Volatile Organic Compound (VOC) measurements in the Pearl River Delta (PRD) region, China

We measured levels of ambient volatile organic compounds (VOCs) at seven sites in the Pearl River Delta (PRD) region of China during the Air Quality Monitoring Campaign spanning 4 October to 3 November 2004. Two of the sites, Guangzhou (GZ) and Xinken (XK), were intensive sites at which we collected multiple daily canister 5 samples. The observations reported here provide a look at the VOC distribution, speciation, and photochemical implications in the PRD region. Alkanes constituted the largest percentage ( > 40%) in mixing ratios of the quantiﬁed VOCs at six sites; the ex-ception was one major industrial site that was dominated by aromatics (about 52%). Highly elevated VOC levels occurred at GZ during two pollution episodes; however, the chemical composition of the VOCs did not exhibit noticeable changes during these episodes, except that the fraction of aromatics was about 10% higher. We calculated the OH loss rate to estimate the chemical reactivity of all VOCs. Of the anthropogenic VOCs, alkenes played a predominant role in VOC reactivity at GZ, whereas the contributions of reactive aromatics were more important at XK. Our preliminary analysis of the VOC correlations suggests that the ambient VOCs at GZ came directly from local sources (i.e., automobiles); those at XK were inﬂuenced by both local emissions and transportation of air mass from upwind areas.


Introduction
The Pearl River Delta (PRD) is located in Southern China, extends from the Hong days in October 1995(Zhang et al., 1998. Between October and December 2001, the highest hourly O 3 average reached 142 ppbv at Tai O, a site on the north-south centerline of the Pearl Estuary (Wang et al., 2003). The daily concentrations of PM 2.5 observed in downtown of GZ reached 111 µg/m 3 in 2002, which is nearly twice the level recommended by the US EPA (65 µg/m 3 , daily) (Li et al., 2005). Such high levels 15 of air pollutants present a serious public health issue. NO x and volatile organic compounds (VOCs) are important precursors of groundlevel ozone. The VOC impact on ozone is closely related to the magnitude and the species emitted from various sources. For instance, liquefied petroleum gas (LPG) leakage played an important role in causing excessive ozone in Mexico City and in 20 Santiago, Chile (Blake and Rowland 1995; Chen et al., 2001). The continuous high levels of atmospheric O 3 in summer in Houston, Texas were caused mainly by reactive VOCs emitted by petrochemical industries (Ryerson et al., 2003;Jobson et al., 2004), and vehicular emissions have contributed more than 50% of ambient VOCs in Beijing city (Liu et al., 2005). Other studies have indicated the importance of biogenic sources 25 of VOCs (Chameides et al., 1988;Shao et al., 2000;Warneke et al., 2004;de Gouw et al., 2005).
In the PRD, VOC speciation and sources have been quite intensively studied. The most representative work, which was conducted in 2000 (Chan et al., 2006) EGU the first snapshot of VOC concentrations in industrial, industrial-urban, and industrialsuburban areas and discussed the importance of industrial and vehicular emissions in shaping the spatial variation of VOCs. The measurements at Tai O (Guo et al., 2006;Wang et al., 2005), a remote site between the PRD region and Hong Kong, illustrated how the characteristics of air masses varied with their point of origin, especially in terms 5 of the differences in regional and local contributions to ambient VOCs at the site. Due to the complexity of VOC variation and the rapid changes in VOC sources in the PRD region, more simultaneous measurements of ambient VOCs with CO, NOx, and O 3 are needed. An understanding of local VOC source profiles will be helpful in interpreting the sources of VOCs in ambient measurements. The PRD air quality 10 monitoring campaign of 2004 represents the first regional study in China designed to gain a better understanding of how ground-level ozone is formed and to determine the sources of fine particles. The measurement of PRD VOCs was a joint effort by the College of Environmental Sciences (CES) of Peking University (PKU); the Research Center for Environmental Changes of Academia Sinica (RCEC), Taiwan; and the De-15 partment of Chemistry of National Central University, Taiwan. Herein we present the data on VOC distribution and speciation obtained at seven PRD sites and we discuss their potential photochemical impacts. We explored the contributions of various VOC sources by analyzing correlations between VOC species as well as the co-variations between VOC species and other gaseous pollutants. 20 2 Field measurements

Sampling sites
We sampled VOCs at seven sites in the PRD during October and November, 2004 (Fig. 1). Two of them -Guangzhou (GZ) and Xinken (XK) -were intensive sites, at which three daily whole air sample (WAS) canisters were collected from 4 October to 3 ences a typical sub-tropical climate. The GZ site is located in the downtown are of the city. We collected canister samples at the roof of a 17-floor building (about 55 m above ground). Xinken lies in a less populated coastal area; it is a rural site located ∼50 km to the southeast of the city center. Ambient air was drawn at the third floor platform of a building (about 10 m above ground). CH is a rural site and HZ is a suburban one, 10 and both are located upwind of the PRD region. We chose DG to examine industrial emissions. FS and ZS, like GZ, are urban sites.
During the PRD air quality monitoring campaign of 2004, abundant sunshine, mild temperature and breeze, and no precipitation characterized the weather. Under the influence of a high-pressure system and stagnant conditions, the boundary layer height 15 was generally within 1 km. At GZ, a northerly wind prevailed (mainly between NNW and NNE) and weakened during the dayime. At XK, a northeasterly wind was dominant (often between N and NE) in the morning, and a sea breeze (a SE or ESE air stream) was observed in late afternoon.

20
We collected WAS in fused silica-lined stainless steel canisters (2 L, 3.2 L, or 6 L). An ozone scrubber (Na 2 SO 3 trap) was installed in the sample line to remove ozone, and a passive capillary (calibrated in advance) was connected to the canister to keep the sampling air flow rate constant. Each

Quantification of VOC species
The analysis of the canister samples was conducted in a laboratory at PKU. Up to 134 species of VOCs were detectable using a cryogenic pre-concentrator (Entech In-5 strument 7100A, SimiValley, CA) and a gas chromatograph (Hewlett Packard, 6890) equipped with two columns and two detectors (see detailed description in Liu et al., 2005). The C 2 -C 4 alkanes and alkenes were separated on a non-polar capillary column (HP-1, 50 m×0.32 mmID×1.05 µm, J&W Scientific) and quantified with a flame ionization detector (FID). The C 5 -C 12 hydrocarbons were separated on a semi-polar 10 column (DB-624, 60 m×0.32 mmID×1.8 µm, J&W Scientific) and quantified uisng a quadrupole mass spectrometer (MS, Hewlett Packard 5973), which was operated in Selected Ion Mode (SIM) with a maximum of six ions being monitored for each time window. First, ambient air samples and internal standards were pumped into the pre-15 concentrator, which has 3-stage cryotraps (Module 1∼3). VOC compounds were initially trapped cryogenically on glass beads of Module 1 at −180 • C by liquid nitrogen; then they were recovered by desorbing at 20 • C to leave most of the liquid H 2 O behind in the first trap. The second cryotrap, which contains Tenax, was cooled to −30 • C, which allows trapping of VOCs while letting CO 2 pass through. From Module 2, VOCs 20 were backflushed at 180 • C then focused again at −180 • C in the Module 3 trap. The Module 3 trap then was rapidly heated to 60∼70 • C in 30 s. Helium was used as the purge gas for the cryogenic pre-concentrator and the carrier gas for the GC. Column HP-1 was initially held at −50 • C for 3 min, then was raised to 164 • C at a rate of 6 • C/min; then to 200 • C at a rate of 14 • C/min, and finally was held for 0.5 min. Column DB-624 25 was programmed to move from 30 • C to 180 • C at a rate of 6 • C/min and then was held for 5 min at 180 • C. Table 1 summarizes the full list of the 134 VOC species that were identified and 14712 the method detection limit (MDL) for each compound is given in EPA TO-15, and the MDL for all measured VOC species ranged from 0.009 to 0.057 ppbv. The response of the instrument to VOCs was calibrated after every eight samples using standard runs of a calibration gas with ambient concentrations.

10
To ensure the quality of the data, we conducted measurement comparison exercises for both standard mixtures and ambient samples. Two planned experiments were involved: 1) analysis at PKU of a known standard gas (provided by D. R. Blake's group from the Department of Chemistry, University of California at Irvine (UCI)); and 2) a blind intercomparison of WAS results measured separately by PKU and RCEC. Table 2 shows the measurements made at PKU for 55 NMHC species in standard gas obtained from UCI; each point represents one species, and error bars were computed from over seven replicate measurements. The correlation between measured concentrations analyzed at the PKU lab and the reference values were good (R 2 =0.96), and the averaged slope was 1.09±0.04. The measured concentrations of alkanes 20 were very close to their reference values, and the relative standard deviation ranged from 0.9% to 9.6%. The relative errors of n-butane, i-butane, n-pentane, 2-methyl pentane, and 2-mehtyl hexane were below 5%; for >C7 alkanes the relative errors were usually between 5.7% and 9.9%. The deviations of 1-butene/i-butene, trans-2-butene, 1-pentene, and 2-methyl-1-butene were 4.5%, 9.1%, 5.9%, and 9.5%, respectively. For 25 isoprene and α-pinene, the deviations from the reference values were relatively larger, reaching 10.7% and 13.4%, respectively. The averaged deviations of aromatics were about 10%. Several scattered points, such as those of cyclopentene, that deviated 14713 Introduction  Figure 3 shows the results for some of the NMHC compounds. For most of the alkanes, the slopes of the linear regression for PKU versus RCEC measurements 5 fell between 0.87 and 1.11, with R 2 values over 0.9. For reactive alkene and aromatics compounds, including butenes, cis-2-pentene, 3-methyl-1-butene, benzene, toluene, xylenes, and trimethylbenzene, the measured mixing ratios calculated by the two labs also agreed well within the combined uncertainties for each system. However, the average α-pinene concentration measured at PKU was about 30% lower than that 10 from RCEC lab. Figure 4 shows the averages of the total quantified PRD VOC mixing ratios and the relative contributions from the major VOC groups. The highest total VOC mixing ratio 15 was measured at DG (an industrial area), followed by the major urban site GZ. The levels at XK, FS, and ZS were quite similar to each other. All three sites lie downwind of industrial areas and/or major urban centers. The two lowest VOC values were recorded in CH and HZ, which lie upwind of the major cities. Figure 4 also shows that alkanes constituted the largest group of VOCs at six (CH, 20 HZ, GZ, FS, ZS, and XK) of the seven sites, accounting for over 40% of the total. In contrast, exceptionally high values of aromatics (about 52% of the total VOCs) characterized DG, the industrial site. The DG aromatics likely resulted from emissions from the plants associated with textiles, furniture manufacturing, shoemaking, printing, and plastics. XK lies downwind of DG; consequently, it had the second highest faction of Introduction  Table 2 summarizes the average concentrations and variations of 54 VOCs at GZ and XK, and Table 3 lists the 10 most abundant species observed at these two sites compared with results from a previous studies in Hong Kong and other Chinese cities (Barletta et al. 2005;Guo et al. 2006). In general, the PRD VOC mixing ratios fell within the ranges reported for other Chinese cities. A pronounced similarity existed between 5 XK site and Hong Kong's Tai O site, a coastal site at the southern tip of the PRD region. Large fractions aromatic compounds, especially toluene, were observed at both sites. And XK and Tai O had similar levels of light alkanes as well. Both sites lie downwind from industrial sources of the PRD region, which might explain the similarities.

Mixing ratios of VOC species at Guangzhou and Xinken
In contrast, GZ had the highest concentration of propane, likely due to the 10 widespread domestic and vehicular use of LPG. High levels of acetylene, toluene, ethylene, and ethane at this site probably originated from several anthropogenic sources such as vehicle exhaust, petrochemical industries, and industrial uses of solvents. Vehicular emissions were clearly identifiable from the significant levels of isobutane, isopentane, and benzene.

EGU
of VOCs may be attributed to different sources or processes. In the case of O 3 , there were 14 days with hourly averages exceeding 80 ppbv, which is the second grade of China's NAAQS. However, a clear relationship between these high ozone days and either VOC levels or NO and CO levels was not evident. This may reflect the fact that ozone level is controlled by both advection and local photochemistry.

5
The observations for XK are displayed as a time series in Fig. 6. The NO levels were significantly lower at XK than at GZ. The XK CO levels, on average, also were lower. In addition, the correlations between NO and CO enhancements at XK were much weaker than those for GZ. Large VOC enhancement episodes, with levels more than a factor of two greater than the typical values, occurred seven times between 7 October 10 and 18 October. Total VOC level peaked at over 277 ppbv at XK on the morning of 12 October, but few corresponding changes occurred in NO and CO (Fig. 6a). The O 3 levels observed in XK exceeded 80 ppbv on 23 days within the study period , and were generally higher than those seen at GZ. Figure 7 compares the episode days versus background (or normal) conditions at GZ 15 and XK. The average of the relative contributions from alkanes, alkenes, and aromatics remained quite constant or fluctuated within a narrow range at GZ and XK (Fig. 7a). This suggests that the high VOC levels during the episode days are likely due to meteorological conditions favorable for accumulation of pollutants. Figure 7b illustrates that during the pollution episodes at GZ, total VOC levels were about 2-4 times higher than 20 those from non-episode days.

Diurnal variation at Guangzhou and Xinken
3.3.1 Guangzhou Figure 8 illustrates the diurnal patterns of primary and secondary pollutants, using data from 21 October at the GZ site as an example. The diurnal trend of total VOCs followed 25 a pattern similar to that of the primary pollutants, such as CO and NO, but it differed from that of O 3 . The NO levels were generally over 50% of the NO y concentrations, time than in the morning and afternoon, which would also contribute to the higher level of pollutants.

Xinken
The diurnal patterns of VOC gases measured at XK were quite different from those at GZ (Fig. 9). CO and VOC tracked each other on 9 October, whereas no consistent diur-10 nal variation for either CO or VOCs occurred on 21 October. Unlike at GZ, ambient NO remained at much lower levels and constituted only a small fraction of NO y , suggesting that the air mass was more chemically aged at XK. The ambient NO and NO y spikes occurred around 10:00-11:00 a.m. on both 9 October and 21 October, causing distinct decreases in O 3 due to titration. As no corresponding enhancement in CO and VOCs 15 occurred and SO 2 displayed a similar trend as NO y , these plumes probably originated from power plant emissions from upwind areas. The observations at XK suggest that advection transport likely has a larger impact on local air quality than do the local traffic sources.
Ozone had higher peak concentrations and much rapid variations at XK than those 20 recorded in GZ. The higher ozone levels at XK were accompanied by lower levels of VOCs and NO, indicating that the ozone did not result solely from local photochemistry. Because XK lies downwind of an urban region, the mixing ratios of VOCs in the early morning were higher than those from the same time period at GZ because of the accumulation of VOCs at night as well as transport from upstream urban areas. This 25 phenomenon appears to be more apparent during periods of northerly wind. The wind vectors at XK display a diurnal pattern; frequently, the northerly wind shifted to the south during the nighttime hours or in the early morning, and the land-sea breeze 14717 Introduction EGU circulation had some effects on the convection and recirculation of air pollutants in the region.

VOC reactivity at Guangzhou and Xinken
OH loss rate (L OH ) is frequently used as a gauge to measure the initial peroxy radical (RO 2 ) formation rate, which might be the rate-limiting step in ozone formation in polluted air (Carter, 1994). While this approach does not account for the full atmospheric chemistry of the compounds considered, it does provide a simple approach to evaluate the relative contribution of individual VOCs to daytime photochemistry (Goldan et al., 2004). L OH is calculated as the product of the OH reaction rate coefficient (k OH i ) and the ambient mixing ratio ([VOC] i ) of a given compound: We used Atkinson and Arey's (2003) published k OH i . Table 4 lists the OH loss frequencies of the main VOC groups at GZ and XK. Of the anthropogenic VOCs, reactive olefins dominated the reactivity at GZ. The alkenes at GZ represented 28.9% of the overall mixing ratios of the measured VOCs and ranged 15 from 24.7 to 305.5 ppbv, and they accounted for over 65% of the overall L OH s. In contrast, the alkanes represented 47.1% of the overall mixing ratios but only a small fraction (13%) of the overall L OH s. The contribution of aromatics to VOC reactivity was ∼20%, which was comparable with its percentage of the total mixing ratios.
At XK, the overall L OH s were lower than those at GZ, and the relative contributions 20 from aromatics and alkenes to VOCs reactivity were similar. At lower mixing ratios of total VOCs, the L OH s of alkenes exceeded those of aromatics, and with an increase of the total mixing ratios, the contributions of aromatics were enhanced. For more polluted air, the roles of aromatics were more important in photochemical processes. Because alkenes and aromatics played a significant role in the reactivity of VOCs 25 at GZ and XK, in the subsequent discussion we focus on the contributions of different 14718 Interactive Discussion EGU species of alkenes and aromatics at the two sites. At GZ, all alkenes were classified into groups by their carbon number (Fig. 10a). The most important contributors to the L OH s was C 4 alkenes (butenes), closely followed by propene and pentenes. Isoprene was not the dominant species as expected; this can be explained by the low emissions from plants in the urban center. In the case of clean air, the contribution of isoprene 5 and monoterpenes was slightly increased. Hexenes and heptenes played a smaller role in OH loss due to their low concentrations. Figure 10b shows the percentages of aromatic groups at XK. Together with xylenes, toluene played a predominant role in the reactivity of VOCs. Although trimethyl-benzenes had larger rate coefficients, they made a minor contribution because of their low concentrations. The contribution of benzene, 10 which was the most inert compound among the observed aromatics, decreased from the clean air to the polluted air.

Identification of VOC sources at Guangzhou and Xinken
Determining the PRD VOC sources was a rather complex task because it involved numerous sources in different cities. To assess the VOC sources for four major groups 15 -alkanes, alkene, isoprene, and aromatics -we examined correlations among the measured ambient VOC species and compared them with the known correlations from primary emission sources. Acetylene usually is associated with sources of incomplete combustion of fossil fuel, including combustion of gasoline, diesel, and LPG in vehicles and domestic use of LPG 20 for cooking (Blake and Rowland, 1995;Goldan et al., 2000). We used methyl tert-butyl ether (MTBE), a gasoline additive used to enhance its octane rating and combustion efficiency, as an indicator for exhaust of gasoline-powered vehicles (Blake and Rowland, 1995;Chang et al., 2003). Figure 11 shows strong correlations of acetylene and ethylene with MTBE at GZ. Thus, it is reasonable to conclude that gasoline-powered 25 vehicles are mostly likely the major sources of acetylene and ethylene at GZ.
The ratios of ambient concentrations of two hydrocarbons with similar reactivity remain constant at the value equal to their relative emission rates from sources ( et al., 2000;Jobson et al., 2004). As mentioned above, the C 4 -C 5 alkenes were the most reactive groups at GZ. Correlations between selected butene and pentene parings with similar k OH values are shown in Fig. 12, compared to the results from Pearl River Tunnel samples . The trans-2-butene and cis-2-butene in the atmosphere at GZ displayed excellent correlation with the tunnel samples; the slope 5 of the regression line of ambient data (1.067) is very close to that of the tunnel samples (1.074). The trans/cis-2-pentenes obtained at GZ and XK correlated to each other very well, and again the regression line fit nicely with the The trans/cis-2-pentenes data points measured from the tunnel experiment (Fig. 12b). The trans/cis-2-pentenes levels obtained at XK were more scattered than that from GZ site at the lower concentrations 10 of these two species, which were likely impacted by other sources. These findings suggest that reactive 2-butenes and 2-pentenes at GZ and XK resulted primarily from vehicle exhaust emissions. The widespread use of LPG can be another significant source of VOCs. Propane is one of the important components of LPG fuel. For LPG-powered vehicles, major emissions include light alkanes (i.e., propane, isobutene, and n-butane) as well as some alkenes (e.g., butenes). The correlations of n-butane and isobutane with propane were significant (Fig. 13) at GZ with slopes of 0.48 (correlation coefficient r=0.97) and 0.28 (correlation coefficient r=0.97), respectively. The values of these two slopes agree well with those measured in Mexico City (0.458 and 0.210), where VOCs originate 20 mainly from LPG leakage (Blake and Rowland, 1995). These correlations suggest that gasoline-powered vehicles and LPG use are two important sources of light alkanes.

EGU
Acetylene and propane have similar photochemical lifetimes but come from different sources: gasoline-powered vehicles and LPG use, respectively. The ratio of these two compounds at a given site can be used to assess the relative importance of gasoline 25 and LPG sources (Goldan et al., 2000;Zhang et al., 2004). The ratios of acetylene and propane at different sites provide an overview on a regional scale of the relative importance of these two sources to ambient alkane species levels. Figure 14a shows the plot of acetylene versus propane at six sites (GZ, XK, CH, HZ, FS, and ZS). The ratios than that measured in the mobile source samples. These findings imply that LPG leakage contributed more to ambient VOCs at GZ than at the other sites, probably due to the higher percentage of LPG used for residential energy and for public transportation in Guangzhou City.
The major source of benzene is vehicular emissions, whereas toluene is associated 10 with industrial emissions, solvent and fuel storage, and vehicle exhaust (Bravo et al., 2002;Wang et al., 2002;Na et al., 2003). Toluene was the most abundant species of VOCs observed in industrial areas of the PRD; it is emitted directly from shoemaking, printing, leather manufacturing, furniture making, coating and chemical bonding agent production, and other chemicals plants (He et al., 2002;Chan et al., 2006). In this study 15 we used the toluene/benzene ratio as a tool to evaluate the relative importance of vehicular and industrial emissions on a regional basis. Figure 14b shows the correlations between toluene and benzene at GZ, XK, and DG compared with those measured from tunnel samples in previous studies (Fu, 2005;Fu et al., 2005). The slopes of toluene versus benzene at XK and DG were similar; in both locales shoemaking is a major in-20 dustry housed in widespread factories. The higher toluene levels at XK were impacted by the additional input of industrial emissions from DG that were advected to XK from DG. The GZ data fell between the linear regression lines of the tunnel and DG data, suggesting that ambient toluene at GZ was affected by both automotive and industrial sources.

25
Isoprene is one of the most reactive hydrocarbon species and is used as a tracer for biogenic emissions. Vehicular exhaust also is a source of isoprene in cities (Borbon et al., 2001). We found a good correlation (r=0.91) between isoprene and 1,3-butadiene in the Pearl River tunnel (Fig. 15). However, the mixing ratios of ambient isoprene did EGU not correlate well with 1,3-butadiene measurement from the GZ site (r=0.51). Therefore, we attribute ambient isoprene at GZ to biogenic sources.

Conclusions
Mixing ratios and chemical speciation of VOCs were measured intensively at GZ and XK as well as at five more sites in the 2004 Air Quality Monitoring Campaign in the PRD.

5
We quantified up to 134 VOCs species, and the total VOC levels varied from 10 ppbv to over 200 ppbv. GZ had a very high level of propane, whereas Xinken, the suburban site, had high mixing ratios of aromatics. The chemical compositions differed greatly among the seven sites, reflecting the heterogeneous distribution of VOC sources in the region.
We used the OH loss frequency to assess the chemical reactivity of VOC species. Reactive alkenes and aromatics influenced the VOC reactivity at GZ and XK, respectively, whereas alkanes, which constituted the largest portion (>45%) of overall VOC mixing ratios, comprised merely <15% of the overall OH loss rate. At GZ, butenes showed the greatest relative contribution, closely followed by propene and pentanes; 15 the heavier alkenes with low mixing ratios accounted for a small faction of total VOC reactivity. At XK, toluene and C 8 reactive aromatics made the largest contribution to the OH loss rate.
Using correlations among VOC compounds, we evaluated the relative importance of local emissions of VOCs at different sites. We attributed the ambient acetylene, 20 ethylene, and other light alkenes at GZ to the local emissions from gasoline-powered vehicles. The high level of propane originated mostly from vehicles that consumed LPG fuel. Aromatic species at GZ were influenced by on-road vehicle emissions, industrial solvent use, and fuel evaporation. Due to the limited data about the compositions of LPG at GZ, we could not quantify the contribution of LPG exhaust and its leakage. The 25 toluene/benzene ratio showed that VOCs were affected by emissions from solvent usage, fuel storage, and industrial emission. Before we draw a clear conclusion, however, pending on prevailing wind. For example, while the reactive butenes and pentenes at XK were primarily from local emissions, the aromatics at XK did not originate solely from local emission and likely were impacted by transport from the upwind industrial area of DG. Thus, controlling ozone levels at XK should not be confined soley to management of local emissions. Detailed investigation at the site (e.g., analysis of the VOC     The correlation between toluene and benzene for GZ, XK, and Dongguan (DG), comparing ambient data to the Pearl River Tunnel study (solid squares). The solid and dashed lines represent the regression lines for the results from tunnel samples and ambient data at DG, respectively.