Key points are not available for this paper at this time.
Combined sewer systems are widespread in America and Europe. They often face limitations in transport or treatment capacity, especially during heavy rain events or thaw periods, resulting in combined sewer overflows (CSOs). Predictive modeling for CSOs is essential in a risk management context, and some studies have presented methods to categorize precipitations based on their potential to generate overflows. However, the precipitation classification is usually based on a few characteristics, and its predictive power is limited. The objective of this study is to present a simple yet powerful method to categorize precipitation for predicting CSO occurrences. A prediction model, based on an optimized classification tree, is proposed to predict CSO occurrences as a function of publicly accessible precipitation data. We fit the model on 9 overflow outlets in Montréal city from 2013 to 2019 and use this model to predict CSOs in 2020. The results showed a very good predictive power of overflows, with a prediction rate of 89%, a sensitivity rate of 83%, and a specificity rate of 91%. The method is also more accurate than the 5-category classification currently used by the City of Montréal. The proposed method could be easily applied to another region where CSO data are available, providing a simple and rigorous method for predicting CSOs across urban drainage networks containing many overflow outlets.
Jalbert et al. (Mon,) studied this question.