This study investigates the use of high-spatial, low-temporal scale visible remote sensing data for calibration of a Soil-Vegetation-Atmosphere-Transfer (SVAT) model for semi-arid perennial grasslands. The SVAT model is driven by meteorological data and simulates plant growth and water budget on a daily time step. The model was combined with a canopy reflectance model to simulate shortwave radiometric temporal profiles. Landsat Thematic Mapper (TM) images obtained during a series of ten consecutive years were used to refine the model to work on a spatially-distributed basis over a semi-arid grassland watershed. Continuous simulations were used to estimate two spatially-variable initial conditions and model parameters through a calibration procedure which minimized the difference between the surface reflectance simulated by the model and measured by the TM sensor. Accuracy of model products such as daily above-ground biomass and soil moisture was assessed by comparison with field measurements. The promising results suggest that this approach could provide spatially-distributed information about vegetation and soil conditions for day-to-day grassland management.