A high-resolution ($\mathrm{1}/\mathrm{20}{}^{\circ }$ ) global ocean general circulation model with graphics processing unit (GPU) code implementations is developed based on the LASG/IAP Climate System Ocean Model version 3 (LICOM3) under a heterogeneous-compute interface for portability (HIP) framework. The dynamic core and physics package of LICOM3 are both ported to the GPU, and three-dimensional parallelization (also partitioned in the vertical direction) is applied. The HIP version of LICOM3 (LICOM3-HIP) is 42 times faster than the same number of CPU cores when 384 AMD GPUs and CPU cores are used. LICOM3-HIP has excellent scalability; it can still obtain a speedup of more than 4 on 9216 GPUs compared to 384 GPUs. In this phase, we successfully performed a test of $\mathrm{1}/\mathrm{20}{}^{\circ }$ LICOM3-HIP using 6550 nodes and 26 200 GPUs, and on a large scale, the model's speed was increased to approximately 2.72 simulated years per day (SYPD). By putting almost all the computation processes inside GPUs, the time cost of data transfer between CPUs and GPUs was reduced, resulting in high performance. Simultaneously, a 14-year spin-up integration following phase 2 of the Ocean Model Intercomparison Project (OMIP-2) protocol of surface forcing was performed, and preliminary results were evaluated. We found that the model results had little difference from the CPU version. Further comparison with observations and lower-resolution LICOM3 results suggests that the $\mathrm{1}/\mathrm{20}{}^{\circ }$ LICOM3-HIP can reproduce the observations and produce many smaller-scale activities, such as submesoscale eddies and frontal-scale structures.

Short-term sea surface temperature (SST) forecasts are crucial for operational oceanology. This study introduces a specialized Transformer model (U-Transformer) to forecast global short-term SST variability and compares its performance with Convolutional Long Short-Term Memory (ConvLSTM) and Residual Neural Network (ResNet) models. The U-Transformer model forecast consistently outperformed the ConvLSTM and ResNet models, especially in regions with active mesoscale eddies. Globally, the U-Transformer model achieved SST root mean square errors (RMSEs) ranging from 0.2 °C at a 1-day lead time to 0.54 °C at a 10-day lead time during 2020–2022, with anomaly correlation coefficients (ACCs) decreasing from 0.97 to 0.79, respectively. However, in regions characterized by active mesoscale eddies, RMSEs from the U-Transformer model exceeded the global averages by at least 40%, with values in the Gulf Stream region reaching more than twice the global average. 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 at a 1-day lead time to 0.73 at a 10-day lead time. Specifically, the ACC value dropped to 0.89 in the Gulf Stream region at a 3-day lead time, while maintaining 0.92 globally. These findings underscore the importance of advanced approaches to enhance SST forecast accuracy in challenging active mesoscale eddy regions.

Sea surface temperature (SST) is critically important for understanding ocean dynamics and supporting various marine activities, making accurate short-term SST forecasting highly significant. However, accurately modeling the multi-scale variability of SST remains challenging for existing deep learning (DL) models. This study introduces the Coupled Transformer-CNN Network (CoTCN), a hybrid architecture designed to leverage the multi-scale variability of SST. The CoTCN combines the strengths of Transformers and convolutional neural networks (CNNs), significantly enhancing SST forecasts’ spatial continuity and predictive accuracy. Compared to five state-of-the-art DL models based on Transformer or CNN that include ConvLSTM, ConvGRU, AFNO, PredRNN, and SwinLSTM, CoTCN demonstrates superior performance in global and local areas of SST forecasting. At 1-day lead time, CoTCN reduces the global average root mean square error (RMSE) by over 15%, with forecast errors ranging from 0.20°C to 0.53°C across 1–10 day lead times. Moreover, the CoTCN effectively mitigates the checkerboard artifacts inherent to the Vision Transformer architecture. These findings highlight the effectiveness of CoTCN in capturing SST’s multi-scale features and underscore the promising potential of hybrid architectures for future DL models.