Effect of Modeling Setup Parameters in SWMM Predictions of Large-Scale Tunnel Filling

Due to its robustness, versatility, and open-source, EPA’s Storm Water Management Model (SWMM) has become one of the most popular applications used for urban hydraulics analysis nowadays. One reason that explains the model popularity is the ability to predict flow in complex stormwater systems with reasonable accuracy in realistic inflow conditions. However, in highly dynamic conditions associated with extreme inflows, recent research indicated that SWMM accuracy can be improved with spatial accuracy and smaller time steps. Yet, even among approaches to representing stormwater systems using spatial discretization, there are alternative approaches regarding pressurization algorithms that are not fully understood. The aim of this work is to evaluate how modeled results of piezometric head and flow hydrograph in a stormwater tunnel are affected by varying discretization and pressurization algorithms in SWMM. Comparisons with a combination of three different spatial discretization (link-node and two different spatial discretizations), three different algorithms to solve pressurized flow (EXTRAN, original SLOT, and a custom SLOT with pressure wave celerity of 250 m/s), and 10 different time step discretizations with Hydraulic Analysis for Sewer and Tunnels (HAST), a model specifically designed to simulate surges in tunnels, are presented. These alternative solutions are placed on a real-world stormwater tunnel subjected to a rain event. SWMM’s outputs are evaluated using flow continuity errors and Nash-Sutcliffe normalized statistic. Results show that in relation to the Richmond Tunnel, the DxD 10 spatial discretization proved to be the most reliable, especially when combined with the SWMM standard Preissmann slot algorithm and a time step factor ranging from 0.01 to 1 from the time step calculated using Courant number. The EXTRAN algorithm only showed good results when combined with DxD 10 spatial discretization and time step factor of 1. The DxD 05 discretization presented acceptable results when used with SWMM standard Preissmann slot algorithm and time step factor of 0.01 and 0.02. Link-node setups did not show satisfactory results. Furthermore, this study demonstrates that ensuring small flow continuity errors is not enough to achieve representative flow rates using SWMM, highlighting that another parameter should be taken into account to evaluate numerical oscillation.