Predicting of Runoff Using an Optimized SWAT-ANN: A Case Study

Abstract

Study region: The Baliqlu Chai Watershed is located in Iran in Ardabil province, with area of 1036 km2 and average rainfall of 280.9 mm. The area has been plagued with mismanagement of water resources. Due to the degradation of vegetation, runoff increases, that it can cause problems. Thus, forecasting runoff over the next few years can help manage water resources. Study focus: In this study, used of the SWAT and ANN models and an improved model of a metaheuristic employed for resolving the explained shortcoming. The method of optimization is based on a Mutated model of the Whale Optimization Algorithm (MWOA) that enhances the expected results by reducing the error in the ANN. First, the runoff estimated by the SWAT model, for this purpose, is used of 30 years of statistical data for the calibration and validation model. simulated runoff transmitted as input to the ANN and evaluated, and the MLP/MWOA algorithm used to improve the accuracy of the predicted runoff. New hydrological insights: The results show that according to the regression model, the points distribution in SWAT-MLP/MWOA model is less and has the best linear fit. The values obtained from statistical indices showed that the SWAT-ANN is better than the SWAT model because it has R2 = 0.80, RMSE = 1.61, NSE = 0.79, RE=-0.11. But SWAT-MLP/WOA model has R2 = 0.84, RMSE = 1.42, NSE = 0.81 and RE=-0.09, so SWAT-MLP/ MWOA model is presented as the best model for runoff prediction.

Keywords

• Runoff
• forecasting
• model calibration
• artificial neural network
• whale optimization algorithm
• Mutation watershed