An Improved LSTM Model with Attention Mechanism for Sea Surface Temperature Prediction Around the Korean Peninsula

Ocean temperature has an important influence on the distribution and migration of marine fishes, and with the improvement of modern remote sensing information acquisition technology, ocean data are constantly being improved. Sea surface temperature (SST) around the Korean Peninsula is influenced by several complex factors. In this paper, an improved long short-term memory network (LSTM) model with attention mechanism is proposed to predict the SST for the next 5 days, which extracts more temporal and spatial information by assigning new weights through the attention mechanism. The experimental results show that the root mean square error (RMSE) of the proposed model on day 1 is 0.2181°C and the prediction accuracy (PACC) is 99.14%, which is a 20% reduction in RMSE compared to the existing similar networks.