[M-IS01] ENVIRONMENTAL, SOCIO-ECONOMIC, AND CLIMATIC CHANGES IN NORTHERN EURASIA

[MIS01-P23] New sea surface salinity data for the Arctic region derived from SMAP and SMOS satellite data using machine learning approaches

*Alexander Savin^1,2, Mikhail Krinitskiy^1,2, Alexander Osadchiev^1,2 (1.Shirshov Institute of Oceanology, Russian Academy of Sciences, Moscow, Russia, 2.Moscow Institute of Physics and Technology, Dolgoprudny, Russia)

Keywords:sea surface salinity, machine learning, SMAP, SMOS, Arctic Ocean

Sea salinity plays a significant role in the physical processes occurring in the World Ocean and is currently considered to be one of major climate parameters. Salinity together with temperature determine the global system of density currents in the World Ocean. Sea salinity is impacted by numerous processes, and therefore serves as an indicator characterizing the physical and chemical processes occurring on Earth. During the past decade, remote sensing measurements are actively used to gather sea surface salinity. Standard retrieving SSS algorithms from remote sensing data were developed and verified for tropical and subtropical temperature and salinity values of the World Ocean. However, they demonstrate significantly lower accuracy in the Arctic Ocean, which is characterized by low temperatures and is influenced by substantial freshwater runoff. In this study, Soil Moisture Active Passive (SMAP) and Soil Moisture and Ocean Salinity (SMOS) salinity data are compared with in situ measurements. This work presents a new algorithm, based on machine learning methods, has been developed to retrieve SSS in the Arctic Ocean during ice-free season. SMAP, SMOS and in situ data, collected during multiple field surveys in the Russian Arctic, are used to train and validate machine learning models. A as result, the error in SSS retrieval of the developed algorithm compared to the standard algorithm reduced and correlation between satellite salinity data and in situ measurements increased. New obtained SSS fields allow to analyze spatial and temporal variability of the major water masses in the surface layer of the Arctic Ocean with high accuracy. It is especially important for detecting spreading areas of river plumes, where the quality of standard algorithms is low.