Calibrating the stormwater management model: an automated, genetic approach

Edward Tiernan and Ben Hodges


Parameter calibration is considered a critical, albeit arduous, step for reliable performance of the Stormwater Management Model (SWMM), one that that stormwater management designers often undertake manually. This research presents an open-source, automated calibration routine that increases the efficiency and accuracy of the model calibration process. The routine first represents the catchment network as a graph object using the NetworkX python package for flexibility in handling real-world calibration data availability. Once the calibratable subset of the system is identified, a multi-objective, genetic algorithm (modified Non-Dominated Sorting Genetic Algorithm II, NSGA-II) determines the Pareto front for the objective functions within the parameter space. A significant benefit of the genetic approach is the highly parallelizable nature of the NSGA-II. Each generation of potential solutions can be simulated by SWMM in parallel, such that the algorithm’s efficiency is only restricted by the number of generations required to achieve convergence to the Pareto front. The solutions on this Pareto front represent the optimized parameter value sets for the catchment behavior that could not have been reasonably obtained through manual calibration. A specific solution among this Pareto set can be chosen by assigning weights to the objective functions.

Permanent link: