【英文摘要】Based on the observed data of monitoring sites of the active layer of permafrost, as well as the intensified observed data in Dongkemadi basin in Tanggula Mountain, three snow parameterization schemes were compared when coupled with the land surface model VIC-3L. The results suggest that both schemes are very sensitive to the early accumulation of snow depth and the accumulation time, however, different stratified schemes only have limited impact on the improvement of simulation results. The observation data of along the Qinghai-Tibet Highway and the upper reaches of Shule River and simulation both indicate that there are only thin snow cover in winter 2007-2009, and the spatial distribution is very uneven, which suggest that the great impact of seasonal snow cover on Tibetan Plateau on climate maybe have previously exaggerated, this point is also were appreciated in other recent studies. It is also suggests that combination of remote sensing data and observed data to more accurate describe the spatial heterogeneity in each grid is more important on the improvement of land surface process models. The model simulation results were improved by calculating the soil thermal physical parameters considering ice content and soil salinity. Studies on model parameter optimization and uncertainty concluded that the integration of different optimization techniques and rules, improve the efficiency and accuracy on probability algorithm, and ensemble different simulation results will improve the simulation efficiency and reduce the uncertainty of the model. On the MODIS satellite remote sensing retrieval of snow cover analysis indicated that snow distribution in Qilian Mountains is extremely uneven. Regional precipitation and runoff analysis suggested that the upper reaches of the Yellow River runoff changes have been strong impacted by the East Asian monsoon, the total runoff in Xinjiang has the strong positive and negative relationship with the western North Pacific monsoon index and the subtropical westerlies index respectively.