Publications

Liu, Y; Li, CY; Liu, DR; Tang, YL; Seyler, BC; Zhou, ZH; Hu, X; Yang, FM; Zhan, Y (2022). Deriving hourly full-coverage PM2.5 concentrations across China's Sichuan Basin by fusing multisource satellite retrievals: A machine-learning approach. ATMOSPHERIC ENVIRONMENT, 271, 118930.

Abstract
High ambient concentrations of fine particulate matter (PM2.5) increase the hazardousness of air pollution. Aerosol optical depth (AOD) retrieved by sun-synchronous or geostationary satellites is valuable for monitoring large-scale air quality. This study sought to combine the AOD of the Himawari-8 and the Visible Infrared Imaging Radiometer Suite (VIIRS) to derive the spatiotemporal distributions of hourly PM2.5 concentrations at-1 km resolution for China's Sichuan Basin by using a machine-learning approach. Two random forest submodels were developed to fill the daytime gaps in these two AOD datasets, which were then used as predictors in another random-forest submodel predicting the hourly PM2.5 concentrations in daytime. One more random-forest sub model was developed to predict the hourly PM2.5 in nighttime based on the auxiliary variables excluding the AOD data. Leveraging the complementary information of Himawari-8 and VIIRS, this approach demonstrated heightened predictive performance, with cross-validation R-2 of 0.840 and RMSE of 15.9 mu g/m(3). The ability to predict nighttime PM2.5 (R-2 from 0.826 to 0.858) was comparable to daytime PM2.5 (R-2 from 0.831 to 0.853). In the Sichuan Basin, the sustained high concentrations of PM2.5 were mainly attributed to the stagnant meteorological conditions associated with the basin topography. Based on this full-coverage dataset, we acterized the spatiotemporal distributions of PM2.5 across the basin while demonstrating the regional patterns and the weekend effect for a central urban area. Provisioning comprehensive datasets of hourly as demonstrated in this study is essential for air quality management and environmental epidemiological analyses.

DOI:
10.1016/j.atmosenv.2021.118930

ISSN:
1873-2844