PhD Thesis Presentation - Machine Learning and Multi-Source Data Fusion for Air Quality Health Risk and Heat Exposure Assessment

This study developed multi-source data fusion and machine learning modelling frameworks to overcome the limitations of current atmospheric monitoring systems. This is demonstrated through two key domains of environmental risk: air pollution through the estimation of ground-level NO2, O3 and PM2.5, and heat through the estimation of neighbourhood-scale air temperature (Tair). The air pollution parameters were further employed to quantify the associated long-term air quality health risk (AQHR) in China. Additionally, the measurements-based global perspective on AQHR was explored. Furthermore, the impacts of temperature spatial heterogeneity on heat exposure in complex urban areas were also assessed.

For NO2, a nested machine learning model was designed to estimate hourly concentrations. For O3, multi-source data fusion framework based on XGBoost produced hourly estimates. PM2.5 was estimated using a satellite-based retrieval approach. In addition, a multistep spatiotemporally weighted machine learning framework was developed to generate neighbourhood-scale Tair estimate.

High NO2, O3, and PM2.5 concentrations were concentrated in densely urbanized and industrialized regions in China, with widespread exceedance of World Health Organization guideline values. AQHR assessment further showed that PM2.5 remained the dominant contributor, but oxidants (NO2 + O3) contributed a substantial and spatially coherent share of the total health burden, particularly in regions where particulate pollution has declined. AQHR was highest in northern and central China and increased with population density. Globally, the highest PM2.5-related risks were observed in South and East Asia, whereas oxidant contributions were comparatively larger in North America and Europe. Additionally, the study also demonstrated that heat exposure is strongly shaped by neighbourhood-scale temperature heterogeneity, with extreme heat burden concentrated in specific urban hotspots.

The findings enhanced our understanding for multi-pollutant air quality health risk and heat exposure management. The study underscores the need for integrated strategies that address both air pollution and heat exposure under a changing climate.