Enhancing Hydrological Model Calibration Using Particle Swarm Optimization and Genetic Algorithm in the Duck River Basin, USA

Rainfall-runoff simulation using hydrological models is highly dependent on accurate model parameters, input data quality, and watershed characteristics. However, identifying optimal parameters for a hydrological model requires significant time with a repetitive trial-and-error process. This study applied Particle Swarm Optimization (PSO) and Genetic Algorithm (GA) to automatic calibration of the HEC-HMS model for the Duck River Basin situated in south-central Tennessee, USA, with the aim of enhancing the performance of the HEC-HMS's internal auto-calibration tool. The HEC-HMS model for Duck River Basin calibrated for the period 2000-2005 gave a Nash-Sutcliffe Efficiency (NSE) equal to 0.827, a Standardized Root Mean Square Error (RMSE) of 0.4, and a percent bias of –4.67%, while for the validation period of 2008-2022, it was an NSE equal to 0.785, a Standardized RMSE of 0.5, and a bias of –2.16%. For PSO- and GA-based calibration, two years of flooding events were utilized in calibration and four years in verification for both single-objective and multi-objective calibrations. Several performance measures were considered in multi objective calibration, and various combinations were attempted. A fuzzy set-based approach was also utilized to convert the multi objective model to a single-objective model for comparative evaluation. The results show that while single- and multi-objective methods improved model efficiency, multi-objective calibration was more computationally demanding due to the complexity of optimizing several criteria at the same time for 21 sub-basins in the HEC-HMS model with varying loss, baseflow, transform, and routing parameters.