Chemical Fingerprint and Source Identification of Atmospheric Fine Particles Sampled at Three Environments at the Tip of Southern Taiwan

The spatiotemporal variation, chemical fingerprints, transportation routes, and source apportionment of atmospheric fine particles (PM2.5) along the coastal region of southern Taiwan were investigated at three environments in the tip of southern Taiwan. Three representative sampling sites at Chien-Chin (urban site), Siao-Gang (industrial site) and Che-Cheng (background site) were selected for simultaneous PM2.5 sampling from December 2014 to May 2015. Regular sampling of 24-h PM2.5 was conducted for continuous 6–9 days in each month. After sampling, the chemical composition, including water-soluble ions, metallic elements and the carbonaceous content of PM2.5, was further analyzed within two weeks. The levoglucosan concentration was further compared to OC and K in PM2.5 originating from biomass burning. Moreover, the potential sources of PM2.5 and their respective contribution were further resolved by backward trajectory simulation, combined with chemical mass balance (CMB) receptor modeling. The field sampling results indicated that the PM2.5 concentrations at the urban and industrial sites were always higher than those at the background site. The most abundant water-soluble ionic species of PM2.5 are SO4, NO3 and NH4, implying that PM2.5 is mainly composed of secondary ammonium sulfate and ammonium nitrate. The most abundant metallic elements of PM2.5 included crustal elements (Al, Fe and Ca) and anthropogenic (generated by humans) elements (V, Ni, As, Cd, Zn and Pb). Moreover, the concentrations of OC and EC at the Chien-Chin and Siao-Gang sites were generally higher than those at the Che-Cheng site, mainly due to the emissions from urban and industrial anthropogenic sources. Vehicular exhausts and industrial emissions were the main sources of PM2.5 at the Chien-Chin and Siao-Gang sites, respectively, while biomass burning and soil dusts were the dominant sources of PM2.5 at the Che-Cheng site.


INTRODUCTION
Anthropogenic emissions are the main cause for rising atmospheric fine particle (PM 2.5 ) concentrations in Southeast and Northeast Asia, mainly due to rapid urbanization and bustling industrialization in the past decades (Fang et al., 2013), although PM 2.5 is commonly emitted from both natural and anthropogenic sources; it significantly affects ambient particulate air quality.
Similar to many Asian cities suffering from atmospheric visibility degradation and severe health problems, high PM 2.5 episodes frequently occur in southern Taiwan, a highly industrialized area as well as an intersectional region among the southern Taiwan Strait, the northern South China Sea and the Basi Channel (Yuan et al., 2002;Wang et al., 2007;Tsai et al., 2010Tsai et al., , 2011;;Li et al., 2013a, b).Poor particulate air quality in southern Taiwan results not only from local sources, but also from long-range transport of Chinese haze, Asian dust storms, industrial emissions from East Asia, and biomass burning from the Indochina Peninsula (Chuang et al., 2012).
PM 2.5 is defined as particles with aerodynamic diameter of 2.5 µm on less, which is approximately 1/28 the diameter of a human hair and 1/35 of the size of river sands, and can be easily inhaled deep in the lungs and directly penetrate the pulmonary alveolar cells in the blood circulatory system (USEPA, 2013).Moreover, PM 2.5 can be mostly attached by toxic substances such as dioxins, polycyclic aromatic hydrocarbons (PAHs) and heavy metals, causing severe respiratory and cardiovascular diseases due to the long-term inhalation of PM 2.5 .
In addition to the typical composition of fine particles, including water-soluble ionic species, metallic elements and carbonaceous contents, anhydrosugars are another important component of fine particles resulting from biomass burning.Anhydrosugars, mainly including levoglucosan (90.1%), polymannosan (6.6%) and galactosan (3.2%), are primarily derived from the thermal breakdown of the plant building materials cellulose and hemicelluloses (Hawthorne et al., 1992;Rogge et al., 1998;Simoneit et al., 1999;Zhu et al., 2015).They can also serve as good tracers for biomass or bio-fuel burning due to their source-specific generation and atmospheric stability (Simoneit et al., 1999;Giannoni et al., 2012).The anhydrosugars can provide valuable information on the influences of biomass burning on a qualitative level, or allow for conservative quantitative estimates of source contributions, as shown by long-term investigations of levoglucosan concentrations on long-range transport (Simoneit et al., 2004;Di et al., 2013).
Therefore, this study aims to investigate the spatiotemporal variation of atmospheric fine particle concentrations in southern Taiwan.The chemical fingerprints of atmospheric fine particles were further characterized.Moreover, this study characterized the monthly variation of levoglucosan in fine particles to ascertain the influence of biomass burning on atmospheric fine particles in southern Taiwan.The potential sources as well as transportation routes of atmospheric fine particles heading toward southern Taiwan were also resolved during the poor air quality periods.

Sampling Sites
The locations of the three sampling sites: Chien-Chin, Siao-Gang and Che-Chen, selected for this study are shown in Fig. 1.The description of the surrounding environments of these three sampling sites is summarized in Table S-1.The Chien-Chin site is characterized as an urban site located on the roof of Chien-Chin Primary School, about 15 m above ground.Its surrounding environments are mainly residential and commercial districts; the air pollutants come mainly from vehicular exhaust.The Siao-Gang site is characterized as an industrial site located on the roof of Siao-Gang Senior High School about 12 m above ground and 1.5 km from Kaohsiung Harbor and near Kaohsiung International Airport.The Siao-Gang region of Kaohsiung City is the largest industrial complex, mainly metallurgyrelated industries, in Taiwan.The Che-Cheng site located at the tip of southern Taiwan is characterized as a background site regarding the cross-boundary transport; it is situated on the roof of the main building of Boli Country Club about 6 meters above ground, and is surrounded by hills and mountains with limited anthropogenic activities in the neighboring environments.

Sampling and Weighing Methods
For this particular study, the sampling of atmospheric fine particles (PM 2.5 ) was undertaken for 6-9 consecutive days in each month from December 2014 to May 2015, which is part of 7-SEAS sampling campaign starting from January to April in 2015.The purpose of 7-SEAS sampling campaign was to investigate the chemical characteristics and source identification of PM 2.5 affected by local sources and/or long-range transport during the polluted seasons (Winter: December-February; Spring: March-May) in southern Taiwan (Lin et al., 2013).
The instrumental methods for sampling and weighing ambient PM 2.5 are summarized in Table S-2.Twenty-fourhour sampling of PM 2.5 was conducted simultaneously at the aforementioned three sampling sites from 8:00 am to 8:00 am of the sequential day.The sampler used for collecting PM 2.5 in the atmosphere at each sampling site was BGI PQ200 with WINS impactor under an air flow rate of 16.7 L min -1 .Before and after PM 2.5 sampling, the quartz fiber filters of 47 mm diameter were conditioned in the desiccators at constant temperature (25 ± 3°C) and relative humidity (45 ± 5%) for at least 24 h (Yuan et al., 2004(Yuan et al., , 2006;;Li et al., 2013a, b), and further weighted by an analytical microbalance (Sartorius MC 5) with a mass precision of 10 -6 gram.

Chemical Analytical Methods
Before conducting the chemical analysis of PM 2.5 , each filter was divided identically for specific chemical species analyses.The chemical analytical methods are summarized in Table S-2.Among the divided filters, one quarter of the filter was put inside a bottle made of polyethylene (PE) containing 30 mL distilled de-ionized (DI) water and vibrated with an ultrasonic vibrator (Branson, Model 5510) for about 3 h to completely dissolve the water-soluble ions from PM 2.5 filtrates.The major cations (NH 4 + , K + , Na + , Ca 2+ and Mg 2+ ) and anions (F -, Cl -, Br-, SO 4 2-and NO 3 -) dissolved in the D.I. water were then filtered and analyzed with an ionic chromatography (IC: Dionex, Series 100) (Li et al., 2013a).
Another quarter of the filter was initially digested in a 20 mL mixed acid solution (HNO 3 :HClO 3 = 3:7) at 150-200°C for 2 h to extract metals from the PM 2.5 filtrates, and then diluted to 25 ml with distilled D.I. water for metallic content analyses.Fifteen metallic elements (Al, Fe, Na, Mg, K, Ca, Ti, Mn, Ni, Cu, Zn, Cd, Pb, Cr and V) were measured with inductively coupled plasma-atomic emission spectroscopy (ICP/AES; Perkin Elmer, Model 400) (Wu et al., 2015).
Moreover, two eighths of the filters were further used to measure the carbonaceous content (EC, OC and TC) of PM 2.5 , respectively, with an elemental analyzer (EA; Carlo Erba, Model 1108) operating in the procedure of oxidation at 1020°C and reduction at 500°C.Before sampling, all quartz fiber filters were preheated at 900°C for 1.5 h to expel the potential carbon impurities from the quartz fiber filters.The preheating procedure could minimize the background carbon in the quartz matrix, which might interfere with the analytical results, leading to an overestimating of the carbonaceous content of PM 2.5 .One-eighth of quartz fiber filters were initially heated at 340-345°C for at least 30 min to expel the organic carbon fraction, while the other one-eighth of quartz fiber filter was analyzed directly without heating.The former organics we obtained was denoted as elemental carbon (EC), while the latter was presented as total carbon (TC).The amount of organic carbon (OC) in the PM 2.5 was then obtained by extracting EC from TC (Yuan et al., 2004).
The final one-quarter of the filter was stored in an environment of 4°C for further pretreatment and chemical analysis of levoglucosan.The filter samples were extracted with D.I. water (> 18.2 MΩ•cm) inside a bottle made of polyethylene (PE), under ultrasonic vibration for 60 min, followed by the filtration of the extracts (for removal of insoluble components) with a syringe filter containing a precombusted quartz filter (0.3 µm pore size).Typical recovery efficiencies of levoglucosan were 100.2 (± 4.7)% based on spiked filter extractions.Filter samples were analyzed with high performance ionic chromatography (HPIC) (Dionex, ICS-3000) (Engling et al., 2013).The high-performance ion chromatograph was utilized and operated in pulsed amperometric mode; it was equipped with an electrochemical detector (ECD), an amperometric cell and a CarboPak MA1 analytical column (4 × 250 mm).A NaOH solution of 400 mM was used as effluent with a volumetric flow rate of 0.4 mL min -1 .Method detection limits for levoglucosan were estimated as 0.58 ng m -3 .Measurement precision was better than 4.9% for levoglucosan (CV; n = 5).The procedural blanks of levoglucosan were below detection limits.The mass concentrations of PM 2.5 were determined gravimetrically from the pre-weighed quartz fiber filters (after conditioning at 25 ± 3°C and 45 ± 5% RH for 24 h) on an analytical balance with a precision of ± 0.01 mg (Sartorius WMC6014).

Quality Assurance and Quality Control
The quality assurance and quality control of both PM 2.5 sampling and chemical analyses were conducted in this study.Prior to conducting PM 2.5 sampling, the flow rate of each PM 2.5 sampler was carefully calibrated with a film flow calibrator (Sensidyne, Model MCH-01).The filters were then carefully handled and placed on the supporting matrix of the PM 2.5 samplers to prevent potential cracking during the sampling procedure.After sampling, aluminum foil was used to fold the filters, which were then temporarily stored in an environment of 4°C and transported back to the Air Pollution Laboratory at National Sun Yat-sen University for further chemical analysis.The sampling and analytical procedure was similar to that described in previous studies (Cheng and Tsai, 2000;Yuan et al., 2006;Tsai et al., 2008Tsai et al., , 2010;;Lin, 2012).
Moreover, both field and transportation blanks were undertaken for PM 2.5 sampling, while reagent and filter blanks were applied for chemical analyses.The determination coefficient (R 2 ) of the calibration curve obtained for each chemical species was required to be higher than 0.995.Background contamination was routinely monitored by using operational blanks (unexposed filters) proceeding simultaneously with the field samples.The blank interference was insignificant in this study, and thus can be ignored.Duplicate analysis was undertaken for every 10 samples.
At least 10% of the samples were analyzed by spiking with a known amount of metallic and ionic species to determine their recovery efficiencies.
The experimental uncertainty dealt with assessing the errors in the field measurement, the experimental data is designed to determine the errors due to instrumentation, methodology, presence of confounding effects and so on.Experimental uncertainty estimates were further used to assess the confidence in the results.To concern about the accuracy of field measurement and chemical analytical data, the quality assurance and quality control (QA/QC) for both sampling and analytical methods were undertaken throughout the entire field measurement and analytical procedure.The QA/QC data obtained for the analysis of chemical species were summarized in Table 1.

Backward Trajectory Simulation
In order to trace the air masses transported toward southern Taiwan, backward trajectories from the reception sites are commonly used to identify their specific source regions.A hybrid single-particle Lagrangian integrated trajectory (HYSPLIT) simulation is a widely used model that simulates the trajectory of a single air parcel originated from a specific source location and height above ground over a period of time.In this study, the 72-h backward trajectories of air parcels transported towards southern Taiwan in different sampling days were plotted to identify the potential sources of PM 2.5 .

Chemical Mass Balanced (CMB) Receptor Modeling
The source apportionment of ambient PM 2.5 was assessed using a receptor model based on the principle of chemical mass balance (CMB) (Kothai et al., 2008;Yatkin and Bayram, 2008;Li et al., 2013;Li et al., 2015).Since the detailed descriptions of CMB receptor model (e.g., CMB8.0) are available elsewhere, only a brief summary is presented below.The CMB receptor model uses the emission profiles of prominent sources to estimate their contribution to a specific reception site.It is assumed that the total concentration of a particular chemical species at the reception site is the linear summation of each individual contribution from various sources.The CMB receptor model simulation uses the results of the least-square regression analysis of the PM 2.5 's chemical composition to resolve the most appropriate contributions of source apportionment.Therefore, this model consists of a least-square solution to a set of linear equations.This solution expresses each receptor concentration of a chemical species as a linear summation of the products of source profiles and source contributions.Source profiles (the fractional amount of each species in the emissions from each source type) and receptor concentrations, each with realistic uncertainty estimates; serve as the input data to the CMB receptor model.The model output consists of the contribution from each source type to the total ambient PM 2.5 mass, as well as to individual chemical species concentration.The CMB8 model results are evaluated by using several fit indices, such as R 2 (≥ 0.8), χ 2 (≤ 4.0), T statistics (≥ 2.0) and the percentage of mass accounted for 80-120%.The source profiles used in this study were reported by USEPA, : determination coefficient of calibration curve; MDL: method detection limit; RPD: relative percentage difference; *: Units of MDL and Blank: µg m -3 .Southern California Air Quality Study, and the research on the chemical composition of PM 2.5 from local prominent sources in southern Taiwan (Chen et al., 1997;Cheng et al., 2000;Yuan et al., 2003;Li et al., 2013b).

Wind Rose Plots
Meteorological conditions are an important factor affecting the transportation of particulate matter suspended in the atmosphere.The wind roses of each month at the three sampling sites are plotted in Fig. S-1.It shows that the wind fields of Chien-Chin and Siao-Gang sites are quite similar to each other.At the tip (Che-Chen) of southern Taiwan, the prevailing winds blow northerly in the winter season (December 2014-February 2015) and then switch to western winds at the end of the spring season.However, compared to the Chien-Chin and Siao-Gang sites, the Che-Chen site show a different wind field, particularly in March and April 2015.The prevailing winds at the Che-Chen site blew from the northeast from December to April and changed to the west winds in May.

Spatiotemporal Variation of PM 2.5
Atmospheric fine particles, as criteria air pollutant in the ambient air quality standard, are mainly considered as anthropogenic pollutants, although they may also be emitted from natural sources.The monthly variaton of PM 2.5 concentration measured in southern Taiwan is illustrated with error bars in Fig. 2. It depicted that the uncertainty of PM 2.5 concentrations was acceptably low during the sampling periods since the errors of the field PM 2.5 measurement were not too high.
Field sampling results indicated that the mass concentrations of PM 2.5 sampled at the Chien-Chin and Siao-Gang sites were at similar levels, and both were higher than those at the Che-Chen site.Although the Chien-Chin and Siao-Gang sites are situated in two different environments (i.e., urban versus industrial environments), there were almost no differences in the PM 2.5 concentrations.Overall, the mass concentrations of PM 2.5 mostly violated the 24-h ambient air quality standard of 35 µg m -3 , except for those observed in May.Additionally, the observation of much lower PM 2.5 concentrations at the Che-Chen site were basically in accordance with its background environment at the tip of southern Taiwan, when compared to the Chien-Chin (urban) and Siao-Gang (industrial) sites.
Moreover, the seasonal variation of PM 2.5 concentration in southern Taiwan is illustrated in Fig. 2. The monthly variation of PM 2.5 concentration at the Chien-Chin and Siao-Gang sites show that the concentrations of PM 2.5 sampled from December 2014 to March 2015 were much higher than those in April and May 2015.However, there were no significant differences in PM 2.5 concentrations from December 2014 to May 2015 at the Che-Chen site.
The almost total lack of differences of PM 2.5 sampled in April and May 2015 at all three sampling sites in southern Taiwan was mainly attributed to the fact that air masses were migrated from the south and southwest, a relatively cleaner marine region.The results show that the variation of wind direction dominated the concentration levels of atmospheric fine particles at the tip of southern Taiwan.In addition to the PM 2.5 concentration, the mass ratios of PM 2.5 and PM 10 (PM 2.5 /PM 10 ) were mostly higher than 0.5 from December 2014 to April 2015, except for those observed in May 2015, indicating that PM 10 was dominated by PM 2.5 in the atmosphere of southern Taiwan.The lower PM 2.5 /PM 10 in May 2015 concurred well with the change of prevailing wind direction from the north to the west.PM 10 comprised fine particles (PM 2.5 ) and coarse particles (PM 2.5-10 ), i.e., PM 10 = PM 2.5 + PM 2.5-10 .While the mass ratio of PM 2.5 and PM 10 (PM 2.5 /PM 10 ) was higher than 0.5, the mass concentration of PM 2.5 was higher than that of PM 2.5-10 , indicating that PM 10 was dominated by PM 2.5 .Oppositely, while the mass ratio of PM 2.5 and PM 10 (PM 2.5 /PM 10 ) was lower than 0.5, the mass concentration of PM 2.5 was lower than that of PM 2.5-10 , indicating that PM 10 was dominated by PM 2.5-10 .Fig. 2. The tempospatial variaton of PM 2.5 concentration at the soutnern Taiwan.

Chemical Fingerprints of PM 2.5
After conducting the field sampling of PM 2.5 at the three selected sites, the chemical composition of PM 2.5 was further analyzed to characterize their chemical fingerprints in southern Taiwan.Table 2 summarizes the sample numbers and chmical composition of PM 2.5 at the three sampling sites.
The mass concentrations of the ionic species of PM 2.5 sampled in southern Taiwan are illustrated in Fig. 3.In terms of water-soluble ions, The mass concentrations of total watersoluble ions accounted for 38.9-54.7% of PM 2.5 , while the most abundant ionic species of SO 4 2-, NO 3 -, and NH 4 + , namely secondary inorganic aerosols (SIA), accounted for 18.7-42.4% of PM 2.5 , which were in accordance with previous researches (Mangelson et al., 1997).Additionally, high Na + and Cl -concentration in PM 2.5 commonly observed in the southern Taiwan accounted for 3.1-8.4% of PM 2.5 that mainly came from oceanic spray, since the three sampling sites were located along the coastal region of the southern Taiwan.Overall, the concentrations of water-soluble ions in PM 2.5 measured at the Chien-Chin and Siao-Gang sites showed relatively higher levels than those at the Che-Cheng site.Similar trend was observed for PM 2.5 and total watersoluble ion concentrations.Similar to PM 2.5 concentrations, the concentrations of water-soluble ions were relatively high from December 2014 to March 2015 at the Chien-Chin and Siao-Gang sites while compared to the Che-Cheng site, suggesting that the polluted air masses were mainly blown from the western side of the Central Range and relatively clean air masses were blown from the eastern side of the Central Range (see Fig. S-1).Moreover, the prevailing winds changed from the north to the west after April 2015 in the southern tip of Taiwan Island.
The metallic contents of PM 2.5 sampled in southern Taiwan are illutrated in Fig. 4. The PM 2.5 contain metallic elements emitted not only from natural sources, but also from anthropogenic sources.Overall, the mass of metallic elements of PM 2.5 covering a total of fifteen analyzed metals accounted for 9.14%-16.12% of PM 2.5 , which correlated quite well with the PM 2.5 concentration.The dominant metals of PM 2.5 included crustal elements (Al, Fe and Ca) and anthropogenic elements (V, Ni, As, Cd, Zn and Pb).High concentrations of As in PM 2.5 were observed at the Siao-Gang site from December 2014 to March 2015,   indicating industrial sources could have emitted As-bearing primary fine particles from iron work (Tian et al., 2010).Moreover, the Siao-Gang site located near an industrial complex had relatively higher concentrations of Fe, Al and K in PM 2.5 than the other two sampling sites.With a lot of anthropogenic sources from human activities surrounding the Chien-Chin and Siao-Gang sites, the concentrations of Zn, Pb and Cr in PM 2.5 were generally higher than those at the Che-Chang site.
The carbonaceous contents (OC, EC and TC) of PM 2.5 sampled in southern Taiwan are depicted in Fig. 5. Overall, the total mass of carbonceous content (TC) accounted for 1.05-12.8% of PM 2.5 .The mass ratios of OC and EC (OC/EC) ranged from 1.16 to 2.60, with the highest OC/EC ratios observed at the Siao-Gang site in February 2015 among the three sampling sites.The concentrations of OC and EC at the Chien-Chin and Siao-Gang sites were generally higher than those at the Che-Cheng site, mainly due to the immense emissions from anthropogenic sources.Monthly variation of carbonaceous concentration of PM 2.5 show them to be much higher (> 7.5 µg m -3 ) from December 2014 to March 2015 at the Chien-Chin (urban) and Siao-Gang (industrial) sites; however, their OC/EC ratios were relatively low (< 2.0).On the countary, the OC/EC ratios in April and May 2015 were relatively higher (> 2.0), although their carbonaceous concentrations were consistently low (< 3.0 µg m -3 ).Higher OC/EC ratios than 2.0 in southern Taiwan in the spring implied the potential formation of secondary aerosols, also known as aged particles (Tsai et al., 2006).

Levoglucosan versus Relevant Indexes
The monthly concentration variation of levoglucosan sampled at the three sampling sites in southern Taiwan is illustrated in Fig. 6.Consistently high levoglucosan levels observed at all sampling sites in December 2014 suggested cross-boundary transport toward southern Taiwan due to the prevailing northeastern wind in winter.Relatively higher levoglucosan levels observed only at Che-Cheng in March and April 2005 were mainly influenced by biomass burning from local sources and/or long-range transport from Indochina Penninsula (Maenhaut et al., 1996).Moreover, a consistent variation of levoglucosan concentration and levoglucosan/OC mass ratio in southern Taiwan was revealed in this study.Further plots of the correlation of organic carbon (OC) versus levoglucosan are illustrated in Fig. 7.The mass ratio of organic carbon to levoglucosan (OC/Levo) can be used as a valuable indicator to properly distinguish the sources of PM 2.5 and further cluster the three sampling sites into two separate groups (Zdráhal et al., 2002;Mazzoleni et al., 2007;Pio et al., 2008;Schmidl et al., 2008;Engling et al., 2009).The lower mass ratios of OC/Levo were observed at the Che-Cheng site, while the Siao-Gang and Chien-Chin had relatively higher mass ratios of OC/Levo similar to each other.
Fig. 8 illustrates the correlation of potasium ion (K + ) versus levoglucosan.Potassium ion has been widely used as the fingerprint of biomass burning in many previous studies (Echalar et al., 1995;Khalil and Ramussen, 2003).It has been commonly applied to identify the transport or existence of particulate matter emitted from biomass and bio-fuel burning sources.This study also tried to correlate potassium ion with levoglucosan, in terns of the mass ratios of potassium ion and levoglucosan (K + /Levo).The mass ratios of K + /Levo at the sampling sites were highly different, with the order: Siao-Gang site > Chien Chin site > Che-Cheng site.The results indicated that the K + /Levo ratio serves as a much more useful indicator for identifying the sources of PM 2.5 at the three sampling sites.
Figs. 7 and 8 are plotted to differentiate the proper indicators of biomass burning by using the correlation coefficient of organic carbon (OC) versus Levoglucosan (Levo) and potassium ion (K + ) versus Levoglucosan (Levo), respectively.Previous literatures reported that OC, K + , and Levo are valuable indicators for biomass burning.In this study, the correlation coefficients are applied to quantify the linear relationships between any two measured indicators of biomass burning for different sampling site.Statistical analysis results showed that the three sampling sites can be highly differentiated by these three indicators based on the correlation coefficient analysis.Most of the correlation coefficients were higher than 0.6 (i.e., medium positive correlation), some were even higher than 0.7 (i.e., strong positive correlation) (Landis et al., 1977).

Transportation Routes of PM 2.5
In addition to the chemical indexes of OC/Levo and K + /Levo, backward trajectory simulation can be also used to identify the routes of air masses transported in the atmosphere.The backward trajectories of air masses tranported toward southern Taiwan are plotted in Fig. 9.The transportation routes can be clustered into three different categories based on the wind direction.The variable n shown in Fig. 9 represents the number of PM 2.5 samples collected at each sampling site in different seasons for this particular study.While, the percentages in the brackets represent the probabilities of the clustered transportation routes, occurring during the sampling seasons, plotted by the backward trajectory simulation starting from each sampling site for different seasons.
In the winter season (December 2014-February 2015), the most frequent transportation routes toward the Chien-Chin, Siao-Gang, and Che-Cheng sites were Route A (63%-68%), followed by Route C (11-27%) and Route B (5-26%).The most frequent occurrence of Route A indicated that PM 2.5 was mainly transported from Northeast Asia in the winter season.In the spring season (March-May 2015), the most frequent transportation routes toward the Chien-Chin and Siao-Gang sites were Route C (53%), followed by Route A (32-36%) and Route B (5-11%), indicating that PM 2.5 came mainly from the South China Sea or the Philippines.The most frequent transportation routes toward the Che-Cheng site were Routes B and C (42-47%), indicating that PM 2.5 probably came from South China, the Indochina Peninsula and the South China Sea.

Source Apportionment of PM 2.5
A chemical mass balance receptor model (CMB8.0)was used for this particular study to resolve the types of sources and their contribution to PM 2.5 sampled in southern Taiwan.The CMB receptor modeling was based on the chemical composition (i.e., ionic species, metals and carbons) of PM 2.5 .Table 3 and Fig. S-2 summarize the source apportionment of PM 2.5 respectively collected at the Chien-Chin, Siao-Gang and Che-Cheng sites located in southern Taiwan.We herein observed obvious seasonal variations of PM 2.5 's source types and their contribution (Fig. S-2) that are described in this section.
In the winter season, vehicular exhaust (17.27 ± 7.45%) and steel plants (20.30 ± 6.81%) were the major sources of PM 2.5 at the Chien-Chin and Siao-Gang sites, respectively, while biomass burning (16.21 ± 2.62%) was the major  source at the Che-Cheng site.It shows that biomass burning was a significant source causing an increase in PM 2.5 at the Che-Cheng site.In addition to biomass burning, soil dusts (15.90 ± 2.57%) were commonly higher at the Che-Cheng site than those at other two sampling sites, which concurred with previous reports stating that atmospheric particles were highly influenced by biomass burning from local sources and/or long-range transport in the winter and spring seasons (Maenhaut et al., 1996).As an urban site in Kaohsiung City, the major source of PM 2.5 at the Chien-Chin site was vehicular exhaust that accounted for 17.27 ± 7.45% mass of PM 2.5 .A higher contribution of vehicular exhausts to PM 2.5 was always observed at the Chien-Chin site compared to the other two sampling sites since the Chien-Chin site is located in downtown Kaohsiung City with heavy traffics during the rush hours.Industrial sources, such as steel plants, petroleum refinery, coal-fired power plant, etc., contributed the largest portion of PM 2.5 at the Siao-Gang site that is located at Taiwan's biggest industrial complex in the Siao-Gang District of Kaohsiung City.An integrated steel plant and several electric arc furnaces located in the Siao-Gang Industrial Complex emit an immense amount of PM 2.5 into the atmosphere of Kaohsiung City.Among these resolved sources, biomass burning and soil dusts were the major sources of PM 2.5 at the Che-Cheng site, while vehicular exhaust and industrial emissions were the major sources in the urban (Chien-Chin) and industrial (Siao-Gang) sites in Kaohsiung City.
Similar to the source apportionment of PM 2.5 in winter, the major sources of PM 2.5 in spring were vehicular exhaust (32.74 ± 6.39%) at the Chien-Chin site, steel plants (30.03 ± 6.80%) at the Siao-Gang site and biomass burning (15.79 ± 0.83%) at the Che-Cheng site.The contribution of vehicular exhaust at the Chien-Chin site in spring was significantly higher than that in winter.Although vehicular exhaust and industrial sources were the dominant sources in urban (Chien-Chin) and industrial (Siao-Gang) sites, secondary aerosol was another important source causing the increase in PM 2.5 in southern Taiwan.The Northeastern Monsoons transport fine particles from the upwind emission sources (i.e., industrial complex and metro Kaohsiung along the coastal region) to the downwind site (Che-Cheng), causing a significant increase in the contribution of secondary sulfate (13.25 ± 2.10%).Overall, the contribution of secondary aerosols to PM 2.5 at the Che-Cheng site was commonly higher than those at the urban (Chien-Chin) and industrial (Siao-Gang) sites in Kaohsiung City, and soil dust was the third important source (13.35 ± 2.18%) contributing to PM 2.5 at the Che-Cheng site.The results showed that the sources contributing to PM 2.5 in spring were different from those in winter.Moreover, the sources of PM 2.5 contributing to the Che-Chen site differ from those contributing to the Chien-Chin and Siao-Gang sites.The chemical composition of PM 2.5 sampled in this study was compared with previous literature as shown in Table 4. Overall, the chemical composition of PM 2.5 was similar to that reported by previous studies with the exception of lower carbonaceous content and OC/EC ratios.

CONCLUSIONS
Atmospheric fine particles in southern Taiwan were sampled to investigate the spatiotemporal distribution and chemical fingerprints of PM 2.5 .Along the coastal region, PM 2.5 concentrations were generally higher at the urban and industrial sites, and lower at the background site.Monthly variation of PM 2.5 concentration shows that the concentrations of PM 2.5 from December 2014 to March 2015 were much higher than those in April and May 2015 at the urban and industrial sites, while there were no significant differences in PM 2.5 concentrations from December 2014 to May 2015 at the background site in southern Taiwan.The PM 2.5 /PM 10 ratios were mostly higher than 0.5 during the sampling period, except for May, indicating that PM 2.5 dominated PM 10 in southern Taiwan.The most abundant ionic species of PM 2.5 were secondary inorganic aerosols (SO 4 2-, NO 3 -and NH 4 + ), accounting for 18.67-42.36% of PM 2.5 .Relatively high Na + and Cl -concentrations of PM 2.5 were commonly observed in the coastal region of southern Taiwan.Metallic elements covering fifteen major metals accounted for 9.14%-16.12% of PM 2.5 .The dominant metallic elements of PM 2.5 included crustal elements (Al, Fe and Ca) and anthropogenic elements (V, Ni, As, Cd, Zn and Pb).The total mass of carbonceous content (TC) accounted for 1.05-12.8% of PM 2.5 .The concentrations of OC and EC at the Chien-Chin and Siao-Gang sites were generally higher than that at the Che-Cheng site, mainly due to the emissions from urban and industrial anthropogenic sources.The OC/EC ratios ranged from 1.16 to 2.60 with the highest OC/EC ratios in February 2015 at the Siao-Gang site among the three sampling sites.These results indicate that the mass concentration and chemical composition of PM 2.5 vary with the prevailing wind direction.
The source identification of PM 2.5 was further undertaken by levoglucosan analysis, backward trajectory simulation and CMB receptor modeling.Consistently high levoglucosan levels observed in December suggest cross-boundary transport toward southern Taiwan due to the Northeastern Monsoons.Relatively higher levoglucosan levels observed only at Che-Cheng in March and April were mainly influenced by biomass burning from local sources and/or long-range transport from the Indochina Pennysula.The OC/Levo and K + /Levo ratios serve as important chemical indicators to identify the origins of PM 2.5 sampled at various sites.The backward trajectory simulation can be also used to identify the routes of air masses transported toward southern Taiwan.In winter, the most frequent transportation route toward southern Taiwan came mainly from Northeast Asia.In spring, the most frequent transportation routes toward the Chien-Chin and Siao-Gang sites were mainly from the South China Sea or the Philippines.The most frequent transportation routes toward the Che-Cheng site weree mainly from South China, the Indochina Peninsula and the South China Sea.Results obtained from CMB receptor modelling indicated that the sources of PM 2.5 differ among the three sites in southern Taiwan.Vehicular exhausts and industrial emissions were the main sources of PM 2.5 at the Chien-Chin and Siao-Gang sites, respectively, while biomass burning and soil dusts were the dominant sources of PM 2.5 at the Che-Cheng site.

Fig. 1 .
Fig. 1.Location of three sampling sites selected for collecting PM 2.5 at the southern tip of the Taiwan Island (Urban site: Chien-Chin, Industrial site: Siao-Gang, and Background site: Che-Chen).

Fig. 3 .
Fig. 3.The concentrations of ionic species of PM 2.5 sampled at the southern Taiwan.

Fig. 4 .
Fig. 4. The concentrations of metallic elements of PM 2.5 sampled at the southern Taiwan.

Fig. 5 .
Fig. 5.The concentrations of carbonaceous content of PM 2.5 sampled at the southern Taiwan.

Fig. 6 .
Fig. 6.The monthly variation of levoglucosan concentration sampled at the southern Taiwan.

Fig. 9 .
Fig. 9.The backward trajectories of air masses tranported toward the southern Taiwan.

Table 1 .
Summary of QA/QC results for the analysis of chemical species in this study.

Table 2 .
Chemical composition of PM 2.5 sampling at the Chien-Chin, Siao-Gang, and Che-Cheng sites located in the southern Taiwan.

Table 3 .
Source apportionment of PM 2.5 sampled at the Chien-Chin, Siao-Gang, and Che-Cheng sites located in the southern Taiwan.

Table 4 .
Comparison of chemical composition of PM 2.5 with previous literature.