Univariate streamflow forecasting using deep learning networks

Streamflow plays a vital role when deciding on water resource planning and management. According to data resources and their availability, streamflow prediction for different regions has been made so far using distinctive models, such as physically based hydrologic models, statistical models and machine learning algorithms. This article describes the applications of recently generated deep learning N-BEATs algorithm by modifying the basic structure with nonlinear predictor coefficient and long short-term memory (LSTM) for univariate streamflow forecasting in the Ponnaiyar River Basin. To develop the model, the model utilised the data of three streamflow stations that contain 40 years of Villipuram discharge and 36 years of Gummanur and Vazhavachanur discharge. The experimental analysis is performed to analyse the performances of the proposed model. From the results, both models performed well during the training and validation period. Similarly, the accuracy estimation of validation conducted by N-BEATs and LSTM Nash-Sutcliff efficiency for upstream (0.827 and 0.792) and midstream (0.9407 and 0.865) have revealed that the modified N-BEATs accomplished superior outcomes than LSTM, respectively. It is concluded that the proposed N-BEATs model can be applied for univariate streamflow forecasting which simplifies the data complexity for model establishment.