Spatial and Interannual Associations Between Vegetation Cover and Land-Surface Temperature Across Urban and Rural Fuzhou, China

Urban greening assessment requires evidence on spatial and annual vegetation–temperature relationships. This study analyzed 5155 warm-season observations from 260 grid cells (300 m × 300 m) across urban and rural Fuzhou, China, during 2019–2025, combining Landsat land-surface temperature (LST) with calendar-year tree and grass classifications derived from Sentinel-2. Generalized additive mixed models separated between-cell differences from within-cell annual deviations, adjusting for meteorology, terrain, building morphology, observation quality, and spatial dependence. The results showed that comparing 60% with 50% mapped vegetation cover yielded adjusted LST contrasts of −0.47 °C in urban centers, −0.42 °C in intermediate settlements, and −0.62 °C in rural areas. Annual mapped cover ten percentage points above a cell’s multiyear mean was associated with a 0.38 °C lower LST (95% CI: 0.30–0.46 °C). Adding tree share increased prediction root mean square error by 0.007 °C in year holdouts and 0.023 °C in buffered spatial holdouts. Against 545 published field references, the combined tree-plus-grass class achieved 95.2% recall and 89.8% specificity, while none of the nine grass references retained the grass label. These adjusted associations support joint assessment of mapped vegetation and surface thermal conditions within the sampled cover range, while annual classification variability and limited composition discrimination constrain interpretation of vegetation changes and tree–grass differences.