Snowmelt runoff is an important component of water resources in the arid mountainous regions of Northwest China, and accurately simulating the snowmelt runoff process is of great significance for the management and planning of water resources in the arid areas of Northwest China. Taking the upper reaches of the Kaidu River Basin in Xinjiang as the study area, the snowmelt runoff model (SRM) model was improved by setting runoff determination thresholds and runoff amplification functions to supplement the consideration of groundwater impact on the runoff process. On this basis, different schemes of model parameters were formulated under combinations of positive accumulated temperature calculation, runoff coefficients, and recession coefficients, using measured runoff data from 2000 to 2009 for parameter calibration and model validation. The results show that the improved SRM model significantly increases the simulation accuracy of winter runoff processes in the study area, and both the positive accumulated temperature estimation scheme based on a segmented function of the highest temperature and the runoff coefficient determination scheme based on a segmented function of precipitation can significantly improve the model's simulation accuracy. The research findings can provide references for hydrological forecasting and water resource management in the arid areas of Northwest China.