Abstract: Cloud phase state identification is of critical importance in the study of climate change, artificial weather modification, and aircraft ice accumulation. In this study, a novel layered cloud phase recognition method based on the random forest algorithm is proposed. A layered cloud phase recognition model for geostationary satellites is constructed using Advanced Himawari Imager (AHI) data from Himawari-8/9, along with the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observations Vertical Feature Mask product CALIPSO-VFM data and aircraft observations. The proximity labeling screening and K-means clustering method were employed to optimize the data quality control. Four infrared spectral channels of AHI (3.9, 8.6, 10.4, and 12.4 μm) were selected as feature inputs, and a cloud phase classification model applicable to all-weather conditions was developed. The model successfully classified clouds at different altitudes (high, medium, and low) and exhibited an overall test accuracy of 97.97% in tests conducted using satellite data. The validation using observed data from the silhouette aircraft in Hubei demonstrates that the model inversion results are highly consistent with the measured cloud phases, especially in the identification of ice and water clouds at the cloud top. This suggests that the satellite cloud phase identification model proposed in this paper has a certain degree of accuracy and reliability in the middle reaches of the Yangtze River, as represented by Hubei.