Abstract
Salinity, as one of the essential physical properties of seawater, is a common tracer differentiating water masses in the ocean, which often require relatively high-resolution datasets. Limited by the coverage of direct observations, however, high-resolution spatial and temporal salinity data are not always available, which hinders the fine application of salinity data in discerning ocean processes and improved modeling of ocean physics and biogeochemistry. To supplement the salinity database, we reconstructed sea surface salinity (SSS) with reasonably high spatial resolution (0.05° × 0.05°) over 2003–2020 in the South China Sea (SCS) with a machine learning algorithm based on a combination of MODIS-Aqua remote sensing data and a large cruise observation-based dataset. The reconstructed SSS has a mean absolute error (MAE) of 0.2 when compared with our underway observations with a corresponding root mean square error (RMSE) of 0.3. The MAE between station-based observations and our reconstruction was 0.5, and the RMSE was 0.7. These validations strongly suggest that our reconstruction is highly adequate, representing at most a quarter of the identified discrepancies compared to the remote sensing SSS or two other prevalent model-derived datasets. Based on our reconstruction, the SSS in the SCS is relatively low in coastal waters, but high in the ocean basin, with a seasonal pattern with a minimum in the summer and a maximum in the winter. This spatio-temporal distribution is well consistent with the observations and is affected by the Pearl River plume, sea surface circulation, and precipitation. Using our reconstructed SSS, we were able to successfully characterize the spreading of the Pearl River and Mekong River plumes and the intrusion of the Kuroshio Current from the Pacific Ocean into the SCS.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have