A novel ensemble learning algorithm integrating WRF-CMAQ and downscaling models for hourly estimation of regional air pollution along with vegetation exposure risk detection

Timely assessment of the spatiotemporal evolution, meteorological drivers, and vegetation exposure risks of regional air pollution is critical for addressing challenges posed by rapid urbanization. To achieve this, an integrated machine learning algorithm was developed by coupling the Weather Research and Forecasting-Community Multiscale Air Quality model with a downscaling model to estimate air pollutants during compound pollution episodes in Beijing. The results show that the integrated learning model improves the estimation accuracy of PM2.5, O3, and SO2 by integrating data from Himawari-8 satellite remote sensing, ground monitoring, and other sources, and estimates three additional pollutants using Himawari-8 for the first time. Furthermore, the spatial distribution of composite pollution events was significantly influenced by topography and regional meteorological conditions, with pollutant concentrations being notably higher in the foothill areas. Vegetation Exposure Intensity (VEI) was further introduced to assess the potential health risks to vegetation under air pollution stress. It was found that VEI values significantly increased during the composite pollution period, indicating that vegetation exposure risks rise under high pollution conditions. This study provides new insights into regional air pollution management and the assessment of vegetation health risks.