Long short-term memory (LSTM) neural networks for short-term water level prediction in Mekong river estuaries

Abstract

This study firstly adopts a state-of-the-art deep learning approach based on a Long Short-Term Memory (LSTM) neural network for predicting the hourly water level of Mekong estuaries in Vietnam. The LSTM models were developed from around 8,760 hourly data points within 2018 and were evaluated using the Nash-Sutcliffe efficiency coefficient (NSE), mean absolute error (MAE), and root mean square error (RMSE). The results showed that the NSE values for the training and testing steps were both above 0.98, which can be regarded as very good performance. Furthermore, the RMSE were between 0.09 and 0.11 m for the training and between 0.10 and 0.12 m for the testing, while MAE for the training ranged from 0.07 to 0.08 m and varied from 0.08 to 0.10 m for the testing. The LSTM networks appear to enable high precision and robustness in water level time series prediction. The outcomes of this research have crucial implications in river water level predictions, especially from the viewpoint of employing deep learning algorithms.

Keywords

• long short-term memory (lstm)
• neural network
• water level
• mekong river estuary