Accurate identification of precipitation phase is crucial for understanding regional water cycle processes and enhancing water resources management. As a mainstream high-resolution global satellite precipitation product, GPM IMERG has undergone multiple version updates, yet its phase recognition capability under complex climatic conditions still requires systematic evaluation. Based on observations from 2424 meteorological stations across mainland China during 2014-2018, this study comprehensively assessed the performance of IMERG V06 and V07 products using detection metric (POD) and quantitative statistics (CC, RMSE, Bias, KGE) under different precipitation phases (liquid, solid, mixed). The results indicate that IMERG exhibits clear phase dependence, with the highest accuracy in detecting and estimating liquid precipitation (POD >0.7, CC > 0.6), the lowest detection capability for solid precipitation (POD <0.5) but relatively better quantitative estimates than for mixed events, and the poorest quantitative performance for mixed-phase precipitation. In addition, IMERG V07 shows significant improvements over V06 in liquid precipitation across most climate regions, largely benefiting from microwave algorithm optimization; however, it continues to suffer from underdetection and systematic underestimation of solid precipitation in cold and arid regions. Furthermore, phase consistency analysis shows high accuracy in identifying liquid precipitation (with PLP > 50% in more than 98% of cases), whereas solid precipitation remains prone to regional misclassification and is strongly influenced by elevation, temperature, and seasonal variability. Overall, mixed-phase events remain the most challenging for IMERG, with substantial uncertainties in both detection and quantitative estimation. These findings provide a scientific reference for applying satellite precipitation products in cold-region hydrology and climate change research.