Efficient SST prediction in the Red Sea using hybrid deep learning-based approach

Abstract

Prediction of Surface Sea Temperature (SST) is of great importance in seasonal forecasts in the region and beyond, mainly due to its significant role in global atmospheric circulation. On the other hand, SST predicting from given multivariate sequences using historical ocean variables is vital to investigate how SST physical phenomena generated. This paper seeks to significantly improve the prediction of Surface Sea Temperature (SST) by combining two machine learning methodologies: short-term memory networks (LSTM) added to Gaussian Process Regression (GPR). We developed a data-driven approach based on deep learning and GPR modeling to improve the prediction of SST levels in the red sea based on meteorological variables, including the hourly wind speed (WS), air temperature at 2m (T2), and relative humidity (RH) variables. The coupled GPR-LSTM model may potentially carry both flexibility and feature extraction capacity, which could describe temporal dependencies in SST time-series and improve the prediction accuracy of SST. It is necessary to indicate that these types of hybrid-based approach architectures have not used before in SST time-series prediction, so it is a new approach to deal with these types of problems. The results demonstrate a significant improvement when this hybrid model is compared to LSTM and the most frequently used ensemble learning models.