Abstract: The correct and accurate estimation of the river flow using different models is a significant issue in water resources research. In this research, two hydrometric stations of Sari-Qomish and Nizam-Abad located in West Azarbaijan province were used to accurately estimate the daily flow of Zarineh-Rood River. To reach this aim, Empirical Mode Decomposition (EMD) preprocessing algorithm was used to deal with the complexity and instability of time series data. EMD is a data analysis method for extracting signals in data generation through non-linear and non-stationary operations. In this research, the method of gene expression programming model and artificial neural network model were used. The results of the research showed that the performance of the gene expression programming model was equal and sometimes less than the performance of the artificial neural network. However, the combination of the two mentioned models with the technique (EMD) increased the accuracy of the model and reduced the error in simulating the river flow in Sari-Qomish and Nizam-Abad stations.
 
Introduction: Streamflow prediction has been one of the most important challenges in water resources management in recent decades, so researchers have used different methods to do so. Typically, physics-based numerical models are used in modeling or better modeling to estimate Streamflow (Partington et al., 2012). In recent years, the use of artificial intelligence methods such as artificial neural network (ANN), support vector machines (SVM), gene expression programming (GEP) to model and solve problems in water engineering due to its considerable benefits, is current. (Rezaie-Balf et al., 2019). The Empirical Mode Decomposition (EMD) was proposed as a noise-based data analysis method by (Huang et al., 1998) which separates the higher frequency input series into the solved (frequency) components with lower frequency. Amir ashayeri et al. Studied 3 stations in West Azerbaijan to predict daily reference transpiration evaporation with ANN models, tree model and combination of these models with Empirical Mode Decomposition. The results showed that the Empirical Mode Decomposition hybrid method with tree model was the best model. Was selected (Amirashayeri et al., 2020). The purpose of this study was to provide an accurate prediction model for Streamflow at Sari-Qomish and Nezam-Abad stations on the Zarrineh River in the Urmia Lake basin, using hybrid Gene Expression Programing (GEP) models and artificial neural network (ANN) models with The Empirical Mode Decomposition to have a more accurate combination for daily flow in the desired stations.
 
Methodology: In this study, to accurately estimate the daily streamflow of the Zarrineh River, two hydrometric stations of Sari-Qomish and Nezam-Abad located in West Azerbaijan province were used. Recently, well-known AI-based intelligent models along with hybrid models have been accepted in all predictions. In this study, to deal with the complexity and instability of time series data, the Empirical Mode Decomposition (EMD) algorithm was used. To extract signals in the information generated by nonlinear and non-static operations. In this study, the gene expression programming model (GEP), which is part of the circulatory algorithm, the artificial neural network model, was used. This model simulates the ability of the human brain to find patterns and learn from trial and error. daily streamfllow of the Zarrineh River in West Azerbaijan province were used, during a 26-year period, are investigated and then estimated. Accuracy of models is evaluated by three indexes. These three indexes are mean absolute error (MAE), root mean squared error (RMSE) and correlation coefficient (R²).
 
Results and Discussion: The results showed that the performance of the gene expression programming model was equal to and sometimes lower than the performance of the artificial neural network (ANN). However, the combination of the two models with the EMD algorithm increased the model's accuracy and reduced the error in simulating streamflow in Sari-Qomish and Nezam-Abad stations. The results showed that in the test phase, combining the GEP method with EMD improved the coefficient of correlation by 7.29% and 2.15% for Sari-Qomish and Nezam-Abad stations, respectively. Also, in the test phase, combining the ANN method with EMD improved the coefficient of explanation. 3.06% and 3.09% for Sari-Qomish and Nezam-Abad stations, respectively.
 
Conclusion: Since river flow forecasting plays an important role in resource planning and management, the correct and accurate estimation of the river flow using different intelligent artificial models is one of the issues that are being investigated by researchers in water resources.  In this research, the daily flow data of two stations of Sari-Qomish and Nizam-Abad in a period of 26 years (1369 to 1395) and up to three-time delays were applied for estimation of the river flow, through artificial network model prediction models (ANN), and gene expression programming (GEP). And the preprocessing method of Empirical Mode Decomposition (EMD) algorithm was used to increase the accuracy of two models. The results showed that pre-processing data using the experimental mode analysis algorithm increases the accuracy of the gene expression programming model by 29.7% and 2.15% for Sari-Qomish and Nizam-Abad stations, respectively. Furthermore, the proposed combined method, for example, at the Sari-Qomish station has far less error than the Nizam-Abad station in daily discharge forecasting, and for the artificial neural network model increases the accuracy of the indicators.