Support vector regression optimized by meta-heuristic algorithms for daily streamflow prediction

Accurate and reliable prediction of streamflow is vital to the optimization of water resources management, reservoir flood operations, catchment, and urban water management. In this research, support vector regression (SVR) was optimized by six meta-heuristic algorithms, namely, Ant Lion Optimization (SVR-ALO), Multi-Verse Optimizer (SVR-MVO), Spotted Hyena Optimizer (SVR-SHO), Harris Hawks Optimization (SVR-HHO), Particle Swarm Optimization (SVR-PSO), and Bayesian Optimization (SVR-BO) to predict daily streamflow in Naula watershed, State of Uttarakhand, India. The significant inputs and parameter combinations for hybrid SVR models were extracted through Gamma Test before processing. The results obtained by hybrid SVR models during calibration (training) and validation (testing) periods, which were compared against observed streamflow using performance indicators of root mean square error (RMSE), scatter index (SI), coefficient of correlation (COC), Willmott index (WI), and by visual inspection (time-series plot, scatter plot and Taylor diagram). The results of comparison demonstrated that SVR-HHO during calibration/validation periods (RMSE = 92.038/181.306 m(3)/s, SI = 0.401/0.715, COC = 0.881/0.717, and WI = 0.928/0.777) had superior performance to the SVR-ALO, SVR-MVO, SVR-SHO, SVR-PSO, and SVR-BO models in predicting daily streamflow in the study basin. In addition, the new HHO algorithm outperformed the other meta-heuristic algorithms in terms of prediction accuracy.