Analysis of Sewer Network Performance in the Context of Modernization: Modeling, Sensitivity, and Uncertainty Analysis

Specific flood volume and degree of flooding are important parameters for evaluating the performance of stormwater networks. Hydrodynamic models are usually used to calculate these important measures, but this task requires the collection of detailed data on land use, the sewer network, rainfall, and flows, which are not always possible to obtain. The present research consists in the development of a methodology, using the USEPA Stormwater Management Model (SWMM), for simulating the performance of a stormwater network to determine whether it is in need of modernization. This determination is based on independent variables including rainfall data, catchment retention, and channel capacity. A logistic regression was developed to assess the sewer network performance based on simulation of specific flood volume and degree of flooding in the context of modernization. An extended sensitivity analysis was also used to assess the influence of rainfall intensity on the results of sensitivity coefficient calculations for the calibrated SWMM parameters. Using the extreme gradient boosting method, a tool has been developed to optimize the combination of SWMM parameters, reducing the uncertainty of simulation results, which can be used in the selection of their measurement methods prior to model development. It has been shown that, using the logistic regression model, it is possible to rapidly simulate the operation of a stormwater system to assess its need for modernization. It was confirmed that an increase in rainfall intensity leads to a significant decrease in the values of the calculated sensitivity coefficients associated with the SWMM parameters. The highest sensitivity coefficient was shown for a correction coefficient for percentage of impervious areas; for rainfall intensity $33–133\mathrm{L}·{(\mathrm{s}·\mathrm{ha})}^{-1}$ varied from 1.45 to 12.38. This result leads to a method for selecting specific rainfall events for calibration of the model, thereby improving the ability to assess the performance of the stormwater system. Interestingly, however, for the exemplary catchment in Kielce, Poland, the generalized likelihood uncertainty estimation (GLUE) method was used, combined with the XGboost machine learning technique, to determine that the reliability of the SWMM parameters has a negligible impact on the probability of a stormwater network failure.