GSSA: A Network for Short- to Medium-Term Regional Sea Surface Temperature Prediction

Accurate prediction of sea surface temperature (SST) is critical in marine sciences and related disciplines. The task is rendered challenging by complex environmental factors and the inherent nonlinearity of time. Current studies typically concentrate on either the structural aspect of sensor networks or their temporal correlations, often overlooking the interplay between these dimensions. To address this issue, this letter proposes a graph spatiotemporal sampling aggregation (GSSA) method. This method comprises two primary components: spatial information aggregation and temporal information convolution. Designed to simultaneously address spatial structural correlations and temporal variations, GSSA offers a comprehensive solution for SST prediction. The method utilizes neighbor sampling to endow the model with inductive capabilities, leveraging spatial information aggregation to capture spatial details and temporal information convolution to grasp temporal dynamics and significantly improving upon the SAGE aggregation’s ability to capture temporal correlations among node sequences. Experiments on SST data from the South China Sea and the Bohai Sea demonstrate that this method surpasses traditional machine learning methods in two different regions and at different prediction scales in the short and medium terms.