Regional Analysis #2

[D2-01] Estimating hourly PM_2.5 using top-of-atmosphere reflectance from Geostationary Ocean Color Imagers I and II

*Hyunyoung Choi¹, Seonyoung Park², Jungho Im¹ (1. Ulsan National Institute of Science & Technology (UNIST), 2. Seoul National University of Science and Technology)

Keywords:particulate matter, geostationary satellite data, machine learning, air quality

To produce real-time ground-level PM_2.5 information, many studies have investigated the applicability of satellite data, particularly AOD. However, traditional methods for retrieving AOD are computationally demanding. Thus, this study proposed a machine learning(ML)-based algorithm to directly estimate hourly PM_2.5 concentrations over South Korea using top-of-atmosphere(TOA) reflectance from the GOCI-I and its next-generation GOCI-II. A light gradient boosting machine(LGBM) was applied as an ML technique. Three schemes were examined based on the input feature composition of the GOCI spectral band. GOCI–II–based schemes 2 and 3 (R²=0.85 and 0.85) performed slightly better than GOCI-I-based scheme 1 (R²=0.83). The TOA reflectance at a new channel (380 nm) of GOCI-II was shown as the most contributing variable, given its high sensitivity to aerosols. The use of GOCI-II enables a more detailed spatial distribution of PM_2.5 concentrations compared to GOCI-I, thanks to its higher resolution of 250 m. Our results indicate that the proposed algorithm has great potential for estimating ground-level PM_2.5 concentrations for operational purposes.