The past few decades have seen decreasing trends of snow-covered regions in the Northern Hemisphere. It remains unknown how these trends affect the spatial and temporal variability of snowpack water storage, a variable with significant implications for managing water resources to meet agricultural, municipal, and hydropower demands. To improve snowpack estimates, this dissertation developed a new snow data assimilation system (SNODAS) through multi-institutional collaborations. The new SNODAS consists of coupling of the Community Land Model version 4 (CLM4) and the Data Assimilation Research Testbed (DART), which is capable of assimilating multi-sensor satellite observations including the Moderate Resolution Imaging Spectroradiometer (MODIS) snow cover fraction (SCF) and the Gravity Recovery and Climate Experiment (GRACE) terrestrial water storage (TWS) anomalies. This dissertation describes the new SNODAS, presents the results of the data assimilation of MODIS SCF and GRACE TWS observations, and assesses the influence of uncertainties from multiple sources on the SNODAS performance. The first two studies compared the open loop run and the assimilation runs to evaluate the data assimilation (DA) performance. Data assimilation results were also evaluated against other independent observation-based snow data on daily and monthly timescales. Both assimilations can improve the snowpack simulations in CLM4; their strengths and drawbacks were discussed. When only MODIS SCF is assimilated, the innovation (i.e. the difference between analysis and forecast) is marginal in the regions where the snow cover extent reaches 100% regardless of snow mass changes. Further assimilation of GRACE TWS anomalies, however, can adjust the modeled snowpack, resulting in noteworthy improvements over the MODIS-only run in high-latitude regions. The effectiveness of the assimilation was analyzed over several Arctic river basins and various land covers. The third study discussed the influences of atmospheric forcing, model structure, DA technique, and satellite remote sensing product within the framework of SNODAS. The atmospheric forcing uncertainty is found to be the largest among the various uncertainty sources examined, especially over the Tibetan Plateau and most of the mid- and high-latitudes. The uncertainty of model structure as represented by two different parameterizations of SCF is the second largest. DA methods and products of GRACE TWS data have relatively less impacts. This study also showed that CLM4.5 produces better TWS anomalies than CLM4, which would have implications for improving the performance of GRACE TWS data assimilation.

Since the 1960's, remote sensing instruments aboard Earth-orbiting satellites have provided scientists with snowfall data. As the technology of these instruments has changed, so have the methods of using remote sensing data to calculate snow cover, snow depth, and snow-water equivalent. Spectral characteristics of snow are highly variable based on a number of factors, including grain size, depth, terrain, and age. Previous research has shown that an index can be generated using reflectance values from remote sensing data that makes classification of snow-covered land much easier.