Numerical simulations were performed using the Colorado State University & "Regional Atmospheric Modelling System" model to show a new tool for regional and local agricultural variables forecats, such as evapotranspiration rate. The RAMS (3b) was configured for a non-homogeneous domain of vegetation and topography and homogeneous initialization of atmospheric fields. The model was used to validate and to provide the evapotranspiration forecast. Because of lack of micrometeorological observation data of turbulent heat fluxes exchanges between soil and atmosphere, the model was calibrated with high horizontal and vertical resolution by using daily climatological data, such as air temperature and. Comparisons between observed (Penman Monteith results) and simulated fields were made over a small domain on the Southwest of Minas Gerais, Brazil. The model showed good result, for 36 hour forecasts, without updating the boundary conditions. For agricultural purposes, this new methodology, which includes the complex physical processes of soil-atmosphere interaction, rather than simple empirical methods, showed to give good estimates not only of evapotranspiration rates, but also of hydrological cycle, energy budget and water stress index, in a regional scale. The results can be improved by updating the lateral boundary conditions during the forecast run. The prognostic aspect of the methodology and the reasonable accuracy of the estimates favor the implementation of this work in a operational Meteorological Center.