Satellite aerosol optical depth (AOD) data assimilation (DA) using numerical air quality forecast models has shown a limited improvement due to large uncertainties in the AOD observation operator. This study employed a machine learning (ML) algorithm to estimate the ground‐level particulate matter (PM) from the Geostationary Ocean Color Imager (GOCI) AOD through the random forest with high accuracy. Analysis fields were subsequently produced by applying PM estimations to the Weather Research and Forecasting‐Chemistry/three‐dimensional variational DA system. Initialization of the model with the new analysis remarkably reduced the analysis error and increased the forecast skill. The PM₁₀ prediction showed significant benefits for up to 24 forecast hours, whereas PM_2.5 prediction was improved for up to six forecast hours. Considering a broad spatial coverage by satellites, the synergistic use of DA and ML can maximize the effectiveness of satellite DA for air quality forecasts at the ground.

This study develops an aerosol data assimilation and forecast system using the Weather Research and Forecasting model coupled with Chemistry (WRF-Chem) and the three-dimensional variational (3D-VAR) data assimilation method. The system assimilates the aerosol optical depth (AOD) from the Geostationary Ocean Color Imager (GOCI) satellite and surface particulate matter (PM) observations. The simulation domain covers Northeast Asia at 15 km horizontal resolution, and the assimilation and forecast skill is evaluated for the Korea–US Air Quality (KORUS-AQ) intensive observing period. Observing system experiments (OSEs) are conducted to examine the changes in quality of assimilation and forecast skills sensitive to the assimilated observational input data. The baseline model simulation underestimates AOD and surface PM concentration in most regions, in which the assimilation of satellite and in-situ data improves the mean biases and spatial distribution. Moreover, it improves the forecast skill of the surface concentration of PM₁₀ and PM_2.5. The results from the OSEs indicate that the assimilation of GOCI AOD only slightly enhances the forecast skill. However, most of the skill improvement comes from the surface PM assimilation, showing a practically useful level of skill until 12 hours from the initial state. The marginal improvement in the PM₁₀ forecasts by the GOCI AOD suggests the non-negligible difference between column-representing AOD and the surface PM concentration.

A numerical simulation system is developed to predict the dispersion of hazardous air pollutants (HAPs) over a populated city due to accidental release. Ulsan, as one of the megacities in Korea, is chosen as an ideal testbed for the simulation, as it is located in complex terrain and hosts a national industrial complex on the outskirts of the city. The system is based on the California puff model (CALPUFF) for simulating a HAP’s dispersion, in which the three-dimensional atmospheric circulation derived from the observed weather station data is specified at a fine horizontal resolution of 200 m. A test scenario is developed for the accidental release of benzene during the daytime and nighttime, respectively, by a fictitious explosion of a storage container, and the injection amount is determined arbitrarily yet comparable to those in the past accidents. In attempting a quantitative assessment and zoning the level of potential risk over the impacted area, multiple simulations have been conducted each day with different hourly varying meteorological conditions in August. The dispersion characteristics of the air pollutant depend largely on the local wind patterns that vary substantially from day to day. Nevertheless, the composite analysis sufficiently identifies the impacted area by the HAP’s dispersion due to the local prevailing wind such as the land–sea breeze circulation. An immediate hazardous area is determined based on the vulnerability map constructed by zoning the level of risk determined by the spatial distribution of the HAPs’ concentration and the harmfulness standard to the human body.

Strong air pollution control policies have been implemented to reduce health damage from heavy air pollution, but the PM2.5 concentration level is still high in Southeast Asia. This study aims to identify the causes of air pollution in East Asia using chemical transport models (CTMs). Using three CTMs (CMAQ, WRF-Chem, and GEOS-Chem), air quality was simulated by season in 2019 in Northeast Asia including China and Korea. The prediction performance of PM2.5 and its major components was evaluated, and the causes affecting the difference between CTMs were analyzed, and ways to improve the prediction performance were considered. The 2019 emission inventory updated to reflect recent changes in air pollution emissions was used in common for all three models.As a result of analyzing the total mass concentration of simulated PM2.5 and major chemical components, the performance of each model was different for each season. CMAQ in January, WRF-Chem in July, and GEOS-Chem in October tended to overestimate the PM2.5 concentrations. For CMAQ, secondary organic aerosol (SOA) was produced by semivolatile/intermediate-volatility organic compounds (S/IVOC), and PM2.5 concentrations was high in winter. This is because CMAQ includes a new pathway for potential combustion secondary organic aerosol (pcSOA). Also, WRF-Chem simulated a particularly high concentration of sulfate because it affected the scavening function according to the aqueous phase chemistry.These multi-model intercomparison of air quality simulations will be helpful in future research to increase understanding of the differences between CTMs in Northeast Asia and to identify the causes of air pollution.