of SCMs exist, each with unique pollutant removal performance and associated costs.;A critical aspect of TMDL projects is modeling alternative SCM implementation strategies to evaluate which strategies offer the greatest opportunity of TMDL compliance at reasonable costs. However, current SCM modeling practice suffers from several deficiencies, particularly as it relates to modeling for TMDL compliance. One problem is that most SCM modeling studies do not incorporate uncertainty analysis (UA), despite recommendations from the National Research Council (NRC) and others. This is generally due to a lack of knowledge for how to perform UA, lack of available models/algorithms that include UA capabilities and/or perceptions by decision makers that UA will not affect the most cost-effective decision. Another problem is that SCM models are typically calibrated and operated on an "event-basis" (assuming steady-state hydraulic conditions), whereas most watershed and receiving water models operate dynamically. This presents practical difficulties for modelers as they link watershed models to SCM models to receiving water models for TMDL studies and can also affect decision making as SCM model results are based on events and many TMDLs are subject to durations of hours, days, months, etc.;This dissertation addresses those problems by providing new tools and knowledge that can improve SCM modeling and decision making for TMDL compliance. In Chapter 2 ("Uncertainty Analysis of a Stormwater Control Measure Model using Global Sensitivity Analysis and Bayesian Approaches"), we compare different UA methods and use global sensitivity analysis to determine the most sensitive parameters in a new pollutant removal model. We conclude that an informal Bayesian approach (the Generalized Uncertainty Estimation Method) provides better estimates of SCM pollutant removal uncertainty compared to a formal Bayesian approach. We also show that the TSS removal in EDBs is most sensitive to the particle size distribution and particle density of solids in the runoff entering EDBs. In Chapter 3 ("Appraisal of Steady-State Stormwater Control Measure Pollutant Removal Models within a Dynamic Stormwater Routing Framework with Uncertainty Analysis"), we evaluate the effects of applying three different event-based (steady-state) SCMs models to a dynamic modeling framework. The linear regression model produces almost identical outputs under both steady-state and dynamic conditions, however the modified Fair and Geyer (MFG) model and k-C* model both produce results that underestimate TSS pollutant removal by 20-90% at the median. Using those same three models, 5-95% percentile prediction intervals (PI) were also evaluated using Monte-Carlo (MC) and first-order variance estimation (FOVE) UA methods. The FOVE method generally produced smaller PIs compared to the MC method, however, the 95th percentile values generated from the dynamically-applied SCM models were closer to the 95th percentile values generated from the steady-state SCM models using MC. In Chapter 4 ("Selecting Stormwater Control Measures to Achieve Total Maximum Daily Loads: The Effects of Performance Measures and Uncertainty"), we evaluate how incorporating UA into SCM modeling can affect decision making to achieve TMDL compliance. Using theoretical TMDL scenarios and three different TMDL compliance measures, our results show that the most cost-effective SCM design/implementation strategy can be different based on the decision maker's risk level, which can only be incorporated into the decision making process through the use of UA. This finding justifies the recommendations from the NRC and others that UA should be included all TMDL modeling studies.