Deep learning: Long short-term memory in hydrological time series

Abstract

Deep learning, defined as a subset of machine learning in artificial intelligence with artificial neural networks that are able to learn without supervision and also known as deep neural network, has been popularly employed in the literature. Among deep learning techniques, long short-term memory (LSTM) models received much attention, especially in hydrological fields, due to their accurate predictive performance for long-term as well as short-term time interval. Developments of these models and their variants are described. Also described is the derivation to estimate the weights (or parameters) of LSTM models, along with training of their networks. Furthermore, hydrological applications of LSTM models are presented. Finally, the difficulty of LSTM models to apply in hydrological studies as well as future possible developments is discussed. It is concluded that LSTM models can be a promising technique in hydrological applications by overcoming the shortcoming that LSTM models have. © 2023 Elsevier Inc. All rights reserved.