Sea surface salinity (SSS) provides information on the variability of ocean dynamics (global water cycle and ocean circulation) and air-sea interactions, thereby contributing to the identification and prediction of significant changes in the global climate. Monitoring global SSS via satellite observations has been possible using L-band microwave radiometers since 2010; however, their performance is limited by their retrieval algorithms under conditions such as radio frequency interference, low sea surface temperatures, and strong winds. This study proposes a new global SSS model using multi-source data based on seven machine learning approaches: K-nearest neighbor, support vector regression, artificial neural network, random forest, extreme gradient boosting, light gradient boosting model, and gradient boosted regression trees (GBRT). Five Soil Moisture Active Passive (SMAP) products, Hybrid Coordinate Ocean Model (HYCOM) SSS, and four ancillary data were used as input variables. All models produced better performance than either SMAP or HYCOM SSS products, with the top performing GBRT model reducing the root mean square difference for the validation dataset from 1.062 to 0.259 practical salinity units compared to the SMAP SSS product. The improved SSS products had increased correlation with the in-situ data for both low-and high-salinity waters across all global oceans, thus further advancing the understanding and monitoring of global SSS.