Reconstructing long-term seasonal and inter-annual dynamics of floating-leaved and emergent aquatic vegetation from Landsat–MODIS fused time series

Floating-leaved and emergent aquatic vegetation (FEAV) is a sensitive indicator of freshwater ecosystem health, yet its long-term dynamics are difficult to capture because of the optical complexity of shallow lakes and temporal gaps in satellite observations. To enable high-frequency monitoring, we developed a framework that integrates spatiotemporal image fusion and fraction-based machine learning to reconstruct intra- and inter-annual FEAV dynamics. By combining Landsat and MODIS observations, we employed a LightGBM regressor to estimate sub-pixel FEAV fractions and implemented a hierarchical state-tracking scheme to map vegetation dynamics across 30 urban lakes in Wuhan from 2000 to 2024. Validation yielded an overall accuracy of 0.931, and the proposed approach demonstrates a reduction in omission and commission errors relative to pixel-based classifiers. The reconstructed record reveals a pronounced regional decline in FEAV coverage, decreasing from a total area of 141.4 km 2 in 2000 to 20.9 km 2 in 2020, followed by a partial increase to 51.6 km 2 in 2024. SHAP analysis reveals that impervious surface expansion is the dominant negative driver, with extreme summer heat further exacerbating FEAV loss. Conversely, dissolved organic carbon and mean growing season temperatures exhibit positive associations with vegetation persistence. This framework provides a practical tool for tracking continuous aquatic phenology and supports retrospective assessment of ecological degradation in urban lakes.