A New Transformer Network for Short-Term Global Sea Surface Temperature Forecasting: Importance of Eddies

Short-term sea surface temperature (SST) forecasts are crucial for operational oceanology. This study introduced a specialized Transformer model (U-Transformer) to forecast global short-term SST variability and to compare with those from Convolutional Long Short-Term Memory (ConvLSTM) and Residual Neural Network (ResNet) models. The U-Transformer model achieved SST root mean square errors (RMSEs) of 0.2–0.54 °C for lead times of 1–10 days during 2020–2022, with anomaly correlation coefficients (ACCs) from 0.97 to 0.79. In regions characterized by active mesoscale eddies, RMSEs from the U-Transformer model exceeded the global averages by at least 40%, with increases exceeding 100% for the Gulf Stream region. Additionally, ACC values in active mesoscale eddy regions declined more sharply with forecast lead time compared to the global averages, decreasing from approximately 0.96 to 0.73. Specifically, in the Gulf Stream region, the ACC value dropped to 0.89 at a 3-day lead time, while the value can maintain 0.92 globally. Compared with the ConvLSTM and ResNet models, the U-Transformer model consistently delivered smaller RMSEs and larger ACCs, especially in regions with active mesoscale eddies. These findings imply the importance of advanced approaches to enhance SST forecast accuracy in regions with active mesoscale eddies.