 This study investigates the improvement of soil moisture active passive, SMAP, sea surface salinity, SSS, over five river dominated oceans globally using three machine learning approaches, random forest, support vector regression, and artificial neural network. The results show that all models improve SMAP SSS product by up to 28% reduced in root mean square error for validation, with random forest yielding better performance than the other two methods. The calibration and validation IMSs by RF were 0.203 and 0.556 practical salinity unit, SU, while those of SMAP SSS were 0.774 SU. The improved SSS well captured spatiotemporal patterns of SSS for both low and high salinity water in all five regions, and the proposed approach can be used operationally to improve global SMAP SSS product including other coastal areas and neopolar regions in the future. This article was authored by Yuna Jang, Young Jung Kim, Jung Ho Im, and others. We are article.tv, links in the description below.