CausalSFNN: A Causality-Aware Spherical Fourier Neural Network for Sea Surface Prediction

Sea surface temperature (SST), sea surface salinity (SSS), and absolute dynamic topography (ADT) are among the most critical Essential Ocean Variables (EOVs) for understanding global ocean systems. Accurate spatiotemporal forecasting of these variables is essential for climate change monitoring, marine ecosystem preservation, and disaster early warning. However, existing deep learning approaches often overlook the causal dynamics of physical systems and the inherent spherical geometry of the earth. In this work, we propose a novel Causality-Aware Spherical Fourier Neural Network (CausalSFNN), which integrates causal convolution, temporal attention mechanisms, and spherical neural operators to effectively learn the multivariate spatiotemporal evolution of global ocean fields. Extensive experiments conducted on multisource satellite remote sensing data from 2011 to 2023 demonstrate that CausalSFNN significantly outperforms existing models in terms of accuracy, highlighting its strong potential for global ocean forecasting tasks.