Long-term simulation of terrestrial water cycle is estimated using a Land Surface Model (LSM). There are no observations long enough to drive LSM except for monthly precipitation and monthly temperature; therefore daily atmospheric forcing which is indispensable to run the LSM is statistically and/or empirically constructed from monthly precipitation and monthly temperature. High correlations of annual runoff variations are obtained at many basins globally, however, the correlations of annual runoff are low in dry areas and cold regions. It was also showed that the model successfully replicated the annual snow covered area in North America and summer soil moisture in Mongolia. Flood and drought indices are then extracted from the estimated 100-year river discharge and compared with those extracted from available runoff observations at large basins. Results showed that the historical large drought and flood were well simulated in the model.