Datasets in a machine learning benchmark for reconstructing subsurface temperature in marginal seas from satellite data

The Datasets are for a machine learning benchmark for reconstructing subsurface temperature (ST) in marginal seas using satellite data. Two marginal seas are included, the North Sea (NS) and the South China Sea (SCS). The selected regions are 1°W–9°E and 50°N–60°N for the NS, and 105°E–121°E and 0°–23°N for the SCS. The satellite data include sea surface temperature (SST), sea surface salinity (SSS), sea-level-related variables (sea level anomaly-SLA, absolute dynamic topography-ADT), and 10-m sea surface wind (SSW) velocity. Subsurface data are from the reanalysis product GLORYS12V1 (four variables are included, horizontal velocity components, temperature, and salinity). All data are stored in NetCDF files per day per layer. For wind data, the original 6-hourly data is averaged to daily data. The data cover the period from 2010 to 2025. The spatial resolutions of SST, SSS, SLA&ADT, and SSW are 0.05°, 0.125°, 0.125°, and 0.25°, respectively. The spatial resolution of the reanalysis data is 1/12°. Data sources are shown in "Data_source.pdf". From reanalysis data, six depths are included for the North Sea (up to 30 m: 5.1, 9.6, 15.8, 21.6, 25.2, and 29.4 m), and eleven depths are used for the SCS (up to ~200 m: 9.6, 21.6, 29.4, 40.3, 55.8, 77.9, 109.7, 130.7, 155.9, 186.1, and 222.5 m). Files NS_1.0W_9.0E_50.0N_60.0N.zip and SCS_105.0E_121.0E_0.0N_24.0N.zip correspond to the reanalysis data in the NS and SCS, respectively. The other zip files starting with "NS" and "SCS" correspond to the satellite ocean surface data. The corresponding downloading scripts are in zip files starting with "download". CCMP_wind.zip file contains wind data for both regions, as well as the scripts for downloading and daily averaging.