07/27/2020 Congratulate Soo Bin Chun on her MS thesis defense

Soo Bin Chun successfully defended her MS thesis today. Congratulations!

Title: Predicting Combined Sewer Overflow Occurrence from Rainfall Data

Abstract

The occurrence of combined sewer overflows (CSOs) is one of the most pressing environmental issues facing many cities with combined sewer system (CSS). Combined sewer often exceeds the capacity of the CSS or wastewater treatment plant and discharges directly into water bodies without treatment. CSO discharges contain a variety of contaminants and degrade receiving water quality. To manage CSO, it is necessary to characterize its occurrence, which generally requires monitoring CSO for long time periods or building complex hydrological models for simulation. However, monitoring CSO is very expensive and hydrological models are often too time-consuming to predict CSO in real time. Thus, this work aims to develop a simpler method for predicting CSO occurrences using the rainfall characteristics and investigate which rainfall characteristics can be the best predictor. The city of Buffalo, New York, is taken as a case study. A calibrated hydrological model was used for one-year continuous simulation to generate CSO discharges at 52 CSO outfalls. An R-language code was developed to analyze the characteristics of rainfall and CSO events. Rainfall characteristic quantities such as duration of the rainfall event, rainfall depth, and maximum rainfall intensity were assessed for their prediction power of CSO occurrence. Prediction accuracy with single characteristics (rainfall depth or maximum rainfall intensity) ranged from 80 to 100%, while rainfall duration was found not a good predictor. Rainfall depth was found a better predictor for sewersheds with larger areas and smaller imperviousness, while maximum rainfall intensity was found better for sewersheds with smaller areas and bigger imperviousness. Also, the decision tree algorithm was utilized to combine more than one rainfall characteristics, aiming for better accuracy. The average prediction accuracy was slightly improved from 93% (using single characteristics) to 95%. These results reveal that rainfall data can yield accurate prediction of CSO occurrence, and the proposed method can be an effective alternative to complex hydrological models and/or expensive monitor for managing CSOs.

Advisor: Dr. Zhenduo Zhu

Committee: Dr. Joseph Atkinson