Process-based ecosystem models simulate carbon, nitrogen and water cycles and they are applicable to assess ecosystem functions and services. Meteorological conditions as key driving forces of ecosystem processes are variable both spatially and temporally, therefore require appropriate downscaling from the coarser resolution data in use with ecosystem models. We examined a weather generator (WG), one of the stochastic downscale methods generating daily precipitation and air temperature in biomass carbon and water balance simulation by Biome-BGC model. Beijing and Urumqi, China having different climatic conditions, were selected as case study sites. The biomass carbon and water balance in Beijing by WG-meteorological data showed close values to those by measured meteorological data. In contrast, the water outflow by WG data showed large relative error (-85%) to one by measured data in Urumqi. To reduce the error of outputs we applied the bias correction to daily precipitation generated by WG. The biomass carbon by corrected WG data showed smaller error. Water outflow error was also reduced, however, it still had -20% relative error to the one by measured data. This error is probably attributed to the increasing trend of daily precipitation with duration of continuous rainy days which was not reproduced by present WG. This study showed the potential of WG as a front end of ecosystem models that downscales low spatio-temporal resolution meteorological data, however it also showed a limitation of WG in such cases where the characteristics of meteorological elements that are not reproduced by WG are important in ecological processes.