Multi-factor deep learning model for sea surface temperature forecasting

Abstract

Accurately predicting sea surface temperature (SST) is crucial for marine environmental monitoring and climate research. However, existing ocean model approaches often struggle to capture complex spatiotemporal patterns and are limited by their reliance on thermodynamic equations to impose oceanographic constraints. To address these challenges, we propose a multi-sensor SST prediction model that integrates Long Short-Term Memory (LSTM) networks, convolutional neural networks (CNNs), and an attention mechanism to directly incorporate physical variables such as temperature, salinity, density, and current velocity. By bypassing the need for explicit physical equation constraints, our model effectively learns complex relationships from multi-source data. Experimental results show that our approach significantly improves predictive accuracy across various ocean regions, providing a robust solution for both short-term and long-term SST forecasting.