Surface water resources in arid zones are mainly derived from rain and snowmelt of mountain. The simulation and evaluation of snowmelt is one of research focuses of water resources in recent years. Estimation of snow water equivalent from remote sensing is important to estimate mountainous water resources in arid areas where have sparse weather stations. Using optical radiation data, snow extent is generally felt to be obtained reliably, but snow water equivalent is less confidence due to the band limited and cloud affected. At present, passive microwave remote sensing is the most efficient way to derive snow properties, but its spatial resolution is relatively low. This project intends to study snow water equivalent from passive microwave data in inhomogeneous land surface taking the central of Tianshan in China as a case. This paper takes field observations, MODIS, AMSR-E and MVRI as data source. The relations between terrains and snow physical properties (snow grain size, density, stratigraphy, snow temperature and temperature at the snow/soil interface) will be obtained based on long-term field experiments a priori. In considering of the spatial resolution merit of visible band, the fusion methods of passive microwave and optical data will be constructed. The estimation methods of snow depth will be modified for complex land surface. The dynamic retrieval model of snow depth based on the temperature gradient index will be built and its applicability will be evaluated. Then, the theory and methodology will be finally established for retrieving snow water equivalent of non-uniform underlying surface with passive microwave remote sensing. The reliably estimation of the snow water equivalent can improve the accuracy of the snowmelt-runoff prediction and the management of the water supply.