Presented at the 2020 Ocean Sciences MeetingSince 2010, the European Space Agency L-Band satellite, Soil Moisture and Ocean Salinity (SMOS), has provided sea surface salinity (SSS) maps of the global ocean. However, SSS retrieval at high latitudes remains a difficult task due to low sensitivity of L-Band measurements at low sea surface temperature (SST) and the presence of ice-inducing SSS biases. Weekly SSS fields are refined using a pseudo-electric constant (ACARD parameter) derived from SMOS measurements allowing to filter SSS pixels polluted by ice and thus to significatively decrease the ice-induced bias. We further show that the accuracy of SST used previously in SSS retrieval in the Arctic Ocean is a significant source of error, especially in the river plume areas where large differences of SST distributions between the products from European Centre for Medium-Range Weather Forecasts (ECMWF, used as prior SST in the SMOS processor) and from Remote Sensing Systems (REMSS) are observed. Comparisons between SSS in-situ measurements (CTD casts, underway TSG and Argo profilers) ranging from 1- and 11-meter depths and SMOS SSS at 1cm depth with a focus on areas of large variability demonstrate that the correction based on the ACARD partially corrects SSS differences between satellite estimates and in-situ measurements. Furthermore, in the Arctic Ocean, the sensibility of the satellite validation to stratification is demonstrated. When the stratification effect is taken into account, SMOS SSS and in-situ salinity measurements compares better, particularly in the stratified Arctic seas. New SMOS weekly estimates present a very good agreement between satellite estimates and in-situ measurements (Standard deviation of differences (STDD) = 1.21pss). SMOS estimates demonstrate its ability to capture the whole scope of SSS variability in the Arctic Ocean with better performances than the TOPAZ reanalysis (STDD = 1.88pss), particularly in seas where river plumes with freshening larger than 10pss may be observed.