Abstract: Based on the daily radiosonde data of Guiyang Station from 2010 to 2024, the ground observation data of Guiyang Airport, and the fine grid numerical forecast product of the European Center from 2023 to 2024, the vertical temperature and humidity characteristics of the atmosphere under the four precipitation phases of snowfall, sleet, freezing rain and rainfall at the airport in winter were statistically analyzed. A phase forecast model was established based on the decision tree algorithm. And conduct quantitative evaluation of the model. The results show that the temperature and humidity profiles of each precipitation phase in winter are significantly different. Although there are differences in common atmospheric physical quantities, it is impossible to effectively distinguish the four precipitation phases based on a single physical quantity value. A decision tree prediction model composed of 2 m temperature, 700 hPa temperature, 500 hPa temperature, cloud top temperature, height difference between 500 hPa and 850 hPa, warm layer top height, and 850 hPa relative humidity was introduced. The accuracy rate of phase discrimination reached 94.83%, which was the optimal prediction model. After evaluation by TS score and comparative test, this model has reference value for actual forecasting work.