OI-SwinUnet reconstructed daily Chlorophyll-a products in the South China Sea

Chlorophyll-a is the most significant pigment for phytoplankton photosynthesis, and its concentration is a useful measure for determining the density of phytoplankton. The ocean has numerous mesoscale and submesoscale processes, including mesoscale eddies, upwellings, and fronts. These ocean dynamic mechanisms influence the growth and extinction of phytoplankton. Satellite observations of phytoplankton dispersion on the ocean surface can help researchers better comprehend localized and complicated dynamical processes. With the advancement of satellite remote sensing technology, satellite remote sensing has become the primary method for obtaining ocean color information. However, due to weather conditions, satellite sensor failures, and other variables, a considerable number of satellite remote sensing sea surface chlorophyll products are intermittently missing. Satellite observation data is incomplete, which makes it difficult to use in maritime research. As a result, the reconstruction and reanalysis of satellite remote sensing data has become a hotspot for research in the discipline.

The OI-SwinUnet model we developed is a deep learning model based on the expected variance of data anomalies, which combines the advantages of the optimal interpolation (OI) method and the deep learning network framework SwinUnet. Among them, the OI module serves to generate the background field using the products of satellite observations, while SwinUnet realizes the filling of missing values by learning the difference between the satellite observation field and the background field. The OI-SwinUNet model achieves a good performance in reconstructing the chlorophyll-a concentration data with high spatial and temporal resolution in the South China Sea.

The dataset we provided is a reconstructed surface chlorophyll-a concentration in the South China Sea using the OI-SwinUNet model, which is based on the merged product of two MODIS satellite observations. The dataset covers the period from January 1, 2013, to December 31, 2017 (and will be continuously updated in subsequent work), with a temporal resolution of 1 day and a spatial resolution of 1 km. The reconstructed dataset not only accurately depicts the seasonal-scale temporal and spatial patterns of surface chlorophyll-a changes in the South China Sea, but it also meticulously recreates the weather-scale processes of fast-changing oceanic phenomena, such as the effects of freshwater flushing on upwelling via nutrient input on surface phytoplankton changes in the upwelling zone. Our reconstructed data can be used to study the main ecological impacts of all mesoscale eddy activities in a specific sea area, as well as depict the chlorophyll perturbations of individual eddies at different life stages, providing researchers with a new and comprehensive perspective on eddy research.

Each netCDF file contains three chlorophyll-a products: 'chl_sat', 'chl_rec', and 'chl_filled'. Among these, 'chl_sat' refers to satellite observed chlorophyll data, which is a merged product of two satellites, MODIS/Aqua and MODIS/Terra, processed using SeaDAS software. 'chl_rec' represents the reconstructed chlorophyll data using the OI-SwinUnet model. 'chl_filled' represents chlorophyll data obtained by filling missing pixels of the satellite observations with reconstructed data. Geographic data is stored in the variables 'latitude' and 'longitude'.