Streamflow forecasting using long short-term memory network

Abstract

The effective prediction of streamflowis important for water resources planning and management. In this study, we explored the potential use of long short-term memory network (LSTM), and proposed an approach based ona network containing two-layer LSTM on top of a dense layer for multi-step ahead streamflow forecasting. Two streamflow datasets are used to evaluate the performance and applicability of the proposed approach. The prediction accuracy of LSTM is compared with that of multi-lay perceptron (MLP). The obtained results indicate that LSTM is applicable for time series prediction and is a superior alternative when longer time steps ahead prediction are expected.