Modeling the discharge of Karun River using a new method based on the joint LSTM and GRU neural networks

Modeling the river discharge is of great importance in water resources and risk management. This is especially important in mountainous areas since most of the low-income people in such areas are heavily dependent on agriculture and commercial activities such as electricity. In this regard, in recent years, machine learning models have received more attention due to their high accuracy in predicting through black box learning. Therefore, in this study, a combined approach has been proposed to predict the average monthly discharge of the Karun River. This method uses a combination of LSTM and GRU neural networks. The LSTM network is a deep learning neural network that has the ability to add the concept of time to modeling; therefore, this method has been considered in this study due to the nature of time series of the data. However, the utilized network in this method is considerably slow due to its large number of gates. Accordingly, to compensate the speed issue, the GRU layer method, as another example of deep learning networks, is used. To predict the average monthly flow of the Karun River in Dubai, the statistical data of Molasani station from April 1th, 1995 to March 20th, 2016, with five combination of river discharge inputs on monthly basis, has been used. The proposed approach is compared with other available methods such as support vector machine, adaptive fuzzy-neural inference system, and multiple linear regression model. The results show the high accuracy of the proposed approach compared to other methods.