The snowfall in Beijing on 12 January 2023 was the snowfall first appearing in early and mid-January in last 42 years, and the precipitation types underwent complex changes during this process. In this paper， we utilize various conventional and non-conventional observations such as cloud radar and micro rain radar (MMR) data as well as ERA5 reanalysis data to analyze this event, with the focus on causes of complex phase distribution and variation. The findings indicate that the low-level warm and wet southerly jet stream provided abundant moisture for the precipitation. However, a lack of uplifting of warm and wet air by cold east winds and weak upper trough resulted in relatively low precipitation amount. Meanwhile, the northeast wind just above the surface did not play the role of cold cushion, so not conducive to the maintenance of snow and the snowfall in the whole city. The disparity at the height of the 0℃-layer between western and eastern parts of Beijing, caused by low-level warm advection, explains the snowfall in the west and rain in the east. Additionally, the cooling effect caused by melting and evaporation processes is the main reason for the rapid decrease of 0℃-layer height to below 500 m, which caused rain to snow in plain areas. The thickened warm layer >0℃ and the decreased snowflake size and density above the melting layer caused snow to turn into rain when reaching the ground. When it comes to forecasting precipitation types, deterministic models failed to accurately depict the snow-melting process, but the short-range ensemble forecast results can compensate for the errors in deterministic models. Integrating the data of cloud radar, micro rain radar and microwave radiometer can enhance our ability to monitor and analyze snow formation within clouds and melting in the boundary layer, and improve the accuracy of now casting of precipitation types.