at SciVerse ScienceDirect

Atmospheric Environment 88 (2014) 297e305

Contents lists available

Atmospheric Environment

journal homepage: www.elsevier .com/locate/atmosenv

The characteristics, seasonal variation and source apportionment ofVOCs at Gongga Mountain, China

Junke Zhang a,b, Yang Sun a, Fangkun Wu a, Jie Sun a, Yuesi Wang a,*

a State Key Laboratory of Atmospheric Boundary Layer Physics and Atmospheric Chemistry (LAPC), Institute of Atmospheric Physics,Chinese Academy of Sciences, Beijing 100029, ChinabUniversity of Chinese Academy of Sciences, Beijing 100049, China

h i g h l i g h t s

� This is the first time to study the VOCs in the remote station in southwest China.� Aromatics and alkanes are the major components of VOC.� The seasonal variation shows higher value in spring and lower value in autumn.� Anthropogenic sources are the most important sources in the remote area.

a r t i c l e i n f o

Article history:Received 1 November 2012Received in revised form4 March 2013Accepted 18 March 2013Available online 29 March 2013

Keywords:VOC sources at Gongga MountainSeasonal variationPMF receptor modelSource apportionment

* Corresponding author. Tel.: þ86 (0)1082080530;E-mail addresses: wys@dq.cern.ac.cn, wys@mail.ia

1352-2310/$ e see front matter � 2013 Elsevier Ltd.http://dx.doi.org/10.1016/j.atmosenv.2013.03.036

a b s t r a c t

The mixing ratio, composition and variability of volatile organic compounds (VOCs) were measured from2008 through 2011 at Gongga Mountain Forest Ecosystem Research Station (102�000E, 29�330N, elevation1640 m), a remote station in southwest China. Weekly samples were collected in the Gongga Mountainarea and were analyzed using a three-stage preconcentration method coupled with GCeMS. An advancereceptor model positive matrix factorization (PMF) was applied to identify and apportion the sources ofVOCs. The results show that the measured VOC mixing ratio at Gongga Mountain is dominated by ar-omatics (35.7%) and alkanes (30.8%), followed by halocarbons (21.6%) and alkenes (11.9%). The generaltrend of seasonal variation shows higher mixing ratios in spring and lower mixing ratios in autumn. Theeffect of alkanes and aromatics on the seasonal variation of total volatile organic compounds (TVOCs) issignificant. Five sources were resolved by the PMF model: (1) gasoline-related emission (the combinationof gasoline exhaust and gas vapor), which contributes 35.1% of the measured VOC mixing ratios; (2)solvent use, contributing 21.8%; (3) fuel combustion, contributing 29.1%; (4) biogenic emission,contributing 5.2%; and (5) industrial, commercial and domestic sources, contributing 8.7%. The effect onthis area of the long-range transport of air pollutants from highly polluted areas is significant.

� 2013 Elsevier Ltd. All rights reserved.

1. Introduction

There is no doubt of the importance of volatile organic com-pounds (VOCs) in the atmosphere, because they are the importantprecursors of secondary air pollutants and secondary organicaerosols in photochemical processes. Moreover, most of VOCs arealso impact the environment and human health directly (Leuchnerand Rappenglück, 2010; Pérez-Rial et al., 2010).

The study of VOCs is still very limited in China because mostsuch studies have focused on mega-cities or city clusters, such asBeijing (Song et al., 2007; Yuan et al., 2009; Su et al., 2011), the Pearl

fax: þ86 (0)1062362389.p.ac.cn (Y. Wang).

All rights reserved.

River Delta (Chan et al., 2006; Guo et al., 2006, 2011; Ling et al.,2011) and the Yangtze River Delta (Cai et al., 2010; Geng et al.,2010; Huang et al., 2011). However, information about the sourcecharacteristics of VOCs in remote areas is insufficient because it ismore difficult and more expensive to observe these areas thanobserve areas in cities. However, the study of VOC emissions fromremote areas is crucial because these areas occupy more area thancities worldwide. Thus, the results of such studies are highly usefulfor studying the global temporal and spatial variation of VOCs.Unfortunately, this type of study is rare in China, especially in theremote area of the underdeveloped southwestern region of China.

Furthermore, it is not sufficient simply to measure the mixingratios of VOCs to develop of an effective control strategy. We alsoneed to obtain and understand accurate information about thesources of VOCs. One effective method for studying VOC sources

J. Zhang et al. / Atmospheric Environment 88 (2014) 297e305298

involves the use of receptor models (Song et al., 2007; Lanz et al.,2009; Chan et al., 2011; Ling et al., 2011), such as positive matrixfactorization (PMF). This receptor model has been successfullytested in comparison with other receptor models for VOCs (Songet al., 2007; Sauvage et al., 2009; Yuan et al., 2009) and particu-late matter (PM) (Leuchner and Rappenglück, 2010; Chan et al.,2011). A very effective statistical method, it can apportionambient concentration data to sources by identifying the intrinsiccharacteristics of the data while limiting all of the elements in thefactor score (source profiles) and the factor loading (source con-tributions) matrix to positive values, making this method especiallyuseful in environmental source analysis (Yuan et al., 2009).

In this paper, the mixing ratios, composition and seasonal var-iations of VOCs at Gongga Mountain, a station in a remote area ofsouthwest China, are first presented. We then use the PMF receptormodel to identify VOC source types and to apportion their contri-butions. To the best of our knowledge, this paper is the first to studyVOCs at a remote station in southwest China.

2. Experiment and methods

2.1. VOC sampling and analysis

The measurements were performed at the standard meteoro-logical field of the Gongga Mountain Forest Ecosystem ResearchStation (102�000E, 29�330N, elevation 1640 m) in Hailuogou ScenicArea, a remote site located in southeast Ganzi Tibetan AutonomousPrefecture in Sichuan province. This scenic area is famous for itslarge areas of glaciers and virgin forest and its large number of rareanimal and plant resources. The forest area in Hailuogou isapproximately 70 km2 and the total resort area is approximately200km2. Theprincipal tourist seasons are spring and summer. Thereare approximately 30,000 inhabitants in the vicinity of the scenicarea. During the tourist seasons, the population is several timesmore numerous than the local residents. This station is approxi-mately 130 km to the northeast of Ya’an and 250 km from Chengdu,the capital of Sichuan province (Fig.1). There are twomajor roads inthenorth and east, 500mand400m fromthe station, respectively. Agas station is located 1000 m to the north of the station.

Air samples were collected twice a day every Tuesday from Jan.2008 to Dec. 2011. A total of 380 samples were collected during themeasurement period (approximately 40 samples were excluded forvarious reasons). Themonitoring site is shown in Fig.1. A pumpwasused to draw, ambient air samples from a gas inlet through a PFA-Teflon tube (OD: ¼ in). These samples were collected in pre-evacuated 1 L electropolished canisters. A 3 min integrated

Fig. 1. Location of the sampling site and some important cities around it.

sample was taken for each canister sample at 8:00 and 14:00.Ambient air was collected at a flow rate of 1 L min�1. After thecanisters were pressurized to 60 psig, the valve was closed and thepump turned off. The samples were returned to Beijing for analysiswithin 7 days of collection.

Measurements of VOCs were made using an Entech 7100Apreconcentrator (Entech Inc., USA) followed by a GCeMS system(Finnigan Trace GC/Trace DSQ). The details of the VOC analysisprocedures have been described previously by Mao et al. (2009)and will be described briefly here.

A 500 mL sample was concentrated in an Entech 7100A pre-concentrator, a three-stage preconcentration was utilized toremove the water, carbon dioxide, nitrogen and oxygen in airsamples. The VOCs were further focused with a capillary focusingtrap for rapid injection prior to the analytical column. A DB-5 MSfused-silica capillary column (60 m � 0.25 mm � 0.25 mm, AgilentTechnologies Inc.) coupled with a quadrupole mass spectrometerdetector (Finnigan Trace 2000/DSQ. Thermofisher Inc. USA) wasused for qualification and quantification.

Forty VOCs were identified and quantified. Each target specieswas identified by its retention time, mass spectrum and USEPAstandard gases. The quantification of the target VOCs was per-formed with multi-point external standard curves and modifiedusing Relative Responsible Factors (RRFs). The calibration curveswere prepared using 100 ppbv external standard gases (ScottSpecialty, TO14 standard; alkanes and alkenes) and 100 ppbv in-ternal standard gases of dibromomethane at five different dilutedconcentrations plus nitrogen (0e100 ppbv). Internal standard gaswas added to each sample to trace the analytical procedure (Maoet al., 2009). Five sets of replicate samples were collected tocheck the precision and reliability of the sampling and analysismethods. The RSD was within 10% for the target compounds in allfive replicates. For most samples, the VOC species were above thedetection limit of 5e10 pptv.

2.2. Positive matrix factorization (PMF)

2.2.1. Principles and application of PMFThe PMF method is comprehensively described by Paatero and

Tapper (1994) and Paatero (1997) and has been used in manyVOC source identification studies (Song et al., 2007; Sauvage et al.,2009; Yuan et al., 2009). In this paper, PMF 3.0 (USEPA, 2008) wasused to apportion the contributions from emission sources. Severalconcepts that are relevant to the understanding of this work arebriefly described here. For additional details about the method, thereader is referred to the publications cited above and to the PMF 3.0user manual.

An ambient data set can be viewed as an i by j data matrix X inwhich i samples and j chemical species are represented. The goal ofmultivariate receptor modeling is to identify a number of sources p,the species profile f of each source and the amount of mass gcontributed by each source to each individual sample as well as theresiduals eij:

Xij ¼Xpk¼1

gjkfkj þ eij

where eij is the residual for each sample/species.The PMF solution minimizes the objective function Q based on

these uncertainties (u):

Q ¼Xni¼1

Xmj¼1

"xij �

Ppk¼1gikfkjuij

#2

J. Zhang et al. / Atmospheric Environment 88 (2014) 297e305 299

To find the number of sources, it is necessary to test differentnumbers of sources and to find the optimal value corresponding tothe most reasonable results. If the number of sources is estimatedproperly, the theoretical Q value should be approximately thenumber of degrees of freedom or approximately the total numberof data points. However, the Q value may deviate from the theo-retical value if the number of sources is not well determined, (Guoet al., 2011).

The PMF 3.0 model also provides the rotational freedomparameter (Fpeak) function, which can control whether moreextreme values are assumed for the factor loadings (by assigningpositive Fpeak values) or for the factor scores (by assigning negativeFpeak values) (Chan et al., 2011). In this analysis, an average ofapproximately 98% of the scaled residuals calculated by PMF wasbetween �3 and 3, indicating a good fit of the modeled results. Thefactors also showed oblique edges, a property that has been pro-posed as an additional check of the rotation (Paatero et al., 2005).

2.2.2. Data pre-processing and the choice of speciesNot all VOC species were used in this source apportionment

analysis. There are three principles for choosing the right speciesfor the model. (1) In the mixing ratios file, species with more than25% of samples missing or below the minimum detection limits(MDLs) were rejected. (2) Species that are highly reactive wereexcluded because they react and are lost quickly in the ambientatmosphere. Including them may bias the model. An exception tothis approach is the inclusion of special species that are importanttracers of certain emission sources, even though they react quicklyand become lost in the ambient atmosphere, such as isoprene, animportant biogenic tracer (Brown et al., 2007). (3) Certain species atlow mixing ratios that are not typically tracers of emission sourceswere also rejected. Eventually, 28 VOC species were selected for thesource apportionment analysis because they are themost abundantspecies and/or are typical tracers of various emission sources.

The PMF 3.0 model requires two input files: one for themeasured mixing ratios of the species and one for the estimateduncertainty of themixing ratios. Data below theMDLwere replacedby half the MDL values, and their uncertainties were set as 5/6 ofthe MDL. Missing data were replaced by the geometric mean of the

Table 1Average mixing ratios of VOCs samples in the air of Gongga Mountain.

Name Mixing ratio (ppbv)

AlkanesButane 0.20 � 0.16Isopentane 0.47 � 0.50Cyclopentane 0.11 � 0.19Hexane 0.09 � 0.122-methylhexane 0.13 � 0.363-methylhexane 0.18 � 0.42Methylcyclohexane 0.14 � 0.323-methylheptane 0.14 � 0.35

Alkenes1-butene 0.38 � 0.47Isoprene 0.40 � 0.59c-2-pentene 0.02 � 0.04Limonene 0.06 � 0.08

AromaticsBenzene 0.72 � 0.64Ethylbenzene 0.28 � 0.42Styrene 0.23 � 0.41Isopropylbenzene 0.07 � 0.141,3,5-trimethylbenzene 0.12 � 0.24

HalocarbonsChloromethane 0.62 � 0.34CFC-11 (trichlorofluoromethane) 0.28 � 0.25Dichloromethane 0.36 � 0.421,2-dichloroethane 0.25 � 0.36

measured mixing ratios of that species, and their uncertaintieswere set at four times the geometric mean (Polissar and Hopke,1998). The uncertainties for the normal data points weresubstituted with 20% of the mixing ratio values, as suggested byBuzcu and Fraser (2006).

3. Result and discussion

3.1. General characteristics of the VOCs

Forty VOC (C4eC12) species were measured in the samplescollected at GonggaMountain from Jan. 2008 to Dec. 2011, includingalkanes, aromatics, alkenes and halocarbons. The means and stan-darddeviations of themixing ratios of these compounds are listed inTable 1. Benzene, chloromethane, isopentane, toluene and isoprenewere themost strongly dominant compounds, with averagemixingratios of 0.72 � 0.64 ppbv, 0.62 � 0.34 ppbv, 0.47 � 0.50 ppbv,0.44 � 0.33 ppbv and 0.40 � 0.59 ppbv, respectively.

The mixing ratio for the total volatile organic compounds(TVOCs) was 8.75 � 5.76 ppbv. This value is higher than that ofMount Tai (Table 2), where the mixing ratio observed was6.95 � 5.71 ppbv (Mao et al., 2009). However, this value is muchlower than the result obtained in Shanghai (32.35 � 19.76 ppbv), amega-city in China. The NMHCmixing ratio was 6.33� 4.63 ppbv atGongga Mountain. This value is higher than that at JianfengMountain (4.78� 1.85 ppbv), a background site in Hainan province,and is lower than that at Dinghu Mountain (23.40 � 9.84 ppbv), aremote station in Guangdong province (Tang et al., 2007). There-fore, the TVOCs or NMHCs at Gongga Mountain were at an inter-mediate level compared with other background or remote sites butwere much lower than the emissions of large cities.

The contributions of the four main hydrocarbon groups to theTVOC mixing ratio at Gongga Mountain are given in Fig. 2. Aro-matics (35.7%) provided the largest contribution to the TVOCs,followed by alkanes (30.8%), halocarbons (21.6%) and alkenes(11.9%). The contributions of the four main hydrocarbon groups tothe TVOCs are similar to the results for Mount Tai. The contributionof aromatics was largest (34%), and the percentage of alkenes wasthe lowest, at only 11% (Mao et al., 2009).

Name Mixing ratio (ppbv)

Isobutane 0.27 � 0.26Pentane 0.22 � 0.232-methylpentane 0.08 � 0.10Methylcyclopentane 0.17 � 0.382,3-dimethylpentane 0.18 � 0.35Heptane 0.14 � 0.302-methylheptane 0.16 � 0.36

1-pentene 0.05 � 0.07t-2-pentene 0.04 � 0.06a-pinene 0.10 � 0.18

Toluene 0.44 � 0.33m/p-xylene 0.28 � 0.42o-xylene 0.31 � 0.43Propylbenzene 0.29 � 0.441,2,4-trimethylbenzene 0.39 � 0.50

CFC-114 (tetrafluorodichloroethane) 0.02 � 0.03CFC-113 (trichlorotrifluoroethane) 0.09 � 0.08Chloroform 0.20 � 0.35Chlorobenzene 0.08 � 0.15

Table 2Comparison of TVOCs or NMHCs measured at Gongga Mountain and at otherstations.

Stationname

Mount Taia JianfengMountainb

DinghuMountainb

Shanghaic GonggaMountain

Stationtype

Background Background Background City Background

TVOCs(ppbv)

6.95 � 5.71 32.35 � 19.76 8.75 � 5.76

NMHCs(ppbv)

4.78 � 1.85 23.40 � 9.84 6.33 � 4.63

a Mao et al. (2009).b Tang et al. (2007).c Cai et al. (2010).

J. Zhang et al. / Atmospheric Environment 88 (2014) 297e305300

3.2. Seasonal variation patterns of VOCs

The seasonal trend characteristics of VOCs are valuable for un-derstanding important processes in atmospheric transport andchemistry. In this study, winter is defined as the three-monthperiod from December to February; spring, from March to May;summer, from June to August; and autumn, from September toNovember.

The seasonal variations of the mixing ratios of the total and thefour main hydrocarbon groups are shown in Fig. 3. As shown, thereis a large and statistically significant variation in VOC levels (onlyconsidering the mean values). The mixing ratios were generallyhigh in spring (12.9 ppbv) and low in autumn (6.44 ppbv). Thisresult is different from those of studies in several cities, which allfound that the mixing ratios of most VOCs were high in winter andlow in summer (Ho et al., 2004; Parra et al., 2009). In addition, notethat the error range for the mixing ratios of the total and the fourmain hydrocarbon groups are wide. The significant difference maydisappear if these errors are considered. The following discussionwill only focus on the difference in mean values.

Several factors affect the seasonal variation of VOCs in the at-mosphere. These factors include the following: (1) Photochemicalremoval (primarily by the hydroxyl (OH) radical). The chemicalremoval of VOCs by OH radicals is faster in warmer seasons than incooler seasons. Because more sunlight and higher temperatures inwarmer seasons will result in higher chemical removal reactionrates (Ho et al., 2004). (2) The dilution due to atmospheric mixing.The mixing layer in warmer seasons is much higher than in coolerseasons. The dilution of airborne pollutants from ground sourceemissions in warmer seasons is stronger than in cooler seasons.However, another, more important factor influences the seasonalvariation of VOCs at Gongga Mountain. Because the site is a touristattraction, the principal sources of VOC emissions will change withthe tourism seasonal variations significant. The highest and thelowest values appeared in spring and autumn as the result ofall of the above factors. Spring is a favorable season for tourism.Many local sources could introduce VOCs into the air. Moreover,

Alkanes30.8%

Alkenes11.9%

Aromatic35.7%

Halocarbons21.6%

Fig. 2. Contributions of four main hydrocarbon groups to TVOC in the air at GonggaMountain.

photochemical removal and dilution are weak due to the lowertemperature. Therefore, the VOCs will accumulate. The highestvalue appeared in this season. Although emission sources are alsopresent in autumn, they are lower than in summer, and the effectsof photochemical removal and the dilution are remain strong. Forthis reason, the lowest values of VOCs appeared in autumn. Inwinter, although the emission sources are less than in autumn, theaccumulative effect of meteorological factors is strongest. The VOCemissions from non-tourism sources will accumulate, and the totalin winter will higher than that in autumn. However, the long-termpresence of air masses could result in the transport of VOCs fromlarge cities to this area in spring than in other seasons, whereas thiseffect could be negligible in autumn. We will discuss this topic indetail in Section 3.4.

Among the four main hydrocarbon groups, the mixing ratios ofalkanes and aromatics were dominant throughout the year andshowed a pattern of variation similar to that of the TVOCs. A cor-relation analysis found that the relationship between TVOCs andalkanes or aromatics was significant (P < 0.01). Consequently, theseasonal variation of the TVOCs was primarily affected by that ofthe alkanes and aromatics. The decisive role of these two hydro-carbon groups has also been identified by other studies (Saito et al.,2009; Geng et al., 2010).

3.3. Source identification

Five factors were resolved at Gongga Mountain through theapplication of the PMFmodel. Fig. 4 shows the explained variations(EVs) for all the identified sources. The EV quantity indicates theimportance of each factor element in explaining the total mass ofthe element and is particularly powerful for identifying tracerspecies if the absolute amounts of chemical species show a signif-icant difference (Yuan et al., 2009).

Source 1 is identified as gasoline-related emissions (the com-bination of gasoline exhaust and gas vapor). Gasoline is the domi-nant vehicle fuel, and VOC emissions from gasoline are transportedalong the following three pathways: (1) gasoline vapor emittedfrom headspace emissions at gas stations and bulk terminals andfrom vehicles as diurnal emissions and resting loss; (2) liquidgasoline arising from spillage, leakage and vehicle operations; and(3) exhaust released from the tailpipes of gasoline-powered vehi-cles during gasoline combustion (Watson et al., 2001; Choi andEhrman, 2004). Isobutene, isopentane and hexane are the mainconstituents of gasoline and are thus good tracers of gasolineevaporation (Xie and Berkowitz, 2006; Brown et al., 2007). Addi-tionally, BTEX and 2-methylhexane are important componentspecies of vehicular exhaust, as shown by many studies (Watsonet al., 2001; Guo et al., 2006, 2007). The Hailuogou scenic area at-tracts large numbers of tourists due to the attractive natural envi-ronment and the favorable climate. The number of visitors reachedto 141,000 between January and July in 2009, and the number ofvisitors per day can reach approximately 10,000 on Labor Day, atraditional festival in China. The two roads cited above are theprincipal routes into the scenic area. For this reason, the gasolineexhaust emissions from the traffic sources on the two main roadsand the gas vapor from the gas station are very important con-tributors to this source at the study site.

High percentages were found for trimethylbenzene and aro-matic hydrocarbons, especially for TEX (toluene, ethylbenzene, m/p-xylene and o-xylene) in Source 2. It is known that TEX is theprimary constituent of solvents (Guo et al., 2004a; Choi et al., 2011).TEX is often used as a solvent in paints, coatings, synthetic fra-grances, adhesives, inks and cleaning agents, in addition to its usein fossil fuel (Borbon et al., 2002; Chan et al., 2006). 1,2,4-trimethylbenzene and 1,3,5-trimethylbenzene are also typical

Jan.Feb.Mar.Apr.MayJun. Jul. Aug.Sep.Oct.Nov.Dec.-1

0

1

2

3

4

5

6

7

8

9

10

Spring Summer Autumn Winter

0

2

4

6

8

10

12

14

16

18

Mix

ing

ratio

(ppb

v)

TVOCs Alkanes Aromatics Alkenes Halocarbons

Fig. 3. Monthly and seasonal average mixing ratios of the total and the four chemical groups of Gongga Mountain.

J. Zhang et al. / Atmospheric Environment 88 (2014) 297e305 301

tracers that are used as solvents (Borbon et al., 2002; Guo et al.,2004a). In the PMF-derived source profile, 1,2,4-trimethylbenzeneand 1,3,5-trimethylbenzene all account for 75% of the total mass.This source is therefore assigned to solvent use. In recent years, the

0

20

40

60

80

100 Gasoline-related emission

0

20

40

60

80

100Solvent use

0

20

40

60

80

100

% o

f sp

ecie

s

Fuel combustion

0

20

40

60

80

100 Biogenic emission

Chl

orom

etha

ne

Isob

utan

e

Isop

enta

ne

1-Pe

nten

e

Pent

ane

Isop

rene

t-2-

Pent

ene

c-2-

pent

ene

Dic

hlor

omet

hane

2-m

ethy

l-Pe

ntan

e

Hex

ane

1,2-

Dic

hlor

oeth

ane

Ben

zene

0

20

40

60

80

100 Industrial, commercial and domestic

Fig. 4. Explained variation of five ident

development of the Gongga area has been rapid. For example, thenumber of visitors in the first seven months of 2009 was 123%greater than that in the same period in 2008. The tourism zonerequires more hotels, restaurants and other related infrastructure

2-m

ethy

l-he

xane

2,3-

dim

ethy

l-pe

ntan

e

Hep

tane

Tol

uene

Chl

orob

enze

ne

Eth

ylbe

nzen

e

m/p

-xyl

ene

Styr

ene

o-xy

lene

Isop

ropy

lben

zene

prop

yl-b

enze

ne

1.3.

5-tr

imet

hyl-

benz

ene

1.2.

4-tr

imet

hyl-

benz

ene

-pin

ene

Lim

onen

e

ified sources at Gongga Mountain.

solvent use,22%

gasoline-relatedemission, 35%

industrual,commercial anddomestic, 9%

fuel combustion,29%

biogenicemission, 5%

Fig. 5. Individual contributions of the five major sources to the measured VOCs.

J. Zhang et al. / Atmospheric Environment 88 (2014) 297e305302

to meet the rapid growth in visitor numbers. Solvent widely areused extensively in the construction industry. They are the primarycomponents of coatings, adhesives, paints and cleaning agents aspreviously stated.

Source 3 was characterized by a significant presence of chloro-methane (91%) and benzene (41%) (Guo et al., 2004b, 2007).Chloromethane is a typical species from biomass burning (Linget al., 2011). Barletta et al. (2009) also considered chloromethanea biomass/coal burning tracer, which is consistent with the resultsof Wang et al. (2005). Moreira dos Santos et al. (2004) found thatcoal combustion could release significant amounts of benzene intothe atmosphere. Source 3 has, therefore, been assigned to fuelcombustion. At Gongga Mountain, local residents burn largeamounts of straw after the harvest. Moreover, the development ofthe tourism industry has promoted the rapid development of manyrelated industries, and coal is used as the principal fuel by many ofthese industries. Therefore, the widespread use of these two typesof fuel is the key reason for their identification as the fuel com-bustion source.

Source 4 was distinguished by a strong presence of isoprene, a-pinene and limonene, the indicators of biogenic emissions. In thederived source profile, isoprene, a-pinene and limonene represent75%, 77% and 83% of their EVs, respectively. Isoprene can also becharacterized by industrial emissions if it is emitted with other in-dustrial VOCs, such as 2-methylpentane, 3-methylpentane and 1,3-butadiene (Buzcu and Fraser, 2006), and if it if emitted by heavyvehicular traffic (Song et al., 2007). This possible contribution can beneglected because the correlation coefficient R2 between isopreneand two important tracers of industrial emissions (1-pentene and c-2-pentene) and two important tracers of gasoline-related emissions(isopentane and hexane)were low (0.03 and 0.158, 0.154 and 0.151).Therefore, this source was identified as a biogenic emission.

Guo et al. (2004a) found that 1-pentene and cis-2-pentene areemitted into the air from industrial, commercial and domesticprocesses that either use or manufacture the materials or that theyare emitted from the sites where they are formed as byproducts. 1-pentene is a product of oil pyrolysis. It can be used for organicsynthesis, dehydrogenation for producing isoprene, and as an ad-ditive in high-octane gasoline. Cis-2-pentene is used in organicsynthesis and as a polymerization inhibitor. This source is alsoassociated with 70% of the total pentane used in themanufacture ofartificial ice, anesthetic agents and the synthesis of pentanol and i-pentane. In source 5, 1-pentene and cis-2-pentene represent 83%and 90% of their EVs, respectively. Buzcu and Fraser (2006) alsofound that 2-methylpentane is an important emission componentfrom industrial processes. This species is an important chemicalrawmaterial and can be used as a rubber solvent and a vegetable oilextraction solvent. It is also an intermediate in organic syntheses.Therefore, this source was identified as industrial, commercial anddomestic.

The sources resolved by PMF were the results of the interactionof local emissions and long-term transport from areas around thesite. The topic of local emissions was addressed at the end of each ofthe preceding discussions of the sources. A discussion of long-termtransport must involve information on other air masses. For thisreason, long-term transport will be discussed in Section 3.4.

The individual contributions of the five major VOC sources tothe measured VOC mixing ratios are shown in Fig. 5. This figureshows that gasoline-related, industrial, commercial and domesticsources and solvent use are estimated to be the major contributorsto the TVOCs at GonggaMountain, representing approximately 86%of the total source. Biogenic emissions play only a minor role,contributing only 5% of the total VOC mass. The basis of this phe-nomenon is that there were only three species of emissions fromthe biogenic we studied and that the mixing ratios of a-pinene and

limonene were only 0.10 � 0.18 ppbv and 0.06 � 0.08 ppbv,respectively. This finding is similar to those of previous studies inremote or background sites, such as those at Lin’an (Guo et al.,2004b) and Yufa (Yuan et al., 2009).

3.4. The effect of long-range transport

The long-range transport of air pollutants (clean air) from highlypolluted areas (clean areas) could increase (decrease) the VOCmixing ratios at the study site. This transport will affect the sourcesat the sampling site in conjunction with the local sources. TheHybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT)model is a useful air trajectory model, especially for studying thelong-range transport of air masses (Mao et al., 2009; Cheng et al.,2010). This model was used to calculate 36 h back-trajectories foreach day during different seasons in 2011. The back-trajectorieswere calculated every 6 h. The air mass direction can be dividedinto four categories in terms of the original direction, namely, 0�e90�, 90�e180�, 180�e270� and 270�e360�.

Only one air mass in the 0�e90� category was present in spring(Fig. 6a). It originated from Xi’an, a large city in central China, andpassed over Hanzhong and several cities of Sichuan, includingGuangyuan, Mianyang, Deyang and Chengdu, before ultimatelyarriving at Gongga Mountain. The transport of pollutants fromthese cities via the air masses could significantly promote the levelof VOCs in the air at the study site. Two air masses were associatedwith the 180�e270� direction. The air masses originated from thenortheastern part of India and passed over northern Myanmar andthe clean air area in Yunnan province, China. Two air masses wereassociated with the 270�e360�direction. These air masses origi-nated from several underdeveloped areas of China, including Qin-hai province and the Tibet Autonomous Region. The effect on thesampling site of the air masses that originate from the latter twodirections was to dilute themixing ratio of VOCs in the atmosphere.

In summer, two air masses were associated with the 0e90�

direction (Fig. 6b). The original and effect of the longer of these twoair masses (7th) was similar to that of the 0e90� air mass in spring.The shorter air mass originated from and passed several smallcounties or towns. Only one air mass was associated with the 90�e180� direction. This air mass originated from Guizhou province, anunderdeveloped province in China. Three air masses were associ-ated with the 180�e270� direction. The third of these air massesoriginated from Yunnan province, a clean area in China. The othertwo air masses originated from northern Myanmar, also a cleanarea. The effect of the air masses originating from the latter twodirections was to dilute the VOCs in the atmosphere.

One air mass originated from the east of China in autumn(Fig. 6c). This air mass was very short and may have transportedsmall amounts of pollutants from nearby towns. However, its effectwas very limited. The major air masses in autumn were associated

Fig. 6. The main air masses of different seasons at Gongga Mountain in 2011. a: spring; b: summer; c: autumn; d: winter.

J. Zhang et al. / Atmospheric Environment 88 (2014) 297e305 303

with the 180�e270� and 270�e360� directions. The origins andeffects of these air masses were similar to those of the air massesidentified in spring and summer and associated with the samedirections.

In winter, the air masses were associated with the 180�e270�

and 270�e360� directions (Fig. 6d). The origins and effects of theseair masses were the same as those of the air masses associated withthe same directions in spring and autumn.

The air masses can be divided into the following two categories:(1) contaminated air masses, it only appearing in spring and sum-mer (the 5th air mass in spring and the 7th air mass in summer)which could transport air pollutants from highly polluted areas andincrease the level of VOCs in the GonggaMountain area; and (2) theother air masses, which originated from other directions anddiluted the VOCs at the sampling site. The heights associated withthese two categories of air masses differed significantly. The heightof the air masses belonging to the first category was always lessthan 500 m. This characteristic was not conducive to the dilution ofpollutants in the vertical direction during the transport process. Incontrast, the height of the air masses belonging to the second

category was not stable. It ranged from more than 3000 m to lessthan 500 m for different directions or seasons.

The differences between air masses in different seasons can alsobe used to explain the seasonal variation at Gongga Mountain. Airmasses carrying high mixing ratios of pollutants only appeared inspring and summer, and the associated proportions were all 10%(the 5th air mass in spring and the 7th air mass in summer). Whilethe scavenging effect is stronger in summer than spring, due to thehigher temperature and mixing layer. For this reason, the effect oflong-term transport on the VOCs in the Gongga Mountain area isstronger in spring than in summer. All of the air masses appearingin autumn andwinter can be viewed as dilute air masses. Therefore,the mixing ratio of VOCs in autumn and winter is lower than inspring (especially) and summer.

Note that the station at which this study was conducted wouldbe considered a remote station in a relative sense. The air massestraveling to the site from several large cities have some effects, asdiscussed above. Therefore, the source tracers cannot be used as thesource apportionment if these tracers experienced a loss duringtransport. Accordingly, we analyzed the correlations between

J. Zhang et al. / Atmospheric Environment 88 (2014) 297e305304

several important tracers belonging to the same source. Forgasoline-related emissions, we analyzed isopentane, hexane and 2-methylhexane; for solvent use, we analyzed toluene, ethylbenzene,xylene, 1,3,5-trimethylbenzene and 1,2,4-trimethylbenzene; forfuel combustion, we analyzed chloromethane and benzene; and forindustrial, commercial and domestic sources, we analyzed 1-pentene, c-2-pentene and 2-methylpentane. The ranges of the R2

statistic for the tracers for gasoline-related emissions, solventsources, fuel combustion sources and industrial, commercial anddomestic sources were 0.43e0.78, 0.51e0.89, 0.34 and 0.45e0.60,respectively. The R2 values for chloromethane and benzene werelower. The reason for this difference is that chloromethane andbenzene primarily represent emissions from biomass and coalburning, respectively, although they are the tracers for the com-bustion source. As the above discussion indicates, we found that thecorrelations for the important tracers corresponding to everysource were still reliable although some of them representedtransport from certain cities. Therefore, these data could be used inthe PMF analysis.

4. Conclusions

Field measurements of volatile organic compounds (VOCs) wereconducted at the Gongga Mountain Forest Ecosystem ResearchStation, an important remote station in southwest China, from2008 to 2011. The VOC mixing ratios at Gongga Mountain weredominated by alkanes and aromatics, followed by halocarbons andalkenes. The VOCs showed obvious seasonal variation, with highermixing ratios during spring and lower mixing ratios duringautumn. The effect of alkanes and aromatics on the seasonal vari-ation of the TVOCs was significant. A positive matrix factorization(PMF) model was used to investigate the contributions of the VOCsources. Five stable sources were identified on the basis of finger-print species in the source profiles by the PMF model: gasoline-related emissions (the combination of gasoline exhaust and gasvapor), solvent use, fuel combustion, biogenic emissions and in-dustrial, commercial and domestic sources. The first three sourceswere found to be the strongest VOC contributors in this area, rep-resenting approximately 86% of the total source material. The effecton this area of the long-range transport of air pollutants fromhighly polluted areas should not be ignored. This process has asignificant effect on sources and seasonal variations. Note that thisstudy reports the results of preliminary research on this topic. Weneed more precise studies in the future, such as more intensivesampling and analysis of the sources in different seasons, becausethe effect of air masses on the sampling site differs over the fourseasons.

Acknowledgements

This work was funded by the CAS Strategic Priority ResearchProgram Grant NO. XDA05100100 and the National Natural ScienceFoundation of China Grants No. 41021004, 41175107 and 41275139.We gratefully acknowledge this financial support. The authors alsoacknowledge the additional support provided by all members ofthe Gongga Mountain campaign science team.

References

Barletta, B., Meinardi, S., Simpson, I.J., Atlas, E.L., Beyersdorf, A.J., Baker, A.K.,Blake, N.J., Yang, M., Midyett, J.R., Novak, B.J., McKeachie, R.J., Fuelberg, H.E.,Sachse, G.W., Avery, M.A., Campos, T., Weinheimer, A.J., Rowland, F.S.,Blake, D.R., 2009. Characterization of volatile organic compounds (VOCs) inAsian and north American pollution plumes during INTEX-B: identification ofspecific Chinese air mass tracers. Atmospheric Chemistry and Physics 9, 5371e5388.

Borbon, A., Locoge, N., Veillerot, M., Gallooa, J.C., Guillermoa, R., 2002. Characteri-sation of NMHCs in a French urban atmosphere: overview of the main sources.Science of the Total Environment 292, 177e191.

Brown, S.G., Frankel, A., Hafner, H.R., 2007. Source apportionment of VOCs in the LosAngeles area using positive matrix factorization. Atmospheric Environment 41,227e237.

Buzcu, B., Fraser, M.P., 2006. Source identification and apportionment of volatileorganic compounds in Houston, TX. Atmospheric Environment 40, 2385e2400.

Cai, C., Geng, F., Tie, X., Yu, Q., An, J., 2010. Characteristics and source apportionmentof VOCs measured in Shanghai, China. Atmospheric Environment 44, 5005e5014.

Chan, L.Y., Chu, K.W., Zou, S.C., Chan, C.Y., Wang, X.M., Barletta, B., Blake, D.R.,Guo, H., Tsai, W.Y., 2006. Characteristics of nonmethane hydrocarbons (NMHCs)in industrial, industrialeurban, and industrialesuburban atmospheres of thePearl River Delta (PRD) region of south China. Journal of Geophysical Research111, D11304. doi:11310.11029/12005JD006481.

Chan, Y.C., Hawas, O., Hawker, D., Vowles, P., Cohen, D.D., Stelcer, E., Simpson, R.,Golding, G., Christensen, E., 2011. Using multiple type composition data andwind data in PMF analysis to apportion and locate sources of air pollutants.Atmospheric Environment 45, 439e449.

Cheng, H.R., Guo, H., Saunders, S.M., Lam, S.H.M., Jiang, F., Wang, X.M., Simpson, I.J.,Blake, D.R., Louie, P.K.K., Wang, T.J., 2010. Assessing photochemical ozone for-mation in the Pearl River Delta with a photochemical trajectory model. At-mospheric Environment 44, 4199e4208.

Choi, Y.J., Ehrman, S.H., 2004. Investigation of sources of volatile organic carbon inthe Baltimore area using highly time-resolved measurements. AtmosphericEnvironment 38, 775e791.

Choi, E., Choi, K., Yi, S.M., 2011. Non-methane hydrocarbons in the atmosphere of aMetropolitan City and a background site in South Korea: sources and health riskpotentials. Atmospheric Environment 45, 7563e7573.

Geng, F., Cai, C., Tie, X., Yu, Q., An, J., Peng, L., Zhou, G., Xu, J., 2010. Analysis of VOCemissions using PCA/APCS receptor model at city of Shanghai, China. Journal ofAtmospheric Chemistry 62, 229e247.

Guo, H., Wang, T., Louie, P.K., 2004a. Source apportionment of ambient non-methane hydrocarbons in Hong Kong: application of a principal componentanalysis/absolute principal component scores (PCA/APCS) receptor model.Environmental Pollution 129, 489e498.

Guo, H., Wang, T., Simpson, I.J., Blake, D.R., Yu, X.M., Kwok, Y.H., Li, Y.S., 2004b.Source contributions to ambient VOCs and CO at a rural site in eastern China.Atmospheric Environment 38, 4551e4560.

Guo, H., Wang, T., Blake, D.R., Simpson, I.J., Kwok, Y.H., Li, Y.S., 2006. Regional andlocal contributions to ambient non-methane volatile organic compounds at apolluted rural/coastal site in Pearl River Delta, China. Atmospheric Environment40, 2345e2359.

Guo, H., So, K.L., Simpson, I.J., Barletta, B., Meinardi, S., Blake, D.R., 2007. C1eC8volatile organic compounds in the atmosphere of Hong Kong: overview of at-mospheric processing and source apportionment. Atmospheric Environment41, 1456e1472.

Guo, H., Cheng, H.R., Ling, Z.H., Louie, P.K., Ayoko, G.A., 2011. Which emissionsources are responsible for the volatile organic compounds in the atmosphereof Pearl River Delta? Journal of Hazardous Materials 188, 116e124.

Ho, K.F., Lee, S.C., Guo, H., Tsai, W.Y., 2004. Seasonal and diurnal variations of volatileorganic compounds (VOCs) in the atmosphere of Hong Kong. Science of theTotal Environment 322, 155e166.

Huang, C., Chen, C.H., Li, L., Cheng, Z., Wang, H.L., Huang, H.Y., Streets, D.G.,Wang, Y.J., Zhang, G.F., Chen, Y.R., 2011. Emission inventory of anthropogenic airpollutants and VOC species in the Yangtze River Delta region, China. Atmo-spheric Chemistry and Physics 11, 4105e4120.

Lanz, V.A., Henne, S., Staehelin, J., Hueglin, C., Vollmer, M.K., Steinbacher, M.,Buchmann, B., Reimann, S., 2009. Statistical analysis of anthropogenic non-methane VOC variability at a European background location (Jungfraujoch,Switzerland). Atmospheric Chemistry and Physics 9, 3445e3459.

Leuchner, M., Rappenglück, B., 2010. VOC sourceereceptor relationships in Houstonduring TexAQS-II. Atmospheric Environment 44, 4056e4067.

Ling, Z.H., Guo, H., Cheng, H.R., Yu, Y.F., 2011. Sources of ambient volatile organiccompounds and their contributions to photochemical ozone formation at a sitein the Pearl River Delta, southern China. Environmental Pollution 159, 2310e2319.

Mao, T., Wang, Y.S., Xu, H.H., Jiang, J., Wu, F.K., Xu, X.B., 2009. A study of the at-mospheric VOCs of Mount Tai in June 2006. Atmospheric Environment 43,2503e2508.

Moreira dos Santos, C.Y., de Almeida Azevedo, D., de Aquino Neto, F.R., 2004. At-mospheric distribution of organic compounds from urban areas near a coal-fired power station. Atmospheric Environment 38, 1247e1257.

Paatero, P., 1997. Least squares formulation of robust nonnegative factor analysis.Chemometrics and Intelligent Laboratory Systems 37, 23e35.

Paatero, P., Tapper, U., 1994. Positive matrix factorization: a non-negative factormodel with optimal utilization of error estimates of data values. Environmetrics5, 111e126.

Paatero, P., Hopke, P., Begum, B., Biswas, S., 2005. A graphical diagnostic method forassessing the rotation in factor analytical models of atmospheric pollution.Atmospheric Environment 39, 193e201.

Parra, M.A., Elustondo, D., Bermejo, R., Santamaria, J.M., 2009. Ambient air levels ofvolatile organic compounds (VOC) and nitrogen dioxide (NO2) in a medium sizecity in Northern Spain. Science of the Total Environment 407, 999e1009.

J. Zhang et al. / Atmospheric Environment 88 (2014) 297e305 305

Pérez-Rial, D., López-Mahía, P., Tauler, R., 2010. Investigation of the sourcecomposition and temporal distribution of volatile organic compounds (VOCs) ina suburban area of the northwest of Spain using chemometric methods. At-mospheric Environment 44, 5122e5132.

Polissar, A.V., Hopke, P.K., 1998. Atmospheric aerosol over Alaska 2. Elementalcomposition and sources. Journal of Geophysical Research 103, 19045e19057.

Saito, S., Nagao, I., Kanzawa, H., 2009. Characteristics of ambient C2eC11 non-methane hydrocarbons in metropolitan Nagoya, Japan. Atmospheric Environ-ment 43, 4384e4395.

Sauvage, S., Plaisance, H., Locoge, N., Wroblewski, A., Coddeville, P., Galloo, J.C.,2009. Long term measurement and source apportionment of non-methanehydrocarbons in three French rural areas. Atmospheric Environment 43,2430e2441.

Song, Y., Shao, M., Liu, Y., Kuster, W., Goldan, P., Xie, S.D., 2007. Source apportion-ment of ambient volatile organic compounds in Beijing. Environmental Science& Technology 41, 4348e4353.

Su, J., Shao, M., Lu, S., Xie, Y., 2011. Non-methane volatile organic compoundemission inventories in Beijing during Olympic Games 2008. AtmosphericEnvironment 45, 7046e7052.

Tang, J.H., Chan, L.Y., Chan, C.Y., Li, Y.S., Chang, C.C., Liu, S.C., Wu, D., Li, Y.D., 2007.Characteristics and diurnal variations of NMHCs at urban, suburban, and ruralsites in the Pearl River Delta and a remote site in South China. AtmosphericEnvironment 41, 8620e8632.

USEPA, 2008. Epa Positive Matrix Factorization (PMF) 3.0 Fundamentals & UserGuide. USEPA Office of Research and Development.

Wang, T., Guo, H., Blake, D.R., Kwok, Y.H., Simpson, I.J., Li, Y.S., 2005. Measurements oftrace gases in the inflowof SouthChina Seabackgroundair andoutflowof regionalpollution at Tai O, Southern China. Journal of Atmospheric Chemistry 52, 295e317.

Watson, J.G., Chow, J.C., Fujita, E.M., 2001. Review of volatile organic compoundsource apportionment by chemical mass balance. Atmospheric Environment 35,1567e1584.

Xie, Y., Berkowitz, C.M., 2006. The use of positive matrix factorization with condi-tional probability functions in air quality studies: an application to hydrocarbonemissions in Houston, Texas. Atmospheric Environment 40, 3070e3091.

Yuan, Z., Lau, A.K.H., Shao, M., Louie, P.K.K., Liu, S.C., Zhu, T., 2009. Source analysis ofvolatile organic compounds by positive matrix factorization in urban andrural environments in Beijing. Journal of Geophysical Research 114, D00G15.http://dx.doi.org/10.1029/2008JD011190.