A Historical-resemblance Machine Learning Approach to PM2.5 Hindcasting and Long-term Exposures in China

Exposure to fine particulate matter (PM2.5) is associated with adverse health outcomes. Over the past two decades, PM2.5 levels in China have substantially changed. Several advanced statistical/machine learning methods have been developed to hindcast PM2.5 concentrations based on the relationship between ground PM2.5 and satellite aerosol optical depth (AOD), but the hindcast estimates outside the monitoring years are less accurate at short timescales and usually have gaps. In this study, we reconstructed a high-quality PM2.5 dataset with full coverage and high spatiotemporal resolution (daily, 1 km) for China's expanse for the period of 2000-2020 and explored long-term changes in PM2.5 exposure. We proposed a novel hindcast modeling method that considers historical resemblances and spatiotemporal features to derive daily PM2.5 variation from ground-level measurements, satellite remote sensing, and numerical simulations. The gaps in PM2.5 estimates due to satellite AOD missingness were addressed through the imputation of high-resolution AOD retrievals. The proposed model achieved superior overall cross-validation (CV) R2 and root mean square error (RMSE) of 0.90 and 12.92 μg/m3 at a daily timescale, improving the by-year CV R2 by up to 0.08 for daily estimates and raising that of monthly estimates to 0.81. Our resultant PM2.5 exposure demonstrates that the national population-weighted annual mean PM2.5 concentrations increased from 49.87 μg/m3 in 2000 to 67.46 μg/m3 in 2011 and then decreased to 35.39 μg/m3 in 2020. Despite the rapid decrease in recent years after the national clean air act, the pronounced seasonal change revealed by the regional monthly population-weighted mean concentrations identified a broad pattern of high PM2.5 exposures (>55 μg/m3) remaining in densely populated regions of eastern China in recent years, with the exception of the Pearl River Delta region. Our long-term PM2.5-attributed premature mortality results further corroborate that the strict air pollution policies in China since 2013 have alleviated the air pollution crisis but suggest more dynamic and comprehensive policies and measures in the future to protect the residents from particle pollution.