Characterization of PM 2.5 Mass in Relation to PM 1.0 and PM 10 in Megacity Seoul

This study examines the PM 2.5 characteristics in Seoul in relation to those of PM 1.0 and PM 10 . Samples were typically collected daily on filters and a few hours sampling were conducted during a few haze events (March 2007 to June 2008). Mean mass concentrations of PM 1.0 , PM 2.5 , and PM 10 were 19.7 μg/m 3 , 26.0 μg/m 3 , and 48.2 μg/m 3 , respectively, and PM 2.5 was reasonably correlated with PM 1.0 (γ=0.79) and PM 10 (γ=0.52). Three mass group types were mainly distinguished. Group 1 (31%): linear increase of PM 1.0 with PM 10 and high OC and NO 3 - ; Group 2 (17%): PM 10 considerably higher than PM 1.0 and high Ca 2 + and SO 42 - ; Group 3 (52%): PM 1.0 relatively more enhanced than PM 10 and highest carbonaceous fraction against mass. The fine mode fraction was lowest (highest) in Group 2 (Group 3). Haze and dust episodes relating to Chinese outflows were mostly evident in Groups 1 and 2, respectively; average PM 2.5 concentrations were visibly higher than in Group 3. Non-Negative Matrix Factorization analysis demonstrated that traffic-related urban primary (28%) and coal-fired industry (27%) emissions equally contributed to the PM 2.5 mass, followed by aged urban secondary (19%), soil mineral (16%), and biomass combustion (10%) sources. Seasonal variations were apparent in air mass trajectories. Urban primary and coal-fired industry factors were predominant in Group 3 under stagnant conditions in the warm season and under a strong northerly wind in the cold season, respectively. However, contributions of the other three factors were higher in Groups 1 and 2. This study shows that the PM 2.5 mass in Seoul is largely dependent on high concentration episodes occurring mostly in cold seasons. It also shows that local emissions contribute considerably during warm months, while the influence of Chinese out-flow predominates during cold months.


INTRODUCTION
Particulate matter (PM) affects visibility and air quality on local and regional scales and exacerbates global climate change (Zhang et al., 2015;IPCC, 2013;Liu et al., 2013;Fowler et al., 2009;Isaksen et al., 2009).PM concentrations are continually increasing, particularly in urban areas, and more than 85% of the world's population now lives in areas where the World Health Organization's Air Quality Guideline is exceeded (WMO/IGAC, 2012).PM has been recognized as one of the major leading causes of death (Lelieveld et al., 2015), but its impact is highly variable and depends on the size, chemical composition, and optical properties of aerosols, which are determined by emission sources and atmospheric processing (Theodosi et al., 2011).While supermicron particles (PM 10 ) generally consist of trace metals (e.g.Al, Fe) and soluble ions (e.g.Ca 2+ , SO 4 2- ) emitted from natural sources such as soil dust and sea salts ( Jacobson, 2012), the major constituents of submicron particles (PM 1.0 ), which include SO 4 2-, NO 3 -, organic carbon, and elemental carbon, are derived from anthropogenic sources, and mostly from combustion processes (Monks et al., 2009;Querol et al., 2009;Brasseur et al., 1999).
Seoul, the capital of the Republic of Korea, is surrounded by heavily populated satellite cities.The Seoul Metropolitan Area (SMA) ranks as the second mostpopulated megacity in the world (WMO/IGAC, 2012) with its combined population of over 20 million (Jeong et al., 2011).In past decades, studies on PM have focused on PM 10 with an emphasis on Asian dust events (e.g., Choi et al., 2001).In Korea, PM 10 has been an environmental standard since 1995, and the annual and daily standard is 50 μg/m 3 and 100 μg/m 3 , respectively in the present (NIER, 2018).The powerful regulation and management base on Clean Air Act contributed to that the annual average PM 10 concentration in Seoul had gradually decreased from 78 μg/m 3 in 1995 to 35 μg/m 3 in 2020.However, it is still higher than in other cities such as Tokyo (16 μg/m 3 in 2019), Los Angeles (29 μg/ m 3 in 2020), Paris (19 μg/m 3 in 2020), and London (16 μg/m 3 in 2020) (NIER, 2021).Recently the focus of concern has shifted from PM 10 to PM 2.5 because of its effect on air quality and human health (Lee et al., 2017;Seo et al., 2017;Ghim et al., 2015;Heo et al., 2009).
The research on PM 2.5 in South Korea has also been carried steadily in keeping pace with policy changes.Heo et al. (2009) investigated the emission sources of chemical compositions of PM 2.5 in Seoul.Ghim et al. (2017) compared the concentration changes of major chemical components in PM 2.5 connected with Asian dust and smog events in Seoul and Deokjeok Island to understand the influence of long-range transport.Park et al. (2013) showed the influence of biomass burning plumes originated from south China and local SO 2 emission on PM 2.5 from the measurement of carbonaceous and inorganic species of PM 2.5 at Gwangju, southwest urban site in Korea.However, these previous studies were of the features of PM 2.5 itself rather than relative character among PM 1.0 and PM 10 .
The westerlies are dominant in northeast Asia, and Chinese outflows significantly affect both the composition and concentration of coarse and fine mode particles (e.g., Shang et al., 2018a).In addition, PM 1.0 has been investigated as a better criterion representing anthropogenic aerosols, because of the impact of soil particles on PM 2.5 (Shang et al., 2018b;Lim et al., 2012).In this context, the mass and chemical characteristics of PM 2.5 need to be thoroughly examined in relation to those of PM 10 and PM 1.0 , to establish an effective abatement policy for PM 2.5 .
In this current study, the mass and chemical compositions of PM 1.0 , PM 2.5 , and PM 10 were measured in Seoul, and their chemical characteristics and source profiles were compared with the ultimate aim of contributing to mitigating air quality for PM 2.5 in Seoul.

1 Sampling and Chemical Analysis
PM 10 , PM 2.5 , and PM 1.0 samples were collected at the Korea University campus in Seoul (N 37.59°, E 127.03°) from March 2007 to June 2008 (Fig. 1).Sampling was basically conducted for 24 h with six days interval but 7 sets of 6 or 12 h intensive sampling in (9% of total samples) carried out for detail understanding of chemical compositions of PMs in three haze events.Samples were collected during times of no precipitation; therefore, only a few samples were collected during the summer monsoon season ( July-August) (Table 1), whereas sampling was more frequent during Asian dust and haze episode, which mostly occurred in spring and winter.
Samples were collected on 37 mm filters using Teflon coated Aluminum sharp cut cyclone (URG, USA) at 16.7 LPM.The cumulative air flow was measured with a dry gas-meter (URG, USA).Following collection by cyclone, the air passed through two annular denuders before reaching the filters: one coated with 1% Na 2 CO 3 to remove acidic gases (SO 2 , HNO 3 ) and the other coated with 1% citric acid to remove basic gas (NH 3 ).Teflon, quartz, and glass fiber filters were used to measure mass concentrations and water-soluble ions, carbonaceous compounds (PM 2.5 and PM 1.0 ), and trace metals (PM 10 and PM 2.5 ), respectively (Pall corp., USA).The Teflon filters were pre-dried during 24 h under constant temperature and humidity condition prior to sampling, and the quartz and glass fiber filters were pre-baked at 500°C for 6 h.
The Teflon filter equilibrated for 24 h under the clean condition after sampling and then the gravimetric masses of PM 10 , PM 2.5 , and PM 1.0 were measured using a micro-balance.The weight of all Teflon filters was averaged over three consecutive measurements.All processes performed under strict quality control to eliminate the possibility of sample contamination.Water-soluble ions, including Na + , NH 4 and SO 4   2-, were analyzed by ion chromatography (Dionex 4500, Dionex, USA), and organic carbon (OC) and elemental carbon (EC) were determined using the Interagency Monitoring of Protected Visual Environments (IMPROVE) thermal/optical reflectance protocol (TOR) method at the Desert Research Institute (DRI) (Chow et al., 1993).OC consists of OC1, OC2, OC3, OC4, and pyrolyzed OC (OP), and EC is the sum of EC1, EC2, and EC3 minus OP.OP was determined as reflectance returning to its initial value after analysis of OC components with injection of O 2 (98% with 2% He) (Chow et al., 2005).
Trace elements, such as Al and Fe, were analyzed by Inductively Coupled Plasma Mass Spectrometer (ICP-MS) at the Korea Basic Science Institute.Reactive gases, including O 3 , NO 2 , CO, and SO 2 , were measured by the National Institute of Environmental Research (NIER) at a monitoring station of Dongdaemun-Gu (37.58°N, 127.03°E) near Korea University (about 1,000 m distance) (Fig. 1).Meteorological factors, including wind speed and direction, temperature, relative humidity, and solar radiation, were obtained every 1 min by an Automatic Weather Station (AWS) at the campus.The backward trajectories of air masses arriving at 1,000 m above sea level (a.s.l.) were calculated for 72 h every 6 h (at 03:00, 09:00, 18:00, and 21:00 local time) using the NOAA Air Resources Laboratory (ARL) Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model (version 4) (Draxler and Rolph, 2012;Rolph, 2012, http://www.arl.noaa.gov/ready/hysplit4.html).

2 Non-negative Matrix Factorization (NMF)
Aerosol sources were examined using non-negative matrix factorization (NMF), which is a receptor model that combines principal components analysis (PCA) and vector quantization (VQ) (Lee and Seung, 1999) and has been widely used for environmental studies in conjunction with positive matrix factorization (PMF) (e.g., Shang et al., 2018b;Laing et al., 2015;Oh et al., 2011;Liang and Fairley, 2006).It is similar to the two-way PMF, in which two solution matrices (source contribution and source profile) with specified dimensions are  (0/6)   9  (1/3)   1  (0/1)  1  5  (0/3)  24   Group 2  4  (2/3)   3  (1/2)   2  (0/1)  0  4  (3/1 calculated by the input data matrix that includes the number of samples and species (Liang and Fairley, 2006).It can be expressed by the following eq.( 1), where X is the input data matrix (n samples × m species), G is the source contribution matrix with p sources, F is the source profile matrix, and E is the residual matrix between observed and calculated values (Xie et al., 2008).The NMF minimizes the conventional leastsquares error (Q) and enables the weighting of variables and samples (Liang and Fairley, 2006).To reduce the effect of outliers on the fitting of contributions and profiles, Q is obtained from a robust mode (Williams et al., 2010;Reff et al., 2007).In this study, the optimum number of factors were determined by Q values using a comparison between Q theory and Q robust at a range of p values (e.g., Williams et al., 2010;Reff et al., 2007); as a result, the five factors were selected.Furthermore, data were prepared using a method employed in previous studies (e.g., Alleman et al., 2010;Reff et al., 2007).Missing data and values that were below the detection limit (DL) were replaced with mean concentrations and DL/3, respectively (Heo et al., 2009).Further details about the NMF and PMF methods can be found in Lee and Seung (1999) and Reff et al. (2007).

RESULTS AND DISCUSSION
3. 1 Characteristics of PM 2.5 and Relationship with PM 1.0 and PM 10 Seventy-eight sets of samples were obtained through-out the entire period, for which the average mass concentrations of PM 1.0 , PM 2.5 , and PM 10 were 19.7 μg/m 3 , 26.0 μg/m 3 , and 48.2 μg/m 3 , respectively.While PM 2.5 was in good correlation with both PM 1.0 (γ = 0.79) and PM 10 (γ = 0.52), the correlation between PM 1.0 and PM 10 was relatively poor (γ = 0.37).The PM 2.5 mass concentrations are presented in relation to those of PM 1.0 and PM 10 in Figs. 2 and 3, where it shows that, in general, PM 2.5 is proportionally increased with PM 10 and PM 1.0 .Note that there are two types of outliers that are either due to more enhanced PM 10 or less enhanced PM 10 relative to PM 1.0 .PM 2.5 is generally defined as a  fine aerosol in mode distribution.However, our results show that PM 2.5 increased with a sharp increase in PM 10 or with a dominant increase in PM 1.0 .It likely indicates that measured PM 2.5 mass can be affected by both fine and coarse particles.
Fig. 2 illustrates the relationship of PM 1.0 and PM 10 mass concentrations in Seoul during the whole experiment period, where PM 2.5 mass concentrations were color-coded.While PM 1.0 and PM 10 mass concentrations are well correlated, sample data points can be grouped mainly by 3 domains; (i) samples within a 95% confidence interval ("Group 1", 31% data over total data), (ii) samples with clearly deviated PM 10 mass concentrations ("Group 2", 17%), and (iii) samples with relatively higher PM 1.0 and lower PM 10 mass concentrations ("Group 3", 52%).Table 1 shows the number of samples in each group in accordance with the seasons.
The mass concentrations and other chemical characteristics are summarized in Table 2 and Fig. 3.Of the three groups, Group 2 is distinguished by the highest PM 10 (103.0 μg/m 3 ) and PM 2.5 (34.6 μg/m 3 ) with the highest ratio of PM 2.5 to PM 1.0 (1.6) and the lowest ratio of PM 2.5 to PM 10 (0.3) (Fig. 3a).Group 1 has similar mean concentrations of PM 2.5 (31.6 μg/m 3 ) to Group 2, with the best correlation between PM 1.0 and PM 10 (γ = 0.89), and Group 3 has the lowest PM 10 mass level, but ratios of PM 2.5 /PM 10 (0.7) is the highest and PM 2.5 / PM 1.0 (1.2) is the lowest, which is considered to be associated with the submicron nature of particles.From the seasonal frequency and mass characteristics of the three groups, it is quite likely that Group 2 represents events when the mass of coarse mode particles was enhanced and that Group 1 includes events with high concentrations of fine mode particles (Table 1 and Fig. 3).These results suggest that the PM 2.5 variation in Seoul can be understood in accordance with the collective figures from these three groups, which have distinct mass and chemical composition characteristics.Therefore, detailed characteristics of the three groups were further examined with respect to the major composition of air masses, air mass origins, and source profiles.

2 Characteristics of Major Composition and
Origin of Air Masses Compared to the other groups, the major aerosol composition of Group 1 consisted of the highest concentrations of OC, EC, and NO 3 -(Fig.3), and the average concentrations of reactive gases, including SO 2 , CO, NO 2 , and O 3 , were also the highest (Fig. 3).For Group 1, high PM 2.5 mass concentrations were observed when air masses had moved through urbanized regions in China (Fig. 4), which implies that high concentrations of secondary ions, such as NO 3 -and trace gases, are associated with Chinese outflows (Zhang et al., 2015;Huang et al., 2014;Zhang et al., 2013;Wang et al., 2006).In particular, EC is strongly correlated with SO 4 2-, which suggests the influence of coal combustion (Hou et al., 2011;Seinfeld and Pandis, 2006).
In Group 2, NO 3 -and SO 4 2-concentrations were highly elevated in PM 2.5 and PM 10 (Fig. 5); however, their fractions against mass were the lowest among the three groups because the highest PM 10 and PM 2.5 mass concentrations exceeded the annual standard (Table 2).Ca 2+ was highly enriched in coarse mode particles in PM 10 −PM 2.5 , which indicates the impact of Asian dust (Kang et al., 2013;Lim et al., 2012;Lee et al., 2007).Backward trajectories analysis shows that air masses were originated from desert regions of Inner Mongolia, passed through urbanized northeastern China, and finally arrived in Seoul with high PM 2.5 loadings (Fig. 4).It is a common pathway of Asian dust plumes coming to Korean Peninsula (Lim et al., 2012).It is thus highly likely that the observed coarse-mode NO 3 -and SO 4 2-concentrations for Group 2 are associated with heterogeneous reactions of anthropogenic oxidized vapors on abundant soil particles during transport (Geng et al., 2009;Lin et al., 2007;Wang et al., 2007;Jacobson, 2002Jacobson, , 2001)).
Although there were only a few samples, the concentrations of EC and SO 4 2-were noticeably increased when air masses had passed over industrial regions in northern and northeastern China.These results imply that Groups 1 and 2 relate to episodes of high PM concentrations that are mainly influenced by Chinese outflows.In particular, high PM 2.5 masses that exceed the national standard are associated with haze occurrence in Group 1 and Asian dust events in Group 2, and these are clearly distinguished by the relation between PM 10 and PM 1.0 mass concentrations.
In Group 3, the carbonaceous fraction (sum of OC and EC) against PM 2.5 mass was noticeably high (up to 0.63) along with a high OC/SO 4 2-ratio of 2.85, and most of the OC and EC was enriched in PM 1.0 (Table 2, and Fig. 5).In addition, OC and EC exhibited a significant correlation with NO 3 -(γ ≥ 0.72), which suggests a fossil fuel combustion source (Zhang et al., 2013;Hou et al., 2011;Seinfeld and Pandis, 2006).Fig. 4 shows that backward trajectories of Group 3 are scattered around the Korean peninsula.Of these, trajectories for high PM 2.5 concentrations, which are associated with high EC, NO 3 -, and SO 4 2-concentrations, linger around the area southwest of Seoul under stagnant conditions (Fig. 4).It is quite likely that domestic and  local influences mostly prevail in Group 3. It is also worth mentioning that high OC, EC, and SO 4 2-concentrations of PM 2.5 were intermittently observed in air that had been rapidly transported from the Liaoning and Jilin regions of China during winter, which indicates that the Chinese influence is prevalent during the cold season.
A comparison between the chemical compositions of the three groups confirms that the characteristics of PM 2.5 can be distinguished by the relation between PM 10 and PM 1.0 .It also highlights that the PM mass concentrations of Seoul are largely dependent on the episodic occurrence of high concentration events associated with Chinese outflows.On the other hand, when local influences dominate under stagnant condition, aerosol mass concentrations are relatively lower but contributions from carbonaceous compounds are greater.

3 Source Apportionment
Table 4 shows that from the NMF analysis of 78 samples of PM 2.5 , five factors were identified in relation to source type as follows: primary urban emissions mostly from traffic related sources, and secondary urban sources that are aged, biomass combustion, coal-fired industry, and soil dust.The five sources are seasonally and geographically distinct and are more evident in high-con-centration episodes.It is known that in northeast Asia, the PM 2.5 source and degree of atmospheric processing is largely dependent on synoptic meteorological conditions (e.g., Shang et al., 2018a;Lim et al., 2012), and these are summarized as plume characteristics in Table 4.In addition, the sources of PM 2.5 are estimated and compared for each mass group.
Urban primary aerosols (Factor 1) are represented by high concentrations of OC, EC1, SO 4 2-, NO 3 -, and trace metals such as Pb, V, Cr, and Mn.As the contribution of V relates to both natural and anthropogenic sou rces, the ratio of V/Mn has been suggested as an indicator of vehicle emissions (Cao et al., 2005;Chow et al., 2004;Rodríguez et al., 2004).The contribution of Factor 1 was the largest in Group 3: 28% and rising to 33% when air had been rapidly transported from urban areas of Beijing, Dalian, and Harbin area during cold months or when the air was stagnant in warm months.
Factor 2 is characterized by high NO 3 -, NH 4 + , and SO 4 2-loading, which relates to carbonaceous components, and accounts for 19% of the PM 2.5 mass.These species were found to be highly elevated during haze events and responsible for a reduction in visibility in China (Zhang et al., 2015).The contribution of Factor 2 was evidently higher in Group 1 (23%) and Group 2 (24%) than in Group 3 (13%).In particular, the contri-  bution of Factor 2 in Groups 1 and 2 was greatest when air masses had been slowly transported from eastern China (including Shanghai) during the warm season.It is thus considered that Factor 2 represents aged and secondary urban aerosols (Shang et al., 2018b;Künzi et al., 2015;Huang et al., 2014).K + , Cl -, NO 3 -, and Mg 2+ were identified as major components in Factor 3, and these are related to biomass combustion emissions (Chen et al., 2017;Han et al., 2007;Chow et al., 2004;Duan et al., 2004).The contribution of Factor 3 was the highest in Group 1 (13%) and particularly related to eastern Chinese outflows from the Shandong region.In eastern China, biomass combustion frequently occurs in association with agricultural clearing fires and the use of biofuel for heating, which have been identified as the major sources of haze events (Chen et al., 2017;Han et al., 2017;Yin et al., 2017;Zhang et al., 2012;Cao et al., 2006;Yang et al., 2005;Duan et al., 2004).
Factor 4 is distinguished by high concentrations of NO 3 -, SO 4 2-, refractory components of OC (OC3 and OC4), and EC with relatively high concentrations of trace metals, and thus represents emissions from coalfired industries (Lim et al., 2014;Duan and Tan, 2013;Heo et al., 2009;Han et al., 2006;Cao et al., 2005;Chow et al., 2004;Rodríguez et al., 2004).Trace elements, Pb, Zn, and Cd are known to be enriched in coal dust (Bozlaker et al., 2013;Duan and Tan, 2013).It is of note that Group 3 had the highest contribution of Factor 4, and the fractions of OC and EC were high with relatively low concentrations of PM 2.5 and reactive gases.For these samples, air masses had been rapidly transported during the cold months from northeastern China, including the Liaoning and Jilin regions, and through North Korea.These regions are recognized for their heavy use of anthracite, bituminite, and brown coal in industries, power plants, and for household heating (Li et al., 2017;Kim, 2015;Han et al., 2009;Zhang et al., 2008;Cao et al., 2006).This result indicates that the influence of Chinese outflows is persistent during the cold season but is not associated with high PM 2.5 concentrations.
With high loadings of Ca 2+ , Na + , Cl -, and some met-  als Cu, Ni, Cr, Co, Mn, Factor 5 represents the influence of soil dust (Lim et al., 2012;Zhu et al., 2012;Lee et al., 2007;Wang et al., 2006).The contribution of Factor 5 was particularly high (18%) in Groups 1 and 2 (Fig. S1).In particular, the concentrations of Cl -and Na + were enhanced when air masses had been transported from Inner Mongolia and northeastern China, where saline dust is commonly observed (Shang et al., 2018b;Zhu et al., 2012;Aldabe et al., 2011).Trace elements such as Cu were likely added to the air stream when passing over the Beijing region on the way to the sampling site (Duan and Tan, 2013).
Overall, NMF analysis shows that traffic-related urban primary (Factor 1) and coal-fired industry (Factor 4) emissions are the main sources of (and equal contributors to) PM 2.5 in Seoul.The contributions of these two factors were greatest in Group 3 (accounting for 67%).In Groups 1 and 2, the contribution of an aged urban source (Factor 2) was as large as Factors 1 and 4, and secondary ions such as SO 4 2-and NO 3 -were highly elevated.In addition, sources with the least contributions, including biomass combustion (Factor 3) and soil dust (Factor 5) were significant in these two groups but not in Group 3.These results imply that the air masses in Groups 1 and 2 were more influenced by Chinese outflows than those in Group 3.
Interestingly, the traffic-related urban primary source was distinct not only in the cold season but also in the warm season when air masses were stagnant around the Seoul Metropolitan Area (SMA).It is thus probably representative of emissions from the SMA, and it caused high-mass episodes during the warm season.The development of stagnant conditions that promote pollution episodes during the warm season (from May to June) have previously been observed in Seoul (Lee et al., 2015).Loadings from the coal-fired industry (Factor 4) were elevated with respect to strong northerly winds during cold seasons.Under these conditions, industrial emissions are transported quickly from Far East China and North Korea.Although the mass concentrations of Group 3 were not as high as those of Groups 1 or 2, the fractions of OC and EC against the PM 2.5 mass were significantly higher in Group 3. In this respect, and with   respect to the harmful effects of carbonaceous aerosols on health, these low PM 2.5 masses should not be disregarded (Atkinson et al., 2015;Künzi et al., 2015;Kennedy, 2007;Sørensen et al., 2003).

CONCLUSIONS
PM 1.0 , PM 2.5 and PM 10 were collected at the Korea University Campus in Seoul, South Korea, during March 2007-June 2008.A total of 78 sets of PM 1.0 , PM 2.5 and PM 10 samples were analyzed for mass and chemical compositions including soluble ions, carbonaceous compounds, and trace metals.
In general, PM 2.5 proportionally increased with PM 10 and PM 1.0 (Group 1), although there were two types of outliers due to either more (Group 2) or less (Group 3) enhanced PM 10 relative to PM 1.0 .In Group 1 (31%), the best correlation was found between PM 1.0 and PM 10 mass concentrations (γ = 0.89).In this group, OC and NO 3 -were discernibly higher (with reactive gases such as CO, SO 2 , and NO 2 ) when air masses had been transported through urbanized regions in Eastern China.The average PM 10 and PM 2.5 concentrations were highest in Group 2 (17%) with highly enriched SO 4 2-and Ca 2+ in coarse modes of PM 10 -PM 2.5 , which indicates the impact of soil dust.The PM 2.5 concentration in Seoul is often highly elevated during haze and dust outbreaks, and samples were mostly found in Groups 1 and 2, respectively.Although the masses of PM 10 and PM 2.5 were lowest in Group 3 (52%), the carbonaceous fraction (OC and EC) against PM 2.5 mass was the highest (63%), and OC and EC concentrations were well correlated with NO 3 -, which suggests fossil fuel sources.The source of PM 2.5 was apportioned using the Nonnegative Matrix Factorization (NMF) receptor model into the following sources: traffic-related urban primary (28%), coal-fired industry (27%), aged urban secondary (19%), soil dust (16%), and biomass combustion (10%).Seasonality is implicit in these five sources and their contributions differed between cold and warm seasons, although traffic-related urban primary emissions were significant in both seasons.Along with the mass and chemical characteristics of PM, the results of source apportionment reveal that the aerosol mass concentration of Seoul is largely dependent on the occurrence of high mass events; these were captured in Groups 1 and 2 and are mainly associated with a Chinese outflow.How-ever, local impacts were distinguished by traffic-related urban emissions under stagnant conditions during the warm months, when aerosol mass concentrations were relatively low, but when the carbonaceous contribution was greater.The high contribution from the coal-fired industry, which was not coupled with high PM 2.5 concentrations, highlights the influence from Far East China and North Korea during the cold months.

Fig. 1 .
Fig. 1.(a) Map showing Seoul (which includes Korea University) (triangle) where aerosols were sampled and Dongdaemun-Gu station (cross) where gaseous species in South Korea were measured.(b) Map of South Korea in relation to China and Japan.

Fig. 2 .
Fig. 2. Correlation between PM 10 and PM 1.0 mass concentrations (color-coded circles represent various PM 2.5 concentrations and dashed and solid lines represent linear regression between PM 10 and PM 1.0 and 95% confidence interval).

Fig. 3 .
Fig. 3.For the three groups: (a) mean concentrations of PM 1.0 , PM 2.5 , and PM 10 mass and ratio of PM 2.5 /PM 10 and PM 2.5 /PM 1.0 ; (b) mean concentrations of SO 4 2-, NO 3 -, of PM 10 , PM 2.5 , and PM 1.0 with standard deviation (error bar); (c) mean concentrations of OC, and EC of PM 10 , PM 2.5 , and PM 1.0 with standard deviation (error bar); and (d) mean concentrations of SO 2 , NO 2 , CO, and O 3 .

Fig. 4 .
Fig. 4. (a) Three-day air mass trajectories for Group 1, Group 2, and Group 3, and (b) their vertical variations.One representative backward trajectory was selected for each sample, and concentrations of PM 2.5 mass are color-coded in trajectories.

Table 1 .
Number of samples by group and season (number of haze or dust* impacted samples is provided in parentheses).

Table 2 .
Average concentrations of major chemical constituents in PM 10 , PM 2.5 , and PM 1.0 for three groups.

Table 3 .
Correlation between soluble ions and carbonaceous species in PM 2.5 .

Table 4 .
Source profiles of PM 2.5 estimated from Non-negative Matrix Factorization (NMF) analysis.
*The geographical region of China is given as the six administrative districts: Northern (N), Eastern (E), Northeastern (NE), Northwestern (NW), Central South (CS), Southwestern (SW) China.Season are given as cold months (October-March) and warm months (May-September).