Natural land surfaces are rarely homogeneous over the resolvable scales of numerical weather prediction models. Therefore, these models must somehow account for the subgrid variability in processes that are nonlinealy dependent on surface characteristics. Because of its complex dependence on soil moisture and vegetation, the flux of latent heat is an acutely nonlinear process, involving variables which can vary widely over very short distances. Thus, if the effects of subgrid-scale surface variability are significant, they should appear most prominently in the prediction of evapotranspiration. In this paper, a simple, explicit model for the computation of grid-cell-average evapotranspiration is presented and tested. The model incorporates a statistical distribution of soil moisture on the subgrid scale, the variance of which is obtained from observed distributions of soil moisture and precipitation. Distributions are also assumed for other surface and vegetation characteristics Water-stressed and... Abstract Natural land surfaces are rarely homogeneous over the resolvable scales of numerical weather prediction models. Therefore, these models must somehow account for the subgrid variability in processes that are nonlinealy dependent on surface characteristics. Because of its complex dependence on soil moisture and vegetation, the flux of latent heat is an acutely nonlinear process, involving variables which can vary widely over very short distances. Thus, if the effects of subgrid-scale surface variability are significant, they should appear most prominently in the prediction of evapotranspiration. In this paper, a simple, explicit model for the computation of grid-cell-average evapotranspiration is presented and tested. The model incorporates a statistical distribution of soil moisture on the subgrid scale, the variance of which is obtained from observed distributions of soil moisture and precipitation. Distributions are also assumed for other surface and vegetation characteristics Water-stressed and...