Diagnosis of dust-and haze pollution-impacted PM 10 , PM 2 . 5 , 2 and PM 1 aerosols observed at Gosan Climate Observatory

4 5 Xiaona Shang, Meehye Lee, Saehee Lim, Örjan Gustafsson 6 Gangwoong Lee, Limseok Chang 7 8 1 Department of Earth & Environmental Sciences, Korea University, Seoul, South Korea 9 2 Department of Applied Environmental Science (ITM) and the Bolin Centre for Climate 10 Research, Stockholm University, 10691 Stockholm, Sweden 11 3 Department of Environmental Science, Hankuk University of Foreign Studies, Seoul, South 12 Korea 13 4 National Institute of Environmental Research (NIER), Incheon, South Korea 14 15 16


Introduction
Dust particles are dominant atmospheric aerosols and account for more than 60% of the total global dry aerosol mass burden (Textor et al., 2006).The emission of dust ranges from 1000 to 3000 Tg yr −1 (Zender et al., 2004).Dust particles are abundant in coarse mode, which is represented by PM 10 .Recently, a new type of dust particle has been observed in submicron size that was long-range transported from dry lake deposits in northern China.These particles are more abundant in salts and mineral constituents compared with typical soil dust (Shang et al., 2018).The high-mass loading of dust adversely affects air quality and causes climate change.In regional to global earth's environment, mineral dust contributes to a radiative forcing of − 0.3 ~ +0.1 W m −2 with a large uncertainty (IPCC, 2007(IPCC, , 2013) ) because emission source, chemical and mineralogical composition, and particle size vary in a wide range (Choobari et al., 2014).Dust particles have also been predicted to reduce the rate of ozone production (Dickerson et al., 1997) and promote new particle formation and growth by mixing with other pollutants (Nie et al., 2014;He et al., 2014).Their role in climate change is further highlighted by modifying cloud microphysical process via cloud droplet activation (Bègue et al., 2015) and CO 2 uptake via ocean fertilization (Pabortsava, et al., 2017).
A large amount of dust is found over arid regions in North Africa, North China and South Mongolia (Choobari et al., 2014).In East Asian desert areas, soil type significantly varies from arid and semi-arid deserts to loess deposits and dry lake deposits, depending on the formation process and climate (Zhang et al., 2009).Dust particles generated from these regions are different in chemical and mineralogical composition (Cheng et al., 2012;Kunwar and Kawamura, 2014;Liu et al., 2011;Tao et al., 2014;Su and Toon, 2011), leading to different environmental effects.For example, African dust contains more iron oxides mineral in the form of hematite as compared with those from Asia (Formenti et al., 2011;2014), implying greater light absorption (Zhang et al., 2015).Among atmospheric aerosols generated from variety of emissions (Geng et al., 2014).EC is a main species absorbing light and measured as black carbon (BC) (Han et al., 2010;Saleh et al., 2014).In Northeast Asia, aerosols from various sources are often mixed together while being transported (Kim et

2007
) and dust plumes have been identified as enhanced concentrations of sulfate, nitrate, ammonium, OC or EC over the Korean peninsula (Shin et al., 2015).
An array of AERONET and satellite observations is a vital tool for investigating atmospheric aerosol at regional and global scales (Zhuang et al., 1992;Zhang et al., 2003;Jin et al., 2016;Nazari et al., 2016).Because these techniques measure the extinction of bulk aerosols, it is crucial to accurately estimate the optical property of major aerosol constituents (e.g., Kim et al., 2007) in order to assess the anthropogenic contribution to total aerosol burden and its climate effect using remote sensing data.However, it is still a big challenge to estimate the optical property of main aerosol types especially dust particles in East Asia, because their property is not only dependent on source regions (Huang et al., 2014) but also modified during transport through aging and mixing with pollutants (McFarlane et al., 1992;Von Salzen et al., 2005;Bäumer et al., 2007;Kim et al., 2011).For instance, the mixture of dust and pollutant caused a significant increase of radiative forcing by 0.06 Wm −2 (Zhang et al., 2013a;Mishra et al., 2010;Li et al., 2012).In this context, it is important to figure out the extent of dust particles in the atmosphere and distinguish them from bulk aerosol particles.
For instance, the ratio of PM 10 /PM 2.5 has been employed to eliminate the effect of soil dust from PM 2.5 when determining the mass absorption coefficient of OC (Chung et al., 2012).
In this study, to diagnose the impact of airborne dust and anthropogenic pollution on atmospheric particulate matter and find out feasible criteria, the five-year measurements of mass, water-soluble ions, and carbonaceous compounds for PM 10 , PM 2.5 , and PM 1 were analyzed using statistical methods.The long-term and composite measurements of particulate matter in different sizes are scarce in the study region.The results of mass mode analysis and principle component analysis will provide insight into the role and significance of mineral dust and haze pollution on particulate matter and further its control strategies in northeast Asia.

Methodology
Aerosol samples were collected separately for PM 1 , PM 2.5 , and PM 10 onto 37 mm Teflon and Quartz filters (Pall, Corp.) using sharp-cut cyclones (URG, USA) at the Gosan Climate Observatory (GCO) from 2007 to 2012.Sampling was undertaken for a period of 24 h from 10:00 to 10:00 the next day.A total of 152 sets of samples were collected and analyzed for water-soluble inorganic ions and carbonaceous compounds.Details about the measurement methodology can be found in Lim et al. (2012Lim et al. ( , 2014)).During the five-year period, five dust and eleven haze events were recorded across Korea by the Korea Meteorological Administration (KMA) mostly during the cold seasons from late fall to spring (Fig. 1).
Teflon filters were conditioned for 24 h in desiccators (SANPLATEC, Japan) under a relative humidity of approximately 30%-40% and weighed before and after sampling using an analytical balance (Denver, Germany).Water-soluble species were extracted from the filters into a solution comprising a mixture of 19 mL distilled water and 1 mL methanol.Watersoluble ions, including Cl − , NO 3 − , SO 4 2− , Na + , NH 4 + , K + , Mg 2+ , and Ca 2+ were analyzed via ion chromatography (IC 25, Dionex, USA).For this analysis, 500 µL of sample was injected by an auto-sampler into the AG11 and AS11 columns for anions or CG11 and CS11 columns for cations (Dionex, USA).The eluent and suppressor were 39 mM KOH and ASRSIIULTRA-4 mm and 20 mM MSA and CSRSIIULTRA-4 mm for anions and cations, respectively.Finally, concentrations were determined using a conductivity detector (Dionex, USA), which was calibrated against eight aqueous standards.The detection limit, defined as O 2 /He atmosphere before the reflected light returned to its initial value (Lim et al., 2012;2014).

Measurement overview
From August 2007 to December 2012, the average PM 10 , PM 2.5 , and PM 1 mass concentrations of all measurements were 30 µg/m 3 , 19 µg/m 3 , and 14 µg/m 3 , respectively (Table 1).The PM 10 mass was almost equally partitioned between <1 µm and 1-10 µm.Moreover, the mass of particles between 1 µm and 2.5 µm was considerable and comprises 26% of PM 2.5 mass.As summarized in Table 1, SO 4 2− and OC were the most abundant, followed by NH 4 + and NO 3 − .These four species accounted for 48%, 58%, and 69% of the PM 10 , PM 2.5 , and PM 1 mass, respectively.Of these species, SO 4 2− , NH 4 + , and EC were pre-dominant in PM 1 , which corresponds to more than 75% of those in PM 10 .In comparison, about 65% of OC was partitioned into PM 1 .It was even less for NO 3 − as 33%.It is well known that NO 3 − is more abundant in coarse mode particles due to high affinity to soil mineral.It is also noteworthy that a substantial amount of OC and EC (20 %) was associated with particles between 1 μm and 2.5 μm.Of OC sub-components, OC3 and OC4 were mainly associated with coarse particles.It is evident that Na + and Cl − were highly enriched in coarse particles between 2.5 μm and 10 μm.
The high PM 10 mass was usually observed in the spring along with increased concentrations of Ca 2+ and Mg 2+ (Fig. 1).In comparison, the mass of PM 1 was higher in the cold season (fall to winter) when the concentrations of SO 4 2− , NO 3 − , NH 4 + , K + , OC, and EC were highly elevated.High PM 2.5 concentrations were observed in both winter and spring periods.Overall, the three particle masses and their major constituents were highly elevated when dust and haze events occurred.were higher than those of dust event days.
3 PCA analysis of PM 10 , PM 2.5 , and PM 1 Principle component analysis (PCA) was conducted for all measured species of PM 10 , PM 2.5 , and PM 1 aerosols including water-soluble ions, OC, EC, and mass for the whole period.PCA analysis identifies the correlation of variables through orthogonal transformation and summarizes the main characteristics of measurement data set.In PCA analysis, two components are usually selected.In this study, the principle component 1 and 2 accounted for more than 60% of the total variance of PM 10 , PM 2.5 , and PM 1 .The principle component 1 (PC1) was composed of high loadings for SO 4 2− , NO 3 − , NH 4 + , K + , OC, and EC, especially in PM 1 (Fig. 2).These six species contributed almost equally to the PC1, which explained 46% of the total variance.In contrast, the principle component 2 (PC2) explained 16% of the total variance and was characterized by high loadings for Na + , Cl − , Mg 2+ and Ca 2+ , mainly in PM 10 and PM 2.5 .Interestingly, the principle component 3 (PC3) comprising 9% of the total variance, was associated with high loadings for NH 4 + and Ca 2+ , particularly in PM 10 (Fig. 2 and 3).
These three independent factors explain more than 70% of the total variance.
As the most abundant species, SO 4 2− and NO 3 − concentrations were highly correlated with PC1 loadings for all three-size particles (Fig. 3), confirming that the PC1 represents the influence of anthropogenic pollution sources (Zhang et al., 2013b).So were OC2 and EC1 that have been reported to originate from biomass combustion sources (e.g., Lim et al., 2012).
In PC2, the loadings for Ca 2+ , Mg 2+ , Na + , and Cl − were the highest and well correlated with OC4 concentration in PM 2.5 and PM 10 , which used to be elevated upon dust events (Lim et al., 2012).In saline dust, the concentrations of Ca 2+ , Na + , and Cl − were enhanced concurrently with OC sub-component (Zhang et al., 2014;Shang et al., 2018;O'Dowd et al., 2004;Griffith et al., 2010).The sea-salt contribution of Ca 2+ was estimated to be 12% in PM 2.5 and 19% in PM 10 , assuming that sodium was derived solely from sea salt.In this study, the measurements of water-soluble ions demonstrate that the contribution of sea salt species was found to reach the maximum in summer when aerosol loading is at its minimum under influence of marine air.Thus, the PC2 represents the impact of dust particles including alkaline soils.NH 4 + concentration was moderately related to PC3 loadings in PM 2.5 and PM 10 .In particular, a relatively good correlation of NH 4 + with Ca 2+ in PM 10 indicates the agricultural influence due to fertilizer use.It is noteworthy that PC3 loading was high in spring and summer when the concentrations of particulate matter were low with reduced continental outflows.In addition, the PC3 loadings increased with time, reaching to the highest in 2010.The recent studies also reported that in China, NH 3 emission was increased due to fertilizer application and NH 4 + concentration was higher in spring and summer than the other seasons (Warner et al, 2017;Kang et al., 2016).
Therefore, the three principle components manifest the main sources of particulate matters in the study region.As anthropogenic sources, PC1 is predominant in PM 1 and PM 2.5 .PC2 demonstrates the influence of soil dust on PM 10 and PM 2.5 .Fertilizer use is likely responsible for the variance of PC3.In order to estimate the contribution of these three factors to the mass of PM 10 , PM 2.5 , and PM 1 at GCO, multi-linear regression analysis was conducted using factor loadings, of which result is given below: PM 10 (µg/ 3 ) = 31.1 + 4.7 PC1 + 3.7 PC2 + 4.1 PC3 (r = 0.89, P = 0.03) PM 2.5 (µg/ 3 ) = 19.2+ 3.1 PC1 + 0.4 PC2 + 1.9 PC3 (r = 0.95, P = 0.03) PM 1 (µg/ 3 ) = 14.8 + 2.5 PC1 − 0.7 PC2 + 1.6 PC3 (r = 0.93, P = 0.04) , where PC1, PC2, and PC3 are factor loadings.The intercepts of these three equations are equivalent to the average concentrations for PM 10 , PM 2.5 , and PM 1 (Table 1 and 2).It confirms that the three PCs are sufficient enough to explain the variation of aerosol masses observed at GCO.It is evident that PC1 is a dominant factor determining the particulate mass of PM 10 (63%) as well as PM 1 (99%) and PM 2.5 (90%).
PC2 was most evident in PM 10 (36%) and not negligible in PM 2.5 (9%).It is worthy emphasizing that NH 4 + factor was distinguished as PC3, even though its contribution was the least.In addition, the very small or negative loading of PC2 for PM 2.5 and PM 1 suggests the scavenging of anthropogenic pollutants on dust particles.

Diagnosis of dust and haze
While soil dust has been recognized as a main driver for high PM 10 mass in northeast Asia (Yang et al., 2009), air pollution events are typically distinguished by the concentrations of PM 2.5 (EPA, 2012).In Korea, the aerosol mass concentrations have been often elevated upon Asian dust or haze occurrence.In this context, mode analysis of PM 10 , PM 2.5 , and PM 1 mass concentrations was conducted to diagnose the impact of dust and haze particles on particulate matter.The frequency distributions of all PM 10 , PM 2.5 , and PM 1 measurements are shown in Figure 4.For the three-size aerosol masses, the main-mode concentrations are comparable to the median concentrations (Table 3 and Fig. 4).The main-mode concentration of PM 10 and PM 2.5 was 25 µg/m 3 and 16 µg/m 3 , respectively, which are much lower than those of national standard of annual mean of 50 µg/m 3 and 25 µg/m 3 , respectively.The main-mode concentration of PM 1 (11 µg/m 3 ) was similar to the air quality guideline of the World Health Organization (WHO) for PM 2.5 (10 µg/m 3 ) (WHO, 2006).Of PM 2.5 mass, the contribution of mineral dust was estimated to be ~10% in previous section, which is equivalent to about 2 µg/m 3 .
The mean concentrations of the three types of particulate matters were higher than their median and main-mode concentrations and the standard deviations were comparable to the median concentrations.These results show that mass concentrations varied in a wide range due to high concentration events.For PM 10 , the mean+σ of 52 µg/m 3 was close to the national standard of PM 10 annual average concentration.While the mean+σ concentration of PM 2.5 was higher by 28% than the national standard of 25 µg/m 3 , the mean+σ of PM 1 (25 µg/m 3 ) met the annual standard of PM 2.5 concentration.The mean+σ concentrations of PM 10 , PM 2.5 , and PM 1 were commensurate with the 90 th percentiles that generally represent the highest concentration of the long-tern measurements.
In Korea, dust occurrence is determined by eye observation and haze is recorded when RH is less than 75 % and visibility is between 1 km and 10 km.The concentrations of individual dust and haze samples are presented in the bottom of Figure 4.While the PM 10 and PM 2.5 concentrations of five dust events are placed in the range above the mean+σ, all of the high PM 10 concentrations were not observed on dust days.In contrast, the concentrations of all haze samples were over the mean+σ of PM 1 (Fig. 4).
When PM 10 and PM 2.5 mass belong to the top 10%, their Ca 2+ and Mg 2+ concentrations were also within the highest 10 % of the entire measurements.Particularly, Ca 2+ concentration (0.7 µg/m 3 ) was 3 times as high as the average concentration for both PM 10 and PM 2.5 (Table 1), implying that dust effect is not negligible in PM 2.5 .
At GCO, the five-year measurements of aerosol mass and chemical composition reveal that the top 10 % of PM 10 , PM 2.5 , and PM 1 mass was affected by dust or haze plumes and their effect is traceable by the 90 th percentile mass concentrations of particulate matter.If PM 10 or PM 2.5 mass concentrations are above the 90 th percentile, airborne dust particles played a substantial role in mass enhancement, regardless of the occurrence of event.Likewise, anthropogenic pollution is a main driver for enhanced PM 1 and PM 2.5 mass concentrations if their concentrations are greater than the 90 th percentile.For PM 2.5 , the impact of mineral dust should be considered in northeast Asia region downwind of the dust belt.

Conclusions
At GCO, filter samples for PM 1 , PM 2.5 , and PM 10 were collected and their mass, watersoluble inorganic ions and carbonaceous compounds were analyzed from 2007 to 2012.For the entire period, the average concentrations of PM 10 , PM 2.5 , and PM 1 were 30, 19, and 14 μg/m 3 , respectively.PM 2.5 accounted for 63% of PM 10 , while PM 1 comprised 74% of PM 2.5 on average.
From the principle component analysis using all measured species for PM 10 , PM 2.5 , and PM 1 , the three principle components (PC1, PC2, and PC3) were distinguished, which explained 46%, 16%, and 9% of the total variances, respectively.The PC1 representing the effect of anthropogenic pollution was characterized by high loadings of SO 4 2-, NO 3 -, NH 4 + , K + , OC, and EC.The PC2 was distinct with high loadings for Ca 2+ and Mg 2+ that originate from soil dust.Although the contribution was low, the PC3 was significant for two reasons.First, the loadings of PC3 showed an increasing tendency over time.In addition, the PC3 loadings were discernible during warm season, in contrast to other two components that explain the variations of mass and major constituents of aerosol during cold season.The multiple regression using the three PC loadings shows that the anthropogenic pollution accounted for 99 % and 63 % of PM 1 and PM 10 mass variation, respectively.The effect of soil dust was the largest on PM 10 (36%) and not negligible on PM 2.5 (~10%).
The mode analysis of PM 10 , PM 2.5 , and PM 1 mass concentrations demonstrates that the main mode was commensurate with the median concentration and the mean + σ was comparable to the concentration of the 90 th percentile.It indicates that the average mass concentration is highly susceptible to high-concentration episodes.Consequently, the mean+σ is suggested as a robust criterion that determines the substantial impact of soil dust or pollution plumes on PM 10 , PM 2.5 , and PM 1 .Furthermore, the results of this study reveal that in northeast Asia, non-combustion sources such as soil dust with impose constraints to the reduction of PM 2.5 as well as PM 10 concentrations and raise questions about the efficacy of yearly average concentrations as environmental standards.
Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-721Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 4 September 2018 c Author(s) 2018.CC BY 4.0 License.Five dust events took place in March, May, November, and December for the entire experiment period.Upon dust incidence, the daily average mass concentrations of PM 10 and PM 2.5 were enhanced by 3 and 2 times, respectively.In comparison, concentrations of mass, secondary ions, and carbonaceous compounds were elevated more than two times in PM 1 during the haze events from October to April.On March 20, 2010, dust and haze event occurred concurrently, leading to a maximum PM 10 concentration of 199 µg/m 3 .PM 10 concentrations were occasionally elevated without an official report of KMA on dust occurrence.For instance, in March 2008, the concentrations of PM 10 mass, Ca 2+ , and Mg 2+ Atmos.Chem.Phys.Discuss., https://doi.org/10.5194/acp-2018-721Manuscript under review for journal Atmos.Chem.Phys.Discussion started: 4 September 2018 c Author(s) 2018.CC BY 4.0 License.

Figure 1 .
Figure 1.Time-series variations of major constituents of PM 10 , PM 2.5 , and PM 1 for the entire experiment [µg/m 3 ].Spring and winter periods are shaded in orange and gray.

Figure 2 .
Figure 2. The results of Principal Component Analysis of all measured species including mass, water-soluble ions, OC, and EC for PM 10 , PM 2.5 , and PM 1 .

Figure 3 .
Figure 3. Correlations between the three principle component loadings and major species concentrations [µg/m 3 ] for PM 10 , PM 2.5 , and PM 1 .

Figure 4 .PM
Figure 4. Frequency distributions of PM 10 , PM 2.5 and PM 1 mass concentrations for all 516 measurements.Mass concentrations are given as ln values in x-axis.The green 517 lines stand for mean+σ.The individual samples collected during dust or haze 518 events are marked as different symbols along the x-axis.519