Under a warming climate, an improved understanding of the water stored in snowpacks is becoming increasingly important for hydropower planning, flood risk assessment and water resource management. Due to inaccessibility and a lack of ground measurement networks, accurate quantification of snow water storage in mountainous terrains still remains a major challenge. Remote sensing can provide dynamic observations with extensive spatial coverage, and has proved a useful means to characterize snow water equivalent (SWE) at a large scale. However, current SWE products show very low quality in the mountainous areas due to very coarse spatial resolution, complex terrain, large spatial heterogeneity and deep snow. With more high-quality satellite data becoming available from the development of satellite sensors and platforms, it provides more opportunities for better estimation of snow conditions. Meanwhile, machine learning provides an important technique for handling the big data offered from remote sensing. Using the overuman Catchment in Northern Sweden as a case study, this paper explores the potentials of machine learning for improving the estimation of mountain snow water storage using satellite observations, topographic factors, land cover information and ground SWE measurements from the spatially distributed snow survey. The results show that significantly improved SWE estimation close to the peak of snow accumulation can be achieved in the catchment using the random forest regression. This study demonstrates the potentials of machine learning for better understanding the snow water storage in mountainous areas.