Reconstructing ocean subsurface salinity at high resolution using a machine learning approach

Abstract. A gridded ocean subsurface salinity dataset with global coverage is useful for research on climate change and its variability. Here, we explore the feed-forward neural network (FFNN) approach to reconstruct a high-resolution (0.25∘ × 0.25∘) (1–2000 m) period 1993–2018 by merging in situ profile observations satellite remote-sensing altimetry absolute dynamic topography (ADT), sea surface temperature (SST), wind (SSW) field data, coarse-resolution (1∘ 1∘) product. We show that FFNN can effectively transfer small-scale spatial variations ADT, SST, SSW fields into 0.25∘ field. The root-mean-square error (RMSE) be reduced ∼11 % global-average basis compared 1∘ reduction RMSE much larger upper than deep because of stronger mesoscale layers. In addition, new reconstruction shows more realistic signals regions strong variations, e.g., Gulf Stream, Kuroshio, Antarctic Circumpolar Current regions, resolution product, indicating efficiency machine learning bringing together observations. large-scale patterns from data are consistent field, suggesting persistence reconstruction. successful application this study provides an alternative complement existing assimilation objective analysis methods. reconstructed IAP0.25∘ freely available at https://doi.org/10.57760/sciencedb.o00122.00001 (Tian et al., 2022).