A Bayesian stormwater quality model and its application to water quality monitoring.

摘要

Stormwater runoff is a topic of research that over the years has increasingly grown due to its impact on our water resources. Treatment systems have been developed to mitigate this impact by preserving the pre-development hydrologic and water quality characteristics of the drainage areas. Understanding of the systems' treatment capabilities is required for stormwater management. The goal of this research was to study the application of a decay treatment model as a conceptual tool for understanding the pollutant removal characteristics of stormwater systems. Three systems were studied in this research: a sand filter, a gravel wetland, and a retention pond. The contaminants under consideration include: total suspended solids (TSS), total petroleum hydrocarbons - diesel range hydrocarbons (TPH-D), dissolved inorganic nitrogen (DIN, comprised of nitrate, nitrite, and ammonia), and zinc (Zn).;The mathematical model was based on the mass balance principle and the assumption that an n-order decay model describes the complex processes of pollutant removal (for example sedimentation, biodegradation, filtration, plant uptake, and chemical precipitation). The model was defined by the parameters of removal rate (k) and the decay order (n). For each treatment system, a collection of storm events was monitored between 2004 and 2006. Monitoring of the treatment systems was performed in a side by side fashion so that each system received the same stormwater quantity and quality. This configuration made possible a comparison of the calibrated parameters obtained for each system. The best set of parameters of the decay model was determined by using a simulated annealing technique as part of the optimization process. Monte Carlo simulations were performed to describe the uncertainty of the estimated effluent concentrations. The gravel wetland achieved the highest median DIN and TSS removal rates. For TPH-D, the highest median removal rates were achieved by the retention pond and gravel wetland. The sand filter and the gravel wetland achieved the highest median Zn removal rates. First and second order decay models were more likely to describe the observed effluent concentrations.;A Bayesian statistical approach for determining parameter uncertainty of the stormwater treatment model is presented. For this model, it was found that a second order decay model was more likely to reproduce estimated effluent concentrations. Mean removal rate values were computed from the posterior distributions. Specifically, for the gravel wetland: kTSS = 59, kZn = 2115, kTPH-D = 88, kDIN = 7; for the sand filter: kTss = 1.7, kZn = 1568, kTPH-D = 57, kDIN = 2; and for the retention pond: kTss = 0.8, kZn = 4645, kTPH-D = 68, kDIN = 8 (k in units of (mg/l)-1/day).