WRF-Chem quantification of transport events and emissions sensitivity in Korea during KORUS-AQ

In recent decades, rapid economic growth has spurred South Korea and China to focus on air quality and associated health issues. Both countries have enacted a series of increasingly stringent air quality regulations, targeting reduction of ambient PM_2.5 concentrations. The most recent regulations in China also recognize the need to reduce ozone (Lu et al., 2020).

Effective air quality management plans (AQMPs) that will achieve goals, particularly in cases of secondary and transboundary pollution, require a good understanding of aerosol formation and transport processes that are active during pollution episodes. The KORea and United States Air Quality (KORUS-AQ) Study field experiment was conducted in South Korea in May and early June of 2016 and was designed to provide measurements to improve the understanding of air pollution in South Korea and to assist South Korea in the development of an updated AQMP (Crawford et al., 2021).

Results from the experiment confirmed meteorology’s important role in air pollution transport, impacting ozone and PM_2.5 levels over Korea. Four phases were designated by Peterson et al. (2019) using synoptic weather conditions. Of particular interest for this article (because of high PM_2.5 and ozone) are the second “stagnant” and third “transport” periods. The stagnant period was characterized by high pressure, dry, and warm conditions with secondary organic aerosol (SOA) formation from local emissions (Nault et al., 2018). The transport period was characterized by cloudy and humid conditions favoring long-range transport (LRT) at low elevations and secondary inorganic aerosol (SIA) formation (Eck et al., 2020; Jordan et al., 2020).

In terms of PM_2.5 source attribution during KORUS-AQ, previous studies found that 76% of PM₁ in Seoul was from secondary formation (Kim et al., 2018), among which SOA was mainly from Seoul emitted precursors (Nault et al., 2018; Lee et al., 2020). Choi et al. (2019) pointed out the contribution of PM_2.5 from China was time dependent, with a high value of 68% in Phase 3. As for SIA, both Korean emissions and LRT from China were found to be important. Bae et al. (2020) estimated China’s contribution for nitrate was 55%–74%, and sulfate was 25%–56%.

Jordan et al. (2020), Eck et al. (2020), and Travis et al. (2022) highlighted the importance of heterogeneous reactions on hydrated particles for nitrate and sulfate production in Seoul, especially during periods of heavy haze. Travis et al. (2022) complemented these with GEOS-Chem modeling showing sensitivity of nitrate formation and the NO_y budget to nitric acid deposition and overly stagnant nocturnal boundary layers, which severely limit N₂O₅ chemistry in the surface layer via ozone titration.

Schroeder et al. (2020), using observationally constrained box modeling, characterized ozone production in Seoul as volatile organic compound (VOC) limited. Among the VOCs, aromatics with 7 or more carbon atoms, which are mainly from anthropogenic sources, were identified as a high-leverage compound class. Simulations of NO_x controls indicated ozone increases in urban areas and reductions in regional ozone. Simpson et al. (2020) found that toluene was elevated in Seoul and suggested VOC controls should target traffic and solvent sources. Oak et al. (2019) explored ozone production efficiency with GEOS-Chem and airborne observations, emphasizing the importance of aromatic chemistry, boundary layer height, and emission controls on ozone production.

Air quality models are important tools in the development of AQMPs. KORUS-AQ provided an opportunity to intercompare 8 models (Park et al., 2021). It was found that each model had appreciable strengths and some weaknesses in terms of prediction skills. The ensemble mean prediction performed better than individual models across a spectrum of air pollutants.

The model used in this article participated in the intercomparison study mentioned above. Within the intercomparison, the model performance for PM_2.5 was among the best in terms of correlation coefficient (r), normalized mean bias (NMB), and root mean square error (RMSE). The good performance was in part attributed to fine horizontal resolution (4 km).

In this article, this model is used to further analyze the KORUS-AQ study period to provide insights into the transport processes that influenced episodic pollution events. After model characterization focused on these goals, we apply traditional tools (e.g., perturbed emissions, trajectory analysis) and a novel analysis of fluxes over the Yellow Sea to estimate and understand the contributions of Seoul, ex-Seoul, and LRT anthropogenic emissions in Seoul, in other cities in Korea, and across the Korean peninsula. Characterization is performed separately by KORUS-AQ meteorological phase and for both average and impaired air quality conditions. We take advantage of the 4-km resolution and relatively high vertical resolution of our modeling system to investigate chemical sensitivities and transport patterns at a number of elevations over the Yellow Sea and across Korea. Quantification of remote source contribution to secondary species is facilitated by the flux analysis.

The Weather Research and Forecasting with Chemistry (WRF-Chem) model Version 3.6.1 (Grell et al., 2005; Skamarock et al., 2008) was used to simulate air pollution in East Asia during the period April 27–June 10, 2016. The model was used in previous studies (Saide et al., 2020; Park et al., 2021). Two nested domains were used (Figure 1), with an outer domain of 20-km horizontal resolution covering East Asia and an inner domain of 4-km horizontal resolution over South Korea. The outer domain was chosen to account for transboundary influences that impact South Korean air quality during KORUS-AQ, including anthropogenic pollutants from mainland China, dust from the Mongolia Plateau and Taklamakan Desert, and biomass burning emissions from Siberia (Saide et al., 2014). The inner domain had 53 vertical layers with 14 layers below 1.5 km and a model top height of 20 km, in order to resolve Korean emission sources, sea-land breeze circulations, and transport events over the Yellow Sea.

The model configuration details are summarized in Table S1 in the Supplemental Materials (SM). The updated regional atmospheric chemistry mechanism (RACM; Ahmadov et al., 2015) with the volatility basis set (VBS; Ahmadov et al., 2012) scheme for the formation of SOA, and the modal aerosol dynamic model for Europe (MADE; Ackermann et al., 1998) aerosol module were used. The Mellor–Yamada–Janjic scheme (Mellor and Yamada, 1974, 1982; Janjic, 1990, 1994) was used to model the planetary boundary layer (PBL), and the Tropospheric Ultraviolet and Visible scheme (Tie et al., 2003) was used for photolysis rates. The National Centers for Environmental Prediction Final aNaLysis (National Centers for Environmental Prediction, 2007) was used for meteorology boundary and initial conditions, and the chemical boundary conditions were from the MOZART global chemistry model (Emmons et al., 2010). Four days were used to spin-up the model run (April 27–May 1).

Anthropogenic emissions were from KORUS-AQ version 5 anthropogenic emissions (Woo et al., 2020). The model of emissions of gases and aerosol from nature (Version 2.04; Guenther et al., 2006) was used for biogenic emissions, and the Quick Fire Emission Dataset (Version 2.4; Darmenov and da Silva, 2015) was used for biomass burning emissions. The Goddard Aerosol Radiation and Transport (Ginoux et al., 2001; Gong et al., 2002; Zhao et al., 2010) scheme was used to calculate natural dust and sea spray emissions. Heterogeneous reactions of SO₂ on hydrated aerosol particles (Sun et al., 2013; Chen et al., 2016; Cheng et al., 2016; Travis et al., 2022) were not implemented in the model. These sulfate reactions provide additional pathways for sulfate production in models, especially in heavy haze conditions.

Results were analyzed for the entire model run, excluding spin-up period. Results were also analyzed separately for each of the 4 synoptic meteorology phases (Peterson et al., 2019). Most relevant for this work is Phase 2 (May 17–24) characterized by high pressure induced stagnation over Korea, and Phase 3 (May 25–31) characterized by haze development and LRT from China. Note that in Peterson et al. (2019), a 2-day gap (May 23–24) existed that was neither in Phase 2 nor in Phase 3. In this article, May 23–24 is included in statistics reported for the stagnant phase. The dates and characteristics of each phase can be found in Table S2.

Surface meteorology was evaluated using data from 94 stations operated by the Korea meteorological agency (KMA; https://www.weather.go.kr/w/index.do). Data from the AirKorea network system, which consists of 321 stations (https://www.airkorea.or.kr/eng/hourlyTrends?pMENU_NO=151), were used to evaluate the modeled ozone, carbon monoxide, nitrogen dioxide, sulfur dioxide, PM_2.5, and PM₁₀. In addition, observations from 8 National Institute of Environmental Research (NIER) sites (https://www-air.larc.nasa.gov/cgi-bin/ArcView/korusaq?) were used to evaluate PM_2.5 composition. Ground observation details for the NIER Olympic Park site are summarized in Table S3.

Modeled vertical distributions of meteorology and atmospheric composition were evaluated against NASA DC-8 observations. During KORUS-AQ, the NASA DC-8 flew 20 research flights (Crawford et al., 2021). NASA DC-8 flights were conducted over Korea during the local daytime, and for each flight, vertical profiles were sampled over the Seoul metropolitan area. The 60-s merged DC-8 data were used for model evaluation. The model was evaluated against the vertical profiles over the Seoul region (37°N–37.6°N, 126.6°E–127.7°E) and at 2 vertical cross sections (Figure S1) over the Yellow Sea (34°N–38°N, 124°E and 126°E). DC-8 observational variables used in this work are listed in Table S4. Modeled values were interpolated in space and time to pair with DC-8 observations.

Meteorological evaluation benchmarks are listed in Table S5 (Emery et al., 2001). For evaluation of pollutants, recommendations of Emery et al. (2017) were used as summarized in Table S6. Emery gives 2 benchmarks, a tighter value (goal) which indicates good/among top 33% model performance and a looser value (criteria) indicating acceptable/among top 66% model performance. NMB, normed mean error (NME), and correlation coefficient (r) were used for evaluation. To focus evaluation on peak ozone events, hours with observed ozone below 40 ppb were excluded. Model-observation evaluation of surface PM_2.5 and its components used 24-h averages. Also reported are MB, RMSE, index of agreement (IOA), mean gross error (MGE), and mean absolute bias (MAB). Surface observations were compared with modeled values using 2D interpolation of the model output to match monitor locations (i.e., KMA, AirKorea, NIER).

Two emission perturbation runs were performed to analyze the contribution of Korean emissions to air quality in Korea. In the first (Seoul-only) perturbation run, anthropogenic emissions were reduced by 50% for all species in the greater Seoul region (37°N–37.6°N, 126.6°E–127.7°E). In the second (Korea) perturbation run, anthropogenic emissions of all species were reduced by 50% across the whole inner domain (areas of emission changes shown in Figure S2). All other emissions and model configuration choices matched those of the base run. The differences in concentration for each species between Seoul-only and Korea perturbation run were used to quantify the Seoul emission impact. The differences between base run and Korea perturbation run were used to quantify the Korea emission impact.

Forward trajectories (starting from DC-8 observation points) were computed from the WRF-Chem base model run using the hybrid single-particle Lagrangian integrated trajectory (Stein et al., 2015) model. In addition, 24-h back trajectories using modeled meteorology were initialized from surface locations and times, where high pollutant events were observed in Seoul.

Mass fluxes of pollutants over the Yellow Sea (at 124°E and at 126°E) were calculated to quantify the east-west transport between China and the Korean Peninsula. Mass fluxes of selected species were calculated through 2 vertical planes (34°N–38°N, 124°E and 126°E, 0–2 km) extending north to south in the Yellow Sea (shown in Figure S1). The model fluxes were calculated hourly multiplying the u component of wind velocity by species concentration.

Modeled mass fluxes were also compared to those calculated using the DC-8 observations at the above 2 vertical planes. DC-8 observations that fell in the range (34°N–38°N, 123.5°E–124.5°E, 0–8 km) and (34°N–38°N, 125.5°E–126.5°E, 0–8 km) were used in the mass flux evaluation for the 124°E and 126°E planes. Matching the treatment of the KORUS-AQ multi-model intercomparisons project (MICP) (Park et al., 2021), we excluded RF7 and RF18 flights, and we excluded measurements falling within 36.4°N–37.15°N and 126°E–126.88°E (local power plant and industrial complex). The mass fluxes were utilized in several ways.

First, hourly mass fluxes per latitude (below 2 km) were averaged by elevation and plotted to visualize mass flux evolution with time during the 4 phases. Second, the mass fluxes through these planes (below 2 km) were first integrated by vertical plane area and then integrated in time for days of interest, permitting comparison to anthropogenic emissions of primary pollutants and secondary precursors over Korea during the same periods (Equation 1 in SM). NO plus NO₂ was used as the calculated NO_y emission. The mass flux of NO_y was summed over NO, NO₂, NO₃, HNO₃, HNO₄, PAN, and N₂O₅.

Third, for ozone, the integrated mass fluxes (calculated as above) were compared to the time average of change of ozone mass over Korea (0–2 km) between the Korea perturbation run and base run during the same time (Equations 2 and 3 in SM). An approximate contribution of Korean emissions to Korean ozone concentration was calculated by comparing the integrated mass flux at the Yellow Sea and the change in ozone mass over the Peninsula, using a mass balance approach (Equations 4 and 5 in SM).

Finally, a transport indicator for all species was defined as mass fluxes integrated over vertical plane area (surface to 2 km) to the mass of pollutant over Korea (integrating up to 2 km) divided by a residence time of pollutants over Korea. Twelve hours was used as the residence time of all pollutants (Equation 6 in SM). The 12 h was chosen to represent the time for air masses from west to east over Korea, supported by trajectory analysis in Section 4.2.1.

During the KORUS-AQ study period, modeled PM_2.5 and ozone concentrations (and observations, not shown) were elevated over East Asia, as shown in Figure 2. The predicted period-wide daytime mean ozone values exceeded the Korean 8-h standard (60 ppb) over east China, Korea, and Japan. Ozone was also elevated over the Yellow Sea, with the highest concentrations along the west coast of Korea, while ozone was lower over urban regions (i.e., Seoul and Busan), reflecting titration by NO emissions. Period-wide averaged PM_2.5 exceeded 30 µg/m³ over the same regions and showed a strong west to east gradient, with values exceeding Korean 24-h standard of 50 µg/m³ in large regions west of 124°E. PM_2.5 was highest along the west coast of Korea. PM_2.5 (24-h averages) and ozone mixing ratios (maximum daily 8-h average, MDA8) are plotted for Seoul in Figure 3, together with their model counterparts.

Pollutant levels, exemplified by ozone and PM_2.5 at Seoul (Figure 3), varied dynamically during the study period. The 4 synoptic phases proposed by Peterson et al. (2019; 1: dynamic weather, 2: stagnant, 3: transport, and 4: blocking patterns) are adopted herein to help interpret the variation and understand its causes. PM_2.5 in Seoul was at its maximum during Phase 3, and this was reproduced in WRF-Chem. Ozone at the Seoul Olympic Park KORUS-AQ supersite was elevated during specific portions of Periods 2 and 3.

Meteorological skill versus surface observations is mapped in Figure 4 and shown as a time series in Figure 5. Statistical metrics of model-observation agreement can be found for the 94 sites across Korea (Table 1) and for Seoul Olympic Park (Table 2).

Model performance is comparable to other contemporary WRF applications to Korea (Mun et al., 2020; Park et al., 2020; Qiu et al., 2020). Temperature was roughly 1°C low, RH was 0.14% high, and wind speed had a 1.3 m/s high bias. The overprediction of wind speed over complex terrain is a common WRF-Chem shortcoming (Yahya et al., 2015; Kumar et al., 2016). The performance criteria targets (described in Table S5) were met for temperature, RH, and wind speed IOA. Criteria targets were also met for MGE and MAB of temperature and were nearly met for RMSE of wind speed. The predictions also captured the observed spatial variability as indicated by the small and largely uniform biases across the Korea Peninsula.

As shown in Figure 5, the model captured the major synoptic and diurnal variations in Seoul. For example, daytime temperatures were the highest and relative humidities were the lowest during phase 2 (the stagnant phase). Precipitation, mainly associated with the passage of the cold fronts, was also well captured. Quantitative targets were met, with the exception of mean gross error at Seoul. Meteorological performance by phase is documented in Table S7.

The variability in the daytime PBL heights between phases was also simulated with skill. Notice that Figure 5 compares model diagnosed PBL height from WRF-Chem to extracted mixed layer height (MLH) from the ceilometer (which is determined from gradients in aerosol backscatter and is thus an estimate of the aerosol layer height). The comparison suffers from 2 difficulties: (1) the layer heights are based on different techniques and (2) especially at night, multiple aerosol layers may exist and distinguishing a meaningful aerosol layer height can be difficult (Fast et al., 2012; Jordan et al., 2020). However, the modeled nighttime mixing is, on average, too weak at the surface, as evidenced most clearly by excessive NO_x and toluene on many nights, including nights in all meteorological phases. Fast et al. (2012), also using 4 km WRF-Chem with the MYJ PBL scheme, reported lower early morning model MLHs versus both sondes and lidar-based aerosol height. This is a widely recognized shortcoming in WRF, WRF-Chem, and many regional models; improvements are needed and are being widely pursued. Strategies for the improvement of nocturnal mixing during KORUS-AQ and the resulting sensitivity of aerosol nitrate formation related to the chemistry of reactive nitrogen and ozone are discussed in Travis et al. (2022).

Figure 6 maps simulated air pollutant concentrations and bias versus observations at AirKorea observation sites, with quantitative model observation comparison for O₃ and 24-h averaged PM_2.5 in Table 3 (other variables in Supplemental Table S8). As described in methods, hours of O₃ less than 40 ppb were excluded to focus on model skill for peak ozone. Statistics for ozone and PM_2.5 met the quantitative targets with the exception of NME for O₃.

The spatial maps (Figure 6) of the primary pollutants (e.g., NO₂, SO₂) show that the largest concentrations occur around the major metropolitan areas, largely located along the coasts. PM, ozone, and CO show the combination of LRT and local pollution, as reflected in the strong west to east gradients in the concentrations, with the highest values in the metropolitan areas and broadly along the west coast of Korea. Ozone has the highest concentrations over the Yellow Sea along the west coast of Korea and in central regions of Korea. The lowest ozone values are found in the large metropolitan regions due to NO_x titration effects.

In general, the model demonstrated skill in predicting O₃, its precursors, and PM_2.5 at the surface level, with skill among the best in model performance as demonstrated in the KORUS-AQ model intercomparison study (Park et al., 2021). As shown in Table S8, NO₂ and SO₂ were overpredicted (biases of 4.8 and 0.9 ppb, respectively), while O₃ and CO were underpredicted (biases of −1.03 and −0.095 ppm, respectively). PM_2.5 was biased high (8.5 µg/m³) and PM₁₀ was underpredicted (bias of −6.8 µg/m³). This may reflect an overprediction of secondary aerosols and/or errors in the primary emission mass size distribution (Saide et al., 2020). This will be considered further in Section 4.

Spatially, the model showed consistent skill across the peninsula. The largest biases lie in (1) the underestimation of CO, especially in the Seoul metropolitan area and Incheon; (2) overestimation of PM_2.5 in Yeosu, Pohang, and Seoul region; and (3) underestimation of PM₁₀ across Korea.

With more than 50% of the Korean population living in the Seoul region, model evaluation in Seoul is important. The richness of observations in Seoul also facilitates a detailed model evaluation. Figures 7 and 8 show the time series of observed and predicted ozone and PM_2.5 related parameters at the Seoul Olympic Park site. The statistics are shown in Tables 4 and 5. The time series of O₃, NO_x, CO, and toluene is shown in Figure 7. Evaluation of the model’s hourly ozone performance at Olympic Park met bias and error criteria as described in Section 2, except for underestimation (NMB = −18%) in Phase 2 indicated by Table S7. Correlation coefficient of hourly predicted and observed ozone at Olympic Park (0.47) fell slightly below the benchmark (0.50). CO was slightly underestimated in Phases 2 and 3. Modeled NO_x and toluene captured observed daytime values and variations between the 4 phases but was biased high relative to observed nighttime values, most noticeably in transport and blocking phases.

Comparison of NO_x vertical profiles, NO_x concentration time series, ceilometer-derived aerosol MLH, and modeled and observed winds in Seoul suggest insufficient vertical mixing of NO_x on nights with significant surface NO_x overprediction. Overprediction was more prevalent during periods of low modeled wind speeds; however, counterexamples of low wind speeds with excellent NO_x skill at night exist, particularly during phase 2. We hypothesize that the actual high pressure during Phase 2 matched the limited vertical mixing in WRF-Chem. Limited vertical mixing is related to the default WRF-Chem configuration, which does not incorporate the anthropogenic heat and turbulence generated by the urban canopy (Chen et al., 2011); lower thresholds on vertical diffusion coefficients can be used over urban areas to improve this (Kim et al., 2021). In terms of entire KORUS-AQ period, the nighttime ozone predictions at Seoul Olympic Park were skillful given that in Seoul there is sufficient NO to fully titrate the nighttime ozone on a majority (61%) of nights. Studies by Eck et al. (2020), Jordan et al. (2020), and Travis et al. (2022) emphasize feedbacks between nocturnal ozone and nighttime processing and fate of NO_y species. The relative significance of insufficient vertical mixing to explain model overestimation of surface NO_x at night requires further investigation. Correct vertical profiles of ozone, NO_x, and aerosols are required for accurate representation of heterogeneous NO_x oxidation pathways and their impacts on the budget and speciation of NO_y.

The time series for PM_2.5 and its components at Seoul Olympic Park are presented in Figure 8 (a similar time series for the KIST monitoring site can be found in Figure S3). In general, the model showed acceptable skill in representing the variations in PM_2.5 mass throughout the study period, including the steady increase in PM_2.5 during May 24 to May 27 in Seoul. Although precipitation on the 24th in Seoul was reproduced in the model and resulted in extensive wet removal, the amount of wet removal was insufficient (WRF-Chem decreased to 15–20 µg/m³ vs. observations at approximately 5 µg/m³).

Overall, WRF-Chem overpredicted PM mass. Black carbon (BC) was well captured during the daytime and overpredicted at night, when the PBL was underpredicted, as discussed earlier in Section 3.4. The variations of the SIA by phase were captured by the model, but with a general overprediction, most prominently for nitrate. Organic carbon (OC) was overpredicted during the transport and blocking phases and slightly underpredicted during the stagnant phase. This suggests that the model did not capture the strong SOA formation occurring from local emission as observed (Nault et al., 2018) and may have overpredicted SOA formation during LRT. Model modifications, not incorporated in this work, have been postulated in Hodzic et al. (2016) and include higher local formation rate of SOA and stronger SOA sinks (more effective dry and wet depositions, photolysis, and heterogeneous oxidation mechanisms).

Jordan et al. (2020) highlighted the importance of SOA production during the stagnant phase (Phase 2) and high sulfate and nitrate concentrations (Phase 3), leading to a compositional shift from a majority organic Phase 2, to a majority SIA Phase 3. This can be seen in the observations of Figure 9a. WRF-Chem (Figure 9b) reproduced some aspects of the composition shift, but did not fully reproduce it, in part from a persistent nitrate overprediction in all phases. The modeled fractional composition shows a notable shift between Phases 2 and 3. SIA increased from 40% to 70% from Phase 2 to 3, while organic mass reduced from 60% to 40%. In the model, Phase 3 had the highest absolute amount of sulfate in both observations and model, and the absolute increase in sulfate between Phases 2 and 3 was approximately correct. As we have discussed elsewhere, the SOA module implemented underpredicted Phase 2 SOA.

In addition to statistical comparison to observations at Seoul, the modeled PM mass and composition were evaluated versus 8 NIER sites distributed throughout South Korea, with results in SM (Figures S4 and S5 and Table S9). Conclusions from this wider evaluation are similar to those from Olympic Park; nitrate overprediction was common, with an NMB being 118%. Furthermore, most of the statistics are better than those for the Olympic Park site. The better performance may relate to some of these sites being less urban (Mt. Taehwa) or coastal (Jeju), thus capturing more background conditions.

Modeled ozone and PM parameters discussed above at the Olympic Park site were also evaluated by phase (Tables S7 and S10). In general, the performances were similar across phases with a few exceptions. For example, during the stagnant phase ozone and temperature predictions degraded slightly, while PM_2.5, NO_x, SO₂, and toluene metrics improved slightly relative to other phases. In the stagnant phase, the SIA observed mass fraction to total PM was the lowest (approximately 30%), but the predicted fraction did not vary much by phase and was around 60%.

Illustrative profiles by phase are presented in Figure 10, with additional vertical profiles in SM (Figures S6 and S7). Temperature and RH were reproduced sufficiently well, with some scattered errors at higher altitudes. For ozone, the vertical profiles show that the underestimation near the surface (shown in the surface comparisons) extended throughout the vertical profile. Daytime NO_x (biased low, NMB of −15% during daytime at Seoul Olympic Park) was also lower in simulation than in observations in the DC-8 profiles. In contrast, CO profiles (Figures S6 and S7) were well predicted in the boundary layer (for Phases 1–3), but systematically overpredicted above the PBL.

BC was slightly overpredicted below 2 km but had little bias from 2 to 8 km. The nitrate vertical profiles show a different behavior. In contrast with comparison to surface observations at Seoul and across Korea (which showed positive model bias for nitrate), DC-8 profile analysis shows observation-model agreement in the lowest layer (below 1 km). However, modeled nitrate was systematically underpredicted at higher altitudes, with model values decreasing with altitude much more quickly than the observations. The same is true for sulfate and ammonium (Figures S6 and S7). This apparent discrepancy is due to diurnal patterns in overprediction, similar to those described above for NO_x.

The bias in vertical profile especially at high elevations suggests that boundary conditions have strong impact on the model prediction, affecting concentrations aloft and at the surface, even in nested configurations like that used in this study. Additional efforts are needed to improve boundary conditions available for use by regional models. This can be accomplished through advancing applications of observation constrained global predictions as boundary conditions.

Figure 11 compares the vertical distributions of ozone mass flux between model and DC-8 observations. Ozone concentration decreased as elevation increased. On the contrary, wind component u increased (in west to east direction) with higher elevations. Mass flux increased with height and demonstrated west to east transport in all elevations, except for at 126°E east to west transport below 1 km mainly due to offshore flow.

Compared to observations, Figure 11a shows that the model had a very good prediction of wind component u at all elevations at 124°E. For ozone, the mass flux bias was small at all elevations. Ozone precursors and PM_2.5 components evaluation are shown in Figures S8 and S9. In general, NO_x and toluene concentrations were overestimated near the surface but were underestimated above 3 km, resulting in their mass flux biases being larger near the surface and smaller at higher altitudes. SO₂ concentrations were underestimated near the surface and overestimated above 3 km, resulting in relatively larger biases at the surface and smaller biases at higher altitudes. Mass flux of CO was predicted quite well below 4 km and had a larger bias at higher elevations. At 126°E (Figure 11b), good agreement of wind component between model and DC-8 observations was achieved from the surface up to 8 km. Ozone concentration was underestimated from surface up to 6 km. Mass flux of ozone was consistent with the observations at all elevations, except for underestimation between 6 and 7 km. Shown in Figures S8 and S9, the mass fluxes of NO_x and toluene were underestimated at the surface, but in good agreement above 2 km. The mass flux for CO was predicted quite well at all level.

Phases 2 “stagnant” and 3 “transport” during KORUS-AQ contrasted in terms of their air quality impact and represent complex and challenging conditions for simulation due to the combination of horizontal transport, vertical transport, and local production. For example, PM_2.5 surface concentrations in Seoul were highest and frequently surpassed the 24-h Korea air quality standard of 50 µg/m³ during the “transport” period, but were lower in the “stagnant” period. As discussed in other KORUS-AQ analyses, aerosol composition shifted from a majority organic composition (Phase 2) to majority inorganic (Phase 3; Jordan et al., 2020). Surface ozone at Seoul Olympic Park, however, was elevated in both phases and exceeded the Korea 8-h standard of 60 ppbv (Kim et al., 2020). The differences between these 2 phases also extended above the surface, as shown in Figure S10a and c. Meteorological variables (temperature, relative humidity, and PBL height) also exhibited appreciable variability between the two phases (Figure 5).

A prominent feature of Phase 2 was high pressure over Korea, with elevated ozone over the Yellow Sea, and anticyclonic flow patterns. As the center of the high pressure system moved eastward, winds near the Korean west coast changed from southward to westward and eventually to northward. Most prominently during May 24–26, and in a series of 4 waves of west-to-east transport, stratified low-elevation plumes containing high concentrations of PM_2.5 and ozone advected from the Yellow Sea over Korea, with differential mixing into the boundary layer that varied in space and time. On many days, coastal locations along the west coast of Korea were also influenced by diurnal land-sea breezes. Key features of the vertical structure of ozone and aerosol extension, associated winds at the surface and aloft, spatial patterns of NO_x, O₃, during these periods, are discussed in Figure S11 in SM.

The contribution of domestic Korean emissions to the air quality during these periods was investigated by the analysis of the simulations with perturbed emissions. By using the difference in concentrations between the Seoul and whole Korea perturbation runs and base run, percentage contributions were calculated and are discussed in the following. Results are divided into influence from anthropogenic emissions of Seoul, of rest of Korea, and of other sources (i.e., LRT and secondary production from sources outside of Korea). These are policy-relevant in that they show the potential influence of Korea-specific emission regulations on KORUS-AQ period concentrations.

Figure 12 shows the source contributions for surface NO_x for 4 cities. For Seoul, local emissions account for more than 85% during the daytime and 70% during the nighttime during Phase 2. Other Korean emissions account for up to 30% during this period. For Seoul in Phase 3, the contributions of other Korean emissions and outside sources play larger roles (up to 50% and 20%, respectively). For the other cities shown, Korea excluding Seoul emissions dominate.

The geographic influence of PM_2.5 sources, shown in Figure 12b, is quite different from that of NO_x. For PM_2.5 in Seoul, the combined contributions of Seoul and Korean ex-Seoul emissions are approximately 50% (30%) during the stagnant (transport) period. Because both BC and OC were underpredicted in the stagnant phase, the Seoul contribution is likely underestimated. Typically, PM_2.5 concentrations were higher in Phase 3 across all 4 locations, as shown in Figure 12b. In Phase 3, higher PM_2.5 values were associated with larger LRT contributions; the LRT contributions were higher at higher elevations (Figure S12 and Table S11). During the transport phase, domestic contributions were higher at east coast regions than west coast, due to the eastward transport direction, which enabled domestic emissions from the west coast emissions to impact the east coast regions. For example, note the Seoul emissions impact at Donghae (Figure 12b). Supporting time series and maps are found in Figures S13 and S14 of SM.

The period-average source contributions for primary and secondary PM_2.5 at Seoul are summarized in Table 6. PM_2.5 had substantial LRT contribution in both phases, with 71% in Phase 3. Primary species (BC and primary organic aerosol [POA]) had substantial (>70%) from Korea emissions. For SIA, nitrate and ammonium contribution varied greatly by phases, with Korea sources dominating in Phase 2. Over 90% of sulfate at Seoul were from sources outside Korea; this should be considered an upper limit on LRT of sulfate, due to the likely occurrence of heterogeneous production of sulfate on hydrated aerosol particles during haze events, a mechanism explored for KORUS-AQ in Travis et al. (2022). After adding this heterogeneous pathway, Travis et al. (2022) found LRT of PM_2.5 dropped from 66% (25 of 38 µg/m³) to 54% (29 of 53 µg/m³) during haze event, and that locally produced sulfate (6 µg/m³) accounted for 25% of the locally produced PM_2.5 (24 µg/m³) after adding heterogeneous reactions_. Representativeness of our results for PM_2.5 nitrate sensitivity is uncertain, particularly for Seoul. This is in part due to the lower model performance for nitrate relative to other species (e.g., Figure 5), coupled with excluded processes (heterogeneous formation of nitrate). Coupling of the Seoul nocturnal boundary layer and nitrate formation during KORUS-AQ, as detailed by Travis et al. (2022), is discussed in Sections 3.2 and 3.4.

Sensitivities of concentrations to Korean anthropogenic emissions (Seoul + rest of Korea) are mapped in Figure 13. These are shown as percent difference in concentrations (Korean emission run minus base run, relative to base run). The NO_x concentrations decreased nearly proportionally to the reduction in Korean emissions (50%). PM_2.5 responses were lower, on order 20%. Ozone (Figure 13c and d) increased over the main urban centers significantly and decreased slightly over rural regions. Influence of Korean emissions on the Yellow Sea varied significantly by phase, due to easterly winds during Phase 2 and westerly winds during Phase 3.

Sensitivities match those, within ±10%, of Choi et al. (2019) for LRT contribution to PM_2.5 in Phases 2 and 3, as well as for the LRT contribution to BC during KORUS-AQ. Eck et al. (2020) reported during May 25 and 26 that 31% of PM_2.5 in Seoul was formed from the oxidation of gases emitted over Korea. To compare, we report 29% of PM_2.5 resulting from Korean emissions, although our averaging window is slightly different, and our value includes a portion from primary emissions within Korea.

High sensitivity of NO_x concentration to local and Korean emissions, and impact from aloft transported ozone, is consistent with previous work reported in Nault et al. (2018) and Crawford et al. (2021), respectfully. For sulfate local and Korean impacts, our reported values should be taken as lower bounds, due to the exclusion of the heterogeneous mechanism for SO₂ oxidation on hydrated particles. Inclusion of that mechanism increased the Korean influence on sulfate to a majority of total sulfate and was responsible for 6 µg/m³ of modeled sulfate.

The vertical variation in the impact of domestic emissions is shown by the curtain plots (Figure 14). NO_x domestic emissions dominated NO_x concentrations; thus, NO_x concentration at all elevations were reduced by nearly 50%. The reduction of NO_x emissions led to increases in O₃ concentration near the surface (0–1 km) as shown in Figure 14b; on the contrary, NO_x reduction led to decrease of O₃ at higher altitude (>1 km). Decrease of PM_2.5 was more prominent in Phase 2 than in Phase 3 (Figure 14c), indicating more contribution to PM_2.5 from Korea domain emissions in Phase 2. During Phase 3, moving from west (126°E) to east (129°E), the color becomes darker (Figure 14a–c) reflecting the influence of domestic contributions transported from west to east. Additional species is shown in Figure S15.

AQMPs often target periods with the highest concentrations in order to manage acute health effects and visibility reductions associated with heavy haze. To support control of the highest concentration periods, we analyzed the sensitivity of the highest 10% of the predicted concentrations of key pollutants to emissions reductions. The results are summarized for Seoul, Gwangju, and Busan in Figure 15 (see SM Figure S16 for other cities). Pollutants, such as NO_x, that are near the 50% circle, have close to a 1:1 sensitivity, that is, a 50% emission perturbation reduces concentrations by 50%. Values near the 100% circle, that is, SO₄⁻ in Seoul, are insensitive to perturbation of emissions in Seoul or in Korea.

Figure 15 shows that peak O₃ is insensitive, and in some cases increases, as a result of emission decreases. This is a combination of the influence of transported ozone and ozone precursors, which are not influenced by the emission perturbations, together with the counterbalancing of O₃ production decreases (from NO_x and VOC decreases) versus decreases in titration by NO. The O₃ sensitivity result is not limited to the 3 selected cities; the same result is seen in cities across Korea (Jeju, Daejon, Andong, Incheon, Daegu, Ulsan, and Donghae, Figure S16).

For Seoul, whether the green line (local emissions reduction in Seoul Metropolitan Area) or the blue line (Korea ex Seoul emission reduction) is farther toward the origin is an indicator of relative sensitivity. For example, NO_x, CO, toluene, and BC are relatively more locally (i.e., Seoul) controlled, while PM₁₀, PM_2.5, NH₄⁺, and NO₃⁻ are relatively more regionally controlled. This is consistent with the large percent of PM from secondary processing as shown in Figure 9 and previous studies (Lee et al., 2020).

Peak PM_2.5 has variable sensitivity among regions. During Phase 3, it appears to be insensitive to Korea emission at western sites, but sensitive at eastern sites, reflecting eastward transport over the Yellow Sea and across Korea.

In general, except for southern sites (Jeju, Busan, and Ulsan), peak/top 10% concentrations are more sensitive to Korea emissions in Phase 2 than Phase 3 on both east coast, west coast, and interior regions. At the southern sites, sensitivity is approximately the same in both phases. The LRT had larger impact in Phase 3 and over west coast regions, especially for BC, CO, sulfate, and PM.

On May 25, a heavy haze event over the Yellow Sea was captured by the NASA DC-8. As we show in the following, this event furthermore had mass transport rates of PM and O_x over the Yellow Sea that equaled or approached maximum values for the campaign. The May 25 haze over the Yellow Sea occurred at the beginning of Phase 3 (transport). In this section, we investigate the May 25 event in detail and use it as an example case to demonstrate metrics for transport. We focus on (i) the altitudes at which transport relevant to Korean air quality occurs; (ii) the time delay between high concentration over the Yellow Sea and over Korea; (iii) representative trajectories of transport connecting air over the Yellow Sea, Seoul, and other locations in Korea; and (iv) quantification of mass fluxes crossing the Yellow Sea during key periods of KORUS-AQ.

We believe these analyses can be generalized to periods outside of KORUS-AQ to show multiyear and seasonal trends of transport (absolute amount and speciation) across the Yellow Sea. We feel that validating these transport metrics on a well-studied episode is necessary and have chosen the May 25, 2016, DC-8 flight and the adjacent Phases 2 and 3 of KORUS-AQ for this purpose. As we show here and elsewhere in this article, not all transported pollution makes its way to the surface. However, a substantial enough fraction does reach surface elevations in Korea to help explain the results from Section 4.1 that indicated high influence of non-Korean sources (e.g., 51%–71% of PM_2.5 at Seoul were from sources outside Korea during Phases 2 and 3).

High concentrations of ozone and aerosols were observed by the DC-8 during the May 25 flight. Model curtains of aerosol extinction during the flight are shown in Figure 16; model-observation comparison during the flight was conducted by Saide et al. (2020). The haze had its peak ozone 139 ppbv at 124.29°E, 35.51°N, shown by the black diamond in Figure 16. This haze layer was observed by the DC-8 three times during the flight (9, 12, and 14 KST), first while flying south at high altitude, and then flying north and descending down through the haze, and finally flying north and ascending up through the haze. The final high extinction layer at 15:30 KST is near Seoul prior to landing. The flight path of the DC-8 on May 25 can be seen in Figure 18d.

To establish the potential for the haze over the Yellow Sea to impact surface concentrations in Korea, forward trajectories from the DC-8 flight path were calculated, as well as backward trajectories from Seoul. These trajectories also establish the altitudes over the Yellow Sea with the highest potential for impact on the surface over the Korean Peninsula. While the trajectories are based on the peak ozone observed by the DC-8, ozone and PM were coincident, and the analyses are relevant to both pollutants.

Figure 16 shows that the haze layer was confined to below 2 km in altitude. A forward trajectory (Figure 17) was initiated at the DC-8 peak ozone location (124.29°E, 35.51°N, 305.2 m, 13:48 PM KST). Other trajectories (shown in Figure S17) were calculated starting from DC-8 flight path locations throughout the day, at altitudes of 0.1, 0.3, 0.9, 1.2, 4.7, and 7.0 km, respectively. Above 4 km, trajectories passed over Korea. Trajectories below 2 km subsided as they passed over the land. The DC-8 observations show high pollution levels at elevations between 0 and 2 km over the Yellow Sea, which was also captured by the model as shown by the high extinction coefficients below 2 km along the DC-8 path (Figure 16).

This shows that the transport with potential for impact on surface concentrations over Korea during the first haze event of Phase 3 occurred primarily at altitudes below 2 km. We further analyzed how the pollution levels over the Yellow Sea impacted Korea by calculating the end points of a cluster of trajectories starting from the region with the highest observed ozone concentrations.

Figure 17 shows the impact of the haze over the Yellow Sea, using forward trajectory footprints at 6, 12, 18, 24, and 48 h after release. Pollutants took 12 h to reach the northwest coast and impact the Seoul region. During hours 12–18, these polluted air masses dispersed over the northern part of South Korea before reaching the eastern coast. Over the next 6 h (18–24 h), these air masses moved clockwise and influenced southeast regions (i.e., Ulsan and Busan). The southwest region of Korea was not impacted by these trajectories throughout the haze event.

These results indicate that the pollutants detected over the Yellow Sea took approximately 12 h to impact Seoul. The timing and altitude of transport is further supported by backward trajectory footprints shown in Figure 18, starting at Seoul on May 26 at 1:48 AM (KST). The start and end times used in the backward trajectory analysis are noted in the time series (Figure S10b and d).

Trajectory analysis graphically depicts transport from the Yellow Sea, but is qualitative. Determination of local versus long-range influence via model perturbation runs is resource intensive. Here, we explore a third method, using the analysis of zonal mass flux over the Yellow Sea. Zonal mass fluxes can be computed from reanalysis atmospheric composition products, which are increasingly constrained by remotely sensed observations. Thus, transported pollutants across the Yellow Sea can be quantified for individual episodes or at longer time scales to investigate seasonal cycles and multiyear trends.

Here, we explore this technique using a flux plane of 2 km in height, running north-south through the Yellow Sea (Figures 17, 18, and S1). Fluxes were calculated at 2 distances from the Korean coast: a more westward plane was located at the center of the Yellow Sea (34°N–38°N at 124°E); a more eastward plane (34°N–38°N at 126°E) was located approximately 50 km off the west coast of South Korea. Evaluation of the fluxes compared to the DC-8 was presented in Section 3.6. The north-south placement of the flux plane (34°N–38°N) and the vertical extent of integration of the fluxes (2 km) are designed to isolate the portion of the Yellow Sea with high impact on South Korean surface air quality based on the trajectory analysis and DC-8 observations in the previous section.

The concept is illustrated in Figure 19, which visualized the mass flux for the entire KORUS-AQ period for O_x (O₃ + NO₂) and for PM_2.5. Meteorological phase is indicated by the Roman numerals in Figure 19a1 and b1. Fluxes from east to west are in blue, while fluxes from west to east are in red. Times and locations where the DC-8 sampled near (±0.5°) the flux planes are shown with black stars. A point of reference is the previously discussed strong west-to-east transport event detected over the Yellow Sea on the afternoon of May 25, which can be seen by dark red colors at the approximately 37°N for both O_x and PM_2.5. Mass fluxes for O_x are in the ±2,000 µg/m²-s range (mostly as O₃), while mass fluxes for PM_2.5 are in the ±500 µg/m²-s. Integrated across the entire flux plane, these amount to 8.6 × 10⁷ kg/day for O_x and 2.3 × 10⁷ kg/day for PM_2.5 during haze event.

While these flux time-latitude figures (Figure 19) contain extensive information, the most obvious features are the occurrence of strong westward (blue) and eastward (red) fluxes, and the rotational flow resulting from high pressure systems with anticyclonic flows (eastward red at the north portion of the flux plane and westward blue at the south portion of the flux plane). These are most prominent below 1 km (Panels 1 and 3).

Other less prominent features can also be seen in Figure 19. Diurnal alternation of land-sea breeze is shown in Figure 19a3 as a sequence of color changes in the green rectangle. Two factors facilitate visualization of diurnal land-sea breeze in this region: (1) Gwangju coastline is close to 126°E and (2) lower synoptic wind speeds during stagnant period permit sea-land breezes to develop. Due to the high model resolution, wind convergence on May 27 (black rectangle) is easily apparent. Further analysis of mass fluxes by elevation is shown in Figure S18 and by PM component in Figure S19.

While the fluxes help visualize zonal transport, normalization of the magnitudes to form a dimensionless indicator of transport strength is necessary for further interpretation. This allows the assessment of whether any given flux represents significant transport. We investigate 3 different normalization methods: emission rates in our Korea domain, daily formation rate of secondary pollutants in the boundary layer over Korea, and mass of pollutant in below 2-km elevation over Korea. The first method works only for primary pollutants; the latter 2 methods work for both primary and secondary pollutants.

The first method (normalization to Korean primary emissions) is demonstrated for 3 KORUS-AQ days, with results in Table 7. May 21 represents a typical day in Phase 2. May 24 had strong zonal transport over the Yellow Sea, and May 26 had high PM_2.5 concentrations over Korea. The average mass fluxes through the flux planes described above were calculated for the days of interest (using KST midnight as the starting time for averaging). These were normalized by the average mass emission rates in the Korean domain as described in methods. Larger absolute values imply larger importance of LRT relative to Korean emissions. Positive (negative) values represent west-to-east (east-to-west) transport. Additional values involved in the calculation can be found in SM (Table S12).

In general, primary pollutants with short lifetimes, for example, NO_x, SO₂, isoprene, NH₃, and toluene, had the lowest absolute values, typically below 1. For primary species with longer lifetime, for example, BC ratios were around 1.5 in Phases 2 and 3, indicating contributions from both local and remote sources. However, during the haze event, Yellow Sea flux for BC was 4–6 times larger than Korean emissions. Similarly, POA ratios were high during Phases 2 and 3. These results are consistent with Seoul results in Table 6. For CO, as expected for compound with a significant hemispheric background, transport was dominant, with absolute values ranging from 6 to 20.

Longer lived species had intermediate values indicating importance of both LRT and domestic emissions. For total sulfate (SO₂ + SO₄⁻), the ratios were under 1.5 in Phases 2 and 3, but 3.0 during the haze event. Total ammonia (NH₃ + NH₄⁺) ratios were below 1.0 in Phase 3 and 1.5–2 during the haze event. SO₂ and NH₃ ratios were lower than ratios for total sulfate and total ammonia, consistent with transport over the Yellow Sea being primarily as secondary aerosol.

On May 24, strong and dominant westerlies passed over the Yellow Sea. On this particular day, value for 124°E is always larger than 126°E, due primarily to entrainment of relatively cleaner air into the 0–2 km range due to frontal dynamics.

Ozone (and other secondary pollutants) requires an alternate method of normalization. Even though the Yellow Sea ozone flux can be large (e.g., 10⁷ kg/day on the 24th at 124°E), ozone precursors (NO_x and representative VOCs) were dominated by Korean domestic emissions. As described in Section 2.8, the ozone fraction from net production from Korean emissions was calculated by comparing the average change in near surface (0–2 km) ozone mass over Korea to the Yellow Sea ozone flux. The results for May 24 and May 26 (Table 8) indicated the fluxes over the Yellow Sea were larger than the ozone production from Korean domestic emissions.

An additional normalization method to interpret the Yellow Sea mass flux’s importance for controlling concentrations was to compare the transport of a given species to its mass at low altitudes (0–2 km). This is referred to hereafter as a “transport indictor” with a convention of positive values for west-to-east transport and negative for east-to-west. Values for Phases 2 and 3 for O₃ and PM_2.5 are shown in Figure 20. Phase 2 has east-to-west transport and in Phase 3 the transport direction is mainly from west-to-east. The highest transport indicator was found during haze event (May 24–25) for both ozone and PM_2.5. There was also a transport peak on May 30.

The transport indicators for key species on May 24 (haze event) are plotted in Figure 21. At the left are the least transport influenced species (i.e., NO_x, isoprene, NH₃); and on the right are the most transport influenced species (i.e., PAN, PM_2.5), while in the middle are mixed influenced species (i.e., O₃). The indicators change by day and also depend on where the mass flux was calculated (for Figure 21 mass flux was calculated at 124°E). However, it is consistent with results in previous sections and enlightens the species-specific transport influence during KORUS-AQ. The transport of PM_2.5 is large, while for ozone, it is intermediate.

WRF-Chem with RACM-MADE-VBS was used to provide insights into the transport processes influencing key KORUS-AQ periods, as well as modeled sensitivity of peak and average concentrations at multiple locations and elevations. The 4-km resolution and relatively high vertical resolution of our modeling system allowed resolving of narrow transport features over the Yellow Sea. Extensive evaluations showed fitness for purpose of investigating processes, such as primary emissions, secondary formation, and transport. However, several model limitations affecting the balance of transported versus local production suggest that estimates of transport influence on PM_2.5 concentrations should be considered an upper limit. These include omission of heterogenous nitrate formation reactions, which will be most prominent in high NO₂ locations, such as the Seoul metropolitan region, lack of heterogeneous sulfate production from SO₂ uptake on hydrated aerosols, underestimation of local SOA formation in the stagnant phase, and overestimation of SOA during the transport phase likely due to insufficient sinks.

Analysis quantified contributions from local, regional sources and LRT during the stagnant and transport phases. Domestic emissions were dominant for NO_x (over 85% in Phase 2) and other primary species with short lifetimes. Analysis highlighted extensive PM_2.5 transported (71% in Seoul during Phase 3), with increases at higher elevations. Sensitivity of pollutant level to Seoul and South Korea domain emissions at 10 cities is conducted. NO_x and toluene were more controlled by local emissions, while SIA (nitrate and ammonium) contributions varied greatly by phases and were more sensitive to regional (South Korea) emission controls. Sulfate decreased slightly in response to Korean emissions, varying by region and by phases. The Korean contribution to sulfate emissions sensitivity may be underestimated by our modeling system, as it lacked heterogenous sulfate production from SO₂ uptake on hydrated aerosols. Peak ozone concentrations in 10 major Korean cities (top 10% of hours during KORUS-AQ) decreased slightly and in some case increased, due to competing effects of ozone increasing in urban NO_x-saturated locations subjected to 50% NO_x reductions, combined with decreased regional production from VOC and NO_x reductions in NO_x-sensitive regions outside of cities.

Low altitude transport (0–2 km) had a critical influence on concentrations in the Korean boundary layer during the haze event of May 24 and 25, 2016. A time lag of approximately 12–24 h existed between the Yellow Sea peak detected by the DC-8 and impacts over Korean receptor sites.

Mass fluxes of primary and secondary pollutants over the Yellow Sea were significant, with both east-to-west and west-to-east contributions. Mass fluxes, relative to a number of normalization quantities (domestic emissions, production rates of secondary species, and masses of pollutants in the Korean boundary layer) peaked at the beginning of the third meteorological phase (transport) of KORUS-AQ. For OA, nitrate, and total ammonium, both domestic and LRT were important. Total sulfate was mainly from remote sources. For ozone, both LRT and local productions were important.

The analysis presented shows the importance of using models with high spatial resolution to capture pollutant transport and mixing around Korea. However, there remain uncertainties in secondary aerosol production mechanisms and indications that local production at times could be higher than those modeled in this analysis. Therefore, the results presented here should be viewed as an upper limit on the importance of LRT.

Future enhancements include investigation of ozone formation chemistry and sensitivity to emissions using the WRF-Chem integrated reaction rate method (Pfister et al., 2019). We also plan to investigate interplay between emission controls in both China and Korea, particularly for joint optimization of air quality health impacts of ozone and PM_2.5.

Observational data from KORea and United-States Air Quality used in this study can be downloaded through the data archive website (https://www-air.larc.nasa.gov/cgi-bin/ArcView/korusaq). Hourly surface observations from the AirKorea network can be downloaded through (https://www.airkorea.or.kr/eng/hourlyTrends?pMENU_NO=151). Observational surface meteorological data via Korea meteorological agency can be accessed through (https://www.weather.go.kr/w/index.do). WRF-Chem modeled data can be accessed through (https://www-air.larc.nasa.gov/cgi-bin/ArcView/korusaq?MODEL=1) under group “PARK.ROKJIN” with filename “korusaq_surface-map_MODEL_20160501_R0_thru20160610-WRFChemUCLA.nc” for surface predictions and “korusaq-MICP_MODEL_2016<DATE>_R0_DC8-meg60.ict” for aircraft data.

The supplemental files for this article can be found as follows:

Supplemental materials. PDF

The authors acknowledge the entire KORea and United-States Air Quality (KORUS-AQ) team for providing critical airborne, surface, and remotely sensed data needed to advance air quality management and science in the region. Surface meteorology data were from Korea meteorological agency (https://www.weather.go.kr/w/index.do). Surface ozone, carbon monoxide, nitrogen dioxide, sulfur dioxide, PM_2.5, and PM₁₀ were from AirKorea network system (https://www.airkorea.or.kr/eng/hourlyTrends?pMENU_NO=151). Surface PM_2.5 components data from NIER stations, DC-8 measurements, and Seoul Olympic Park KORUS-AQ data were from the campaign data archive (https://www-air.larc.nasa.gov/cgi-bin/ArcView/korusaq?).

This research has been supported by the NASA KORea and United-States Air Quality Study grant NNX15AU17G, NASA Health and Air Quality Applied Science Team grant NNX16AQ19G, and NASA ACMAP grant 80NSSC19K0946.

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this article.

Conceptualization: GRC, COS, BT.

Base model configuration and development: PES, MG.

Perturbed emission model runs: BT.

Formal analysis and visualization: BT.

Writing of original draft: BT, GRC, COS.

Review and editing: GRC, COS, PES, MG, BT.

How to cite this article: Tang, B, Saide, PE, Gao, M, Carmichael, GR, Stanier, CO. 2023. WRF-Chem quantification of transport events and emissions sensitivity in Korea during KORUS-AQ. Elementa: Science of the Anthropocene 11(1). DOI: https://doi.org/10.1525/elementa.2022.00096

Domain Editor-in-Chief: Detlev Helmig, Boulder AIR LLC, Boulder, CO, USA

Associate Editor: Alex Guenther, Department of Earth System Science, University of California Irvine, Irvine, CA, USA

Knowledge Domain: Atmospheric Science

Part of an Elementa Special Feature: Korea-United States Air Quality (KORUS-AQ)