swmm-breach: Probabilistic dam-breach hydrograph forecasting integrated with EPA SWMM and PCSWMM

Michael Brian Flynn

The U.S. EPA Storm Water Management Model (SWMM) and its commercial extension PCSWMM are among the most widely deployed open-source urban-hydrology engines worldwide, but neither provides a native facility for simulating embankment-dam, detention-basin, or lagoon failure. Practitioners working on dam-adjacent SWMM models typically generate a breach hydrograph in HEC-RAS, losing their SWMM network model in the process, or hand-construct a boundary condition in spreadsheets and paste it into the .inp file. Although Wahl (2004) demonstrated that empirical breach parameter regressions carry standard errors of estimate of approximately 0.10-0.40 in log10 units on breach-geometry parameters - propagating through the broad-crested-weir relationship to factor-of-two to factor-of-four multiplicative uncertainty on peak discharge - and recommended Monte Carlo simulation as the appropriate response, breach analyses in engineering practice continue to report a single deterministic peak. We present swmm-breach, an MIT-licensed Python package providing end-to-end probabilistic dam-breach hydrograph forecasting for EPA SWMM and PCSWMM. The package implements the Froehlich (1995, 2008) regressions with Wahl-style log-normal residual sampling and multi-model ensemble averaging, level-pool routing through a developing trapezoidal weir, and binary .inp/.out integration. Validation against three historical or independently-modeled cases spanning more than three orders of magnitude in reservoir volume (Anson lagoon 7.9 x 10^4 m^3, Lawn Lake 8.0 x 10^5 m^3, Teton Dam 3.1 x 10^8 m^3) shows that the probabilistic 5-95 percentile envelope brackets the reference peak in every case, while the deterministic single-model point estimate misses the observed Teton peak by approximately 80%.