Modeling River Flow Forecasting In Medium And Small-Scale Watershed

River flow rate estimation is a widespread activity among hydrologists,which takes into account several meteorological factors.While some approaches need a thorough understanding of the hydrological process and the relationships among the different parameters that lead to runoff,others do not require it.This research work essentially proposes an approach based on a particular type of artificial neural network called Long Short-Term Memory(LSTM)to forecast the Jinghe river flow rate.LSTM is a very powerful prediction tool that takes into account temporal dependencies.The model is going to predict the river flow rate of each day based on the previous day flow rate and weather conditions.The model's performance highly depends on the so-called hyperparameters values.To find the best combination of hyperparameters the random search method has been used,that is,testing combinations of randomized hyperparameters until the best possible ones are achieved.Then the built model is compared with a watershed model called GWLF(Generalized Watershed Loading Function).As the GWLF works with monthly values,daily observed and forecasted flow were converted to total monthly flow in order to be able to compare the performances.The use of two performances indicators that are the RMSE and the NSE coefficient,indicated that the proposed LSTM model is much better in both calibration and validation period.