Source Apportionment of PM 2 . 5 Particle Composition and Submicrometer Size Distribution during an Asian Dust Storm and Non-Dust Storm in Taipei

Asian dust storm (ADS) not only increase the coarse particle concentrations, but also bring the fine and ultrafine particles to Taiwan. In this study, A PCA model was applied to identify the potential source categories, obtained through measuring ambient 10–500 nm particle number concentrations, size distributions and composition data, during an ADS and non-dust periods. The three factors estimated with rotational components during an ADS were vehicular emissions (52%), dust storm (24%), and primarily gasoline vehicles (12%). During non-dust periods, the three factors were vehicular emissions, secondary sulfate and nitrate (40%), combustion processes and traffic-related emissions (29%), and road dust (25%). In addition, vehicular emissions and road dust were the main sources (78%) during particulate matter episodes. The results showed that, along with wind direction and wind speed, PM composition and size distribution can be used to determine the locally dominant source categories, and to identify ADS episodes.


INTRODUCTION
The effects of atmospheric particulate matter (PM) on public health in urban areas have been serious concerns over the past decade.Short-term exposure to PM can cause health problems including respiratory illnesses, pulmonary diseases, heart strokes (Kwon et al., 2002;Pope et al., 2002;Ho et al., 2004;Pope et al., 2004;Miller et al., 2007;Brook and Rajagopalan, 2009;Kazunari et al., 2012).In addition, Chen et al. (2004) showed the strongest probable effects increased 7.7% of respiratory disease 1 day after ADS in Taipei.Most Asian Dust storms(ADS) from northwestern China and Mongolia occur in winter and spring, and dust particles mixed with anthropogenic pollutants such as sulfates and nitrates (Talbot et al., 1997;Wang et al., 2007;Huang et al., 2013), heavy metals (Kim et al., 2003;Zhang et al., 2010), carbonaceous aerosols (Ramana et al., 2010;Stone et al., 2011), biomass burning (Tsai et al., 2012) are transported to China, Korea, Japan and Taiwan.Previous studies have not only found that an ADS greatly increases PM 2.5 and PM 2.5-10 concentrations, but also that gaseous pollutants transported by the northeast monsoon seriously affect Taiwan (Chen et al., 2004;Lee et al., 2006).An analysis of aerosol chemical properties was determined at the Gosan site during ADS, and discovered that the organ carbon (OC) and elemental carbon (EC) concentrations in the pollution episode were approximately 90% and 30%, higher than those in the dust episode ( Kim et al., 2005).
Numerous studies analyzed the crustal and anthropogenic elements of atmospheric aerosols, but few studies simultaneously analyzed particle composition and size distributions, especially within the range 10 nm-500 nm, during Asian Dust storms and non-dust periods.The ADS not only increased the coarse particles concentrations, but also brought the fine and ultrafine particles to Taiwan.The impact of the dust on the ultrafine particle number concentrations has been shown that the large majority of ultrafine particles in urban settings are combustion aerosols from vehicular emissions or local industrial pollution.Thus, understanding the source apportionment in urban area between ADS and non-ADS periods is important.
The present study was conducted over a long period, and data were carefully collected to ensure their reliability.Principal component analysis (PCA) was utilized to reduce the numbers of complicated variables, including mass concentrations of PM 2.5 and PM 10 , EC, OC, total carbon (TC), sulfate, nitrate, black carbon, total PAHs, scattering coefficient, and the number of aerosols and their size distribution.The purpose of the study is to understand PM 2.5 particle composition and submicrometer size distribution during Asian Dust storm and non-dust storm, and to better identify the high-pollution episodes.Meteorological data can be used in conjunction with the methods outlined in this study to determine potential sources in the area.

Taipei Supersite
The Taipei supersite was established by Taiwan Environmental Protection Administration in 2000 and began operations in November of 2001.Fig. 1 displays the location of sampling site.The monitoring station is located between the two major traffic hubs in the Hsinchuang area: the Chungshan Expressway and Highway Two intersection in the north and the Highway One and Chungzheng Road intersection in the south.Both were determined to be potential major sources of traffic pollution.About 9 km northwest of the station are the Wugu industrial area and Linkou's petrochemical park, both of which were believed to be potential sources of industrial pollution.
The PM 2.5 and PM 10 concentrations were measured every six minutes from February 10 to 13 and April 22 to 30, 2009 using a tapered element oscillating microbalance (TEOM R & P 1400a).The TEOM, which was operated at 30°C to reduce volatilization, had a diffusion drier that removed particle-bound water (Lee et al., 2005a, b, c;Lin et al., 2008).The concentration of sulfate was measured every 30 minutes using an R & P8400S, by a method described by Harrison et al. (2004).The concentration of nitrate was measured every 30 minutes using an R & P 8400N.The sulfate and nitrate data were corrected every two weeks for Zero-Span gas calibration and conversion efficiency.Therefore, monthly efficiency correction factors were applied to correct the semi-continuous sulphate and nitrate data.Daily OC/EC measurements were made by the manual thermo-optical analysis of filter samples using a Sunset Laboratory Carbon Aerosol Analysis device.An AE21 Aethalometer (Magee Scientific) was used to collect the PM on a rolled quartz filter with a cellulose fiber backing, and the transmission intensities of light beams at wavelengths of 880 and 370 nm.The photoelectric aerosol sensors (PAS) were Model PAS2000 Real-Time PAHs Monitors, from EcoChem Technologies, Inc. (West Hills, CA.) The measurement techniques on which this instrument was based have been described in detail elsewhere (Burtscher and Siegmann, 1992).The concentrations of NH 3 , NO x , and NO were determined by a Thermo Model 17C NH 3 analyzer.The number concentration and size distribution of particles were measured using a Scanning Mobility Particle Sizer (SMPS 3936,TSI Inc).Table 1 summarizes the data set that has been used in this study.

Methodology
This study defines ADS episode as April 25-26, 2009;April 22-24 and 27-29, 2009 for non-ADS period; 10-13 February, 2009 for PM episode.This paper focuses mainly on ultrafine particle composition and size distribution of particles in ADS episode and non-ADS period, whereby a rotational method was applied on the data set.The original numbers of variables for sampling data were 19, including 12 items of air pollutant concentrations, 5 particle number concentrations and size distribution in the size range from 14 to 500 nm and 2 other variables.Hourly values at individual stations are normalized as Eq. ( 1) where Z ik denotes normalized value of k th observation at i th station, C ik represents k th values at i th station, μ i is mean value at i th station, and S i denotes standard deviation at i th station.The model for rotational principal component analysis is (2) where n denotes number of stations, L ij is factor loading of the i th station on the j th rotational component, and P jk represents component score of k th station for j th rotational component.The rotational principal component can be derived by inverting Eq. ( 2).The varimax rotation method developed by (Kaiser, 1958.)increases the segregation between component loadings, which can more easily define a distinct clustering of data clearly.This technique rotates the predetermined principal components while retaining the constraints that the individual components remain orthogonal to each other (Sylvaine et al., 2011).The rotational technique allows us to divide data into five principal components, accounting for 90% of total variance.Five factors at the supersite were isolated based on the following criteria.
Firstly, the number of factors was selected such that the cumulative percentage variance explained by all the chosen factors is more than 90%.Secondly, the factors with eigenvalue greater than one may undergo varimax rotation to obtain the final factor matrix.High factor loading (> 0.70) of ultrafine particle composition and size distribution of particles in a factor can help identify potential sources, a rotational method was adopted to find the simplest possible factor structure by using Statistical Product and Service Solutions 12.0 software to analyze relevant data.
The second component (PC2) obtains higher loadings for PM 2.5 , SO 4 2-, NO 3 -, OC.Some researches have reported that high concentrations of mineral elements and pollutants, such as sulfate and nitrate, were produced by industry in China (Talbot et al., 1997).Lee et al. (2006) showed that sulfate was the major constituent of PM 2.5 in Taiwan during an ADS event in 2002, indicating that this factor may be influenced by ADS events.China's dust storms brought not only dust but also ultrafine particles and pollutants that were by-products of combustion.Most of these particles are smaller than 200 nm (Shi et al., 1999;Wahlin et al., 2001).Fig. 2(b) also demonstrates that the particle size distribution in the range 250-552 nm obtains higher loadings for PC2.Thus, PC2 can be identified as an ADS source.
The third component (PC3) explains 12% of the total variance and Fig. 2(c) shows higher loadings for particulates in the size range 4-20 nm.Most fresh particles emitted by motor vehicle exhausts are ultrafine particle (< 0.1 µm), and tend to exhibit a size distribution with a nucleation mode (< 20 nm) (Palmgren et al., 2003;Pey et al., 2009), indicating that motor vehicle emissions is the potential source of the third component.

Non-dust Periods
In non-dust periods, 94% of the variance is explained by three factors, which account for 40%, 29%, and 25% of the total variance.The first component explains 40% of the total variance and shows higher loading for NH 3 , NO, NO x , BC, OC, TC and PAHs, (factor loading > 0.70).Fig. 3(a) presents higher loading on particle size distribution at 14 nm-130 nm and indicates that vehicle exhaust emissions are potential source.The second component explains 29% of the total variance and shows higher loading for PAHs.Fig. 3(b) presents higher loading for particle sizes in the range 130 nm-500 nm.This factor may also be associated with combustion processes and traffic-related emissions.PAHs in motorcycle exhaust measured two significant peaks appearing at 56 nm-100 nm and 180 nm-320 nm for most middle and higher molecular weight PAHs.The two peaks may be caused by the combustion process associated with PAHs found mainly in the nucleation mode, and the noncombustion process (unburned fuel) leading to the formation of PAHs primarily in the accumulation mode, respectively (Yang et al., 2005).The third component explains 24% of the total variance and shows higher loadings for PM 10 , PM 2.5 , NO 3 -, SO 4 2-, and the scattering coefficient, indicating that dust from the road.The major sources of PM 10 were traffic dusts, road dusts, coal burning dusts in Harbin (Huang et al., 2010).Shen et al. (2010) noted that Water-soluble ions constituted 14.0% of PM 10 and the dominant ionic species were SO 4 2-and NO 3 -in the roadside samples.Besides, In another PM episode, 83% of the variance is explained by three factors, which account for 48%, 30%, and 5% of the total variance.The first component (PC1) is strongly correlated with concentrations of PM 2.5 , NO 3 -, NH 3 , NO x , OC, EC, TC, BC and scattering (factor loading > 0.70), which mostly come from transportation.Figs. 4(a) and (b) show that the particle size distribution in the range of 60-500 nm and 14-60 nm (factor loading > 0.70) is strongly related to PC1 and PC2.Previous researches have found that the aerosol number distribution may be characterized by three modes at diameters 20 nm, 100 nm and 2 μm.These particle size ranges are strongly related to vehicle exhaust emissions (Schauer et al., 1996) and have been observed on and near the roadway (Whitby et al., 1975;Joumard and Perrin, 1988;Harrison et al., 1999;Kittelson et al., 2000).

Analyzing Component Scores
Component scores are analyzed by applying a weight to each concentration measurement that is proportional to its factor loading.Fig. 5 displays the time series of the three rotated PC scores.According in Figs.5(a) and 5(b), NO x and PAHs concentrations peaked together on 24 April, as did the PC1 score in Fig. 5(c).Therefore, the mainly source of PC1 is vehicular emission.According to Nikolaou et al. (1984), NO x pollution comes mainly from mobile sources while roughly 80% of annual total PAHs emission in urban areas has stationary sources, including industrial processes, power generation, residential heating and open burning.
The PM 10 concentration exceeded 850 μg/m 3 on 25 April 2010, during the arrival of a dust storm.Comparing Fig. 5(c) with Fig. 5(d) shows that principal component score2 peaked at the same time.The principal component score2 accurately reflect ADS periods.Besides, when the wind speed exceeds 2 m/s and the wind comes from the East, the result shows a strong relationship between the second component and the particles sizes in the range 200-500 nm, as shows in Fig. 6.Therefore, ADS not only bring dust but also bring the small particles to Taipei.Fig. 7 shows that pollution levels change with the wind direction.The high levels of pollution 25 April coincided with a change in the wind direction to northerly, and a decrease in wind velocity of less than 1 m/s, because the change in wind direction change brought pollution from the North road, while the reduction in speed caused the air to stagnate.These factors both contribute to high pollution.

CONCLUSIONS
This paper discusses PM 2.5 chemical composition, the size distribution of ultrafine aerosols, gaseous pollutants concentrations and meteorological parameters, and utilizes PCA to identify potential source categories during ADS, PM and non-ADS periods.The major source categories are vehicular emissions, dust storm and primarily gasoline vehicle during Asian Dust periods.In addition, when the wind speed decreases, episodes of high pollution occurs because the vehicular emissions and other local pollutants accumulated.Asian Dust storms coming from China, is strongly related to rising sulfate, nitrate and the particle size range in the 200 nm-500 nm.The results, by coupling wind direction and wind speed parameters, can be used to   better identify the potential source categories.During non-ADS periods, vehicular emissions are also the major potential sources of PC1 and PC2, and the particle sizes are in the range 14 nm-500 nm.Another potential source is road dust.Future work should include an analysis of ionic species and heavy metals in aerosol and seek a more detailed understanding of the contribution of individual sources to air pollution.The procedure used herein could help to quantify the PM contribution of specific sources, and to construct a better index for the study of Asian dust storms.
Fig. 2(a) presents that the particle size distribution in the range 20-200 nm (factor loading > 0.70) is strongly related to PC1.Emissions from diesel fuel vehicles produce particles primarily in the 20-130 nm size range, while gasoline-fueled vehicular emissions produce particles primarily in the 20-60 nm size range

Fig. 2 .Fig. 3 .
Fig. 2. Correlations of three factors (PCs) with size distributions of particles during Asian dust storm periods.

Fig. 4 .
Fig. 4. Correlations of three factors (PCs) with size distributions of particles during PM episode periods.

Table 1 .
Summary of data used in this study.

Table 2 .
Factor loadings between principal components and measured concentrations during ADS and non-ADS periods.