It is not yet possible to determine whether global snow mass has changed over time despite collection of passive microwave data for more than thirty years. Physically-based, but computationally fast snow and soil models have been coupled to form the basis of a data assimilation system for retrievals of snow mass and soil moisture from existing and future satellite observations. The model has been evaluated against observations of snow mass and soil temperature and moisture profiles from Reynolds Creek Experimental Watershed, Idaho. Simulation of snow mass was improved early in the season due to more realistic representation of soil heat flux, but led to an overestimation of snow mass later in the season. Soil temperatures were generally simulated well; freezing of the surface layers was not observed but was simulated, which affected soil water transport. Limited knowledge of the soil lower boundary conditions is acceptable for snow mass and surface soil moisture retrievals, although improvements are required for more accurate simulations of deeper soil moisture at this site. Development of a data assimilation framework to retrieve snow mass and near-surface soil moisture is discussed.