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

Dataset Title

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

1. Dataset Overview

This dataset provides a daily, high-resolution (0.25°) reconstruction of global sea surface circulation spanning a 31-year period from January 1, 1993, to December 31, 2023. The reconstruction accurately models ocean dynamics by explicitly accounting for key physical drivers, including wind stress, non-linear effects, and the Coriolis force. It separates the total flow field into geostrophic, Ekman, and non-linear advective components, providing a comprehensive view of global surface currents.

2. Data Format and CF Standard Compliance

• Format: NetCDF-4 Classic (netcdf4_classic)

• Standard Compliance: The NetCDF files have been formatted to comply with the Climate and Forecast (CF) Metadata Conventions (CF-1.8). All spatial and temporal dimensions, variable names, units, and missing values conform to standard conventions to ensure seamless integration with standard Earth science software (e.g., Panoply, CDO, Python xarray, MATLAB).

3. File Naming Conventions and Directory Structure

The dataset contains daily files spanning 31 years. The dataset is systematically organized into subdirectories by year and month. Each daily file follows a standard path structure:

• 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

The data variables are defined on a 1440 (longitude) × 720 (latitude) grid. Missing data points (e.g., over landmasses) are designated with a standard _FillValue of NaN.

Missing Data handling: All un-computable grid points or land areas utilize a standard missing value indicator (_FillValue = NaN).

5. Global Metadata

Each NetCDF file contains global attributes to track provenance and creation details:

• creation_date: Automatically generated date of file creation.

• author: Zhongya Cai and Guangxi Cui et al., University of Macao.

• Conventions: CF-1.8

6. Data Access and Download Method

Due to the large volume of the full 31-year dataset (exceeding 500 GB), which surpasses standard repository upload limits, we have uploaded a representative sample data file alongside the README.txt directly to this Zenodo repository.

To access and download the complete dataset, all individual files are hosted on a high-speed remote S3 server.

• Base URL: https://s3.jn1.is.shanhe.com/umdr2026

Users can retrieve specific files by appending the standardized directory structure to the Base URL:

[Base_URL]/[YYYY]/[MM]/[YYYY]_[MM]_[DD].nc

Download Examples:

• https://s3.jn1.is.shanhe.com/umdr2026/1993/01/1993_01_01.nc

• https://s3.jn1.is.shanhe.com/umdr2026/1993/01/1993_01_02.nc

Please refer to the attached README.txt file on this page for a comprehensive list of direct download links for every individual file in the 31-year series.