Improving predictions of ground-level PM2.5 concentrations in Asia with deep learning and the world’s first geostationary air pollution satellite

Ambient particles raise worldwide concerns due to their impediments on human health, and important roles in the Earth’s weather and climate system via altering radiation and clouds. Particles with diameter less than 2.5 micrometers (PM2.5) are small enough to enter deeply into human lungs, posing the greatest short-term and long-term risks to human health. Accordingly, sources of PM2.5 and particle precursors are highly regulated in most industrialized countries. PM2.5 can linger in the atmosphere for days and exhibit substantial spatiotemporal variations. An accurate depiction of the dynamic evolution of PM2.5 remains a challenge, but urgently needed for better regulation of air quality and health risk assessment. Commonly, ground-based monitoring networks are established to characterize the PM2.5 concentrations in highly populated regions and protected areas such as national parks, but large gaps exist in spatial coverage. Satellite-derived aerosol optical properties serve to complement the missing spatial information of monitoring networks. However, such attempts are hampered under cloudy/hazy conditions or during nighttime. In this study, we aim to overcome the long- standing restriction that surface PM2.5 cannot be constrained with satellite remote sensing under cloudy/hazy conditions or during nighttime. We propose to build a deep learning-based model to fill these observational gaps with data from the newly launched world’s first geostationary air pollution satellite GEMS (Geostationary Environment Monitoring Spectrometer, launched on February 18, 2020 in South Korea). Cloud droplet number retrievals will be also used to constrain below-cloud PM2.5 concentrations, and this approach would provide ground-level PM2.5 concentrations with high spatial resolution and 24-hour temporal coverage. Better constrained spatiotemporal distributions of PM2.5 concentrations will help improve health effects studies, atmospheric emission estimates, and predictions of air quality.