1km-resolution terrestrial model over Japan is developped to build a high quality river discharge forecasting system. Effects of model resolution and soil type dataset are investigated by a sensitivity experiment concerning resolution and soil type dataset. The experiment period is 2 years of 2014, 2015 and analysis is conducted for 6 river basins in Japan. The terrestrial model is consists of MATSIRO, a land surface model and CaMa-Flood, a river routing model. Peak discharge and low discharge representation are improved by using a more detailed soil type dataset. In 9 of 12 experiments, ratios of model discharge to observation are improved by 1-14% and in all experiments, RMSE are decreased . Also, using high resolution experiment settings showed lower RMSE in 10 of 12 experiments.