SeaNet: A Deep Learning Architecture for Enhanced Sea Surface Temperature Forecasting

Sea Surface Temperature plays a crucial role in the Earth's climate system, impacting weather patterns, ocean circulation, and marine habitats. Accurate SST forecasting is vital for various environmental and climatic applications, aiding in informed decision-making. This research proposes two neural architectures for automated time series forecasting of sea surface temperature, SeaNet-1 which is a combination of Long Short-Term Memory and Gated Recurrent Unit networks, and SeaNet-2, which is a transformer-based model with a relatively low number of parameters as compared to SeaNet-1. Experimental findings showed that the suggested models can predict sea surface temperature effectively and precisely. The experimental results achieved a remarkable R-squared score of 95.16% and an incredibly low RMSE of 27.28%. The proposed models perform better in terms of performance and computational efficiency when compared with other existing state-of-the-art models. The proposed models offer promising tools for oceanography, climate studies, and environmental management, supporting informed decision-making and sustainable resource utilization.