Recently, Researcher He Xianqiang and his collaborators published a research paper entitled Physically-constrained flow learning reveals diurnal submesoscale surface currents from geostationary satellite observations in the top international remote sensing journal ISPRS Journal of Photogrammetry and Remote Sensing. The first author is Postdoctoral Researcher Ding Xiaosong from the East China Sea Laboratory, and Researcher He Xianqiang serves as the corresponding author. Co-authors include Senior Engineer Gong Fang and Dr. Ye Feng from our laboratory, Associate Researcher Li Hao and Postdoctoral Researcher Zhao Min from the East China Sea Laboratory, Dr. Hemant Khatri from the University of Liverpool, UK, and Postdoctoral Researcher Li Jiajia from the Research Center for Eco-Environmental Sciences, Chinese Academy of Sciences.
Submesoscale processes with a spatial scale of 0.1 to 10 kilometers play a vital role in ocean energy dissipation and material transport, and satellite remote sensing has become an important means to observe submesoscale processes at global and ocean-basin scales. At present, satellite radar altimeters are the mainstream tools for monitoring global ocean mesoscale dynamic processes. Launched in December 2022, the new-generation Surface Water and Ocean Topography (SWOT) satellite has lifted the spatial resolution to approximately 2 kilometers, promoting the transition of satellite-based ocean dynamic field observation from mesoscale to submesoscale processes. Nevertheless, its nearly 20-day revisit period limits its capability to capture rapid hourly-to-daily variations of submesoscale processes.
In contrast, geostationary ocean color satellite sensors such as GOCI and GOCI-II can provide all-day hourly observational data with meter-level spatial resolution (500 m for GOCI and 250 m for GOCI-II), which provides valuable data support for refined observation of sea surface current structures of submesoscale eddies. However, affected by strong rotational motion and complex deformation characteristics of submesoscale processes, it remains challenging to retrieve sea surface currents of submesoscale features from geostationary ocean color satellite data.
Traditional sea surface current retrieval methods based on Maximum Cross-Correlation (MCC) are easy to implement, but they show large uncertainties in capturing currents associated with eddies, frontal filament structures and other fine features. To address this problem, this study establishes an hourly sea surface current retrieval model named BAPDE-RAFT based on physically-constrained deep learning for ocean submesoscale processes.
The BAPDE-RAFT model integrates two core perception modules, namely the Boundary Attention Module and the Detail Enhancement Module, to improve the recognition ability of submesoscale structures. Meanwhile, a divergence elimination module based on the Poisson equation is introduced to satisfy the incompressible flow condition by solving physically constrained flow fields, thus ensuring the physical rationality of retrieved current fields.
Tests based on the MITgcm LLC4320 benchmark dataset show that compared with the traditional MCC method, the BAPDE-RAFT model reduces errors of current velocity magnitude and flow direction by 44% and 38% respectively. The minimum resolvable spatial scale of dynamic processes is improved from about 70 km to 3.9 km, realizing high spatiotemporal resolution retrieval of sea surface currents for submesoscale processes.
When further applied to GOCI-II chlorophyll-a remote sensing products, the model can obtain diurnal hourly variations of submesoscale sea surface currents at 250 m resolution, effectively compensating for the spatiotemporal resolution limitations of SWOT observations. Using the sea surface currents retrieved by BAPDE-RAFT, this study preliminarily analyzes multi-scale interactions as well as cross-scale cascade behaviors of energy and tracers, and reveals dynamic spectral characteristics including upward cascade of kinetic energy spectra and downward cascade of tracer spectra. The outcomes greatly enhance satellite observation capacity for submesoscale dynamics and ecological process research.
This research is jointly funded by the National Key R&D Program of China, the National Natural Science Foundation of China, Zhejiang Provincial Natural Science Foundation and other research projects.
Fig.1 Overall framework of BAPDE-RAFT: Learning sea surface velocity fields from MITgcm LLC4320 simulation datasets. (i) Spatial distribution of MITgcm LLC4320 simulation data for model training and testing. (ii) Core components of the BAPDE-RAFT architecture. (iii) Innovative modules including Boundary Attention Module, Detail Enhancement Module and PDE physical constraint correction module.
Fig.2 Comparison of sea surface current retrieval results obtained by different models.
Fig.3 GOCI-II-derived sea surface currents on February 27, 2023 retrieved via the BAPDE-RAFT model. (a) AVISO geostrophic current products with sea level anomaly (SLA) as the background field. (b) Daily mean sea surface currents retrieved from GOCI-II with GOCI-II chlorophyll-a concentration as the background field. (c-h) Hourly sea surface currents retrieved from GOCI-II with original 250 m resolution, displayed at an interval of 25 pixels (6.25 km spatial resolution) for visual simplification.
Fig.4 Comparison of diurnal hourly spatiotemporal features of sea surface velocity fields derived from altimeter data and new-generation geostationary ocean color satellite observations on February 27, 2023.
Citation
[1] Ding, X., He, X.*, Hemant, K., Li, J., Ye, F., Li, H., Zhao, M., Gong, F. (2026). Physically-constrained flow learning reveals diurnal submesoscale surface currents from geostationary satellite observations. ISPRS Journal of Photogrammetry and Remote Sensing, 232, 223–237. DOI: 10.1016/j.isprsjprs.2025.12.005.
Related Publications
[1] Ding, X., He, X.
, Bai, Y., Ma, W., Li, J., Ye, F., Yu, S., Hu, Q., Gong, F., Wang, D., & Li, T. (2025). Geostationary ocean color satellite observations reveal the fine structure of mesoscale eddy dynamics. Remote Sensing of Environment, 320: 114652. DOI: 10.1016/j.rse.2025.114652.
[2] Ding, X., He, X., Bai, Y., Li, J., & Gong, F. (2024). Using geostationary satellite ocean color data to map diurnal hourly velocity field changes in oceanic mesoscale eddy.
IEEE Transactions on Geoscience and Remote Sensing, 1-19. DOI: 10.1109/tgrs.2024.3428851.
