One method for interpreting the variability in total water storage observed by GRACE is to partition the integrated GRACE measurement into its component storage reservoirs based on information provided by hydrological models. Such models, often designed to be used in couple Earth System models, simulate the stocks and fluxes of moisture through the land surface and subsurface. One application of this method attempts to isolate groundwater changes by removing modeled surface water, snow, and soil moisture changes from GRACE total water storage estimates. Human impacts on groundwater variability can be estimated by further removing model estimates of climate-driven groundwater changes. Errors in modeled water storage components directly affect the residual groundwater estimates. Here we examine the influence of model structure and process representation on soil moisture and groundwater uncertainty using the Community Land Model, with a particular focus on basins in the western U.S.