The ocean temperature and salinity (T/S) profiles play an important role in the oceanic processes and climate change. However, there is still a lack of in situ data, especially salinity data, to depict the interior structure of ocean. On the other hand, the development of satellite remote sensing technology provides high spatial-temporal resolution data. Therefore, it is of great significance to reconstruct the three-dimensional ocean temperature and salinity fields by combining the satellite remote sensing data with the in situ measured data.

The reconstruction of three-dimensional temperature and salinity profiles from sea surface data essentially requires the statistical relationship between surface and subsurface. With sufficient historical data, machine learning has great advantages in regression statistics. However, the quantity and quality of features have a great impact on the quality of the machine learning methods. Previous studies only took sea surface temperature (SST), sea surface salinity (SSS) and sea surface height (SSH) as sea surface inputs, this paper introduces the K2 algorithm based on information flow optimization to select the characteristics of 11 sea surface parameters that may affect the reconstruction of temperature and salinity profiles, and obtains the best characteristics of sea surface parameters, and discusses the role of longitude and latitude position information and time information in the reconstruction of temperature and salinity profile. The results show that the feature selection of K2 algorithm based on information flow optimization better captures the main characteristics of sea surface parameters, and the selected optimal parameter combination can better reconstruct the temperature and salinity profiles.