The Community Land Model (CLM) exhibits biases in evapotranspiration (ET) and total water storage (TWS) in semiarid regions. The TWS seasonal cycle amplitude is too low, while evapotranspiration is too strong and variable. These biases are consistent with excessive soil evaporation when the canopy is sparse or absent, which reduces moisture inputs into the ground. Here we improve the simulation of soil evaporation by replacing CLM's existing empirical soil resistance parameterization with a more mechanistically based formulation in which soil evaporation is controlled by the rate of diffusion of water vapor through a dry surface layer (DSL). The thickness of the DSL is parameterized as a soil-type dependent function of top layer soil moisture. Soil resistances are calculated from the DSL thickness combined with a soil tortuosity factor. Compared to the existing CLM soil resistance parameterization, the DSL-based soil resistances for a given soil moisture value are generally larger, especially for moister soils. CLM simulations using the DSL-based soil resistance expression have reduced biases of ET relative to the FLUXNET-MTE (Model Tree Ensemble) data set and TWS relative to Gravity Recovery and Climate Experiment (GRACE) observations. Averaged over global semiarid regions, the CLM mean annual amplitude of TWS increases from 26.3 mm to 32.5 mm when the new soil resistance parameterization is used, in good agreement with the GRACE mean annual amplitude for these regions of 32.2 ±2.1 mm. CLM mean annual ET, averaged over global semiarid regions, decreases from 27.8 mm/month to 23.5 mm/month, closer to the FLUXNET-MTE mean annual ET of 21.1 ± 0.9 mm/month. The seasonal amplitude of CLM ET also decreases, from 23.3 mm/month to 18.3 mm/month compared to 14.9 ± 1.1 mm/month for FLUXNET-MTE.