In a modification of the distributed hydrological model, LISFLOOD-WB, a two-source canopy scheme is used to predict both the canopy transpiration and soil evaporation. A revised soil storage capacity curve from the Xinanjiang model is applied to take into account the sub-grid heterogeneity. The modified model is used to estimate the long-term (1980-1997) water budget of the Lushi basin (4423 km²), China. All the input data fields are integrated in a four-dimensional GIS data structure with a raster grid spacing of 1-km. The basin channel network is determined from digital elevation data, and the spatial pattern of canopy leaf area index (LAI) is retrieved from NOAA/AVHRR NDVI images. Generally, the model efficiency for discharge prediction is acceptable, but the discharges are overestimated in the driest years and underestimated in the wettest years. The results indicated that the influence of inter-annual variation of spatial patterns of LAI detected by NOAA/AVHRR NDVI on the estimates of annual evapotranspiration is negligible. Annually averaged ratios of overland flow, infiltration and canopy interception to precipitation are 24±, 56±10% and 20±2%, respectively. The inter-annual variations of precipitation and predicted evapotranspiration are relatively high with standard deviations of 5.1 mm day^-1 and 2.4 mm day^-1, respectively, whereas the inter-annual variation of the net radiation is much less. Monthly temporal patterns of soil moisture follow precipitation strongly. Spatially precipitation and LAI are both significantly correlated with evapotranspiration, although precipitation has a slightly more dominant control. The linear relationship between water yield and LAI is weak on a grid by grid basis.