Gap-filling processes for GOCI-based daily surface salinity product using environmental variables and machine learning approach

Abstract

Ocean surface salinity dataset is useful for research on climate change and its variability. In particular, a gridded daily ocean surface salinity product with high spatial resolution can provide information of short-term variability in the East China Sea (ECS) and the Yellow Sea (YS). Here, we conducted gap-filling of daily surface salinity product based on the Geostationary Ocean  Color Imager (GOCI) for the period 2011-2020 with spatial resolution of 500 m using machine learning approach. For this, we used GOCI-based daily surface salinity preoduct as ground-truth data with envrionemntal variables such as sea surface temperature (SST), sea surface height (SSH), eastward seawater velocity (uo), northward seawater velocity (vo), and seawater salinity (SS) as input data of machine learning model. To identify importance between daily surface salinity and environmental variables affecting daily surface salinity, feature importance ranking was used. Our model shows gap-free daily surface salinity product based GOCI. In addition, the successful application of machine learning model provides the information of long-term variation of daily surface salinity at high spatial resolution in the ECS and the YS.