We have investigated microclimatic controls on the spatiotemporal variations of net primary production (NPP) of a rugged forested watershed using the process-based biogeochemical model (BIOME-BGC). To validate the model simulation of water and carbon cycles at the plot scale, we have conducted field survey over deciduous broadleaf forest (DBF) and evergreen needleleaf forest (ENF) since 2000. The modeled values of soil temperature, soil moisture and soil respiration showed high correlation with those from the field measurements. The modeled seasonal changes of NPP showed high correlation with air temperature but no significant correlation with water related parameters. The precipitation frequency turned out to be the best climatic factor to explain the annual variation of NPP. Furthermore, NPP of ENF was more sensitive to precipitation frequency than that of DBF. With changes in vegetation cover and topography, the spatial distribution of NPP was of great heterogeneity, which was negatively correlated with the magnitude of NPP. Despite the annual precipitation of 1,400mm, NPP at the study site was constrained by the amount of water available for the vegetation. Such a modeling result should be verified by the field measurements.