A key task of a land model coupled in an Earth System modeling framework is the partitioning of moisture inputs from the atmosphere into evapotranspiration, runoff, and changes in land water storage. The Community Land Model (CLM) exhibits biases in evapotranspiration (ET) and total water storage (TWS) in semiarid regions. The biases in TWS, whose seasonal cycle amplitude is too low and ET, which is too strong and variable, are consistent with excessive soil evaporation when the canopy is sparse or absent. In this study, we use observations of TWS from the Gravity Recovery and Climate Experiment (GRACE) satellite and ET estimates from the FLUXNET-MTE data set [Jung et al., 2009] to assess the simulation of the terrestrial water budget of the Community Land Model (CLM).  We introduce in CLM a new dry surface layer (DSL) parameterization which constrains the model's soil evaporation in a manner that more faithfully reflects the underlying processes. Compared to the soil resistances in the standard version of CLM, the soil resistances calculated from the DSL parameterization are significantly larger. Using the resulting soil resistances in CLM simulations simultaneously reduces the ET and TWS biases in the semiarid study region. Globally, annual mean soil evaporation decreases and annual mean transpiration increases, making the CLM partitioning more similar to recent ET partitioning estimates.