Satellite-Derived PM2.5 concentrations over South Korea Using GOCI Aerosol Products and a Machine Learning Method

Particulate matter with an aerodynamic diameter less than 2.5 μm (PM_2.5) causes an adverse effect on human health including cardiovascular and respiratory system problems. Ground-based PM_2.5 measurements in Korea have limitations in assessing the impact of PM_2.5 on human health because nation-wide PM_2.5 monitoring has been available since 2015 at limited sites mostly around urban areas. To overcome the spatial limitations, we estimated ground PM_2.5 concentrations from the aerosol optical properties observed from GOCI during daylight from January 2015 to December 2018 over the South Korea. The ground PM_2.5 concentrations are estimated from random forest model by using satellite-based remote sensing data including aerosol optical depth (AOD) at 550 nm, the fine mode fraction (FMF) at 550nm, and the surface reflectance at 660nm, together with ECMWF reanalysis data for meteorological variables including boundary layer height (BLH) and relative humidity (RH) as well as land-related information and ancillary data. Those data were collocated to each GOCI YAER V2 Product pixel (6 km × 6 km). We used the 10-folds cross validation (CV) to validate our model. Our prediction accuracy is good with mean cross-validation R² of ~0.74 and root mean squared error (RMSE) of ~14 μg/m³. Our model was able to capture spatiotemporal patterns in PM_2.5 concentration. Also, we evaluated historical PM_2.5 concentrations from March 2011 to December 2014 using the model developed based on data from 2015, which were evaluated at hourly, daily and monthly scales using ground based PM_2.5 measurements. Finally, we analyzed the spatiotemporal trends of PM_2.5 concentration from 2011 to 2018 in the study region. The estimated PM_2.5 concentrations have wider spatial and temporal coverages compared to in-situ measurements, enabling its application to research on the health impacts of long-term PM_2.5 exposure even at rural area.