Characteristics and Receptor Modeling of Atmospheric PM 2 . 5 at Urban and Rural Sites in Pingtung , Taiwan

Suspended particles of PM2.5 in air were sampled concurrently at an urban site and a rural site in Pingtung County in southern Taiwan, in the spring, the summer and the fall of 2005. All samples were analyzed to identify eight water-soluble ions, carbonaceous contents, and 19 metal elements. Measurements reveal that the overall means of PM10 (and PM2.5) are 59.2 (47.4) μg/m at Pingtung (urban) site, and 63.6 (45.7) μg/m at Chao-Chou (rural) site. Although both sites exhibited strong correlations (R = 0.98 at Pingtung, and R = 0.78 at Chao-Chou) between PM10 and PM2.5 masses, the mean PM2.5/PM10 ratio was 0.81 at Pingtung, higher than 0.68 at Chao-Chou, suggesting that relatively large bare lands and outdoor burning on farms may have caused more coarse particles to be present in PM2.5 at a rural site (Chao-Chou) than at an urban site (Pingtung). Results of CMB (chemical mass balance) modeling show that the main contributors to PM2.5 mass at Pingtung are vehicle exhaust (49.3–62.4%) and secondary aerosols (SO4, NO3 and NH4) (31.2–37.8%), while those at Chao-Chou are the outdoor burning (25.3–50.4%) of agricultural waste, secondary aerosols (27.2–34.3%) and vehicle exhaust (12.0–26.9%), depending on the seasons.


INTRODUCTION
anthropogenic sources (e.g., industry, vehicles and coal combustion) and natural sources (e.g., volcanic eruption, wildfire and marine aerosols).However, secondary aerosols have particles that are formed in the atmosphere by chemical reactions of gaseous components (gas-toparticle conversion) (Hinds, 1982).The most important secondary aerosols are sulfate (SO 4 2-), nitrate (NO 3 -) and ammonium (NH 4 + ), formed mainly by photochemical reactions of the precursor gases, including sulfur dioxide (SO 2 ), nitrogen oxides (NO x = NO + NO 2 ), ammonia (NH 3 ) or nitric acid (HNO 3 ).Some of the particulate organic carbon is formed from reactions that involve volatile organic compounds (Meng et al., 1997;Chow et al., 1998;Seinfeld and Pandis, 1998).Importantly, fine particles PM 2.5 (particles with an aerodynamic diameter of under 2.5 m), have attracted much interest since 1990s, because they can easily be inhaled and deposited in the respiratory organs of the human body and are thus very detrimental to human health (Needleman et al., 1990;Little, 1995;Oberdorster et al., 1995;Schwartz, 2000).The speciation and concentrations of PM 2.5 are therefore critical in understanding their source contributors and developing effective methods of reducing their atmospheric levels.Currently, Taiwan's EPA (Environmental Protection Administration) has been establishing several PM 2.5 monitoring networks around the country and expects to regulate PM 2.5 in the near future.
Pingtung County (with a population of around 0.91 million and an area of around 2,775 km 2 ) is located at the southern end of Taiwan (Fig. 1).Except in some densely populated areas in the northern parts of the county, with several small industrial parks, it is mainly agricultural, with touring and sightseeing.However, the air quality in the northern (e.g., Pingtung city) and central parts (e.g., Chao-Chou town) of Pingtung County is as bad as that in Kaohsiung metropolitan area, despite the fact that their population densities and emissions are much lower than those of the neighboring areas, mainly because the northern and central parts of Pingtung County are south or southwest of, and thus downwind of, the Kaohsiung City and Kaohsiung County, whenever a northerly or north-easterly wind prevails, including in autumn and winter (Chen et al., 2003;2004).However, the concentrations and compositions of PM 2.5 in Pingtung County have not been measured, which are essential for estimating possible source contributions and/or assessing potential health risks.
This work presents measurements of concentrations and constituents in PM 2.5 , including water-soluble ions, carbonaceous contents and metal elements, at an urban site and an agricultural town in Pingtung County, made in the spring, summer and fall of 2005.The CMB (chemical mass balance) receptor model (Watson et al., 1997) was then applied to identify the potential source contributions to PM 2.5 mass.

Sampling sites and periods
The sampling sites are located in Pingtung city (urban site) and Chao-Chou town (rural site), in the northern and central parts of Pingtung County, respectively (Fig. 1).Pingtung city, with a population of about 216,222 and an area of 65.1 km 2 , at a distance of about 22 km from the west coast, is the capital of Pingtung County.Chao-Chou is primarily an agricultural town, with a population of about 57,189 and an area of 42.4 km 2 and is at a distance of about 14 km from the west coast.Taiwan's EPA has set up air-quality monitoring stations at the two sites.

Methods of sampling and analysis
At the Pingtung and Chao-Chou sites, a manual dichotomous sampler (G241 Model, Graseby Anderson) and a Universal Air Sampler (Model 310, MSP Corporation), packed with Teflon and Quartz filters, respectively, were adopted.The dichotomous sampler was maintained at a total air flow rate of 16.7 L/min-15 L/min for PM 2.5 and 1.7 L/min for PM 2.5-10 (particles with diameters of between 2.5 and 10 m).The Universal Air Sampler was maintained at a total air flow rate of 285 L/min-270 L/min for PM 2.5 and 15 L/min for PM 2.5-10 .Notably, the concentration of PM 10 is the sum of those of PM 2.5 and PM 2.5-10 .Before sampling, filters were placed in box at temperature of 25 ± 3°C and relative humidity of 40 ± 5%, and then weighed after two days.An electrical balance with a precision of 0.01 mg was adopted to weigh the blank filters and particles.
Eight water-soluble ions including Na Recovery efficiencies were measured and analyzed using diluted samples spiked with known quantities of the studied ions.Recovery efficiencies of 93-107% were obtained.Both field and laboratory blank samples were prepared and analyzed; all data were corrected with reference to a blank.The coefficient of determination, R 2 , of the calibration lines made by electrical conductivity measurements exceeded 99.9% for all species.The detection limits of the analysis for each species, when converted to atmospheric concentration in ng/m 3 , were 28 (Na + ), 48 (K + ), 52 (Mg 2+ ), 42 (Ca 2+ ), 108 (NH 4 + ), 186 (Cl ), 114 (NO 3 ) and 246 (SO 4 2 ).
Total carbon (TC) and elemental carbon (EC) were analyzed using an elemental analyzer (TOC-500A, Shimadzu).A quarter of each sample filter was heated in advance in a 340°C oven for 100 min to expel the organic carbon (OC) content, and then fed into the elemental analyzer to determine the EC content.The heating time adopted herein (100 min) was close to that (120 min) adopted by Cachier et al. (1989) and longer than the reduced heating time (43 min) used by Lavanchy et al. (1999).Another quarter of each sample filter was fed directly into the elemental analyzer without any pretreatment to determine the TC concentration.OC was then determined by subtracting EC from TC: OC = TC EC.Similar methods were used by others, for example, Cadle and Mulawa (1990), Harrison et al. (1997) andChoi et al. (2004).
The detection limits of the analysis for OC and EC were 0.024 g.Notably, since O associated with OC mostly exists in carbonyl compounds in atmosphere, their concentrations are usually much smaller than those of alkane compounds.Also, since mass contribution by H is also much smaller than that by C, the mass contribution to OC due to H and O was therefore neglected here.Nineteen metal elements were analyzed by Then, the digested solution was diluted to a volume of 25 mL using ultra-pure water (specific resistance 18.3 M cm) to perform the metal analysis of Ag, Al, Ba, Ca, Cd, Cr, Cu, Fe, K, Mg, Mn, Na, Ni, Pb, S, Si, Sr, Ti and Zn by ICP-MS and ICP-AES.The calibration was made using multi-element (metal) standards (certified reference materials (CRMs); Spex, Metuchen, USA) in a 1% (v/v) HNO 3 solution.
Every tenth sample was spiked with liquidstandards with particular amounts of identified metal elements (Baker Co.).The efficiency of acid digestion was in 90-95% (Allen et al., 2002;Lin et al., 2005).The CRMs were also used as quality control standards.
Analytical drift was monitored throughout the procedure.Recovery efficiencies were determined and analyzed using a diluted sample spiked with a known quantity of metal.Recovery efficiencies from 80 to 120% were achieved.The detection limits of the analysis for metal elements, when converted to atmospheric concentration in ng/m 3 , were 67.15 Calculations of the contributions of sources were based on the CMB8 receptor model (Watson et al., 1997).This model comprises a solution to linear equations that express the concentration of each species at a receptor site as a linear sum of products of source profile abundances and source contributions.The source profile abundances (or fingerprints) and the concentrations of the species must first be (Na), 0.45 (Mg), 2.88 (Al), 83.50 (Si), 250.0 (S), 3.85 (K), 9.93 (Ca), 0.63 (Ti), 0.26 (Cr), 0.20 (Mn), 1.06 (Fe), 0.17 (Ni), 0.07 (Cu), 0.62 (Zn), 1.25 (Sr), 0.16 (Ag), 0.17 (Cd), 0.19 (Ba) and 0.13 (Pb).

Source profiles and sensitivity analysis
Multivariate factor analysis was adopted to help identify dominant source categories, from which three principal factors were obtained that explained about 61% and 56% of the total variance at the Pingtung and the Chao-Chou sites, respectively (Chen et al., 2006).Based on the dominant species in individual factors and current source profiles, the three principal sources are considered to be vehicle exhaust, marine aerosols and incinerators in Pingtung, and vehicle exhaust, outdoor burning of agricultural plants and marine aerosol in Chao-Chou.A pool of 12 profiles associated with these three pollution sources was selected (Table 2); five files (Marin, Auto1, Auto5, and Incer1) were obtained from the local data, and others were obtained from the SPECIATE 3.2 library files (USEPA, 2002).
Prior to performing receptor modeling, sensitivity analyses were performed to select quasi-optimal combinations of source profiles from the pool to generate overall best model performance (Fujita et al., 1994).Table 3 presents typical sensitivity results for the daytime (07:00 19:00) data on 8 March 2005 at Pingtung and Chao-Chou sites, indicating that the "Base" case yielded better results than the other five cases.Therefore, the CMB results presented herein are all derived from the six source profiles (Marin, Auto2, Auto5, Amsul, Amnit, Incer2) in the "Base" case at the Pingtung site in Table 3; and from the six source profiles (Marin, Auto1, Auto5, Amsul, Amnit, Burn2) in the "Base" case at the Chao-Chou site.Notably, the mass fractions in Amsul (ammonium sulfate, (NH 4 ) 2 SO 4 ) are 72.7% for SO 4 2 and 27.3% for NH 4 + , while those in Amnit (ammonium nitrate, NH 4 NO 3 ) are 77.5% for NO 3 and 22.5% for NH 4 + .

Concentrations of PM 10 and PM 2.5
Figs. 2(a) and 2(b) plot the concentrations of PM 10 and PM 2.5 and the PM 2.5 /PM 10 ratio vs. the sampling date, for the Pingtung and Chao-Chou sites, respectively.The results were obtained over three seasons (spring, summer, and fall) in 2005.The highest concentrations of PM 10 and PM 2.5 were 104 and 91.6 g/m 3 on March 9 at Pingtung site, and 140 and 122 g/m 3 on March 8 at Chao-Chou site.The overall means of PM 10 and PM 2.5 were 59.2 and 47.4 g/m 3 respectively at Pingtung, and 63.6 and 45.7 g/m 3 respectively at Chao-Chou.The concentrations of PM 10 and PM 2.5 in the early spring, fall and winter were about two to three times those in the late spring and summer, because fall, winter and/or early spring are typically the worst periods of air quality in southern Taiwan, in which dryness, little rain and low mixing heights reduce the ability of the atmosphere to dilute the airborne pollutants (Chen et al., 2004).
However, unlike the PM 10 or the PM 2.5 mass, the PM 2.5 /PM 10 ratio varied insignificantly with the season.On average, PM 2.5 /PM 10 was 0.81 ± 0.07 at Pingtung site, and 0.68 ± 0.  reported results.For instance, the PM 2.5 /PM 10 ratios were 0.41-0.81at 14 sites in the Central California (Chow et al., 1996;1999), 0.71-0.77at eight sites in and around metropolitan Philadelphia (Burton et al., 1996), 0.57-0.71at seven sites in Kaohsiung (Lin, 2002), about 0.72 at an urban site and a remote site in the eastern United States (Vukovich and Sherwell, 2002), 0.61-0.78at two sites in Hong Kong (Ho et al., 2003), 0.63-0.77at five sites in Nanjing (Wang,  (Gehrig et al., 2003), 0.63-0.73 in Taichung (Fang et al., 2003), 0.45-0.73 at three sites in Beijing (Sun et al., 2004), and 0.4-0.6 in Santiago in Chile (Koutrakis et al., 2005).Many factors, including traffic and industrial activity, emission patterns/strengths and meteorological conditions affect the PM 2.5 /PM 10 ratios.The PM 2.5 /PM 10 ratios are typically (though not always) low (< 0.6) near traffic sites, in relatively dry areas or in the seasons with relatively strong winds such that coarse particles (particles with diameter larger than 2.5 m) are inclined to being re-suspended in the air (Sun et al., 2004;Koutrakis et al., 2005).Notably, although daytime concentrations of PM 10 , PM 2.5 or PM 2.5 /PM 10 ratios may differ from those at night (not plotted), the bulk differences between the two periods were negligible (with a difference of about ± 3%) at each site.

Water-soluble ions and carbonaceous contents in PM 2.5
Figs. 3(a) and 3(b) present the concentrations of eight water-soluble ions vs. the sampling date at the Pingtung and Chao-Chou sites, respectively, indicating that the total concentrations of water-soluble ions generally followed the variation in PM 2.5 .Table 4 presents the mass fractions of individual constituents (including OC and EC) in PM 2.5 mass for all samples at both sites.Both sites had the same leading abundant species in PM

CMB model
Figs. 4(a) and 4(b) are box plots of the percentage source contributions to the PM 2.5 mass for three seasons (spring, summer and fall) at Pingtung and Chao-Chou sites, respectively.The figures reveal that six sources were identified with each site, although source categories and their contributions differ somewhat between the sites.They are marine salt, gasoline vehicles, diesel trucks, ammonium sulfate, ammonium nitrate and incinerators at Pingtung site, and marine salt, diesel trucks, ammonium sulfate and ammonium nitrate at Chao-Chou site, along with motorcycles and outdoor burning, rather than gasoline vehicles and incinerators.
At the Pingtung site, as in Fig. 4

CONCLUSIONS
Measurements reveal several similarities and differences between an urban site and a rural site.At both sites, the concentrations of PM 2.5 (also PM 10 ) varied seasonally, being worst during the fall, the winter or early spring, but the PM 2.5 /PM 10 ratios varied insignificantly with the season.The abundant species in PM 2.5 mass were secondary aerosols (SO 4 2 , NO 3 and NH 4 + ) and carbonaceous species (OC and EC), followed by metals; together, they represented about 82% of the PM 2.5 at each site.
Although both sites exhibited strong correlations (R = 0.98 at Pingtung, and R = 0.78 at Chao-Chou) between PM 10 and PM 2.5 masses, the mean PM 2.5 /PM 10 ratio was 0.81 at Pingtung-higher than 0.68 at Chao-Chou, indicating that relatively large bare lands and outdoor burning on farms may have caused more coarse particles to be present in PM 2.5 at a rural site (Chao-Chou) than at an urban site (Pingtung).
The similarities and differences indicated above were also revealed by the CMB calculations.For instance, at both sites, vehicle exhaust and secondary aerosols were dominant contributors to PM 2.5 mass, while marine aerosol contributed only 3.8 to 8.2%.However, the contribution of vehicle exhaust (49.3-62.4%)at Pingtung (urban) site was about 50 to 100% higher than (12.0-26.9%)at Chao-Chou (rural) site, depending on the season; the contribution (25.3-50.4%) of the

Fig. 2 .
Fig. 2. Mean concentrations of PM 10 and PM 2.5 and PM 2.5 /PM 10 vs. sampling date in 2005 at (a) the Pingtung site, and (b) the Chao-Chou site.

Fig. 3 .
Fig. 3. Mean concentrations of eight water-soluble ions in PM 2.5 vs. sampling date in 2005 at (a) the Pingtung site, and (b) the Chao-Chou site.

Fig. 5 .
Fig. 5. Scatter plots of calculated vs. measured concentrations for five species at (a) the Pingtung site, and (b) the Chao-Chou site.Solid line represents the best fit.
Fig. 4(b), outdoor burning (25.3-50.4%)and ammonium sulfate (16.5-25.3%)are the two leading contributors in the three seasons.In particular, the contribution of outdoor burning in the summer (47.6%) and the fall (50.4%)are thought to be related to the field-burnings of agricultural waste, usually observed during the reaping periods in Chao-Chou town.The next contributors are ammonium sulfate (7.8-11.3%),diesel trucks (1.8-19.4%),motorcycles (5.7-12.3%)and marine aerosol (6.3-8.2%) at the Chao-Chou site.In summary, outdoor burning (25.3-50.4%),secondary aerosols (27.2-34.3%)and vehicle exhaust (12.0-26.9%)from diesel trucks and motorcycles are the most important contributors at the Chao-Chou site.The scatter plots in Figs.5(a) and 5(b) present the correlations between the calculations and measurements of five dominant species, EC, OC, NO 3 , SO 4 two sites.The R 2 is 0.71 0.99 at Pingtung, and 0.74 0.99 at Chao-Chou, indicating strong consistency between the calculated and the measured values.

Table 1 .
Meteorological conditions at the two sites in Pingtung County in 2005.Data were collected from the Hsiung-Kong station of Central Weather Bureau in Kaohsiung city. *

Table 2 .
Possible PM 2.5 source profiles in CMB model.

Table 3 .
Sensitivity of CMB results to combinations of source profiles.

Table 4 .
Mass fraction and standard deviation (SD) of eight water-soluble ions and carbon contents (OC and EC) in PM 2.5 at the Pingtung and Chao-Chou sites.

Table 5 .
Mass fractions and standard deviation (SD) of 19 metal elements in PM 2.5 at the Pingtung and Chao-Chou sites.