Due to inaccurate parameter estimation, there is a high degree of uncertainty associated with hydraulic and hydrologic models performance. In this study, the posterior probability distributions of the Stormwater Management Model (SWMM) model parameters are generated using the Bayesian optimization algorithm and used for model calibration. A new tool, BayesOpt-SWMM, is developed using Gaussian process-based Bayesian optimization algorithm to make faster convergence to optimal model parameter performance. The tool provides iterative runs functionality which enables the multi-task processing, specifying the objective function, and constraining model parameter range through a set of meta-heuristic optimizers including, the standard Gaussian Process (GP) model and Markov chain Monte Carlo (MCMC) algorithms for inference in GP models (GP_MCMC) enhanced parameter searching process of SWMM model. Moreover, three acquisition functions, expected improvement (EI), maximum probability of improvement (MPI), and Lower Confidence Bound (LCB) are combined with GP and GP_MCMC to develop a total of six optimizers for the SWMM model. The model parameters for optimization are classified into two sets of a) fixed, and b) dynamic. Fixed parameters are rainfall event-independent parameters, and the dynamic parameters vary from an event to another event to improve model accuracy. This tool has been tested for the Rocky Branch Watershed located in Columbia, South Carolina. A total of 11 SWMM model parameters for various sub-catchments and conduits, including 14 fixed and 47 dynamic parameters, are optimized respectively for different sub-basins to maximize the overall accuracy using Nash–Sutcliffe model efficiency coefficient (NSE). Running the optimization tool results in updating of the posterior distribution function along with optimal sets of parameters. The validation results show that the model optimizers converge to optimal solutions through 40-50 iterations. BayesOpt-SWMM will serve as a new optimization and uncertainty analysis tool for faster development and deployment of more reliable SWMM models in future.
