Spatiotemporal mapping of lakes across climatic zones using a node-based random forest approach (2000–2023)

• Analyzed 2000–2023 aquatic vegetation dynamics across four climatic zones. • Integrated Random Forest and SHAP to enhance classification interpretability. • Subtropical and tropical lakes showed strong negative water-vegetation correlation. Wetlands are vital for biodiversity conservation and the provision of critical ecosystem services, yet lakeshore wetlands worldwide are increasingly threatened by climate change and human disturbances. Despite extensive studies on aquatic vegetation classification, limited knowledge exists regarding how its dynamics respond to hydrological variability across diverse climatic regions. This study hypothesized that the relationship between water level variations and aquatic vegetation extent differs significantly among climatic regions. To test this hypothesis, we employed a Random Forest (RF) classification model combined with Landsat imagery and Google Earth Engine to analyze aquatic vegetation dynamics from 2000 to 2023 across four representative lakes—Great Salt Lake, Poyang Lake, Tonle Sap Lake, and Ayakkum Lake—spanning semi-arid, subtropical, tropical, and cold desert climates. Model interpretability was enhanced by integrating SHAP values and feature importance metrics. Results showed high classification accuracy, with overall accuracy ranging from 89.67% to 91.22%. Subtropical (Poyang Lake) and tropical (Tonle Sap Lake) lakes exhibited strong negative correlations between aquatic vegetation area and water level (R 2 up to 0.6963), whereas semi-arid and cold desert lakes demonstrated weaker associations due to more stable hydrological regimes. By integrating remote sensing and interpretable machine learning, this study delivers the first cross-climatic zone analysis of aquatic vegetation dynamics, offering valuable insights into wetland ecosystem responses under diverse hydrological and climatic scenarios.