Water resources are key factors of ecological environmental security in northwest arid region of China. They are also the most important factors for socio-economic development against the background of global warmer, especially in arid regions. It is necessary for arid regions to calculate total water resources because it can provide a reference for the government with which to formulate strategies. Water resources may have a large area and be supplied by runoff from mountain snowmelt and precipitation. The goal of this paper was to determine the suitable method to simulate runoff in arid areas. The Soil Conservation Service Curve Number (SCS-CN) developed by the U.S. Department of Agriculture National Resources Conversion Service (NRCS) is the most popular and widely applied model for direct runoff estimation. This method was modified by accounting for the static portion of infiltration and the antecedent moisture. This model has stimulated a great deal of discussion among scientists and hydrologists. The model is based on the water balance equation and curve number CN, which is derived from the tables given in the National Engineering Handbook for catchment characteristics, such as soil texture, land use, hydrologic condition, and initial soil moisture condition. Based on the spatiotemporal differences among watersheds, international and domestic academics have developed different methods to improve the SCS-CN model. One option is to improve its mechanism and another is to improve the parameter calculation methods. Because there is considerable scope to improve the SCS-CN model, we discuss a parameter algorithm to improve the method for snowmelt and precipitation mix and large-scale basins in arid regions as a solution to a major problem. This study focused on the Kuitun River Valley. We explored the adoption of an SCS model runoff simulation in arid and semi-arid regions with snow-melt and rainfall in spring by modifying the calculation method of SCS model parameters. To satisfy the characteristic of mix supplied runoff, precipitation was revised to represent the sum of rainfall and snowmelt. The snowmelt was calculated by the degree-day model. This was the first time MODIS satellite products with approximately 1km resolution were used to invert the Land Surface Temperature and Normalized Difference Vegetation Index. Then, we used the surface temperature/vegetation index (T_S/VI) constructed in a 2D scatter plot. The combined soil moisture absorption balance principle was used to calculate the moisture-holding capacity of soil. We used cluster analysis to modify the initial abstraction computing methods. The calibration and validation periods of Nash-Sutcliffe efficiency were 0.92 and 0.64, respectively. Relative errors were 0.7% and -1.3%, respectively. This indicated that the improved model was effective in simulating spring runoff in the Kuitun River Valley. Using remote sensing parameter information technology to improve the SCS model can indirectly implement data conversion from point to plane. Establishing a database of the initial abstraction can improve the precision in effectively simulating runoff in large-scale basins in arid and semi-arid regions. To circumvent the bottleneck caused by lack of data, reference to simulated runoff can be used under similar basin conditions in data-lacking regions.