Rainfall-Runoff modeling using Deep Learning model (Case Study: Galikesh Watershed)

Artificial neural networks (ANN) are one of the data mining methods applied by many researchers in different fields of studies such as rainfall runoff modeling. To improve the performance of these networks, deep learning neural networks were developed to increase modeling accuracy. This study evaluated deep learning networks to improve the performance of artificial neural networks in Galikesh watershed and to predict discharge for 1, 3, 6 and 12-month time scale based on 1 to 5 month time scale lags made in precipitation and temperature data. Based on 70% and 30% of the data used for training and test respectively the results demonstrated that in all time steps, the deep learning neural network improved the performance of artificial neural network and on average RMSE decreased in both training and test from 0.68 to 0.65 and 0.84 to 0.73 respectively. Moreover, R-square was increased on average from 0.57 to 0.62 and 0.51 to 0.67 respectively in training and test. We can also denote the effect of temperature on the increase of accuracy of rainfall-runoff modeling.