Journal Archive

The metropolitan city of Ulsan, located in the southeastern part of the country, is the largest industrial city in Korea with a population of more than 1.1 million. There are three national-scale industrial complexes (ICs) in Ulsan, including a petrochemical IC, a non-ferrous metal IC and a mechanical and shipping IC, as well as four smaller local-scale ICs.

Airborne PM samples were collected from a residential area located within 100 to 120 m of an eight lane highway and a busy traffic rotary (Fig. 1). The average speed of the vehicles on the highway was 90 km/hr. The traffic rotary had an average traffic density of 5,000 vehicles/hr during the day. The rotary and the highway also had relatively high densities of diesel trucks and buses. There were many apartment complexes located within 100 to 200 m of the sampling site.

Fig. 1. 
A location map of the study area.

The PM samples were obtained using an eight-stage cascade impactor (Tisch Environmental, Inc.) placed 12 m from the ground on the roof of a local official building and set with a constant flow rate of 28.1 L/min. Daily PM sampling was based on a 24 hr measurement period and 16 daily samples were collected each month during the spring and summer (April to August, 2008). Glass fiber filters were used to collect particles in nine size intervals (<0.4, 0.4 to 0.7, 0.7 to 1.1, 1.1 to 2.1, 2.1 to 3.3, 3.3 to 4.7, 4.7 to 5.8, 5.8 to 9.0 and 9.0 to 10.0 μm).

The filters were stored in an environmentally controlled chamber (20°C±2°C, 45±5% RH) for 24 hrs prior to weighing before and after sampling. The weights of the filters were accepted only after triplicate consecutive weights within ±2 μg were obtained. The gravimetry of the samples was determined using an analytical balance with a precision of 1 μg. In order to determine the concentrations of the metallic elements, a mixture of hydrochloric (HCl) and nitric acid (HNO₃) was used to extract the metals from the filters. Each filter was cut into smaller pieces (0.5mm×2 mm) and put into a vial with 30 mL of the mixture (HCl : HNO₃=1 : 1). Metallic elements were extracted using an ultrasonic bath at 90°C for 120 min. The extracts were filtered into beakers using a micro filter with 45 μm pores and brought to a final volume of 50 mL using de-ionized water. The concentrations of 13 metals, including Al, Ca, Na, K, Mg, Mn, Cd, Ni, Cr, Pb, Zn, Cu and Fe, were analyzed using an Inductively Coupled Plasma-Atomic Emission Spectrophotometry (IPC-AES). The relative standard deviation (RSD) was less than 2% for the concentrations of all of the elements measured using the prepared standard solutions of the trace reference materials accredited by the US National Institute of Standards and Technology (NIST). Each solution was measured four times to obtain the concentrations of the elements. In addition, the detection limits of the metals were 1.67, 0.33, 0.58, 1.52, 1.14, 0.26 and 4.17 (ng/mL) for Ni, Cd, Cr, Cu, Fe, Mn and Pb, respectively, and 7.31, 0.03, 5.95, 1.64, 1.32 and 1.76 (ng/mL) for Al, Zn, K, Na, Mg and Ca, respectively. The reproducibility of the analyzed metals were 1.247, 0.797, 1.524, 0.941, 0.840, 0.990 and 1.867 (%) for Ni, Cd, Cr, Cu, Fe, Mn and Pb, respectively, and 0.639, 0.806, 0.817, 0.825, 0.887 and 0.697 (%) for Al, Zn, K, Na, Mg and Ca, respectively. Blank filters and dynamic blanks were exposed to the ambient air for approximately 10 sec at the sampling site and then were handled, stored and treated identically to the collected PM filters. Triplicate blank tests were conducted using the same analysis method as that for the actual samples.

The enrichment factors (EF) of the metallic elements were utilized to identify the origin of the elements in the atmosphere. Fe was selected as a reference element for the crust based on previous studies (Wang et al., 2006; Samara and Voutsa, 2005; Manoli et al., 2002). Natural sources contribute significantly to the composition of PM when the EF goes to unity and non-crustal sources are the predominant sources when the EF >10 (Wang et al., 2006). The crustal average concentrations were obtained from Taylor’s study (Taylor, 1996). The EF_crust of element M in PM was obtained using the following formula:

where M equals the concentration of the metal of concern, [M/Fe]_Air is the concentration ratio of each metal depending on the size of each sample and [M/Fe]_Crust is the concentration ratio of each metal in the crust.

Principle components analysis (PCA) was used to obtain knowledge of emission inventories prior to this study (Chio et al., 2004). In the present study, the number of factors was selected so that the cumulative percentage variance explained by all of the chosen factors was greater than 56% and only those factors with an eigenvalue greater than 1 were chosen. A varimax rotation that maximized the variance of the squared elements was completed using SPSS software (version 13).

Cluster analysis (CA) was used to study the emitting sources of the different heavy metals found in the samples. Even though it is difficult to interpret dendrograms of unclear sample groups, we compared CA results with PCA results in order to be more confident in the final identification of sources. In this study, the groups were chosen based on dendrograms using Ward’s method including the squared Euclidean distance and an agglomeration schedule.

Fig. 2 reveals seasonal changes in the average PM concentrations of nine fractions. The highest PM levels during the spring (April and May) were observed in the largest PM (9 to 10 μm) followed by the finest size (0.4 to 0.7 μm), one fine particle fraction (0.7 to 1.1 μm) and one coarse fraction (5.8 to 9.0 μm). In the summer (June to August), the highest PM levels were also observed in the size range of 9 to 10 μm followed by the fine particle fractions (0.7 to 1.1 and 1.1 to 2.1 μm).

Fig. 2. 
PM concentrations for each size group during the spring and summer periods.

In a comparison of wind characteristics between the spring and summer sampling periods (Fig. 3), stronger winds (with a wind speed of greater than 4 m/s) and more fractions of southern and southwestern wind series occurred during the spring. Therefore, the PM collected during the spring sampling period was expected originate from a greater variety of sources than the PM collected in the summer when a southeastern series of winds prevailed. In addition, the presence of some western series of winds during the spring indicated that the PM collected during the spring would be affected by the Asian dust that usually originates from the west or northwest of the sampling site. In a comparison of PM size distribution between the spring and summer periods, the PM collected in the spring had increased concentrations of coarse particles due to the Asian dust. However, the PM collected in the summer would be affected more by the winds that pass through industrial areas.

Fig. 3. 
Average wind rose at the sampling site during the spring and summer periods.

Table 1 presents the coefficients of determination (r²) between each PM size fraction and the PM₁₀ concentrations during two seasons. In the spring, strong correlations existed between the PM₁₀ and coarse particles. In particular, the PM₁₀ concentrations and the coarse particles with concentrations in the range of 4.7 to 5.8 μm showed the strongest correlation (r²=0.87). The weakest correlation (r²=0.18) was obtained between the submicron particles (0.4 to 0.7 μm) and PM₁₀. These results suggest that coarse particles have a strong influence on the observed variability in PM₁₀ concentrations during the spring. While high correlations were obtained between most fine particles, the coefficients of determination between PM₁₀ concentrations and 0.7 to 1.1 μm and 1.1 to 2.1 μm particles in the summer were 0.76 and 0.73, respectively. The weakest correlation of concentration was between PM₁₀ and coarse particles in the range of 9 to 10 μm (r²=0.30). These results revealed that fine particles have a strong influence on the observed variability in PM₁₀ concentrations in the summer season.

Recent studies have indicated that PM_2.5 is the main component of PM₁₀ from urban locations. Harrison et al. (1997) found that up to 60% of PM₁₀ in Birmingham was composted of PM_2.5. Clarke et al. (1999) reported that approximately 60 to 70% of urban PM₁₀ was composed of PM_2.5. Kim et al. (2000a) and (2000b) studied variations in the PM_2.5 to PM₁₀ ratio within the south coast air basin of southern California and found that 52% to 59% of the PM₁₀ was composed of fine particles (PM_2.5). Fernandez Espinosa et al. (2001) studied the size distribution of ambient aerosols in Seville, Spain, and found that PM_2.5 represented 72% of the PM₁₀. In the current study, however, 41.4% and 41.2% of the PM₁₀ concentrations consisted of PM_2.5 in the spring and summer seasons, respectively (Fig. 4). The fractions of fine particles in this study were much lower than those in other studies. These results indicated that the fraction of coarse particles in the PM₁₀ was much higher than the fractions recorded in other studies. The increase in coarse fractions indicated higher contributions of crustal elements, particles from mechanical sources and suspended particles from the movement of vehicles. The PM collected at the site could easily be affected by traffic related sources such as the abrasion of brake systems, tires and road surfaces, since the sampling site was located near the highway and a busy traffic rotary area.

Fig. 4. 
Contribution of size-based PM components to PM₁₀.

The PM correlations among different size fractions were determined by regression analysis as shown in Table 2. During the spring sampling period, there were strong correlations (r=0.60-0.90, P<0.01) between all of the coarse particles above 2.1 μm. However, all of the fine particles below 2.1 μm and some of the coarse particles less than 3.3 μm were also highly correlated (r=0.57-0.87, P<0.01) during the summer sampling period. Previous studies (Li and Lin, 2003; Vallisu et al., 2000) found that PM_2.5 and PM₁ were highly correlated and therefore could have similar sources. These results indicate that high correlation coefficients for various particle size groups can be related to the same sources. Therefore, these results indicate that there were common or similar PM sources among the coarse particles in the spring and among the fine particles in the summer. For both of the seasons, strong correlations were identified between the size group 5.8 to 9.0 μm and both fine particles (0.4 to 0.7 and 0.7 to 1.1 μm) and the size groups 1.1 to 2.1, 2.1 to 3.3 and 3.3 to 4.7 μm. However, very weak correlations (r=0.08-0.28, P<0.01) were noted in the summer between the size group of 9-10 μm and the other particle size groups. These results indicated that the particles in the 9-10 μm size group have different sources than the other PM size groups. The 9-10 μm size group showed strong correlations with coarse particles in the spring. These results indicated that there was a common source of coarse particles, such as Asian dust, in the spring.

Table 3 shows the results of a factor analysis of the size fractionated PM in PM₁₀. In spring, factor 1 explains 61.24% of the variance with high loading (0.72) of a fraction of fine particles (1.1-2.1 μm) and also with high loading (0.82-0.96) of coarse particles (2.1 to 10 μm). Coarse particles have been used in the past as markers for road and soil dust (Houck et al., 1990). Therefore, this factor can be identified as road dust and soil dust. Factor 2 explains 18.79% of the variance with high loading of submicron particles in the size groups of 0.4 to 0.7 μm (0.91) and 0.7 to 1.1 μm(0.85). Submicron particles are used as indicators of emissions from diesel truck exhaust and oil combustion (Houck et al., 1990). Therefore, this factor can be used to identify both diesel truck exhaust and oil combustion sources. Similarly, two factors were also found in the summer. Fuel combustion was responsible for high loading (0.79 to 0.88) of the fine particles (0.4 to 2.1 μm) and most of the coarse particles (2.1 to 9 μm), with the exception of the 9 to 10 μm size group (Kenneth et al., 1998). Factor 2 only explained 13.01% of the variance with high loading of 9.0 to 10 μm (0.94) in the summer, indicating that the factor could be identified as road dust (Houck et al., 1990).

Table 4 shows the EF of the light metals including each classified particle size. The EF values of the light metals (Al, Ca, K, Mg and Na) in all of the size fractions were less than 10. The lowest EF values were for Al (0.33 to 0.78) and Ca (0.66 to 1.8); these values were deficient compared to the minimal enrichment (Sutherland, 1999). K (1.14 to 3.14), Mg (0.55 to 2.48) and Na (3.16 to 8.61) had moderate EF values. EF values below 10 imply that crustal sources are predominant for those particles (Kao et al., 1995). Manoli et al. (2002) reported that emissions from natural crustal sources were found mostly in coarse particle sizes. Wang et al. (2006) and Berg et al. (1994) reported the enrichment of Mg indicated a contribution from sea salts.

Most of the heavy metals (Cd, Pb, Zn, Cu, Cr and Ni) in all of the PM size fractions and Mn in the fine particle size fraction had very high EF values. Fig. 5 shows the change in the EF values of heavy metals as a function of particle size. The highest EF values for the heavy metals were observed within the fine particle size group of 0.4 to 0.7 μm, with the exceptions of Pb and Cu. The highest EFs for Pb and Cu were in the size group of 2.1 to 3.3 μm. The EF values generally decreased with increasing particle size. Cd (EF: 6,971-60,966) was the most enriched element in all of the particle size fractions, followed by Pb (EF: 233 to 1,151), Cr (EF: 69 to 908), Cu (EF: 145 to 362), Ni (EF: 14 to 152) and Mn (EF: 12 to 31). In particular, the EF values of Cd in the fine particles that were smaller than 2.1 μm ranged from 25,000 to 60,966, which indicated a very high level of enrichment. Therefore, urgent countermeasures should be provided to reduce Cd levels in the atmosphere. For the heavy metals noted above, non-crustal sources such as vehicular exhaust, fossil fuel combustion and several industrial processes are considered major emission sources (Sabbioni and Zappia, 1993). Gao et al. (2002) also reported high EF values ranging from 200 to 20,000 for Cu, Pb, Zn and V in PM from the New York-New Jersey harbor estuary.

Fig. 5. 
Enrichment factors (EF) of heavy metals in PM as a function of the PM size.

As indicated by the results of the EF analysis above, the EF values of the heavy metals Cd, Cr, Cu, Ni, Pb and Zn in all of the size fractions and Mn in the size fraction of 0.4 to 2.1 μm were much higher than 10. The majority of these metals in the air came from the emissions of non-crustal sources such as anthropogenic sources. Therefore, we reanalyzed the anthropogenic origins using the principle component analysis and cluster analysis for these elements in the current study. A factor analysis method using the varimax rotated principal component was applied in order to identify the major sources of the heavy metals.

As shown in Table 5, three factors were identified in the PM_3.3-10 (coarse particles). Factor 1 explains 56.37% of the variance with high loading of Cd (0.94) and Cu (0.95). Cd, Pb and As have been used as maker compounds for non-ferrous metal smelters and automobile emissions (Eddie et al., 1999; Fung et al., 1995) and Cu has often been used as an indicator of air emissions from the smelting industry (Harrision et al., 1997). Therefore, this factor can be identified as a non-ferrous metal smelter source. Factor 2 explains 22.15% of the variance and contains high loading of Mn (0.95), Cr (0.67) and Pb (0.65) which can indicate dust from traffic sources (Almeida et al., 2006; Fang et al., 2006). Factor 3 showed high loading of Ni (0.82) and Zn (0.88). Ni and V are used as indicators of emissions from oil combustion (Kowalczyk et al., 1982; Zoller et al., 1974). Zn has been used as a marker for heavy metals from waste combustion in incinerators (Wiles, 1996). Therefore, this factor can be identified as both oil combustion and incineration.

Three factors were also found in the case of PM_1.1-3.3 (fine and coarse particles) accounting for a total of 75.27% of the variance. Incineration and traffic dust were responsible for high loadings of Cr (0.88) and Zn (0.8) in factor 1 (Viana et al., 2008; Fang et al., 2006). Factor 2 in PM_1.1-3.3 could represent a non-ferrous metal smelter source which has a high loading of Cd (0.7), Cu (0.74) and Mn (-0.75) accounting for 19.30% of the total variance (Harrision et al., 1997; Fung et al., 1995). Factor 3 only explained 16.54% of the total variance with high loading of Pb (0.9) and Ni (0.72), suggesting that this factor can be identified as oil combustion and vehicle emissions (Balacchandran et al., 2000; Pio et al., 1989).

For PM_0.4-1.1 (fine particles), two factors were obtained with eigenvalues that were greater than 1 and explained 69.62% of the total variance. Factor 1 explained 39.47% of the variance with high loading of Cd (0.92) and Cu (0.94). Non-ferrous metals smelting, such as that of Zn or Cu, was indicated as the source for this factor (Harrision et al., 1997; Fung et al., 1995). Factor 2 explained 30.14% of the total variance with a high loading of Cr (0.71), Mn (0.86) and Pb (-0.74). The dust from traffic sources was responsible for factor 2 (Fang et al., 2006).

In this study, we tried to compare our cluster analysis results to our principal components analysis results in order to obtain more confident results for the source identification of heavy metals. Fig. 6 shows the dendrograms that resulted from using the Ward method for a cluster analysis of heavy metals. The results of the CA based on the dendograms basically agreed with the results of the PCA. However, there was a slight difference between the two methods for the identification of the source groups (Table 6). For example, factor 2 in the CA included Cr, Ni and Zn (PM_3.3-10) and factor 2 in the PCA only included Ni and Zn. In general, the PCA analysis reduced the number of variables while capturing as much as possible of the original information. Therefore, the variables in PCA, which have similar characteristics, could be arranged into the same factors (Fang et al., 2004). For the cluster analysis, the variables which have short distances in the dendograms could be basically classified into similar or the same source groups. Sometimes, it is difficult to interpret the CA results clearly. Even though the variables look like a similar or the same group based on the CA dendogram feature, they can be located in some distances which may not be considered as a similar source group. This case brings us unclear source identification. However, the CA classification can also provide simplicity when interpreting the relationship among the components using the distance information. Therefore, PCA and CA are both unique methods for the source identification of the materials of interest in terms of providing the possibility of various sources.

Fig. 6. 
Cluster analysis of heavy metals in different size PM.