Abstract: Snow cover is a kind of valuable freshwater resources,and the change of snow coverage rate has a profound impact on the development of agriculture and animal husbandry.There has been little research on the prediction of such coverage rates so far.In order to improve the accuracy of such prediction,this study uses machine learning algorithms to construct Support Vector Regression (SVR),PSO-optimized SVR,Random Forest (RF),XGBoost and optimized XGBoost prediction models,which are utilized to predict snow cover in Xinjiang,while comparing and analyzing the prediction accuracy of the models.The results show that the R² values of both RF and optimized XGBoost models are greater than 0.9;the R² values of all traditional SVR models is less than 0.8;the R² values of all PSO-optimized SVR models are greater than 0.8,with some greater than 0.9;and the R² values of all XGBoost models are lower than 0.4.These data indicate that the RF,optimized XGBoost and PSO-SVR models can deliver high prediction accuracy for snow cove;the XGBoost models have the poorest prediction results;and it is necessary to optimize traditional models with different algorithms.