In this study, the performance of SWAT hydrological model and three computational intelligence methods used to simulate river flow are investigated. After collecting the data required for all models used, the calibration and validation stages were performed. Using the SWAT model and three methods of the Extreme Machine Learning (EML), the Support Vector Regression (SVR), and the Least Squares Support Vector Regression (LSSVR), Maharlu Lake Basin stream flow was simulated and the results obtained at Shiraz station were used for this study. A noise reduction filter was employed to improve the results from the computational intelligence methods, and SUFI-2 algorithm was used to analyze the uncertainty of the SWAT model. Finally, in order to evaluate the models developed and the SWAT model, three statistics (RMSE), (R²), and (NS) coefficient were used. The results indicated that the SWAT model and the machine learning models were generally appropriate tools for daily flow modeling, but the LSSVR model showed less errors in both learning and testing, with the coefficients NS = 0.997 and R² = 0.997 in the calibration stage and NS = 0.994 and R² = 0.994 in the validation stage, which prove their better performance compared to the other methods and the SWAT model.
In this study, we investigated the performance of SWAT hydrological model and three computational intelligence methods for simulating a river flow.
The calibration and validation stages were performed after collecting the data required for all models.
Using SWAT model and three methods of Extreme Learning Machine (ELM), Support Vector Regression (SVR) and Least Squares Support Vector Regression (LSSVR).