Abstract: Sea surface temperature (SST) is an important factor in the study of global climate and marine environment, and accurate SST prediction is of great significance in the fields of marine ecological protection and research.The current SST prediction methods have the problem of not being able to make full use of the spatial correlation information and temporal evolution information of the data.To solve this problem, this paper proposes a multi-scale large kernel gated attention network(MuLKGANet), which consists of an encoder, a spatio-temporal translator and a decoder.Its encoder and decoder are used for spatial feature extraction and reduction. The core spatio-temporal translator aims to fully extract the global and local spatial dependencies of the data using multiple convolutional kernels, and learn the temporal evolution of the data using the gated attention mechanism. The experimental results show that, compared to the widely used ConvLSTM model, the present method improves the mean-square error metric by about 83% for 12-hour sea surface temperature prediction, and the prediction error is significantly smaller near the coast.

Key words: sea surface temperature, attention mechanism, multiscale, gated mechanism, spatial and temporal prediction