Reconstructing 1-km-resolution high-quality PM2.5 data records from 2000 to 2018 in China: spatiotemporal variations and policy implications

Abstract

Abstract Exposure to fine particulate matter (PM2.5) can significantly harm human health and increase the risk of death. Satellite remote sensing allows for generating spatially continuous PM2.5 data, but current datasets have overall low accuracies with coarse spatial resolutions limited by data sources and models. Air pollution levels in China have experienced dramatic changes over the past couple of decades. However, country-wide ground-based PM2.5 records only date back to 2013. To reveal the spatiotemporal variations of PM2.5, long-term and high-spatial-resolution aerosol optical depths, generated by the Moderate Resolution Imaging Spectroradiometer (MODIS) Multi-Angle implementation of Atmospheric Correction (MAIAC) algorithm, were employed to estimate PM2.5 concentrations at a 1 km resolution using our proposed Space-Time Extra-Trees (STET) model. Our model can capture well variations in PM2.5 concentrations at different spatiotemporal scales, with higher accuracies (i.e., cross-validation coefficient of determination, CV-R2 = 0.86–0.90) and stronger predictive powers (i.e., R2 = 0.80–0.82) than previously reported. The resulting PM2.5 dataset for China (i.e., ChinaHighPM2.5) provides the longest record (i.e., 2000 to 2018) at a high spatial resolution of 1 km, enabling the study of PM2.5 variation patterns at different scales. In most places, PM2.5 concentrations showed increasing trends around 2007 and remained high until 2013, after which they declined substantially, thanks to a series of government actions combating air pollution in China. While nationwide PM2.5 concentrations have decreased by 0.89 μg/m3/yr (p