Evaluation of Classical, Conceptual IHACRES and Hybrid ARMA-ANN Models in Simulation and Prediction of Daily Discharge of Maroun River

The objective of this research is to compare the performance of linear time series models of Box-Jenkins and IHACRES, multilayer perceptron ANN and hybrid ARMA-ANN in order to simulate and predict the daily discharge of Maroun River. For this purpose, daily discharge data of (1991-2006) were used for calibration and data of (2007-2017) were used for verification of the models. Schwartz (SBC) and Akaike information criterion (AIC) were used to select the best model. Different scenarios, learning algorithms and transfer functions with various neuron structures were used to develop the ANN model. The first scenario with less parameters and delay time was selected as the best ANN model in prediction of daily flow rate. Evaluation indices showed that the conceptual model performance in verification stage was better than that in calibration stage. Also, the 4th order moving average model with R2=0.61 had the weakest performance as compared to the other Box-Jenkins models. Evaluation indices indicating a relative promotion for ARMA-ANN hybrid model as compared to the other proposed models.  As, ARMA-ANN hybrid model obtained the highest  R2=0.86 and Nash-Satcliffe coefficient equal to 0.81. The results prove the ability of ARMA-ANN hybrid model for simulation and prediction of daily discharge, as compared with other models.