Abstract: Studying the distribution and variation trends of aquatic vegetation is of great significance for guiding the restoration of degraded lake ecosystems. To address the difficulty of obtaining large quantities of ground samples in degraded lakes, this study integrates the K-means algorithm with the random forest algorithm to develop an automatic classification method for aquatic vegetation based on phenological characteristics. Using Harmonized Landsat Sentinel-2 (HLS) satellite data, a remote sensing classification study of aquatic vegetation in Liangzi Lake was conducted. The results show that: (1) In terms of extracting aquatic vegetation using phenological characteristics, the overall accuracy of the K-means, using 13 phenological parameters as initial clustering centers was only 37.78%. However, the random forest algorithm which utilized pseudo-invariant samples selected from three consecutive years of K-means results as training data, reaches an overall accuracy of 86.67%. The model using the top 5 phenological parameters, ranked by importance, achieves the highest overall accuracy. (2) From 2017 to 2024, the total area of aquatic vegetation in Liangzi Lake showed a decreasing trend: it declined significantly from 2017 to 2020, and the decline slowed after 2020. The area of emergent/floating vegetation decreased continuously, with its distribution center shifting northwestward and notable losses in the southeastern lake area. The area of submerged vegetation first decreased, increased slightly after 2020, and then decreased slowly, with no obvious trend in its distribution center. (3) Both the total vegetation area and the area of emergent/floating vegetation in Liangzi Lake show a significant positive correlation with ammonia-nitrogen concentrations (average correlation coefficient r=0.868 across the lake), positive correlations with water transparency (with r ranging from 0.288 to 0.806 for total vegetation and from 0.338 to 0.847 for emergent/floating vegatation), generally negative correlations with chlorophyll-a and dissolved oxygen concentrations, and significant positive correlations with the area of artificially planted grass (r=0.752 and r=0.759, respectively). The area of submerged vegetation showed a negative correlation with dissolved oxygen concentration and a significant negative correlation with annual precipitation (r=−0.706). The artificially planted grass areas were basically consistent with the remaining emergent/floating and submerged vegetation (from 2017 to 2024) in Nanwan, Shawan waters, and the waters around Zhushan and Liangzi Island, which alleviated vegetation degradation to some extent. The study indicates that combining phenological characteristics with machine learning algorithms can effectively realize long-time-series dynamic monitoring of aquatic vegetation in degraded lakes and provide support for revealing the mechanisms of vegetation degradation and migration and for developing restoration strategies.