Accurate forecasting of precipitation phase transitions is a key focus and challenge in winter meteorological services. Using multi-source observational data, including ground-based automatic weather stations, ERA5 reanalysis data, wind profiler radar, microwave radiometer, cloud radar, and X-band dual-polarization radar, this study comprehensively analyzes the evolution characteristics of phase transitions during two heavy rain-snow events in Shenyang: from November 5 to 7, 2023 (Event I) and from February 18 to 19, 2024 (Event II). The study also extracts indicative metrics for phase transitions based on multi-source observations. The results show: (1) Event I was controlled by a deep, slowly moving cold vortex system, with stronger dynamic and moisture conditions (maximum vertical velocity > 8m/s, liquid water content reaching up to 12 g·m?3), resulting in large cumulative precipitation and a relatively gradual phase transition process. Temperatures retrieved by the microwave radiometer agreed well with the operational phase transition thresholds. In contrast, Event II was dominated by a fast-moving upper-level trough, with rapid cold air intrusion, leading to a more abrupt phase transition. Furthermore, the microwave radiometer-retrieved temperatures were lower than those from sounding, causing a 1–2 hour deviation in precipitation phase identification based on traditional thresholds. (2) Observations from cloud radar and wind profiler radar indicate that both events exhibited a significant weakening of vertical updrafts (decreasing to 1–2m/s), a sharp decline in liquid water content retrieved by the microwave radiometer, and a continuous drop in the 0°C layer height during the rain-to-snow transition phase. The descent of the 0°C layer bright band and the low correlation coefficient region (0.8–0.92) to the ground, as detected by X-band dual-polarization radar during the rain-to-snow transition, represents the most intuitive and reliable near-term warning indicator for phase transitions. (3) Based on the above multi-source observational characteristics, this study proposes an analytical framework consisting of "dynamic attenuation diagnosis—thermal stratification tracking—dual-polarization radar phase identification," shifting the forecasting mindset from "threshold-based" to "trend-based" analysis. This provides a more operationally feasible technical pathway for the nowcasting of winter precipitation phase transitions.