Characterization, sources and reactivity of volatile organic compounds (VOCs) in Seoul and surrounding regions during KORUS-AQ

Air pollution is an environmental and human health concern throughout the world. As well as impacting respiratory and heart diseases, the International Agency for Research on Cancer (IARC) has classified outdoor air pollution as a carcinogen (IARC, 2013). Components of air pollution such as ozone (O₃) damage vegetation and are harmful to humans (Ainsworth et al., 2012). While emission controls on precursor gases such as volatile organic compounds (VOCs) and nitrogen oxides (NO_x) have led to decreases in O₃ levels and extreme O₃ events in North America and Europe, parts of Asia including South Korea have seen increasing long-term O₃ trends (Chang et al., 2017; Fleming et al., 2018). South Korea has an 8-hr O₃ standard of 60 ppbv, and the number of ozone exceedance days has increased from a low of 5 days in 2002 to 58 days in 2016 (Seo et al., 2018).

In an effort to improve air quality, the Korean Ministry of Environment (KMOE) has ongoing policies to tighten environmental standards and reduce the emissions of major air pollutants (KMOE, 2017; Shin et al., 2013). For example, multiple strategies have been used to target emissions from motor vehicles and gasoline evaporation in South Korea over the past two decades. Registered vehicles in South Korea are fueled primarily by gasoline (10.1 million vehicles), diesel (9.2 million) and liquefied petroleum gas (LPG; 2.2 million), with a smaller number fueled by compressed natural gas (CNG; ~39,000), electric (~11,000) and hybrid technology (~233,000) (MOLIT, 2016). Gasoline-fueled vehicles in South Korea have been subject to California’s Non Methane Organic Gases Fleet Average System since 2009, and diesel-fueled vehicles have been regulated under Euro 6 emission standards since 2014. South Korea began a program to replace diesel-powered public buses with buses fueled by compressed natural gas (CNG) in 2000, and all transit buses in Seoul were fueled by CNG by 2010 (ESCI, 2013). To further improve air quality, about half the Seoul bus fleet will switch from CNG to electric during 2018–2025 (Korea Herald, 2018). VOCs have additional sources in South Korea such as organic solvents and production processes, and emission control strategies include the establishment of nation-wide VOC content limits for paints in 2013 (KMOE, 2017). While these emission control measures have led to an overall reduction in VOC levels in Seoul over the past twenty years, understanding the contributions of VOCs to ozone and aerosol formation in the Seoul Metropolitan Area (SMA) remains a high research priority (Kim and Lee, 2018).

In order to better understand the evolving factors that contribute to air pollution in the Korean peninsula, South Korea’s National Institute of Environmental Research (NIER) and the United States National Aeronautics and Space Administration (NASA) led the Korea-United States Air Quality Study (KORUS-AQ), a cooperative field study that deployed in late spring 2016 (May 2 to June 10). The study included measurements of trace gases, aerosols and meteorological parameters from aircraft, ground-based stations and ships, together with modeling analysis of air quality and meteorology (Crawford et al., 2020). South Korea has a temperate climate, with four distinct seasons that often exhibit large variations in meteorology. The mean wind direction is southwesterly in summer and northwesterly in winter (KMA, 2009). Dynamic mid-latitude meteorology facilitates transport of Asian dust to Korea in March and April, and the Asian monsoon wind shift brings heavy rains and occasional tropical disturbances to Korea, beginning in late June. The late spring measurement period of the KORUS-AQ mission was chosen to minimize long distance pollution transport and instead target local photochemical pollution. However, periods of dynamic transport were still encountered during the mission (H. Kim et al., 2018a; Peterson et al., 2019), which allowed inflow from other regions to be characterized. Designated flights over the West Sea, upwind of Korea, enabled sampling of Chinese emissions in isolation from Korea, even when these air masses were not expected to arrive in Korea. KORUS-AQ therefore offers a comprehensive suite of measurement data and modeling products from local and regional sources, influenced by variable springtime meteorology.

As part of KORUS-AQ, the University of California, Irvine (UCI) measured 82 speciated VOCs aboard the NASA DC-8 research aircraft using discrete whole air sampling (WAS) followed by multi-column gas chromatography. The measured VOCs include reactive gases such as toluene that can readily form secondary pollutants such as O₃ and secondary organic aerosol (SOA), and compounds that can have direct impacts on human health such as benzene. Limited ground-based measurements were also collected in Seoul in spring 2015 (the year before KORUS-AQ) using the same technique (H. Kim et al., 2018; S. Kim et al., 2016, 2018). High resolution instruments provided OVOC measurements during KORUS-AQ, including methanol and formaldehyde. Here we characterize VOC concentrations over Seoul, the Korean peninsula, and surrounding waters, including plumes from the Daesan petrochemical facility and in air originating from China. Additional comprehensive analysis of the Daesan petrochemical plumes, including estimates of emission factors, will be presented in a companion paper by Fried et al. (2020). We also estimate the ozone formation potential (OFP) of the VOCs measured over Seoul and identify the major VOC sources that impact Seoul using source apportionment and other techniques. These VOC sources are compared with the KORUS-AQ emission inventory (EI). Finally we discuss strategies to reduce emissions of the most reactive VOCs in Seoul.

The KORUS-AQ mission included 20 local science flights aboard the NASA DC-8 research aircraft, which was based at the Osan Air Base about 50 km south of Seoul (Al-Saadi et al., 2015; Crawford et al., 2020). The flights occurred from May 2 to June 10, 2016 (local Korean time) and were typically about 8 hours long for a total of 154 flight hours.

The 6-week KORUS-AQ mission was influenced by four major synoptic weather patterns (Peterson et al., 2019; Table 1). The dates below are based on local Korean time (UTC + 9 hrs):

1. May 2–15: Dynamic period featuring the passage of five strong frontal boundaries through the study region, large vertical motion, rapid storm movement, and winds generally from the W to N.

2. May 16–23: Stagnant period under a persistent anticyclone that resulted in light wind speeds, warm surface temperatures, and minimal impact from long-distance transport, with winds from the ESE, when they were strong enough to measure.

3. May 24–31: Second transport period following a frontal passage on May 24, with a return to dynamic conditions but featuring weaker frontal boundaries, slower storm movement, and reduced vertical motion when compared with the first dynamic period. Persistent flow from the WSW supported transport of air from China in a moist boundary layer, ultimately influencing some of the worst pollution and visibility restrictions observed during campaign.

4. June 1–10: Rex blocking pattern that resulted in minimal pollution transport and the general dominance of local pollution sources, but featured larger variations in wind speed and direction when compared with the previous stagnant period. The flights on June 9 and 10 were not technically associated with blocking, since a weak front reached the study region on June 8, but there was not a major pattern change in its wake.

The KORUS-AQ study occurred during neutral El Niño/Southern Oscillation (ENSO) and near-neutral Arctic Oscillation (AO) conditions. The average meteorology during KORUS-AQ was therefore representative of a 30-year climatology for this region in late spring (Peterson et al., 2019).

Each of the 20 research flights had a different flight track that best accommodated the research objectives according to the meteorological conditions (Figure 1). During periods of transport the aircraft often flew a north-south flight track over the West Sea to sample outflow from China in isolation from Korean air (Figure 1b). Most flights included a Seoul-Jeju or Seoul-Busan jetway to capture air masses over the southern portion of the Korean peninsula (Figure 1c). With the exception of Flight 13, all flights also included a Seoul standard route or “stereo route” that was repeated over Seoul, with a missed approach over the Seoul Air Base in Seongnam (about 15 km southeast of Seoul) that allowed the aircraft to fly at low altitude over Seoul (100–1000 m), including Olympic Park in central Seoul (Figure 1d). All flights occurred during the daytime, and the Seoul stereo route was typically flown for 15–45 minutes near the beginning, middle and end of each flight (8 a.m., noon and 3 p.m. local time, respectively), allowing for comprehensive sampling of air over Seoul (Nault et al., 2018). The example in Figure 1d also shows downwind sampling of the Seoul plume and point source emissions under an easterly wind. Three flights also targeted industrial sites near Daesan about 30 km southwest of Seoul (Figure 1e).

A total of 2554 whole air samples were collected during the 20 research flights of KORUS-AQ, for an average of 128 samples per flight. Each air sample was collected during a 20 s to 3 min fill time into an electropolished, 2-L conditioned stainless steel canister equipped with a Swagelok Nupro metal bellows valve. The average fill time was 40 s and the longest fill times were at the highest altitudes (11 km). To collect an air sample an operator manually opened the valve, allowing a dual-head metal bellows pump to draw air from outside the aircraft into a ¼” forward-facing inlet, through an air sampling manifold and into the selected canister until it was filled to 40 psig (roughly 3 atm or 2000 Torr), at which time the valve was closed. Samples were collected on average every 3.5 mins. The average ± 1σ pressure altitude of the WAS samples was 2.0 ± 2.1 km, and the median altitude was 1.1 km. At the completion of each flight, the filled canisters were offloaded and couriered to UCI for laboratory analysis (Section 2.3).

The procedures to clean and prepare each canister for field use are described in Simpson et al. (2010). Briefly, the canisters were first conditioned by baking them in humidified air at 225°C for 12 hr in order to form an oxidative layer inside the canisters that passivates the canister walls. Next, after a series of pump-and-flush procedures in clean mountain air, the canisters were evacuated to 10^–2 Torr, filled to 1000 Torr with ultra-high purity helium, evacuated once again to 10^–2 Torr, then humidified by adding 17 Torr of purified water (the approximate vapour pressure of water at room temperature) in order to minimize surface adsorption.

The airborne KORUS-AQ samples were supported by limited ground-based whole air sampling in Seoul in spring 2015 (n = 23), a year before KORUS-AQ during the Megacity Air Pollution Study-Seoul (MAPS-Seoul) campaign (H. Kim et al., 2018b; S. Kim et al., 2016, 2018). The samples were collected into evacuated 2-L canisters (the same canisters as described above) at 09:30 and 15:00 local time on most days from May 23–39 and June 3–10 at the Korean Institute for Science and Technology (KIST; 37.60°N; 127.04°E), at a site located 250 m southeast of a freeway and 150 m northwest of a research forest. The KIST samples were filled to ambient pressure over a sampling period of about 3 min. The canister preparation procedures and analysis at UCI were the same as for KORUS-AQ. Although limited, the ground-based measurements indicate whether the VOC results are robust from year-to-year and whether the low-altitude airborne observations during KORUS-AQ are generally consistent with ground-based measurements. Ground-based VOC measurements were also made by another research team during KORUS-AQ and will be reported separately.

The analysis in this paper also uses selected measurements from other KORUS-AQ teams using fast-response, real-time instruments (Tables S-1, S-2). All fast-response measurements were averaged over the longer fill time of the UCI WAS samples, called the WAS data merge, and these averages are used here. The calibration procedures for selected fast instruments are described in Section 1 of the Supplemental Material (SM).

1. Several VOCs and OVOCs were measured using a proton-transfer-reaction time-of-flight mass spectrometer (PTR-ToF-MS) at 1 Hz frequency (Müller et al., 2014). While some PTR-ToF-MS OVOCs were reported as a combined measurement (acetone + propanal, MVK + MACR + ISOPOOH), others were reported separately and are presented here (methanol, acetaldehyde, methyl ethyl ketone or MEK; Table S-2). However, while MEK is free of interferences in remote regions, other species such as butanal could contribute to the signal in the Seoul region. The measurement precision depends on the signal integration time and the concentration. The precision at 3 ppbv is estimated to be 0.418 ppbv for methanol, 0.122 ppbv for acetaldehyde, and 0.087 ppbv for MEK. Toluene was measured by PTR-ToF-MS without interference, with a precision of 0.091 pptv at 3 ppbv. The reduced electric field strength (E/N, with E being the electric field strength and N being the gas density) in the drift tube was ~110 Td (1 Td = 10^–17 V cm²). Section 2 of the SM compares WAS and PTR-ToF-MS toluene measurements, which shows that data obtained using different techniques and independent calibration methods agree well (Figures S-1, S-2).

2. Carbon monoxide and methane (CH₄) were measured using the NASA Langley Differential Absorption Carbon monoxide Measurement (DACOM), an in situ diode laser spectrometer system (Warner et al., 2010). The DACOM time response is 1 s and the measurement precision is 0.1% for CH₄ and 1% for CO.

3. Formaldehyde (CH₂O) was measured using a Compact Atmospheric Multispecies Spectrometer (CAMS) that uses mid-infrared laser light to detect CH₂O with an overall uncertainty of 4–6% (Richter et al., 2015; Spinei et al., 2018). This instrument also acquired fast ethane measurements with a similar uncertainty range. The WAS and CAMS measurements of ethane are compared in Figure S-2.

4. Nitric oxide (NO), nitrogen dioxide (NO₂), total reactive nitrogen (NO_y) and O₃ were measured using the NCAR 4-Channel NO_xyO₃ Chemiluminescence Instrument, which uses O₃ as a reagent to measure the nitrogen compounds, and NO as a reagent to measure O₃ (Weinheimer et al., 1994). The precision is 1–5% for nitrogen compounds and 1% for O₃.

5. Hydrogen cyanide (HCN) was measured using the Caltech Chemical Ionization Mass Spectrometry (CIMS) instrument, which measures the reaction product ion CF₃O^–·HCN at m/z = 112 (Crounse et al., 2009). HCN is measured at 1 Hz with a precision of 20 pptv and an accuracy of 30%.

During KORUS-AQ the filled WAS canisters were analyzed within 7 days of collection. Each reported VOC has been tested to ensure that its growth or destruction in the canisters is negligible during the interval between sample collection and analysis (Simpson et al., 2010) and the VOCs we report were stable in the canisters during transit. The canisters were analyzed at UCI using multi-column gas chromatography (GC) for 82 C₁–C₁₀ VOCs, namely 14 alkanes (C₂–C₁₀), 4 cycloalkanes (C₄–C₇), 10 alkenes (C₂–C₁₀), ethyne, 14 aromatics (C₆–C₉), 29 halocarbons (C₁–C₂), 8 alkyl nitrates (C₁–C₅), carbonyl sulfide (COS) and dimethyl sulfide (DMS). The limit of detection (LOD), measurement precision, accuracy and atmospheric lifetime of each VOC is given in Table S-2. The measurement precision is determined from replicate analyses of pontoons of air, and the values in Table S-2 are conservative. For example, the precision of toluene is 3% at lower values (~100 pptv) and 1% at higher values (~15,000 pptv), and so a conservative precision of 3% is given in Table S-2.

The detailed GC set-up and analytical procedures for the WAS measurements are described in Colman et al. (2001) and Simpson et al. (2010). During KORUS-AQ three Hewlett-Packard GCs (HP-6890) and five column/detector combinations were used with flame ionization detection (FID), electron capture detection (ECD) and a mass selective detector (MSD) operating in selected ion monitoring (SIM) mode. The five column/detector combinations were DB-1 + FID, PLOT/DB-1 + FID, Restek-1701 + ECD, Restek-1701/DB-5 + ECD, and DB-5 ms + MSD. Most compounds eluted with highest precision on one column/detector combination. When a high-precision peak eluted on two column/detector combinations the results were averaged (e.g., propane, propene, benzene, toluene, ethylbenzene, xylenes). Most hydrocarbons were measured with FID, except the pinenes, branched alkanes, and aromatics ≥C8, which were measured with MSD. Alkyl nitrates and most halocarbons were measured using ECD, except HFCs and HCFCs which were measured using MSD. COS and DMS were also measured using MSD. To analyze a sample, the canister was connected to the analytical system and 2410 cm³ of sample air was passed through molecular sieve traps to remove contaminants. Target VOCs within the sample air were then cryogenically pre-concentrated as the sample air flowed over glass beads immersed in a liquid nitrogen cooled loop. The VOCs were volatilized when the loop was immersed in hot water, and then the VOCs were flushed into a helium carrier flow to a splitter box that separated the sample flow into five streams, each directed to one of the column/detectors. The detector signal was digitally recorded and each chromatographic peak was individually inspected and manually integrated in order to increase the precision.

The UCI group has been making VOC measurements since the late 1970s and early 1980s, and our calibration standards and procedures date back to this time (Colman et al., 2001; Simpson et al., 2010). For hydrocarbons we use a 0.99 ppm propane standard purchased in 1978 from the National Bureau of Standards (NBS, now the National Institute for Science and Technology or NIST) to calculate our per carbon response factor (PCRF), which converts detector response in area units to mixing ratios. This PCRF is compared to those calculated from less expensive commercial NIST and Scott-Marrin standards, and then the PCRFs from propane to decane are assigned based on the less expensive standards. The UCI halocarbon standards are either NIST-traceable or were made in-house from static dilutions prepared at UCI. The hydrocarbon and halocarbon measurements agree well with those from other laboratories in both formal and informal comparison experiments (Apel et al., 2003; Hall et al., 2014; Hornbrook et al., 2011). The calibration procedures for the WAS measurements are further described in Section S1 and Table S-1 of the SM.

Major VOC sources in Seoul were investigated using Positive Matrix Factorization (PMF), a widely used multivariate receptor modeling technique (Brown et al., 2015; Hopke, 2016; Norris et al., 2014; Paatero et al., 2014). Here we used the U.S. Environmental Protection Agency (EPA) PMF Version 5.0 (Norris et al., 2014). The PMF program decomposes an m × n matrix (x_ij) of sample data—in this case the mixing ratios of n = 21 VOCs measured in m = 177 samples collected at low altitude over Seoul (Section 3.3.2)—into two matrices called factor contributions (G) and factor profiles (F). The factor contributions express the mass or percent contribution of each factor to the measured concentration of each VOC. The factor profiles are species profiles and need to be interpreted by the user based on their existing knowledge of VOC source profiles, in our case using experience together with relevant VOC profiles from the literature (e.g., Du et al., 2016; Guo et al., 2011a, b; Kwon et al., 2007; McDonald et al., 2018; Na et al., 2004; Ou et al., 2015; Tang et al., 2014; Wang et al., 2014). The chemical mass balance equation is expressed as (Norris et al., 2014):

where g is the mass contributed by each factor, f is the species profile of each factor, e_ij is the residual for each VOC j in each sample i, and p is the number of factors. The user chooses the number of factors that the PMF program will output. Here we ran PMF solutions for k = 3–6 factors and the results are discussed in Section 3.3.2.

The PMF factor contributions and factor profiles are derived by minimizing the “goodness of fit” or objective function Q, which is a critical parameter for PMF (Bon et al., 2011; Brown et al., 2015; Norris et al., 2014):

where u_ij is the estimated uncertainty of each VOC in the sample matrix. Here the uncertainty matrix was constructed using the precision of the VOC concentration within each sample (Table S-2), and an extra modeling uncertainty of 20% was added when the model was run. For example, a toluene value of 1531 pptv at a precision of 3% was assigned an uncertainty of 1531 × 0.03 = 46 pptv in the uncertainty matrix. Because the precision values are conservative (Section 2.3), the uncertainty is overestimated for some mixing ratios. Of the 21 × 177 = 3717 data points in the sample matrix, only three (two n-heptane samples and one isoprene sample) were below their limit of detection (LOD). They were assigned a value of 5/6 the LOD (so 2.5 pptv) and their uncertainty was assigned as 50% (so 1.5 pptv) (Norris et al., 2014). PMF calculates a value called Q(true) using all the data points in the sample matrix, and a value called Q(robust) by excluding data points with uncertainty-scaled residuals (e_ij/u_ij) > 4 (Norris et al., 2014). Here Q(true) and Q(robust) had the same value because the scaled residuals were all <4. This means that even though PMF was operated in its “robust mode”, which automatically downweights the impact of samples with scaled residuals >4, this was not an issue here.

The PMF model offers several ways to understand the viability of the PMF solution and the correct number of factors to choose. This includes classic bootstrap (BS), displacement of factor elements (DISP), and bootstrap enhanced by displacement (BS-DISP), which are complementary ways to estimate the random errors and rotational ambiguity of the factor solutions (Brown et al., 2015). PMF also calculates a value called Q/Q(expected), or Q/Qexp, which is >2 when a species is not well modeled by the PMF solution (Norris et al., 2014). These PMF error estimation methods are applied in Section 3.3.2. Previous field studies have further discussed the uncertainties and limitations of PMF (e.g., Bon et al., 2011; Gao et al., 2018; Yuan et al., 2012). For example, Bon et al. (2011) used PMF to identify three dominant factors in a study of VOC sources in Mexico City, and cautioned that more PMF factors caused major factors to be split rather than identifying separate minor emission sources. The uncertainties and limitations of the PMF analysis are discussed in Sections 3.3.2 and 3.3.3.

The VOC sources inferred for Seoul using PMF were compared with the KORUS-AQ version 5 (KORUSv5) emission inventory for 2015, which was developed at Konkuk University and is a modified version of the official Korean EI called Clean Air Policy Support System (CAPSS) for South Korea (Lee et al., 2011). The KORUSv5 EI geographically covers all of Asia and uses the Sparse Matrix Operator Kernel Emissions (SMOKE-Asia) emissions processing (Woo et al., 2012) and SAPRC-99 chemical mechanism (Carter et al., 2000). The inventory reports CO, NH₃, NO_x, PM_2.5, PM₁₀, SO₂ and VOCs on a 3 km grid resolution for South Korea, and does not include biogenic VOC sources. The VOCs are based on 30 VOC groups including 12 VOC groups measured in common with WAS, namely 5 alkane/alkyne groups, 5 alkene groups and 2 aromatic groups that are based on reactivity. The VOCs measured in common with WAS account for 98% of the KORUSv5 VOC emissions, with OVOCs comprising the remaining 2%.

Back trajectories along selected flight tracks were calculated using the Lagrangian FLEXible PARTicle dispersion model (FLEXPART) version 9.0.3 (Stohl et al., 2005; https://www.flexpart.eu). 25,000 particles were released per trajectory calculation to estimate air mass history and uncertainty. The 48-hour backward simulations were driven by 3 hourly wind fields from the National Centers for Environmental Prediction (NCEP) Global Forecast System (GFS) model on a 0.25° × 0.25° grid.

Global model simulations of tropospheric chemistry, as well as CO-like tracers from various regions for the KORUS-AQ period were performed using the Community Atmosphere Model with chemistry (CAM-chem; Tang et al., 2019). CAM-chem is a configuration of the global Community Earth System Model (CESM) with detailed tropospheric and stratospheric chemistry (Tilmes et al., 2015). The results shown here are from a simulation at 0.5° horizontal resolution and are nudged to meteorological reanalysis fields from the Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2).

The air samples collected during KORUS-AQ showed distinct VOC distributions and signatures depending on their source origins. The spatial distributions of three VOC tracers are shown in Figure 2 as an example. Isoprene, a short-lived biogenic tracer (Guenther et al., 2012) was elevated over less populated areas of Korea along the Seoul-Jeju and Seoul-Busan flight tracks (Figure 2a). Toluene, an urban tracer (Guo et al., 2006), was primarily elevated over the SMA in northwest Korea, with mixing ratios up to 10.4 ppbv during the missed approach over Seoul (Figure 2b; Table 2). By contrast, COS, which has sources including domestic coal combustion and oxidation of CS₂ emitted from rayon production (Campbell et al., 2015; Du et al., 2016) was strongly enhanced in air sampled along the West Sea flight track, with mixing ratios up to 1.5 ppbv or more than double background values (Figure 2c; Table 2). Based on FLEXPART back trajectories and CAM-chem model simulations, the elevated COS levels measured over the West Sea originated from northeast China, including Shandong Peninsula and Hebei province (Figure 3). Overall, these results show that isoprene is a biogenic tracer for KORUS-AQ, toluene is one of the tracers of air from Seoul, and COS is a tracer of air masses originating from China. Enlarged spatial distributions for central areas of interest are shown in Figure S-3, including ethene which was strongly elevated at the Daesan petrochemical facility (Section 3.1.3).

In addition to surface spatial distributions, the VOC enhancements were inspected using altitude-concentration plots that were colour-coded by longitude to make the different source regions apparent, namely Seoul (yellow), Daesan (green), the Busan region (red), and air from China sampled over the West Sea (blue) (Figure 4). The latitude ranges of these different source regions are provided below and shown in Figure 2d. Figure 4 shows where each VOC showed it maximum enhancement and how the mixing ratios compare from region to region. For example, even though ethene was clearly elevated in Seoul (yellow), its highest levels during KORUS-AQ were from the Daesan petrochemical facility (green) which produces ethene (Figures 4g, S3d). Additional time series plots show whether the elevated VOC concentrations occurred during stagnant and/or dynamic periods (Figure 5). Based on this analysis each source region showed specific and distinct VOC signatures, as discussed below (Sections 3.1.2–3.1.4).

Like toluene, many VOCs showed their strongest enhancements of the KORUS-AQ mission during the missed approaches over Seoul. These gases comprised mostly of C₇–C₉ aromatics and non-brominated halocarbons (Figure 4a–d) and their mixing ratios were typically highest during the stagnant periods (Figure 5a–e). The most abundant VOCs measured over Seoul by WAS were, in descending order by average mixing ratio, ethane, propane, toluene, n-butane and ethyne (Figure 6b). (The selection of the Seoul plumes is described in Section 3.2.1.) Similar results were found for the ground-based WAS measurements measured at KIST in 2015 (Figure 6a). When OVOCs are included, methanol was the most abundant VOC in Seoul, and the predominantly oxidation products formaldehyde and acetaldehyde were also in the top 10 species (Table 2). The ground-based measurements in Seoul in 2015 also found abundant methanol as well as acetone (H. Kim et al., 2018). The Seoul VOC results are further discussed in Sections 3.2 and 3.3 in terms of ozone formation potential and VOC sources, respectively.

The Seoul results are also notable in terms of the behaviour of some halocarbons, in particular CFC-11 which is not declining in the atmosphere as quickly as expected based on the Montreal Protocol, with observational evidence for ongoing emissions from eastern Asia (Montzka et al., 2018; Rigby et al., 2019). During KORUS-AQ, while air originating from China was clearly elevated in CFC-11, stronger CFC-11 enhancements were measured over Seoul including during the stagnant periods (Figures 4d, 5e). This is different from other long-lived halocarbons such as CFC-113, CFC-114 and CCl₄, which were more elevated in air arriving from China and were not elevated over Seoul during stagnant periods (Figure 5g–h). Correlations between CFC-11 and China tracers such as CCl₄ and COS also show relatively elevated CFC-11 levels over Seoul compared to China (CCl₄ shown in Figure 7a). The CFC-11 enhancements over Seoul will be fully presented in a separate paper.

Whole air samples collected over the Daesan petrochemical complex showed a distinct chemical signature compared to VOCs collected over Seoul or in air arriving from China (Figure 4e–h). Of the 82 VOCs measured by WAS, 33 (or 40%) had their maximum values of the KORUS-AQ mission over the Daesan complex, including all alkanes, cycloalkanes and non-biogenic alkenes. Four of the 33 VOCs are either known carcinogens (benzene, 1,3-butadiene) or possible carcinogens (styrene, CHCl₃). The Daesan samples were selected for further analysis based on a geographic box that encloses 36.8–37.2°N, 125.9–126.5°W, and altitudes less than 500 m (n = 63). Based on this selection, the WAS Daesan samples were richest in ethene, C₂–C₆ alkanes and benzene (Figure 6c). Several OVOCs were also elevated in the Daesan plumes, including the highest levels of formaldehyde and acetaldehyde observed during the KORUS-AQ mission (Table 2). As noted in Section 1, the Daesan results will be fully presented in a companion paper by Fried et al. (2020).

Because emission signatures from China evolve over time, each NASA airborne mission needs a new characterization of the current fingerprint of air from China. For example, during the 2001 Transport and Chemical Evolution over the Pacific (TRACE-P) mission, the fire suppressant Halon 1211 (CF₂ClBr) and the biomass burning tracer methyl chloride (CH₃Cl) were characteristic tracers of air from China (Blake et al., 2003), and COS was also elevated in air from China (Blake et al., 2004). A larger suite of gases (COS, CH₃Cl, 1,2-dichloroethane, ethyl chloride, Halon 1211) was recommended as tracers of air from China during the 2006 Intercontinental Chemical Transport Experiment, Phase B (INTEX-B; Barletta et al., 2009). During KORUS-AQ, COS, CFC-113, CFC-114, CCl₄ and 1,2-dichloroethane all had strongly elevated levels in air originating from China (Figures 4i–l, 5f–i), and we recommend these five compounds as tracers for China during KORUS-AQ. Notably absent from this list is Halon 1211, which used to be produced primarily in China until its production was phased out in 2006 (Fraser et al., 1999; Vollmer et al., 2009). During KORUS-AQ Halon 1211 showed minimal enhancements over background values in air originating from China, and instead was most enhanced over Seoul (Figure 7b). Therefore, unlike previous missions such as TRACE-P and INTEX-B, Halon 1211 is no longer a useful tracer of air from China. While CH₃Cl was elevated in air from China during KORUS-AQ (up to 2.2 ppbv) and is still a tracer of air from China, it was much more elevated in the industrial Daesan plumes (up to 6.2 ppbv) (Figure 5j; Table 2).

The air samples originating from China and sampled at low altitude along the West Sea jetway were selected for further analysis based on a geographical box enclosing 30.2–37.4°N, 124.2–124.4°W and altitudes <500 m. The data were further selected to only include flights during dynamic periods, in other words flights 3, 7, 8, 13, 15 and 16 (Table 1), for a total of n = 68 samples. Of the 82 VOCs measured by WAS, these air samples were richest in ethane, propane, ethyne, CH₃Cl and COS (Figure 6d). The air from China was also rich in primary and secondary OVOCs including methanol, formaldehyde and acetaldehyde (Table 2).

A few halocarbons, notably CCl₄, H-2402 and HCFC-141b showed strong enhancements in southeast Korea near Busan (CCl₄ is shown in Figures 4k, 5h). The CCl₄ results have been presented in Lunt et al. (2018) and the remaining halocarbons will be addressed in a separate paper.

VOCs react with nitrogen oxides (NO_x) in the presence of sunlight to form secondary products including O₃ and SOA. The potential of individual VOCs to form O₃ can be roughly characterized using reactivity scales such as OH reactivity (k_OH) and Maximum Incremental Reactivity (MIR). Because photochemically active VOCs are more efficient at forming O₃, policies based on ozone formation potential (OFP) are helpful for guiding VOC control measures, rather than measures based on VOC abundance (Wu and Xie, 2017). Below we investigate which VOCs are most effective at forming O₃ in Seoul.

The Seoul data used in this analysis were selected based on a geographical box around Seoul (37.3–37.7°N and 126.7–127.3°E) and low altitudes (<0.5 km), which yielded 177 samples that covered all four synoptic-scale meteorological conditions during the mission (Figure S-4). Importantly, this geographical boundary excluded the Daesan petrochemical facility since its VOC signature is distinct from that of Seoul (Section 3.1.3) and winds in Seoul were rarely from Daesan during KORUS-AQ (Section 2.1.1). Altitudes below 500 m were chosen to minimize the effects of transboundary sources on VOCs in Seoul (Lamb et al., 2018; Section 4 of the SM). The average altitude of the Seoul data was 245 ± 126 m. Average atmospheric boundary layer heights in the Seoul region during spring, 2016 were 548 ± 180 m at night and 1182 ± 540 m during the day (H.-J. Lee et al., 2019). The earliest Seoul sample was collected at 7:40 a.m. Local Standard Time (LST), about 2 hours after sunrise, and the latest at 4:10 p.m. Therefore, while it is possible that a few higher-altitude samples closer to 500 m could have some impact from a more aged residual layer, especially early in the morning, most samples are expected to represent a well-mixed boundary layer.

Based on this data selection, OVOCs, alkanes and aromatics were most abundant in Seoul (Section 3.1.2). Excluding OVOCs, the contributions to VOC abundance (based on average mixing ratios for the 177 Seoul samples), were alkanes (46%), halocarbons (20%), aromatics (19%), alkenes (6%), alkynes (5%), sulfur compounds (3%) and alkyl nitrates (0.6%) (Figure S-5a). This is similar to previous work in South Korea which found that alkanes followed by aromatics were the largest contributors to VOC abundance in Seoul (halocarbons and OVOCs were not included; Shin et al., 2013). When primary OVOCs without interference are included in the KORUS-AQ Seoul data (i.e., methanol), the weightings shift to methanol (51%), alkanes (23%), halocarbons (10%), aromatics (9%), alkenes (3%), alkynes (2.5%), sulfur compounds (2%) and alkyl nitrates (0.3%) (Figure S-5b). By comparison, the percentages for May data during the MAPS-Seoul campaign were OVOCs (57%), alkanes (29%), aromatics (9%) and alkenes (4%) (H. Kim et al., 2018; halocarbons were not reported). That is, the overall percentages were similar during KORUS-AQ and MAPS-Seoul, though we note that the MAPS-Seoul data included speciated OVOCs besides methanol such as acetone.

The ‘total’ OH reactivity (k_OH, in s^–1) is a rough indicator of the potential of trace gases that react with OH to form O₃. In the case of VOCs, it is the product of VOC concentration ([VOC_i], in molec cm^–3) and the rate coefficient for VOC reaction with OH (k_OH+VOCi, in cm³ molec^–1 s^–1) (Kovacs et al., 2003; Yang et al., 2016):

Rate constants were taken from Atkinson and Arey (2003) and Atkinson et al. (2004, 2006, 2008), and average VOC mixing ratios over Seoul were used (Table S-2). Even though formaldehyde and acetaldehyde were present at appreciable mixing ratios in Seoul (Table S-2), they were not included in the O₃ analysis below because they were found to have a dominant photochemical source during KORUS-AQ (Schroeder et al., 2020).

VOCs that are both abundant and reactive with OH have a high potential to form O₃. For example during KORUS-AQ isoprene was a strong contributor to k_OH in Seoul because it is reactive rather than abundant, whereas toluene was a strong contributor because it is abundant rather than reactive (Figure S-6). We note that m,p-xylene was a combined WAS measurement during KORUS-AQ, even though m-xylene and p-xylene have different reaction rate constants with OH and therefore different lifetimes (Table S-2), which is important for their individual O₃ formation potential and for policy considerations. However the m- and p-xylene isomers were speciated by WAS during MAPS-Seoul in 2015, and this information was used in the KORUS-AQ analysis. The average and median ratios of m:p-xylene during MAPS-Seoul were 471:318 and 344:231 pptv/pptv, respectively, or a ratio of about 3:2. This 3:2 ratio was used to speciate the KORUS-AQ m,p-xylene measurements for purpose of the OH reactivity calculations, in order to apply their different OH reaction rate constants. Because m-xylene was both more abundant and more reactive than p-xylene, it was a much stronger (~3×) contributor to OH reactivity in Seoul during KORUS-AQ (Figure S-6).

Of the VOCs measured by WAS, the strongest individual contributors to OH reactivity in Seoul were isoprene, toluene, m-xylene, ethene and propene (Figure 8a). Consistent with this, isoprene contributed most to daytime OH reactivity during MAPS-Seoul based on the ground-based WAS samples in spring 2015 (S. Kim et al., 2018). We note that the contribution of isoprene to k_OH in Seoul is expected to be lower in winter as compared to the spring timing of KORUS-AQ and MAPS-Seoul. Considering anthropogenic VOCs, the KORUS-AQ results are similar to Wu and Xie (2017), who found that the xylenes, ethene, toluene and propene had the largest potential to form O₃ in the major urbanized regions of China. Previous year-round VOC measurements in Seoul from 2004–2008 found that toluene, ethene, n-butane and propane were the main contributors to O₃ formation (Shin et al., 2013). Although primary OVOCs such as methanol are not always measured in urban studies, the OH reactivity of methanol was higher than isoprene during KORUS-AQ (Figure 8a).

As a group, aromatics contributed most to VOC OH reactivity in Seoul during KORUS-AQ (44%), followed by non-isoprene alkenes (19%), isoprene (17%), heavy alkanes ≥C₄ (16%) and light C₂–C₃ alkanes (5%) (Figure 8b). These results show generally good agreement with 0-D photochemical box modeling estimates of O₃ production by VOCs over Seoul during KORUS-AQ, which were constrained by DC-8 observations (Schroeder et al., 2020). The box model used zero-out simulations, whereby selected VOCs or VOC groups were removed from the model to test the model sensitivity to these VOCs. The model found that O₃ production in Seoul is VOC-limited and most sensitive to changes in aromatics, followed by isoprene and anthropogenic alkenes. For example, removing aromatics caused a 32% reduction in mean ozone production in the Seoul region, and removing isoprene and non-isoprene alkenes caused 15% and 14% reductions, respectively (Schroeder et al., 2020). When expressed in pie chart format, to directly compare with the OH reactivity calculations, the model was most sensitive to aromatics (49%) followed by isoprene (23%), other alkenes (22%), heavy alkanes (5%) and light alkanes (2%) (Figure 8c). That is, the OH reactivity calculations and box modeling agreed in their major findings, namely the predominant impact of aromatics followed by alkenes in ozone formation in Seoul, with poorer agreement for heavy alkanes (16% and 5%, respectively). While the photochemical box modeling is more complex in its handling of radical chemistry, the OH reactivity calculations are able to give speciated VOC results (for example, the box model does not speciate the aromatics). The overall good agreement between the two suggests that results from the OH reactivity calculations can be reasonably used to identify which anthropogenic VOCs are most likely to form O₃ in Seoul and could be targeted for VOC emission reduction strategies (i.e., toluene, m-xylene, ethene).

The box modeling also assessed the impact of methanol on O₃ formation (Schroeder et al., 2020). Removing methanol caused a 6% reduction in O₃ production in the zero-out calculations, or 8% when expressed as a pie chart together with the VOC groups (Figure S-7b). By comparison, methanol contributed 20% to OH reactivity based on VOCs + methanol (Figure S-7a). This suggests that the OH reactivity calculation over-weighted the contribution of methanol to ozone formation, most likely because it does not take into account the subsequent radical chemistry in which methanol oxidation produces relatively less O₃ than a higher-chain molecule such as toluene.

VOC source influences can be diagnosed using a combination of techniques, both qualitative (e.g., presence of known VOC tracers, the ratio of one VOC to another, VOC correlations) and quantitative (e.g., source apportionment). We applied all these techniques to the KORUS-AQ WAS samples collected over Seoul, as discussed below.

The suite of VOCs measured during KORUS-AQ includes known tracers of both biogenic and anthropogenic sources that are relevant for Seoul.

Biogenic: Isoprene is a well-known biogenic tracer that also has a minor traffic source (Borbon et al., 2001; Guenther et al., 2012; Khan et al., 2018). Here isoprene did not correlate with traffic tracers such as CO, ethene and 1,3-butadiene in the urban data, either in low-altitude Seoul data (r² < 0.03) or in the ground-based data at KIST (r² < 0.02), suggesting negligible impact of traffic on the isoprene emissions in Seoul. Other biogenic tracers such as α- and β-pinene were minor during KORUS-AQ, with average mixing ratios <15 pptv (Table S-2). Methanol also has a major biogenic source (Jacob et al., 2005; Tie et al., 2003), although in the urban Seoul data methanol correlated much better with solvents such as ethylbenzene (r² = 0.78) than with isoprene (r² = 0.14), suggesting multiple source influences.

Natural gas: Ethane is a useful tracer of fossil fuel sources including natural gas (Kille et al., 2019; Smith et al., 2015). Natural gas is used for heating and cooking in South Korea, with five times more consumption in winter than in summer (Na and Kim, 2001; Na et al., 2004). Residential natural gas is mostly methane (89% by mass), and its non-methane components are primarily ethane (71%), propane (22%) and the butanes (5%) (Na et al., 2004). Natural gas is also used in South Korea’s public transport sector, and all transit buses in Seoul have been fueled by CNG since 2010 (Section 1). During KORUS-AQ ethane and propane showed some correlation in Seoul (r² = 0.57) and the propane/ethane ratio was 0.99 ± 0.03 (Figure 7c). This 1:1 ratio is greater than the 1:3 ratio that would be expected if natural gas were the main source of ethane and propane, indicating other source impacts (Section 3.3.2). Likewise, while 1,3-butadiene has been reported in the exhaust of CNG-fueled buses (Kado et al., 2005), 1,3-butadiene levels were low in both the airborne and ground-based Seoul samples (9 ± 19 and 24 ± 16 pptv, respectively).

Liquefied petroleum gas (LPG): Propane and the butanes are the major components of LPG. In South Korea propane-rich LPG (>90%) is used for residential and commercial cooking and heating, while butane-rich LPG mixed with 5–30% propane is used for autogas in taxis and other LPG-fueled vehicles (KLPG, 2004). Less propane is added to autogas in summer than winter, and the springtime autogas mix in Seoul would have about 95% butanes and 5% propane + other minor components (Na et al., 2004). Industrial LPG in South Korea is rich in i-butane (>85%; KLPG, 2004). During KORUS-AQ the i-butane/n-butane ratio in Seoul was 0.53 ± 0.01 and the butanes showed excellent correlation (r² = 0.93; Figure 7d). Similar results were observed in the ground-based data at KIST (slope = 0.47 ± 0.02; r² = 0.98). An i-butane/n-butane ratio of around 0.5 is consistent with an LPG influence, whereas a lower ratio of 0.2–0.3 indicates biomass burning and a higher ratio of 0.6–1.0 can indicate natural gas (Russo et al., 2010) or, in South Korea, use of industrial LPG. For example the i-butane/n-butane ratio in the industrial Daesan samples (1.08 ± 0.08; r² = 0.77) was richer in i-butane than the Seoul samples (Figure 7d). Overall, the butane ratios in Seoul point to impact from LPG rather than biomass burning, natural gas or industry.

Gasoline evaporation: Toluene, pentanes and methylpentanes are dominant species in gasoline evaporation (Dai et al., 2013; Wang et al., 2016; Yue et al., 2017). Similar to the butanes, the pentanes are useful for distinguishing various combustion and fossil fuel sources, including gasoline evaporation. A i-pentane/n-pentane ratio around 0.5 indicates biomass burning (Akagi et al., 2011), 0.8–1.0 represents oil and natural gas (Abeleira et al., 2017; Gilman et al., 2013), 1–3 suggests traffic (Baker et al., 2008; Fraser et al., 1998; Na et al., 2006), and 3–4 indicates gasoline evaporation (McGaughey et al., 2004; Na et al., 2004; Simpson et al., 2014). The i-pentane/n-pentane ratio in Seoul was 1.25 ± 0.03 (r² = 0.88), with similar results for the ground-based samples (1.34 ± 0.09; r² = 0.88), suggesting that gasoline evaporation is not a major influence in Seoul. By comparison, the i-pentane/n-pentane ratio in Mecca, Saudi Arabia, a city heavily impacted by gasoline evaporation, was 3.52 ± 0.02 (Simpson et al., 2014). The pentane ratio in Seoul (1.25–1.34) is lower than traffic ratios in North America (2–3; Baker et al., 2008; Fraser et al., 1998) but similar to results from a Seoul tunnel study in 2000 (1.22–1.25; Na et al., 2006).

Vehicle exhaust: Many VOCs are emitted in vehicle exhaust, with different emission profiles depending on fuel type. For example ethene, propene, aromatics and heavier alkanes are prominent in diesel exhaust (Shen et al., 2018; Tsai et al., 2012; Wang et al., 2017; Yao et al., 2015); light alkanes, light alkenes, toluene and benzene in gasoline exhaust (Cao et al., 2016; Guo et al., 2011a; Wang et al., 2017); and propane and butanes in LPG exhaust (Guo et al., 2011a). Among diesel-fueled vehicles, the proportion of alkenes is highest for light-duty diesel trucks followed by heavy-duty diesel trucks, with little alkene emission from light-duty diesel cars (Wang et al., 2017). The ratio of ethene/ethyne helps to distinguish vehicle exhaust from other sources. The ethene/ethyne ratio is 1–3 for vehicle exhaust, with values <1 for older vehicles due to higher ethyne emissions, and >10 for petrochemical sources (Pang et al., 2014; Ryerson et al., 2003; Simpson et al., 2014). During KORUS-AQ the ethene/ethyne ratio was 1.02 ± 0.07 in Seoul (and likewise 1.26 ± 0.13 for the ground-based Seoul samples) compared to 11.3 ± 1.9 in Daesan (Figure 7e), clearly showing their respective vehicular and industrial influences.

Solvents: Many cities in Asia are heavily impacted by solvents such as toluene (e.g., Cai et al., 2010; Guo et al., 2011b; Song et al., 2018; Li et al., 2019). Previous work in Seoul has also found an abundance of toluene (Na et al., 2005; Nguyen et al., 2009). The ratio of toluene to benzene (T/B) is a useful indicator of the relative impact of traffic versus solvents: a T/B ratio of 1–2 indicates traffic while >2 indicates solvent influence (Barletta et al., 2005, 2017; Wang et al., 2016). For example, based on measurements in the Pearl River Delta region of China, Liu et al. (2008) reported a T/B ratio of 1.5 based on tunnel measurements, 2–5 in Guangzhou, and 6–15 in the manufacturing city of Dongguan. During KORUS-AQ the average T/B ratio over Seoul (calculated as the average ± 1σ of individual T/B values) was 7.6 ± 4.9 (median of 6.3), with ratios of 10–19 for toluene levels greater than 6 ppbv, suggesting a prominent influence of solvents on toluene (Figure 7f). By contrast the T/B ratio was 1.22 ± 0.05 in Lahore, Pakistan, a city heavily impacted by traffic emissions (Barletta et al., 2017), and 1.05–1.27 in suburban areas of Beijing, also suggesting vehicle emissions (Wang et al., 2012, 2016). While toluene is associated with both paint solvents and non-paint solvents (e.g., consumer products, printing), ethylbenzene and the xylenes are primarily associated with paint solvents (Kwon et al., 2007; Ou et al., 2015; Wang et al., 2014). Consistent with this, ethylbenzene and the xylenes formed a tight correlating group in Seoul (r² = 0.85–0.99), with poorer correlation with toluene (r² = 0.53–0.75) and even poorer correlation with the vehicle exhaust tracers ethene and propene (r² = 0.23–0.42). Likewise, toluene showed poor correlation with ethene and propene (r² = 0.20–0.36). This suggests that paint solvents were a primary influence on ethylbenzene and the xylenes, and that toluene had mixed solvent impacts (both paint and non-paint).

The solvent influence on toluene in Seoul was further investigated by limited tunnel sampling during the blocking period on June 2, 2016 (15:12 local time) and June 5, 2016 (07:40 and 07:48), by filling WAS canisters and then analyzing them in the field using PTR-ToF-MS. The June 2 tunnel sample was taken within 15 mins of the DC-8 performing a missed approach over the Seoul Air Base during Flight 17, providing an excellent opportunity for direct comparison. Four WAS samples were collected aboard the DC-8 during the missed approach from 14:54–14:57 local time (alt. = 100–470 m), with an average T/B ratio of 8.4 ± 1.5 compared to 2.2 in the June 2 tunnel sample, clearly showing the additional impact of non-traffic sources on toluene in Seoul. A similar T/B ratio was measured in the Sangdo tunnel in Seoul in May 2000 (2.5 at the entrance and 2.2 in the middle of the tunnel; Na et al., 2006).

Overall the analysis in Section 3.3.1 suggests that VOCs in Seoul during KORUS-AQ were impacted by biogenic sources, LPG, vehicle exhaust and solvents, with less prominent impact from gasoline evaporation, natural gas or biomass burning. Source apportionment was used to further investigate the sources of VOCs in Seoul, as discussed below.

Source apportionment was performed on the KORUS-AQ Seoul data (n = 177) using EPA PMF Version 5.0 (Section 2.4). Multiple runs were performed using different numbers of gases and factors in order to select an optimal data set to use. VOCs were chosen based on their abundance (Section 3.2.1), their value as tracer gases (Section 3.3.1), and the ability of PMF to reproduce the observed mixing ratios. Sixteen VOCs were excluded because they had low mixing ratios and/or numerous values below detection (the butenes, α- and β-pinene, styrene, and all C₉ aromatics). Methane and long-lived halocarbons (lifetime >6 mo) were not used, and the alkyl nitrates, which are secondary products in urban areas, also were not used. Other gases including shorter-lived halocarbons (e.g., CHCl₃, C₂Cl₄) were tested and rejected because the model did not reproduce the data well (r² of 0.31 to 0.50 for observed vs. predicted data). Likewise n-nonane, n-decane and the cycloalkanes were not used because, with the exception of cyclopentane, they included scaled residuals (e_ij/u_ij) outside the range of –3 to 3. Methanol was tested but not used because of a low signal-to-noise (S/N) ratio of 5.7.

Based on this data screening the designated PMF model run used 21 trace gases that were abundant in the SMA and represented tracers from a wide range of sources. This includes the biogenic tracer isoprene; combustion tracers CO and ethyne; vehicle exhaust tracers ethene and propene; natural gas tracer ethane; LPG tracers propane, i-butane and n-butane; gasoline evaporation tracers i-pentane and n-pentane; solvent tracers n-hexane, n-heptane, toluene, ethylbenzene, m,p-xylene, o-xylene, and the China tracer COS. Benzene, 2-methylpentane and 3-methylpentane were also included because of their abundance. n-Heptane was used despite having a data point with a scaled residual of 3.8 because of its value for distinguishing between paint and non-paint solvents (Ou et al., 2015). Likewise propene, i-butane, m,p-xylene and o-xylene each had 1–2 data points with scaled residuals of 3.02–3.13, but were kept because of their value as tracers. All 21 compounds yielded a strong S/N ratio (9.9 for isoprene and n-heptane, and 10.0 for the remaining compounds) and the model reproduced the data well (r² of 0.78 to 0.96, where n-heptane had the weakest r² value and ethylbenzene had the strongest). The scaled residuals were normally distributed and were between –3 and 3 for most compounds as described above, indicating that they were well modeled (Brown et al., 2007; Wu et al., 2016). That is, PMF was able to model VOCs with a range of lifetimes and mixing ratios, including VOCs with log-normal distributions such as toluene.

The PMF model was run using between 3 and 6 factors, and each solution was assessed for how robust and reasonable it was. The 3-factor solution was muddled and failed to resolve a clear biogenic source. Indeed, isoprene had Q/Qexp = 5 in the 3-factor solution, whereas all species had Q/Qexp < 2 in the 4–6 factor solutions (from Section 2.4, Q/Qexp > 2 indicates that a species was not well modeled). Like the 3-factor solution, the 6-factor solution was also muddled, in its case because three factors split to create a new factor. For the 4-factor and 5-factor solutions, in general the 4-factor solution was more robust while the 5-factor solution was more reasonable. Both solutions are presented below, followed by a discussion of the limitations of the PMF analysis.

Factor 1: Biogenic. The 4-factor and 5-factor solutions both had a factor (called Factor 1) that was dominated by isoprene and was assigned a biogenic source (Figure 9a, 9e). This factor showed no rotational ambiguity for both the 4- and 5-factor solutions (no swap counts in the DISP diagnostics; Brown et al., 2015). In reducing from 5 to 4 factors, the contribution of isoprene to Factor 1 remained similar (84% and 87%, respectively) whereas the contributions of non-biogenic VOCs increased by an average of 2.2 (Table S-3). For example, 7% of the combustion tracer ethyne was associated with Factor 1 with 5 factors, versus 13% with 4 factors. This suggests that the 5-factor solution is more reasonable for the biogenic source assignment, because non-biogenic VOCs contributed less to Factor 1 in the 5-factor solution. Factor 1 accounted for 13% and 7% of the source apportionment in the 4- and 5-factor solutions, respectively. Of the VOCs used in PMF, isoprene is the shortest-lived and the most sensitive to photochemistry (the isoprene lifetime is 3 hrs, followed by propene at 10 hrs; Table S-2). The sensitivity of the PMF results to photochemistry was investigated by increasing the amount of isoprene used in the PMF calculations. The biogenic contribution to the 4- and 5-factor solutions changed by +0.1 and +0.4%, respectively, with a 20% increase in isoprene, and by +1.7 and –0.6% with a 50% increase in isoprene. This suggests that potential impacts of photochemistry were relatively minor compared to other uncertainties in the PMF analysis.

Factor 2: Long-range transport. The 4- and 5-factor solutions both had a factor (Factor 2) that was characterized by the China tracer COS and longer-lived combustion tracers such as CO, ethyne and benzene, which together suggest long-range transport from China (Figure 9b, 9f). Similar to Factor 1, Factor 2 showed no rotational ambiguity for both solutions (no DISP swap counts). The contributions of the main VOC components were similar in both solutions (e.g., 67% for COS in both solutions; Table S-3), whereas the contributions of the minor components increased when reducing from 5 factors to 4. For example the contribution of toluene/ethylbenzene/xylenes (TEX) increased from 1–3% to 10–13% (Table S-3). That is, like the biogenic factor, switching from 5 to 4 factors caused some shorter-lived VOCs to contribute more to the long-range factor, suggesting that 5 factors is more reasonable. While Factor 2 includes tracers of incomplete combustion that could be associated with biomass burning (e.g., CO, ethyne), the Ministry of Environment Protection of the People’s Republic of China reported very few wheat burning hotspots in China in May and June 2016 (Y. Zhu, pers. comm., 2018; http://www.zhb.gov/cn/). Even though levels of the biomass burning tracer HCN were higher in air from China than in air from Seoul (Table S-3), the elevated HCN was likely associated with other fuel use such as domestic coal burning (Crounse et al., 2009; Du et al., 2016) given the insignificant biomass burning in the region. Factor 2 accounted for 24% and 20% of the source apportionment in the 4-factor and 5-factor solutions, respectively.

Factor 3: Vehicle exhaust. The remaining factors in the 4- and 5-factor solutions were associated with traffic and solvents. Both solutions had a factor (Factor 3) with highest loadings of the vehicle exhaust tracers ethene and propene, as well as other compounds associated with vehicle exhaust such as benzene and light alkanes (Figure 9c, 9g). Unlike Factors 1 and 2, Factor 3 had some rotational ambiguity in the 5-factor solution (6 DISP swap counts), meaning that caution should be used when interpreting the 5-factor solution. Factor 3 accounted for 57% and 52% of the ethene and 71% and 60% of the propene in the 4- and 5-factor solutions, respectively (Table S-3). The prominence of ethene and propene suggests that Factor 3 captured Seoul’s diesel emissions (Section 3.3.1). The combustion tracer ethyne did not associate especially strongly with Factor 3 (20% and 15% in the 4- and 5-factor solutions, respectively), though this has been previously observed in Hong Kong where ethyne emissions from gasoline-fueled cars were negligible and caution was suggested when using ethyne as a tracer of vehicle emissions (Guo et al., 2011a). Even though aromatics are emitted by vehicle exhaust, the aromatics except benzene had a relatively weak contribution to Factor 3 because solvents were a much stronger influence than traffic on the TEX aromatics in Seoul (Section 3.3.1; also see below). The combustion tracer CO also had a relatively small association with this factor (18% and 12% in the 4- and 5-factor solutions, respectively), and instead CO was most associated with Factor 2, the long-range transport source (53% and 51%, respectively; Table S-3). This is broadly consistent with analysis by Halliday et al. (2019), who found that ΔCO/ΔCO₂ showed regional differences during KORUS-AQ, with a lower ratio (~1%) over Seoul reflecting efficient combustion, and a higher ratio (2–4%) at higher altitudes and over the West Sea reflecting less efficient combustion in air arriving from China. The vehicle exhaust source accounted for 27% and 21% of the source apportionment in the 4- and 5-factor solutions, respectively.

Factor 4: Mixed source. The fourth factor (Factor 4) was more complex than the first three factors and was different for the 4-factor and 5-factor solutions. The 4-factor solution had highest loadings of TEX compounds (67–69%), followed by C₃–C₇ alkanes and C₂–C₃ alkenes (Figure 9d; Table S-3). That is, the 4-factor solution showed a mix of VOCs that can be related to solvents, traffic and cleaner fuels. In broad terms, this factor split into two factors in the 5-factor solution (Factors 4 and 5). While Factor 4 of the 5-factor solution showed less rotational ambiguity (1 DISP swap) than Factor 5 (6 DISP swaps), both factors are interpreted with caution. Factor 4 of the 5-factor solution retained most of the xylenes (65–70%), ethylbenzene (56%), C₃–C₅ alkanes and C₂–C₃ alkenes (Figure 9h; Table S-3), and this factor appears to be a mixed source. The strong contributions from ethylbenzene and the xylenes suggest impact from paint solvents, as discussed in Section 3.3.1. The presence of traffic tracers ethene and propene together with light alkanes suggests contributions from gasoline and LPG fuels. The ratio of i-pentane/n-pentane is higher in this factor (1.7) than in the other traffic and solvent factors (1.20–1.25), suggesting that gasoline evaporation may contribute to this factor even if its overall impact in Seoul is relatively small (Section 3.3.1). Therefore the alkenes and alkanes associated with Factor 4 are consistent with cleaner fuels and a mix of combustive and evaporative emissions. This mixed fuel and solvent source accounted for 36% and 32% of the source apportionment in the 4- and 5-factor solutions, respectively.

Factor 5: Solvents (non-paint). Factor 5 had highest loadings of toluene (57%), n-hexane (57%) and n-heptane (45%), followed by methylpentanes, ethylbenzene and the xylenes (Figure 9i; Table S-3). In other words, consistent with the correlation analysis is Section 3.3.1, it is notable that toluene associates more strongly with a different factor than ethylbenzene and the xylenes, implying different source influences for individual TEX species. Previous work in Asia has found that the printing industry is toluene-rich (Tang et al., 2014) and that n-hexane and n-heptane are also tracers of the consumer products and printing industry, for example disinfectants, air fresheners, furniture polishes, sealants and offset printing (Kwon et al., 2007; Ou et al., 2015; Shen et al., 2018). As noted in Section 3.3.1, though aromatics can be emitted from vehicle exhaust, the high T/B ratio and poor correlation of toluene with traffic tracers both suggest a predominant impact from solvents rather than vehicle exhaust on toluene in Seoul. While recognizing that there can be significant regional differences in VOC source emission signatures (Yuan et al. 2010; Zheng et al., 2013), Factor 5 is consistent with non-paint solvents such as consumer products and printing. The association of methylpentanes with this factor suggests some solvent influence on methylpentanes in Seoul in addition to traffic. Factor 5 accounted for 20% of the source apportionment.

A combination of complementary techniques was used to investigate the most likely sources of VOCs in Seoul. VOC tracers, correlations and ratios identified contributions from vehicle exhaust, solvents, LPG and biogenic sources, and ruled out major impacts from gasoline evaporation, natural gas and biomass burning. Source apportionment identified vehicle exhaust, solvents, cleaner fuels such as LPG, biogenic and long-range transport sources, with a separation of paint and non-paint solvents depending on how many factors were used. While PMF can quantify the contributions of each source to the measured VOCs, these quantitative results should be interpreted with caution. On the one hand, although the 4-factor solution was more robust than 5-factors, it was not necessarily more reasonable because several VOCs became mis-assigned. On the other hand, while the 5-factor solution was able to separate paint and non-paint solvents, which is useful for informing policy, it showed some rotational ambiguity. Because the separation of paint and non-paint solvents was also evident in simple correlation analysis, and because the 5-factor solution was more reasonable for the biogenic and long-range source factors, we suggest that the 5-factor solution can be used with appropriate uncertainty considerations. The average difference between the 4 and 5-factor solutions was about 5% (Table S-3), and this is likely the minimum uncertainty for interpreting the PMF results. Therefore, bearing in mind this uncertainty, the measured VOCs in Seoul are represented by biogenic sources (7%), long-range transport (20%), vehicle exhaust (21%), non-paint solvents (20%) and a mix of paint solvents and cleaner fuels (32%).

While recognizing that the mixed source provides a range of uncertainty, the 5-factor PMF results suggest roughly equal contributions of vehicular emissions (21–53%) and solvents (20–52%) to VOC levels over Seoul, with a maximum solvent contribution of 52% for the mix of VOCs that was used (though this value is unlikely as it would require fuel tracers to be accounted for by solvents). These PMF results yield relatively less solvent impact on total VOCs than the KORUSv5 EI for Seoul (82% solvents, 15% mobile). However it is important to recognize additional uncertainties in this comparison. For example the KORUS-AQ source apportionment is based on CO and 20 VOCs using focused measurements during springtime, while the KORUSv5 EI is based on 30 VOC groups that were speciated from annual NMVOC estimations in the CAPSS inventory. The uncertainties in the chemical speciation process, which converts total NMVOCs to speciated model-ready data, and in the temporal allocation process, which converts the 2015 annual estimation to monthly values, could be other sources of uncertainty. Overall, vehicle emission control measures and fuel substitution strategies are well established in South Korea, and the PMF results and KORUSv5 EI both point to solvents as a major source of reactive VOCs that could be further pursued for VOC emission reductions.

Using multiple data analysis techniques based on 82 VOCs measured in whole air samples and selected OVOCs measured using PTR-ToF-MS, this study has identified light alkenes (isoprene, ethene, propene), C₇–C₈ aromatics (toluene, xylenes and ethylbenzene) and methanol as major precursors to ozone formation in Seoul. The aromatics also contribute to SOA formation (Nault et al., 2018). Ozone formation in Seoul is more sensitive to VOCs than NO_x (H. Kim et al., 2018b; S. Kim et al., 2018), similar to many cities in Asia (Guo et al., 2017). Because VOC reductions in South Korea have been slower than NO_x and because air pollution in Seoul may worsen under increasing ratios of VOCs/NO_x (Kim and Lee, 2018), aromatics and alkenes are attractive targets for VOC emission reduction strategies. Vehicle exhaust was identified as the main source of ethene and propene emissions, and solvents were the main sources of aromatics, primarily paint solvents for ethylbenzene and the xylenes, and non-paint solvents (e.g., consumer and printing products) for toluene. In other words, aromatic (TEX) reductions can best be achieved by targeting solvents rather than traffic, with a focus on multiple solvent products including both paint and non-paint solvents. This step would complement and expand on existing strategies to limit VOC content in paints, in addition to long-established policies in South Korea to limit emissions from vehicle exhaust.

A similar strategy has been successfully applied in Hong Kong, a city where toluene used to be the most abundant VOC and where policies that regulate the VOC content in solvent products have led to a decline in toluene levels (Guo et al., 2006; Lyu et al., 2017). Specifically, the Hong Kong government prescribed VOC content limits for 172 types of regulated products including architectural paints, printing inks, consumer products, vehicle refinishing coatings, vessel and pleasure craft coatings, adhesives and sealants, fountain solutions, and printing machine cleansing agents (P.K.K. Louie, pers. comm., 2018).

The 2016 KORUS-AQ study provided a focused opportunity to study the characteristics, sources and ozone-forming potential of VOCs in Seoul and its surrounding regions during spring. Based on VOC measurements in 2554 whole air samples, distinct VOC signatures were characterized for major emission regions, including the Seoul region and the Daesan petrochemical facility in northwest Korea, the Busan region of southeast Korea, and air originating from China during periods of dynamic transport. In addition to methanol, formaldehyde, ethane and propane, Seoul air was particularly toluene-rich and Daesan air was rich in ethene and benzene. Levels of CCl₄ and H-2402 stood out in Busan. Recommended tracers of air from China during KORUS-AQ include COS,1,2-dichloroethane, CFC-113, CFC-114 and CCl₄. While Halon 1211 used to be a good tracer of air from China, this is no longer the case because its production in China has been phased out. Although CFC-11 was elevated in air from China, it was surprisingly more enriched in air from Seoul including during stagnant periods.

Low altitude air samples collected over Seoul were analyzed using a combination of techniques. Isoprene, toluene, xylenes and ethene were strong individual contributors to OH reactivity in Seoul, similar to results from major urban areas of China. While methanol was also a strong contributor to OH reactivity, photochemical box modeling showed a lower impact of methanol on ozone formation in Seoul, which is attributed to the model’s more realistic handling of radical chemistry. Both the OH reactivity calculations and photochemical box modeling agreed that ozone production in Seoul is most sensitive to aromatics, followed by isoprene and anthropogenic alkenes. These results highlight the importance of source apportionment studies and other techniques that can identify specific source sectors for VOCs, particularly for individual aromatics. Correlations, ratio analysis and source apportionment all pointed to stronger influence from solvents than traffic on the reactive TEX aromatics. Separation of the solvents source provided valuable information to guide strategies for reducing emissions of aromatics that are implicated in O₃ and SOA formation. Specifically, ethylbenzene and the xylenes were more associated with paint solvents and toluene with non-paint solvents. Therefore, in addition to current programs that limit VOC emissions from vehicles and the VOC content in paints, this work suggests that content limits on additional volatile chemical products, for example consumer products and printing agents, could help achieve the goal of reducing aromatic levels in Seoul, especially toluene, as part of ongoing efforts to improve air quality.

All observational data from the KORUS-AQ mission, including the WAS VOC data, are archived at: https://www-air.larc.nasa.gov/cgi-bin/ArcView/korusaq.

The KORUS-AQ mission was jointly funded by NASA and the Korean National Institute of Environmental Research (NIER). We gratefully acknowledge the KORUS-AQ crew and science team, and the technicians and staff at UC-Irvine who supported the WAS effort (Barbara Chisholm, Gloria Liu Weitz, Brent Love, Nick Vizenor, Rafe Day). The NO_y and O₃ data were provided courtesy of Andrew Weinheimer and Denise Montzka (NCAR), and John Crounse provided HCN data. The PTR-ToF-MS measurements were supported by the Austrian Federal Ministry for Transport, Innovation and Technology (bmvit) through the Austrian Space Applications Programme (ASAP) of the Austrian Research Promotion Agency (FFG). The PTR-ToF-MS instrument team (P. Eichler, L. Kaser, M. Müller) is acknowledged for their support in the field and during the mission preparation phase, and Ionicon Analytik is acknowledged for instrumental support. We also thank Daniel Bon (Colorado Dept. of Public Health & Environment) for helpful discussions on PMF, Hannah Halliday (NASA Langley) and Meehye Lee (Korea University) for helpful discussions on regional source influences, Saewung Kim (UCI) for coordination during the MAPS-2015 campaign, and Yanhong Zhu (Jinan University) for providing wheat count hotspots during the KORUS-AQ timeframe. We thank the UC White Mountain Research Center for use of the Crooked Creek Station for gathering air samples for use as working standards. Finally we thank numerous colleagues for valuable discussions and feedback on the interpretation of the source apportionment results.

The authors have no competing interests to declare.

IJS interpreted the data and wrote the manuscript. DRB oversaw the WAS contribution to KORUS-AQ and performed data analysis and interpretation. NJB oversaw WAS field operations, and NJB, SCH and LTF acquired WAS data. SM oversaw WAS laboratory operations and performed data analysis. BB oversaw WAS data management. JHC and JRS performed photochemical box modeling calculations and data interpretation. GSD and SEP provided CO and CH₄ measurements. HG, DW and IZ supported the PMF analysis. DAP oversaw the meteorological analysis. AW and TM performed PTR-ToF-MS analysis. JHW, JK and YK oversaw the KORUSv5 emission inventory and provided data for the paper. AF led the CH₂O analysis. LKE, JB and BG provided CAM-chem model simulations. CK provided the FLEXPART back trajectories. POW and MJK provided HCN data. All co-authors revised the manuscript and approved the submitted version for publication.