Mapping and Understanding the Synergy Between Land Surface Temperature and PM₂.₅ at 250 m Resolution in Wuhan: Implications for Climate Adaptation and Air Quality Management

Abstract. Urban PM₂.₅ pollution and land surface temperature (LST) are key indicators of urban environmental quality, yet their fine-scale coupling over decadal periods is still unclear. This study maps PM₂.₅ at 250 m resolution in central Wuhan, China, for 2015, 2020 and 2025, using summer, winter and annual composites. A three-stage downscaling framework was developed. First, multi-source predictors, including satellite PM₂.₅, MODIS products, ERA5 meteorology, topography, population and road data, were screened by Variance Inflation Factor analysis, reducing 24 variables to 18. Second, Random Forest, XGBoost and LightGBM were trained on nine year– season datasets. Random Forest performed best in all cases, with test-set R² ranging from 0.531 to 0.936 and near-zero bias (|Bias| ≤ 0.10 μg/m³). Third, the 1 km RF models were applied to 250 m grids, with station-based IDW residual correction where bias exceeded 1 μg/m³. Leave-One-Out Cross-Validation showed an average RMSE reduction of about 35%.

The results show that annual mean PM₂.₅ decreased by 47.5% from 2015 to 2025 (69.28 to 36.39 μg m⁻³), while summer LST increased by 1.39°C. This “cleaner but hotter” pattern indicates a weakening of aerosol-related cooling as PM₂.₅ declined. Pearson correlation analysis at 250 m and 1 km shows that the LST–PM₂.₅ relationship shifted from weak or negative in 2015 to positive by 2025, with annual r₂₅₀ₘ reaching 0.335. The highest PM₂.₅ quintile showed the strongest coupling, reaching r = 0.51 in 2025. These findings support coordinated air-quality and heat-risk management.