Concentration Variations in Particulate Matter in Seoul Associated with Asian Dust and Smog Episodes

Particulate species were measured in April–May 2003 at two sites, one in the megacity of Seoul and the other in the Deokjeok Island to the west of Seoul, to examine the effects of long-range transport under the influence of prevailing westerlies. The effects of Asian dust were observed in April, and a severe smog episode occurred in late May. During May, all air pollutants showed negative correlations with wind speed at Seoul, particularly for NO2 and CO, a large proportion of which result from vehicle emissions. SO4 continued to increase in association with an inflow of air pollutants from China with heavy use of coal, with fluctuations depending on wind speed. The smog episode in late May occurred as emissions from Siberian forest fires were superimposed on pollutant inflows from China that had persisted since early May and local emissions accumulated under stagnant conditions. During the episode, Siberian forest fires increased K while local emissions primarily from vehicles increased NO3, OC, and EC. The effects of an inflow of air pollutants from the outside were significant at Deokjeok, with small local emissions, resulting in substantial increases in Ca during the Asian dust event and SO4 during the smog episode, compared to those at Seoul. Because both sites were strongly influenced by the effects of long-range transport in May, PM2.5 along with SO4 and K exhibited a strong correlation between Seoul and Deokjeok.


INTRODUCTION
The so-called dust belt extends from the west coast of North Africa and over the Middle East to China (Prospero et al., 2002).Biomass burning, including forest fires, is active immediately north and south of the dust belt (Giglio et al., 2006a).As Northeast Asia is located in downwind areas under the influence of prevailing westerlies, the effects of biomass burning emissions along with fugitive dust are significant.In general, the timing of their effects on the Korean Peninsula slightly varies due to differences in spatial distributions of the emissions (Bates et al., 2004;Clarke et al., 2004;Maxwell-Meier et al., 2004).However, their effects could occur at the same time when emissions lie along the same air trajectory, as illustrated by a sharp spike of organic carbon (OC) concentrations during an Asian dust (AD) event (Nguyen et al., 2015).
The smog episode on May 20-24, 2003 was rare in Korea in that large-scale combustion emissions from Siberian forest fires were introduced into Korea, leading to significant degradation in visibility.Lee et al. (2005) first raised the possibility of smog occurrences caused by Siberian forest fires based on satellite and ground-based remote sensing data along with back-trajectory analyses.Jeong et al. (2008) and Youn et al. (2011) reported changes in radiative forcing and meteorological variables such as surface temperature, pressure, and precipitation rates due to the effects of Siberian forest fires.In and Kim (2010) considered biomass burning and fugitive dust emissions from other areas such as China and Korea as well, to explain the variations of concentrations in May 2003.Recently, Jung et al. (2016) demonstrated that the haze episode in late July 2014 was consecutively impacted by anthropogenic emissions from China and smoke plumes from Siberian forest fires using remote sensing data and chemical analyses.
While various studies have dealt with AD in Korea (Choi et al., 2001;Arimoto et al., 2006;Kim, 2008;Choi et al., 2014), studies for air pollution episodes dominated by forest fires are less common.Except for the Siberian forest fires in May 2003 and July 2014, only Park et al. (2013a, b) reported elevated concentrations of PM 2.5 due to forest fires that occurred in southern China in early February 2011 and early April 2012.Associated with Siberian forest fires, a maximum daily-average PM 10 of 138 µg m -3 was observed at Gwangju in the southwestern part of the Korean Peninsula in late May 2003 (Lee et al., 2005), which was higher than 117 µg m -3 in early February 2011 and 94 µg m -3 in early April 2012.Of note is that the distance to this region from the general vicinity of Lake Baikal, presumed as a source of emissions caused by Siberian forest fires, is more than twice the distance from southern China.It is known that emissions from huge forest fires rise to the upper troposphere or lower stratosphere by pyroconvection and have a considerable influence on the hemispheric scale (Fromm and Servranckx, 2003;Damoah et al., 2006;Monks et al., 2009).Many studies have predicted that due to global warming, boreal forest fires at high latitudes will increase, suggesting that their potential impacts are highly likely to become an important issue in the future (Tchebakova et al., 2009;Pechony and Shindell, 2010;Shvidenko and Schepaschenko, 2013).
In this study, concentrations of particulate species were measured at two sites in the spring of 2003: one in Seoul and the other in Deokjeok Island as an upwind site of Seoul given the general prevailing westerlies.The variations in pollutant concentrations at Seoul in May 2003 were first analyzed to examine how the effects of forest fires evolved, as air pollutants from China were transported into the study area by prevailing westerlies of seasonal character.The concentration variations in the megacity of Seoul and in Deokjeok Island were primarily influenced by AD in April, and by anthropogenic emissions from China as well as Siberian forest fires in May.Next, these concentration variations were compared to differentiate the effects of long-range transport and local emissions.

Measurement Sites
As shown in Fig. 1, Deokjeok Island is located about 330 km east of the Shandong Peninsula in China, about 70 km west-southwest of Seoul.The island has an area of 21 km 2 and, as of 2013, had a population of about 1,400.Forest land accounts for about 90% of its area, and there is a peak at an elevation of 300 m to the north.The measurement site is located at the rooftop of a one-story building at an elevation of 160 m above sea level (37°13′ N, 126°09′ E) around the top of the easternmost hill.The island has a small-scale diesel-fired power plant and an incinerator on the lower ground, and there are mobile emissions from a small fleet of vehicles and marine vessels.However, in most cases, the effects of local emissions on the measurement site are not noticeable because of its high elevation and relatively small emissions in and around the island (Ghim et al., 2003).
Another monitoring site is located at the rooftop of a three-story building within the Korea Institute of Science and Technology (3760N, 12705E, 52 m above sea level), in the northeasten part of Seoul (see Fig. 1).Vehicle emissions can be assumed to affect the measurement site, considering that there are several roads nearby including a four-lane road each way about 350 m to the northwest (Kim et al., 1999;Ghim and Kim, 2004).However, although it is located in the megacity of Seoul, the measurement site is not directly influenced by vehicle emissions from the nearby roads because a 140-m hill lies to the east and an intervening 50-m hillock lies between the site and the road to the northwest.

Sampling and Analysis
Samples were collected using a low-volume sampler with a flow rate of 16.7 L min -1 for 24 hours starting from 9 a.m.To determine total mass and ion concentrations, PM 2.5 was sampled by a Teflon filter (Zefluor) using a filter pack (Savillex #6T-473) equipped with a cyclone (URG 2000-30EH) having a cut size of 2.5 µm.Total suspended particles (TSP) were also sampled by a Teflon filter (Zefluor) using a filter pack (Savillex #0-47) to investigate the effects of AD and/or other fugitive dust.PM 10 was sampled for OC and elemental carbon (EC) analysis by a quartz filter (Gelman) using a filter pack (Savillex #6T-471) equipped with a cyclone (URG 2000-30ENB) having a cut size of 10 µm.While PM 10 was sampled, the concentrations of OC and EC were assumed to be the same as those of PM 2.5 because most of these two fractions are present in PM 2.5 (Watson et al., 1998).
Teflon filters used to determine mass concentrations were kept in a desiccator at 40% relative humidity for more than 24 hours before and after sampling, and they were weighed using an electric balance (Mettler MT5) with a resolution of 1 µg.Three anions (Cl -, NO 3 -, and SO 4 2-) were analyzed by ion chromatography (Dionex 2000i/sp), while four cations (Ca 2+ , Mg 2+ , K + , and Na + ) were analyzed by atomic absorption spectroscopy (Hitachi ZR8200).The ammonium ion (NH 4 + ) was analyzed by a spectrophotometer (Spectronic Genesys2) after color development via the indophenol method.When analyzing the standard solution repetitively, the relative error for the ammonium ion was highest at 6.6%, while those for other ions using ion chromatography and atomic absorption spectroscopy were 0-4.7%.Carbonaceous components, such as EC or OC, were analyzed via the selective thermal oxidation method using MnO 2 catalyst at the AtmAA in Calabasas, California, an accredited laboratory in the United States (Fung, 1990).

Other Data Use
One-hour average concentrations of gaseous pollutants from 25 stations of the urban air monitoring network scattered across Seoul were used in this analysis.Measurement methods were: the non-dispersive infrared method for CO, chemiluminescent method for NO 2 , and pulse ultraviolet fluorescence method for SO 2 .Based on the KME (2011) guidelines, CO concentrations were rounded to one decimal place (0.1 ppm), while NO 2 and SO 2 concentrations were rounded to the whole number (1 ppb).The daily average over Seoul was obtained by averaging the daily average at each monitoring station.While daily averages for particulate species were obtained from sampling from 9 a.m. for 24 hours, those for gaseous pollutants were calculated from 1hour averages from midnight for 24 hours.For meteorological A backward trajectory analysis was conducted to track air masses into the Korean Peninsula using the National Oceanic and Atmospheric Administration (NOAA) Air Resources Laboratory's Hybrid Single Particle Lagrangian Integrated Trajectory (HYSPLIT) 4 model (Draxler et al., 2012).The meteorological data fields used were from the National Centers for Environmental Prediction (NCEP) final (FNL) dataset on 2.5° by 2.5° grids at 6-hour intervals.The vertical motion of trajectories was calculated using an isentropic process.Trajectories starting at a height of 1.5 km above the measurement site at 0000 UTC (09:00 local time) were traced back every hour for 4 days.

RESULTS AND DISCUSSION
Fig. 2(a) shows the variations of annual-average PM 2.5 in Seoul during the 2002-2008 period (Ghim et al., 2015).Annual-average PM 2.5 levels rank the highest in 2002, when AD was the most severe in intensity and frequency, but the highest in 2003 if AD days are excluded.An AD day is defined as the day that meets the KMA (Korea Meteorological Administration) criteria for an AD event (explained later in this paper).Monthly-average PM 2.5 is compared in Figs.2(b) and 2(c) for including and excluding AD days, respectively.The monthly averages in March-April 2002 are higher than the 2002-2008 average, regardless of the inclusion of AD days.However, monthly averages for excluding AD days are lower (by more than 15 µg m -3 ) than those for including AD days, which shows that a significant proportion of higher concentrations was influenced by AD.Monthly-average PM 2.5 levels are mostly high in the first half of 2003, notably with a peak in May, but low in April despite the influence of AD.
Table 1 summarizes measurement data for particulate species at Seoul and Deokjeok for April and May 2003.The number of measurement days is identical to that of samples at Seoul because samples were collected for 24 hours every day.In contrast, samples were collected for 48-96 hours three times at Deokjeok, and one day in April and five days in May fall short.With some exceptions, mean concentrations of particulate species are higher in May than in April, at Seoul than at Deokjeok, and thus concentrations are generally highest at Seoul in May and lowest at Deokjeok in April.

Analysis of Concentration Variations at Seoul in May 2003
Fig. 3 shows variations in daily-average concentrations   ) biomass burning (Watson et al., 1998).Concentrations of EC and gaseous pollutants increased but not as much as the other species noted here (Figs.3(c) and 3(d)).Small variations of EC are attributed to contributions of vehicle emissions, showing small variability of emission patterns in urban areas (Bond et al., 2004;Wang et al., 2012).In Fig. 3, one of the distinctive features in the variations of air pollutants is that increases of concentrations under low wind speeds occur periodically at intervals of 7-10 days.To look at the variations more closely, correlations between key meteorological parameters and concentrations of air pollutants were examined in Table 2. Wind speed is negatively correlated with all air pollutants, especially with gaseous pollutants.Among gaseous pollutants, negative correlations with NO 2 and CO are evident, a large proportion of which is emitted from vehicles.The correlation of particulate species with wind speed is mostly weak; a lack of correlation between K + and wind speed in particular indicates lesser effects of local emissions.
PM 2.5 exhibits a strong correlation with secondary ions, such as NO 3 -, SO 4 2- , and NH 4 + , because they account for a large proportion of PM 2.5 .Among particulate species, K + is  highly correlated with PM 2.5 , following secondary ions.In general, K + shows a high correlation with secondary ions, and more so with SO 4 2-than with NO 3 -.In contrast, correlations of PM 2.5 with NO 2 among gaseous pollutants and with EC among particulate species are relatively weak, both of which are mainly emitted from vehicles.PM 2.5 seems to be more affected by secondary formation and/or long-range transport than by local emissions.
It has been reported that the incidence of forest fires and affected areas in Russia has increased since 1990 (Davidenko, 2000).Forest fires occurred far more frequently in May 2003, compared to the same period of 2001 and 2002 (Giglio et al., 2006b), and notably in Siberia to the north of the Korean Peninsula (Fig. 4).Fig. 5 shows the synoptic meteorology at 850 hPa from May 17 to 22. On May 17-18, pressure gradient intensified behind the low pressure system due to blocking of a following high pressure system.Wind speeds around Lake Baikal, where numerous forest fires broke out, increased along tightly packed isobars, and air pollutants from forest fires were rapidly transported toward Manchuria.On May 19-20, strong northwesterly wind brought these pollutants to the Korean Peninsula.On May 20, a high pressure system positioned over central China and slowly migrated to the Korean Peninsula.On May 21-22, air    pollutants mostly became trapped during stagnant conditions caused by subsidence inversion associated with the high pressure system over the Korean Peninsula.Meanwhile, some persistent westerly wind transported air pollutants from China as well.
Fig. 6 shows backward trajectories that arrive at Seoul in May 2003.Before the smog episode of May 20-24, trajectories continued to flow into Korea from the direction of China.Trajectories passed directly over the areas of forest fires in the north of Manchuria on May 21-22, the most severe smog episode.In the latter part of the smog episode, trajectories slowly passed over the areas of smallscale fires around the Korean Peninsula (Fig. 4; In and Kim, 2010).After the smog episode, air pollutant concentrations declined substantially due to an inflow of clean air from the southeastern direction, i.e., Pacific Ocean and a small amount of precipitation (Fig. 3(a)).
In summary, the smog episode in late May 2003 occurred as emissions from Siberian forest fires were superimposed on emissions from China that persisted into the episode period, and on local emissions that accumulated under stagnant conditions caused by light winds.This episode contrasted with Park et al. (2013a) (2) biomass burning-induced pollution with high K + ; and (3) complicated pollution that combined the previous two pollution types.The smog episode in this study can apparently be classified as complicated pollution.However, during the episode described by Du et al. (2011), despite a strong correlation between secondary ions and K + , K + concentrations were not high due to long transport distances of emissions from biomass burning.In this study, K + concentrations were high because the Korean Peninsula was directly affected by biomass burning albeit a long-range transport.
The OC/EC ratio that was used as an indicator of biomass burning (Park et al., 2013a;Jung et al., 2016) was not that high, averaging 4.34 with a range of 2.99-5.86 during the smog episode.This was because greater effects of local emissions under stagnant conditions led to higher EC concentrations as well, although OC concentrations were substantial.

Comparison of Concentration Variations between Seoul and Deokjeok Island
In Korea, the occurrence of an AD event is determined by examining several elements, including visual identification, PM 10 concentration spikes, movement of air currents, and dust generation in source regions (KMA, 2007).Since the turn of the century, the occurrence of an AD event was reported to be least frequent in 2003(NIER, 2015)).In 2003, the AD event was observed at Seoul for three days: one in March and two in April (April 12 and 13).
The AD of April 2003 originated from the Tengel and Ordos Deserts in north-central China and affected the Korean Peninsula from noon on April 12 to the afternoon of April 13 (KME and NIER, 2004).The cold low pressure system that settled around Manchuria migrated south to the Korean Peninsula (Fig. 7).High winds on the back side of the system and westerly or northwesterly wind direction constitute general features of the occurrence of the AD event.However, its strength was very weak, and consequently only weak AD was observed around the central area of the Korean Peninsula in the afternoon of April 12, which persisted to late afternoon of April 13 nationwide, and then began to dissipate at night.Fig. 8 shows relative concentrations of TSP, PM 2.5 , and key species by period at Seoul and Deokjeok, with respect to overall averages for April and May 2003.First, the level of contribution of sea salt was examined by analyzing the composition of seawater because Deokjeok is an island.However, sea salt accounted for only a small portion of the components, such as SO 4 2-, Ca 2+ , and K + , at 0.4, 4.4, and 1.4%, respectively, indicating that they were mostly of anthropogenic origin (Kim et al., 1998;Lee et al., 2001).Thus, concentrations of these species were used without subtracting portions of sea salt, as shown in the previous figures and tables.
As illustrated in Fig. 8, relative concentrations at Deokjeok are notably higher than those at Seoul for Ca 2+ during the AD event and for SO 4 2-during the smog episode.Ca 2+ is a key species of AD, while SO 4 2-is known to represent the effects of coal use in China (Lu et al., 2010;Liu et al., 2015).Thus, Deokjeok with its small local emissions is more affected by the inflow from the outside areas than Seoul, which has substantial local emissions.Several coal-fired power plants and shipping on and near the west coast could also be sulfur sources, but these are not considered to contribute a similar variation at Seoul and Deokjeok, as shown in Fig. 8, because these emissions are located between the two sites.
During the smog episode, relative concentrations of PM 2.5 , NO 3 -, OC, and EC at Seoul are greater than not only those at other periods but also those at Deokjeok, indicating the influence of local emissions on these species.It was presumed previously that OC/EC ratio during the episode was not high due to EC increase caused by local emissions, despite effects of biomass burning.However, Fig. 8 showed that OC was also highly influenced by local emissions.A further study might be warranted to estimate the relative contributions of direct emissions and secondary formation to OC level (Lim and Turpin, 2002;Kim et al., 2012;Shon et al., 2012).
It is noteworthy that relative concentrations of TSP and Ca 2+ become elevated during the smog episode at Seoul, although not as great as those related to AD.This is because fugitive dust generation is considerable even during the smog episode with low wind speed, which is opposed to the common understanding that only high wind conditions result in higher fugitive dust generation.However, it was reported that, in European cities, fugitive dust generation tends to be inversely proportional to wind speed due to dilution effects when wind speed is not high (Harrison et al., 2001;Barmpadimos et al., 2012).Choi et al. (2014) explained that, under low wind conditions, dust accumulated on the roads results in increasing fugitive dust generation as dust deposited on the roads is resuspended by vehicle traffic rather than wind.
Table 3 shows the correlation of concentrations between Seoul and Deokjeok for April and May 2003.In April, correlation coefficients for TSP and Ca 2+ are large due to the effects of AD and/or fugitive dust that occurred under high wind and dry weather conditions on the AD days at both Seoul and Deokjeok (Ghim et al., 2015).In May, PM 2.5 along with SO 4 2-and K + , exhibits a strong correlation because of the widespread effects of long-range transport over the study area (Fig. 6).TSP and Ca 2+ in May and PM 2.5 along with SO 4 2-and K + in April also show relatively high correlations, albeit not as high as those in April and May, respectively.It is presumed that both Seoul and Deokjeok were considerably influenced by fugitive dust in May (In and Kim, 2010) while the contribution of coal combustion in China and biomass burning was still significant even in April.
Because of the different surrounding environments between Seoul and Deokjeok, NO 3 -, OC, and EC are mainly  influenced by local emissions and, therefore, show less correlation between the two sites.However, in April, NO 3 showed a negative correlation.Fig. 9 shows correlations between Seoul and Deokjeok for SO 4 2-and NO 3 -in April and May.It can be seen that a higher correlation of SO 4 2in May than in April in Table 3 was because the correlation was maintained at high concentrations during the smog episode.As can be seen from the correlation coefficient of 0.35, the correlation of NO 3 -between Seoul and Deokjeok was low in May.In April, there was a negative correlation between the two sites, as concentrations at Deokjeok decreased but concentrations at Seoul increased, which was also shown during the AD event in Fig. 8. Won et al. (2010) reported a decrease of NO 3 -in PM 2.5 during the severe AD event when concentrations of coarse particles increased sharply because of its migration toward coarse particles (Song and Carmichael, 1999).It can be seen that a similar situation occurred at Deokjeok where the effects of AD were pronounced.

SUMMARY AND CONCLUSIONS
In April and May 2003, particulate species were measured in the megacity of Seoul and Deokjeok Island to the west of Seoul.The influence of prevailing westerlies in April and May was not as strong as that in winter but still strong enough, so Deokjeok was chosen as an upwind site of Seoul.While air pollutants from China moved into the Korean Peninsula along with westerly winds, a severe smog episode occurred in late May and the effects of AD were observed in April.
The smog episode in late May occurred as emissions from Siberian forest fires were superimposed on pollutants inflow from China that had persisted and local emissions primarily from vehicles accumulated under stagnant conditions.Although fluctuating depending on the change in wind speed, SO 4 2-continued to increase in association with an inflow of air pollutants from China with heavy use of coal.During the episode, Siberian forest fires and, to some extent, small-scale forest fires around the Korean Peninsula increased K + , while local emissions from vehicles increased NO 3 -, OC, and EC under stagnant conditions.In general, a significant portion of OC results from biomass burning but the ratio of OC/EC was not that high as a result of larger effects of local emissions during the smog episode of this study.
The effects of an inflow of air pollutants from the outside were significant at Deokjeok Island with small local emissions, resulting in increases in Ca 2+ during the AD event and SO 4 2-during the smog episode.Because both sites were strongly influenced by the effects of longrange transport in May, PM 2.5 along with SO 4 2-and K + exhibited a strong correlation between Seoul and Deokjeok.The correlation of concentrations between Seoul and Deokjeok also showed that the effects of coal combustion in China and biomass burning were important even in April and those of fugitive dust were considerable even in May at both Seoul and Deokjeok, albeit not as prominent as those in May and April, respectively.

Fig. 1 .
Fig. 1.Locations of measurement sites in Seoul and Deokjeok Island along with emission source areas.Approximate locations of the Tengel and Ordos Deserts are also shown.

Fig. 2 .
Fig. 2. Variations in PM 2.5 concentrations during the 2002-2008 period: (a) annual averages; (b) monthly averages when including Asian dust (AD) days; and (c) monthly averages when excluding AD days.Concentrations were averaged over 14 monitoring stations during the study period, where PM 2.5 was continuously measured in Seoul.

Fig. 3 .
Fig. 3. Variations in meteorological parameters and concentrations of air pollutants at Seoul in May 2003.Meteorological parameters were obtained from the Seoul Weather Station.Daily averages of gaseous pollutants were calculated from hourly averages from urban air monitoring stations scattered throughout Seoul.

Fig. 5 .
Fig. 5. Weather map for synoptic meteorology at 850 hPa for 0000 UTC during the period of May 17-22, 2003.The number in the upper left corner denotes the day in 17 , in which the episodes caused by emissions from wildfires in China and local emissions consecutively occurred in February-April 2011, and with Jung et al. (2016), in which the episodes caused by emissions from China and Siberian forest fire consecutively occurred in July 2014.Du et al. (2011) identified three types of episodes for Shanghai in the summer of 2009: (1) secondary pollution with elevated secondary ions;

Fig. 6 .
Fig. 6.Backward trajectories starting from Seoul before, during, and after the smog episode of May 20-24, 2003.Trajectories, starting at a height of 1.5 km over the measurement site at 0000 UTC, were tracked at one-hour interval for four days.Starting dates are indicated for selected trajectories.

Fig. 8 .
Fig. 8. Relative concentrations of particulate species at Seoul and Deokjeok by period, with respect to overall averages for all periods.

Fig. 9 .
Fig. 9. Correlations between Seoul and Deokjeok for SO 4 2-and NO 3 -in April and May 2003.Solid and dashed lines denote the best fit lines for April and May.

Table 1 .
Statistical summary of particulate species concentrations (µg m -3 ) for April and May 2003.for particulate inorganic ions, OC, EC, and gaseous pollutants along with meteorological parameters at Seoul in May 2003.During the smog episode (May 20-24), the visual range decreased from more than 10 km on a typical day to the level of 3-4 km (see Fig. 3(a)).During the same

Table 2 .
Correlation matrix between meteorological parameters and pollutant concentrations at Seoul for May 2003 a,b .

Table 3 .
Correlations of concentration variations by key species between Seoul and Deokjeok for April and May 2003 a .