Improve the capability of physical model for runoff and sediment yield modelling with a hybrid artificial intelligence-based error updating system

Uncertainty of physical models is caused by the model's complexity as well as the accuracy of input variables. In present study, an error update model was developed based on artificial intelligence, namely Wavelet Multilayer Perceptron (WMLP), Wavelet Support Vector Machine (WSVM), Wavelet Multilayer Perceptron-Genetic Algorithm (WMLP-GA), and Wavelet Support Vector Machine-Genetic Algorithm (WSVM-GA). Sensitivity findings demonstrate that groundwater recession factor (ALPHA_BF), plant uptake compensation factor (EPCO), and curve number were the most sensitive for runoff in the Shivnath River basin, India, as well as four distinct parameters were shown to be highly sensitive to sediment yield at the Sigma station. During the calibration and validation phase, the SWAT model overpredicts runoff at the Simga and Pathardhi stations while under-predicting it at the Kotni station. SWAT model performance for runoff modelling is enhanced via the use of statistical indicators like as Percent of Bias (PBIAS) decreased from-35.355 to-20.503, Root Mean Square Error (RMSE) decreased from 319.905 to 25.191 m(3)/sec, and coefficient of determination increase from 0.712 to 0.904. At Simga station, the Nash-Sutcliffe efficiency (NSE) increase from 0.315 to 0.745, the PBIAS decreased from-39.502 to -22.946, the RMSE decreased from 186.711 to 24.432 m(3)/sec, and the coefficient of determination increased from 0.733 to 0.996. Further, NSE increased from 0.305 to 0.985 at Kotni station and PBIAS reduced from-114.026 to-9.214, RMSE reduced from 82.205 to 31.082 m(3)/sec, coefficient of determination increased from 0.779 to 0.976, NSE increased from 0.523 to 0.927 at Pathardhi station after applying an error update model. For sediment yield prediction, similar outcomes to runoff were obtained. The WSVM-GA model was reported to be the optimum error update model both for runoff and sediment yield modelling. It was recommended that the SWAT model might be applied for runoff and sediment yield simulation in the future.