Seasonal snowpack serves as natural reservoir by storing winter precipitation and releasing it as snowmelt. In in the Western U.S., this is a crucial water supply that supports agriculture, hydropower, ecosystems, and millions of users. Rising temperatures are causing reduced snow storage, earlier melt, and increased drought risk. Future climate models predict these trends will continue and intensify, posing challenges for water management. Accurate snow water equivalent (SWE) estimates are essential for water managers in snowmelt-reliant regions, but characterizing the spatial distribution of snow is an ongoing challenge. In situ measurements are not always representative nor widespread and remote sensing of mountain SWE remains elusive. Modeling can fill space-time gaps in the observational record but is impacted by biases in forcings and uncertainties in model physics. To address the issue of uncertainties in model physics for simulating snow, we evaluate how altering the configurations of a land surface model (Noah-MP) affects its ability to recreate observed SWE across 199 stations in the Western U.S. The base case configuration, which matches that for the National Water Model, overestimates SWE at 90% of sites. Adjustments to model physics for precipitation partitioning, snow albedo formulation, and surface resistance cause significant changes in SWE predictions that vary by season and site climate and geography. No single configuration works best everywhere, but four alternatives outperform the base case at most sites. To address the challenge of biases in mountain precipitation products, we leverage a historical snow reanalysis dataset to develop, apply, and test a novel precipitation bias correction and downscaling method towards modeling SWE in a real-time context. Over a test domain, this precipitation bias correction is effective in reducing error in April 1st SWE (-58%) and streamflow forecasts (-52%). Assimilating remotely-sensed snow depth observations further reduces errors. To explore the impact of future shifts in snowpack on water resources, we apply hydrology projections driven by downscaled global climate models (GCMs) and a simple reservoir operations model to 13 major reservoirs in the California Sierra Nevada. Region wide, snowpack reductions (-44%) and earlier snowmelt (11 days) lead to earlier inflow and drops in water deliveries (-19%) and year-end storage (-18%). Reservoir storage and rainfall help offset the impact of snowpack losses, but the extent and mechanisms of this vary on the reservoir’s operations, characteristics, and upstream climate and hydrology. Current operating rules are not well-suited to let reservoirs store earlier inflow under future climate conditions.