Daily Global Sea Surface Ageostrophic Current Dataset from 1993–2023 via Physics-Informed Deep Learning (Code)

AC-PIDNN: Daily Global Sea Surface Ageostrophic Current Dataset from 1993–2023 via Physics-Informed Deep Learning Dataset DOI: https://doi.org/10.5281/zenodo.19966716 Zenodo Record URL: https://zenodo.org/records/19966716 Authors / Creators: Cui, Guangxi (Data collector & Developer, Department of Ocean Science and Technology, University of Macau) Cai, Zhongya (Corresponding author & Contact person, Department of Ocean Science and Technology, University of Macau) 1. Dataset Overview The AC-PIDNN dataset provides a daily, high-resolution ($0.25^\circ \times 0.25^\circ$) reconstruction of global sea surface circulation spanning a 31-year period from January 1, 1993, to December 31, 2023. Generated via a physics-informed deep learning framework, this dataset integrates satellite observations with ocean momentum constraints—explicitly accounting for geostrophic balance, wind stress (Ekman dynamics), nonlinear advection, and the Coriolis force. It decomposes the total sea surface flow field into geostrophic, Ekman, and nonlinear advective components, offering a physically consistent and comprehensive view of global surface currents and ageostrophic dynamics. 2. Data Format and CF Standard Compliance Format: NetCDF-4 Classic (netcdf4_classic) Standard Compliance: All NetCDF files strictly comply with the Climate and Forecast Metadata Conventions (CF-1.8). Software Compatibility: Spatial and temporal dimensions, variable names, units, and missing values conform to standard conventions, ensuring seamless integration with common Earth science analysis tools (e.g., Python xarray/netCDF4, MATLAB, CDO, NCO, and Panoply). 3. Directory Structure and Naming Conventions The dataset consists of daily files spanning 31 years (1993–2023), systematically organized into subdirectories by year (YYYY) and month (MM). Each daily file follows a standardized path and naming convention: Plaintext Structure: [YYYY]/[MM]/[YYYY]_[MM]_[DD].nc Examples: 1993/01/1993_01_01.nc 1993/01/1993_01_02.nc 4. Variables, Units, and Missing Values All 2D spatial variables are defined on a global $1440 \text{ (longitude)} \times 720 \text{ (latitude)}$ regular grid ($0.25^\circ$ resolution). Missing data points (e.g., over landmasses, ice cover, or uncomputable grid cells) are flagged with a standard missing value indicator (_FillValue = NaN). Variable Name Long Name / Definition Units Dimensions Data Type lon Longitude degrees_east lon (1440) double lat Latitude degrees_north lat (720) double time Time days since 1993-01-01 time (1) double sla Sea Level Anomaly m lat, lon double uadv Zonal (u) component of the nonlinear advective term m/s lat, lon double vadv Meridional (v) component of the nonlinear advective term m/s lat, lon double utotal Zonal (u) component of the total flow field (ugeo+uekman+uadv) m/s lat, lon double vtotal Meridional (v) component of the total flow field (vgeo+vekman+vadv) m/s lat, lon double resu Residual of the zonal (u) momentum m/s2 lat, lon double resv Residual of the meridional (v) momentum m/s2 lat, lon double ugeo Zonal (u) component of the geostrophic current m/s lat, lon double vgeo Meridional (v) component of the geostrophic current m/s lat, lon double uekman Zonal (u) component of the surface (0 m) Ekman current m/s lat, lon double vekman Meridional (v) component of the surface (0 m) Ekman current m/s lat, lon double 5. Global Metadata Each NetCDF file includes global attributes to track data provenance and creation details: creation_date: Automatically generated date of file creation. author: Zhongya Cai and Guangxi Cui et al., University of Macau. Conventions: CF-1.8 6. Data Access and Batch Download Method Due to the large total volume of the full 31-year daily dataset (exceeding 500 GB), which surpasses standard repository single-deposit storage limits, a representative sample data file alongside the README.txt documentation is hosted directly in this Zenodo repository. To facilitate direct access to the complete 1993–2023 archive, all individual daily NetCDF files are hosted on a high-speed remote S3 storage server: Base URL: Plaintext https://s3.jn1.is.shanhe.com/umdr2026 Users can retrieve specific daily files via HTTP/HTTPS (using wget, curl, or Python requests) by appending the standardized directory path to the Base URL: Bash # Download Format [Base_URL]/[YYYY]/[MM]/[YYYY]_[MM]_[DD].nc # Command-line Download Examples (via wget) wget -c https://s3.jn1.is.shanhe.com/umdr2026/1993/01/1993_01_01.nc wget -c https://s3.jn1.is.shanhe.com/umdr2026/1993/01/1993_01_02.nc 7. Code Availability The complete source code for the AC-PIDNN physics-informed deep learning framework (including data preprocessing pipelines, physical momentum constraint modules, model training scripts, and global velocity reconstruction routines) is available from the authors upon reasonable request. Researchers interested in obtaining the full code package or discussing technical implementations are welcome to contact: Dr. Guangxi Cui: [email protected] / [email protected] Prof. Zhongya Cai (Corresponding Author, University of Macau): [Please insert Prof. Cai's email address here, e.g., [email protected]] 8. Citation If you use this dataset in your research, please cite both the dataset deposit and the corresponding Earth System Science Data (ESSD) publication: Dataset Citation: Cui, G. and Cai, Z.: Daily Global Sea Surface Ageostrophic Current Dataset from 1993–2023 via Physics-Informed Deep Learning, Zenodo [data set], https://doi.org/10.5281/zenodo.19966716, 2026. Cui, G., Yuen, K.-V., Chen, Y., Liu, Z., Yang, D., Liu, G., and Cai, Z.: Daily Global Sea Surface Ageostrophic Current Dataset from 1993–2023 via Physics-Informed Deep Learning, Earth Syst. Sci. Data Discuss. [preprint], https://doi.org/10.5194/essd-2026-341, in review, 2026.