Evaluation of statistical models and modern hybrid artificial intelligence in the simulation of precipitation runoff process

To date, the rainfall-runoff process is among the most significant and complicated hydrological phenomena, regarding taking appropriate measures in terms of floods and droughts and surface water resources management. A proper understanding of the basin's behavior can play an effective role in model selection, such that simulation may become time saving. Providing the water of several large rivers in Iran, the Karkheh catchment is of vital significance in order for its precipitation runoff processes to be modeled. In this study, statistical and artificial intelligence (AI) approaches, i.e. multivariate linear regression (MLR), artificial neural network (ANN), support vector regression (SVR), wavelet SVR (WSVR), black widow optimization-SVR (BWO-SVR), and the algorithm of innovative gunner-SVR (AIG-SVR), were used to simulate the runoff process of the Karkheh catchment on a daily time scale during the statistical period 2010–2020. To evaluate the simulation performance, statistical indices were employed, including coefficient of determination (R²), root mean square error (RMSE), mean absolute error (MAE), Nash–Sutcliffe efficiency coefficient (NSE), and percentage bias (PBIAS). As it was demonstrated, the studied models exhibited better performance in composite structures. Additionally, AI models have less error and better performance than statistical models. Further, the results highlighted that the AIG-SVR has the greatest efficacy with the least error in comparison with other models (R² = 0.978–0.985, RMSE = 0.004–0.008 m³/s, MAE = 0.002–0.004 m³/s, NSE = 0.984–0.991, and PBIAS = 0.001). Finally, the use of hybrid AI models is an effective approach in the rainfall-runoff processes and can be considered as a suitable and rapid solution in water resources management.