Answers are grounded in this paper and cite their sources.
Sea surface temperature (SST) exerts a critical influence on diverse physical and biological processes within the marine environment. Nevertheless, accurate short-term SST forecasting still remains challenging. Current deep learning (DL) models have two key limitations: first, the convolutional neural network (CNN)-based methods fail to capture global spatiotemporal dependencies; second, Transformer-based approaches suffer from relatively high computational costs. To address these issues, this study introduces a VMamba–CNN fusion network (VMCF), a dual-branch architecture designed to effectively characterize the multiscale temporal and spatial variability of SST. The VMCF combines the powerful local feature extraction capability of CNNs with the efficient linear-computability global spatiotemporal representation of VMamba, further strengthened via channel-wise feature fusion. Extensive experiments on the OISST dataset demonstrate significant performance improvements across all evaluation metrics, clearly validating the superiority of VMCF in SST forecasting. Specifically, it achieves root-mean-square error (RMSE) values ranging from 0.17 °C to 0.51 °C across 1–10-day lead times, outperforming five state-of-the-art DL models.
No takes yet. Share an insight, caveat, or question.
Zhan et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: