High Spatiotemporal Resolution Estimates and Trends of Fine Particulate Matter (PM2.5) in East Asia as Inferred from a Geostationary Instrument, 2011-2020

We applied a random forest (RF) algorithm to 2011-2020 aerosol optical depth (AOD) observations from the Geostationary Ocean Color Imager (GOCI) instrument over East Asia to infer 24-h daily surface fine particulate matter (PM_2.5) concentrations at continuous 2x2 km² resolution over eastern China, South Korea, and Japan. Along with AOD, we train on meteorological data, land use indices, precursor emission inventories, and chemical transport model (CTM) output. Missing AOD data are imputed by a separate RF fit where representations of missingness are used as a training input. Similarly, we infer PM_2.5 prior to 2015 in South Korea using an RF trained on surface monitor data for pollutants measured at that time including coarse particulate matter (PM₁₀). We account for sampling bias across land use types and improve representation of rural areas by reweighting training data, and also follow a similar approach to account for rapid growth of surface networks in the study period. The predicted 24-h GOCI PM_2.5 concentrations correlate highly with unseen pixels in crossvalidation (daily R² = 0.82), improving further at annual resolution (annual R² = 0.93). We find that emissions reductions in the region have led to considerable reduction in pollution exposure: in 2012, 49% of the population in the study domain was exposed to a PM_2.5 annual average of 50 ug/m3 or higher, as compared to 29% in 2019. Annual PM_2.5 has followed a statistically significant decline across the study domain from 2012 through 2020, ranging in South Korea between 2 and 5% annual declines relative to 2015 concentrations. However, improvements slowed considerably in the second half of the decade, with insignificant reductions in the Seoul Metropolitan Area, rural Japan, and much of East China. Preliminary results suggest that during the COVID-19 lockdowns in early 2020, PM_2.5 decreased across much of the study domain but increased in rural Japan. We use surface monitoring stations of PM_2.5 composition to offer physical interpretations of these trends and their relationship with declining emissions in the region.