Enhancing Accuracy of Forecasting Monthly Reservoir Inflow by Using Comparison of Three New Hybrid Models: A Case Study of The Droodzan Dam in Iran

Predicting the reservoir inflows plays a central role in the control and management of water resources and the related activities, such as the reservoir exploitation and flood/drought control. The complex nature of the hydrological systems and difficulties in their application processes have urged the researchers to look for more efficient reservoir-inflow modeling methods. Aimed at this objective, the current study developed three SVM-GA, ANFIS-GA and ARIMA-LSTM hybrid models, compared their performances with one another as well as with LSTM, SVM, ANFIS and ARIMA models and SWAT hydrological model in predicting the inflows to the Droodzan Dam reservoir in Fars Province, Iran, and evaluated their results using such statistical criteria as the RMSE, MAE, MAPE, MSE and R². In short, the results revealed that the combined methods performed better than the single models and ARIMA-LSTM and LSTM predicted the monthly reservoir inflows more accurately (R² = 0.9272, 0.8805, training, and R² = 0.9097, 0.8148, testing). Among the three combined and four single studied models, ARIMA-LSTM showed higher accuracy and had better potential in predicting the monthly reservoir inflows in arid and semi-arid regions.