Streamflow Prediction Using a Hybrid Methodology Based on Convolutional Neural Network and Long Short-Term Memory
Juan F. Ramirez Rochac, Nian Zhang, Tolessa Deksissa, Wagdy Mahmoud
Abstract
This paper proposes a real-time, long short-term memory (LSTM) based low flow forecast system, while utilizing historical streamflow to make prediction of the probability of flows dropping below drought trigger levels for the Potomac River basin. The proposed recurrent neural network learns to predict the value of the next time step of the time sequence. We evaluate the prediction accuracy of the proposed LSTM-based model with real-world data and compare it to other state-of-the-art baseline models as well as other LSTM variants. The experimental results show that the prediction accuracy of the proposed method outperforms other methods. This design will help improve the performance of the decision support system for drought management.