Accurately estimating fine particulate matter (PM2.5) is challenging because many data-driven studies either rely on coarse or low-accuracy datasets, weakening prediction reliability, or focus on ultra-fine resolution mapping that requires dense monitoring, detailed emissions data, and high computational cost, making them impractical for large regions. To address this gap, we developed a spatiotemporal random forest (ST-RF) model to reconstruct monthly PM2.5 concentrations at 1 km resolution across the Yangtze River Delta (YRD) during 2015–2020. The model integrates satellite-derived aerosol optical depth (AOD), meteorological variables, vegetation index (NDVI), digital elevation model (DEM), and spatiotemporal features to capture both spatial autocorrelation and seasonal dynamics. Evaluation results showed robust predictive ability (R² = 0.764, RMSE = 11.135 µg/m³, MAE = 7.238 µg/m³), outperforming conventional random forest, XGBoost, and support vector machine models. AOD and spatial context were the most influential predictors, followed by seasonal effects, temperature, and NDVI. The reconstructed dataset reveals a significant decline in PM2.5 from 2015 to 2020, with winter peaks, summer troughs, and persistent hotspots in inland industrial cities. This study provides a practical and transferable framework that balances accuracy and feasibility for regional PM2.5 mapping and long-term air quality assessment.
Ge et al. (Wed,) studied this question.