A Lagrangian stochastic model was used to estimate the fetch and latent heat flux above a nonuniform grass-covered forest clearing site at the Duke Forest, in Durham, North Carolina, and an irrigated bare soil patch at the University of California in Davis. The latent heat flux predictions by the Lagrangian model compared well with eddy correlation flux measurements. In order to apply the Lagrangian model to a nonuniform grass-covered forest clearing, the surface was treated as an imaginary equivalent uniform terrain subjected to surface roughness and turbulence statistics (i.e., mean, variance, and covariance) of velocities and scalars identical to those measured above the nonuniform terrain. At the irrigated bare soil site the equilibrium distance of the air from dry to wet was well defined, and its influence on the water vapor flux internal boundary layer was considered. In the Lagrangian model, five different schemes to account for inhomogeneous turbulent flows were compared in terms of estimating scalar fluxes. Our comparisons demonstrate that the five different schemes produce similar scalar fluxes despite the fact that some of them do not satisfy the well-mixed criterion. Also, the analytical solution to the advection-diffusion equation was used to predict the fetch and latent heat flux under neutral conditions and compared to the Lagrangian model.¿ 1997 American Geophysical Union