Monoaromatic compounds in ambient air of various cities: a focus on correlations between the xylenes and ethylbenzene

Speciation of o -xylene, m -xylene, p -xylene and ethylbenzene was performed by gas chromatography from ambient air and liquid fuel samples collected at various locations in 19 cities in Europe, Asia and South America. The xylene ’ s mixing ratios were compared to each other from the various locations, which included urban air, tra $ c air and liquid fuel. For all samples, the xylenes exhibited robust correlations, and the slopes remained constant. The m -xylene/ p -xylene ratio was found to be 2.33 $ 0.30, and the m -xylene/ o -xylene ratio was found to be 1.84 $ 0.25. These ratios remain persistent even in biomass combustion experiments (in South America and South Africa). Comparing the xylenes to toluene and benzene indicate that combustion, but not fuel evaporation, is the major common source of the xylenes in areas dominated by automotive emissions. Although a wide range of combustion types and combustion e $ ciencies were encountered throughout all the locations investigated, xylenes and ethylbenzene ratios remained persistent. We discuss the implications of the constancies in the xylenes and ethylbenzene ratios on atmospheric chemistry. ( 2000 Elsevier Science Ltd. All rights reserved.


Introduction
Hydrocarbons are signi"cant components in urban air because of combustion, solvent and fuel evaporation and tank leakage. Among them, most of the aromatic compounds are listed as toxic air contaminants (e.g. benzene) or potential toxic air contaminants (e.g. toluene, xylenes) (Hanson, 1996). During daytime hours, once released into the atmosphere, aromatic components undergo OH oxidation, and thus participate in the formation of urban and suburban photochemical smog. The total ozone attributable to each organic compound is in#uenced greatly by the relative concentration of each species. Based on 1987 non methane hydrocarbon emissions in the United Kingdom, Derwent and Jenkin (1991) calculated that m-xylene, trimethylbenzenes and C }C alkenes produce as much or more ozone than ethylene. Furthermore, the reaction products from the atmospheric oxidation of individual alkylbenzenes include potential toxic and mutagenic compounds such as aromatic aldehydes, quinones, dicarbonyls, epoxides (Kwok et al., 1997;Kleindienst et al., 1999) and secondary organic aerosols (Odum et al., 1997;Kourtidis and Ziomas, 1999;Kleindienst et al., 1999).
Because of the potential hazards associated with the alkylbenzenes, it is important to accurately determine the atmospheric mixing ratios of these gases and to identify their main sources. Although a large number of studies have investigated these issues, only a few of the laboratories making the measurements were able to resolve the three xylenes and ethylbenzene (Wathne, 1983;Zweidinger et al., 1988;Olson et al., 1992;Tsujino and Kuwata, 1993;RappengluK ck et al., 1998). However, each of the above-mentioned study investigated one particular site at a time, and the number of samples was limited. All the other studies reported the sum of m-and p-xylene mixing ratios. Because of the lack of speciated xylene data, in calculating the photochemical ozone creating potential (POCP) and photochemical PAN creating potential (PPCP) values, Derwent and Jenkin (1991) and Derwent (1995) assumed that m-and p-xylene were equally present in the atmosphere in the UK.
The current study presents hydrocarbon speciation including clean separation of the xylenes and ethylbenzene by multi-column gas chromatography. The purpose is to determine the concentrations of m-and p-xylene and other alkylbenzenes concentration ratios in a large number of samples taken at various locations throughout the world. This allows us to con"rm and improve the method proposed by  to determine the extent of photochemical reactivity in ambient air by measuring the ratios between (m#p)-xylene (taken together) and ethylbenzene. The "nal purpose is the identi"cation of the main source of these gases.

Sampling
Air samples were collected in 2-l electropolished stainless-steel canisters. Nineteen cities in Europe, Asia and South America were investigated at various times of the year in 1994, 1996, 1997 and 1998 (Table 1).`Urban aira samples were collected throughout each city, in parks and aerated areas, far removed from roadways or any known local source in#uence. The collections were performed within the city limits, not in the suburbs.`Tra$c aira samples were collected on roadways or in motorway tunnels with #uid tra$c. A relatively long sampling period was used for the tra$c samples in order to damp out any in#uence of one particular car passing the sampling site. Each sampling period was integrated over several minutes.

Sample analysis
The whole air samples collected during the city studies were analyzed for NMHCs, halocarbons and alkyl nitrates using a four gas chromatograph (GC), six column, six detector system. A 250 ml (STP) aliquot from each canister was trapped on a glass bead "lled -inch stainless-steel loop immersed in liquid nitrogen. No traps were used to remove CO or water from the samples. The total volume sampled was measured by pressure di!erence using a capacitance manometer. Once the sample was trapped, the preconcentration loop was isolated and warmed to 803C. When the four independently programmed GCs were at their appropriate initial temperatures, they were allowed to equilibrate for exactly 20 s, then the sample was injected. The hydrogen carrier gas #ushed the sample loop to the splitter which quantitatively and reproducibly split the #ow in six ways to the respective columns. Flame ionization detectors (FIDs) were used with a 30 m;0.53 mm ID Alumina PLOT column (J&W Scienti"c, Folsom, CA) for the light hydrocarbons (C }C ), a 60 m;0.25 mm ID, 1 m "lm thickness DB-1 column (J&W Scienti"c, Folsom, CA) for C }C NMHCs, and a 60 m;0.25 mm I.D., 0.25 m "lm thickness Cyclodex-B column (J&W Scienti"c, Folsom, CA) for the C }C NMHCs. Connected to the electron capture detectors (ECDs) were a 60 m;0.25 mm ID, 1 m "lm thickness DB-1 column (J&W Scienti"c, Folsom, CA) for C }C halocarbons, a 60 m;0.25 mm ID, 0.5 m "lm thickness DB-5MS column (J&W Scienti"c, Folsom, CA) for C }C halocarbons and methyl nitrate, and a 60 m;0.25 mm ID, 0.25 m "lm thickness Rtx-1701 (Restex Corporation, Bellefonte, PA) column for C }C halocarbons and C }C alkyl nitrates. For this system con"guration, 27.0% of the carrier #ow was directed to the PLOT/FID, 18.9% to the DB-1/ECD, 16.2% to the DB-1/FID, 16.1% to the DB-5MS/ECD, 11.0% to the Cyclodex-B/FID, and 10.8% to the Rtx-1701/ECD.
For air sample analysis, the working standard was a pressurized whole air sample contained in an Aculifetreated Luxfer cylinder that was collected at Niwot Ridge, CO. It was assayed after every four samples in the same manner used to analyze the canister samples. The time required for one complete cycle of sample trapping, injecting, and chromatographic separation, was 20.5 min. To monitor any drift in the standard or the analytical system, four other pressurized whole air standards were also assayed at daily and weekly intervals throughout the analysis period. Mixing ratios of all quanti"ed gases in the standards exhibited no statistically signi"cant changes (less than 1 ) over the duration of the analysis periods. The measurement precision for the xylenes was 5% while the limit of detection was 10 pptv for the 250 ml (STP) sample size. The absolute calibration of these standards and further analytical details are described in Blake et al. (1992Blake et al. ( , 1994 and Sive (1998).
Liquid gasoline samples were diluted in n-pentane, then analyzed by direct injections into a Hewlett-Packard gas chromatograph equipped with a mass spectrometer (G1800A GCD) utilizing electronic impact. The apparatus was equipped with a Cyclodex B (0.25 mm I.D., 60 m, 0.25 m "lm thickness) capillary column.

Gas chromatography performances
In order to measure as many hydrocarbons as possible, many research groups rely on analyses performed using a single-gas chromatography column system, containing a very versatile stationary phase, allowing for separation of a wide range of gases typical of urban air. However, using these versatile phases (mostly DB-1 (100% dimethylpolysiloxane), or PLOT Al O or DB-5 MS (5%-phenyl-methylpolysiloxane)), it is not possible to obtain baseline resolution for the m-and p-xylene. Thus, these gases are often reported as (m#p)-xylene (Table 2) and it is usually assumed that they are of equal concentration.
The Cyclodex-B column, which has a much di!erent stationary phase ((14%-cyanopropylphenyl)-methylpolysiloxane blended with cyclodextrin), is primarily used to resolve chiral compounds. However, we have found that baseline resolution of the xylenes is attainable using this column. The chromatograms from a DB-1 column, DB-5MS column, and a Cyclodex-B column of a synthetic standard containing the xylenes and other NMHCs are shown in Fig. 1. This "gure illustrates how di!erently these three columns separate the xylenes. Of these columns, only the Cyclodex-B is capable of baseline resolution for the xylenes. Furthermore, because of retention time shift, and because of coeluting unidenti"ed and identi"ed peaks, the quality assurance of hydrocarbon analysis is largely improved with a multidimensional column analytical system including at least one speci"c phase for high volatile hydrocarbons, and one speci"c phase for low volatile hydrocarbons. It is clear from Table 2 and Fig. 1 that such a system is necessary for the analysis of a large number of hydrocarbons (C }C ), and in particular for m-and p-xylene speciation with good resolution in urban samples, which are heavily loaded.

Ambient air and liquid fuel analysis
As seen in Table 3, median urban mixing ratios of NMHCs vary over a wide range. Because mixing ratios are a!ected by the boundary layer height ratios between several NMHCs including C }C monoaromatics have been calculated.

The xylenes and ethylbenzene ratios
3.2.1.1. The sum of m-and p-xylene versus ethylbenzene mixing ratios.  have shown that this ratio was constant throughout di!erent sources such as vehicle exhaust, solvent petrol and fuel evaporation. They found a (m, p)-xylene/ethylbenzene (X/E) Wathne ( Capillary columns when not speci"ed. HC"hydrocarbons. Equivalent to DB-1 phase. ratio of 3.6 in the Sydney area, which compared well to a few other studies. Table 4 summarizes this ratio calculated from more recent data found in the literature, and compared to our results. The X/E ratio in di!erent source samples is relatively constant (from 2.8 to 4.6) except from the work by Siegl et al. (1999), who found a ratio of 15.5. However, these authors found extremely low emissions of xylenes and ethylbenzene, likely leading to large uncertainties. Including our values, the overall average value for the X/E ratio in di!erent source samples is 3.5$0.5. Far from local source in#uence, in urban and suburban areas, the X/E ratio is lower, and spreads over a larger range, especially for low values (from 1.3 to 4.5). This indicates that the xylene and ethylbenzene are emitted by the same major sources, but decay at di!erent rates from OH-oxidation in the atmosphere. Therefore,    as mentioned by , the X/E ratio is a tool to investigate the photochemical age of an urban plume. However, the di!erence in oxidation rate between m-xylene and ethylbenzene is even larger, making this ratio an even better tool to determine the photo chemical age of an urban plume (see below).

m-Xylene versus p-xylene mixing ratios
In our urban air samples, p-and m-xylene mixing ratios vary from 0.13 to 3.25 ppbv and from 0.32 to 7.05 ppbv, respectively. In each city, these mixing ratios are enhanced in tra$c samples, indicating a signi"cant vehiclar source. Although m-and p-xylene concentra-tions span a large range, they are exceptionally well correlated (R"0.99) in urban air throughout the di!erent locations studied and during the di!erent times of the year during the years of the study (Fig. 2a). This indicates that p-and m-xylene sources are identical throughout the di!erent urban environments. In tra$c air samples, these compounds are also well correlated throughout the different locations (Fig. 2b), and the slope is not signi"cantly di!erent from the one obtained from ambient urban air samples.
The composition of fuels varies widely by city and type of gasoline (Sigsby et al., 1987;Jemma et al., 1995). However, in our data, there is a very good correlation between m-and p-xylene concentrations throughout the di!erent fuels from the di!erent cities studied (Fig. 2c). The slope is also in very good agreement with those obtained for both urban air and tra$c air samples. Because p-and m-xylene's boiling points are very close (Table 5), the good slope agreement may indicate that in tra$c air and urban air, p-and m-xylene concentrations result from fuel evaporation. However, this is only partially true because xylenes are also generated during combustion processes (Sigsby et al., 1987). Therefore, we have examined the alkylbenzene versus combustion and evaporative markers (see below).
We also compared the m-and p-xylene mixing ratios in totally di!erent types of combustion. We considered three di!erent biomass burning data sets where background concentrations of the xylenes were below detection limit. Airborne samples collected near "res in Brazil } 1995 (Ferek et al., 1998) and during the Transport and Atmospheric Chemistry near the Equator}Atlantic (TRACE A) project over both Brazil and Africa } 1992 (Blake et al., 1997). A good correlation between m-and p-xylene was found (Fig. 2d), with a slope not signi"cantly di!erent from those obtained in urban air, tra$c air and fuels. The slightly lower correlation coe$cient can be attributed to the much lower concentrations observed, closer to the detection limit, thus, larger measurement error bars.
To our knowledge, this is the "rst time that m-and p-xylene are analyzed on a large number of samples collected at various places around the world. Their relative mixing ratios remain surprisingly constant whatever the location, type of combustion and the type of fuels used: using all our 706 samples, we obtain a value of 2.33$0.30 for the m-xylene/p-xylene ratio.

m-Xylene versus o-xylene mixing ratios.
As with m-and p-xylene, o-and m-xylene are also well correlated, and their relative mixing ratios remain constant (within the estimated error) for urban air, tra$c air, biomass combustion and fuel samples. Using all of our 706 samples (including urban and tra$c air, fuel samples and "re samples), we obtain a value of 1.84$0.25 for the mxylene/o-xylene ratio.

m-Xylene versus ethylbenzene mixing ratios. m-
Xylene and ethylbenzene are also very well correlated for heavy loaded urban, tra$c, and liquid fuel samples, and the slope obtained remains constant: using all the samples taken close to the sources (all except`urban lowa and`biomass burninga samples), a value of 2.24$0.33 for the m-xylene/ethylbenzene ratio is obtained. The`urban low concentrationsa data sets exhibit slope values signi"cantly lower. Therefore, m-xylene and ethylbenzene are emitted by the same sources, but because m-xylene's atmospheric lifetime is signi"cantly shorter than that of ethylbenzene (respectively, 11.8 h and 1.6 days; see Table 5), their relative mixing ratios decrease rapidly far from their common sources. This indicates that the mxylene/ethylbenzene ratio provides a good tool to investigate the age of an urban plume.

Toluene versus ethylbenzene mixing ratios
It is clear from above that the three xylenes and ethylbenzene are emitted by the same sources in all the urban locations investigated during this study. Thus, we compared one of them, ethylbenzene, to a ubiquitous Table 6 Correlation between the C , C and C monoaromatic compounds, and with ethene (combustion marker), and n-pentane (solvent evaporation marker) (in this study) aromatic compound in urban air, toluene. We chose ethylbenzene because its atmospheric lifetime towards OH radicals is comparable to that of toluene (Table 5). Toluene versus ethylbenzene plots exhibit good correlation in tra$c samples (R"0.94) but are poorly correlated in liquid fuel samples (R"!0.50) ( Table 6). Although toluene and ethylbenzene are known to be emitted by both fuel evaporation and combustion processes (Sigsby et al., 1987), the good correlation in tra$c samples and the poor correlation in fuel samples indicate that combustion is the dominant common source for these compounds in areas dominated by automotive emissions.
Although toluene and ethylbenzene have comparable atmospheric lifetimes, the`urban aira samples exhibit a signi"cantly poorer correlation between toluene and ethylbenzene and a higher slope value as compared to tra$c samples. This may be the consequence of extra sources of toluene in cities, such as architectural surface coatings, graphic arts, industrial solvents and chemical feedstock (Aronian et al., 1989;Harley et al., 1992).
Toluene and ethylbenzene are also very well correlated in biomass combustion samples (Table 6), but the slope is signi"cantly higher than in tra$c air samples.

Benzene versus ethylbenzene mixing ratios
Benzene is also a ubiquitous compound in urban air, and there is a strong desire to reduce its emissions because it is carcinogenic. The correlation between ethylbenzene and benzene lead to the same conclusions as for toluene versus ethylbenzene except that the correlation coe$cients are lower, in part because their atmospheric lifetimes towards OH radicals are di!erent (Table 5).

Toluene versus benzene mixing ratios
Toluene and benzene are well correlated throughout tra$c, biomass burning and fuel samples, while the slopes vary widely: toluene is more concentrated in all the city samples, whereas benzene is dominant in biomass burning samples. The correlation is poorer in`urban aira samples, likely because of extra sources other than combustion, and also because of the di!erence in their atmospheric lifetimes towards OH radicals.

Inyuence of catalytic converters
It has been previously shown that concentrations correlation between benzene and toluene and between benzene and C -alkylbenzenes are very good in car exhaust without catalytic converters whereas they are poor in car exhaust equipped with catalytic converters (Heeb et al., 1999). One measure of catalytic converter e$ciency is the ethene/ethyne ratio: it is 3 or above for well maintained catalyst-equipped vehicles whereas it is closer to 1 for non-catalyst vehicles (Hoekman, 1992;Du!y and Nelson, 1996). In our samples, the ethene/ethyne ratio is closer to 1 than to 3 in all the locations investigated. This suggests that the proportion of active catalytic converters was small or that the majority of emissions came from non-catalyst equipped vehicles. Therefore, the good correlations observed in tra$c samples between toluene and ethylbenzene, between benzene and ethylbenzene and between toluene and benzene are in good agreement with the study by Heeb et al. (1999). However, C -alkylbenzenes ratios should not be signi"cantly a!ected by the quantity of active catalytic converters. The reason for this is that only benzene/alkylbenzenes ratios are signi"cantly a!ected by active catalytic converter use because it is likely that alkylbenzenes are dealkylated (to form benzene) by the converters (Hoekman, 1992;Du!y and Nelson, 1996;Heeb et al., 1999). Furthermore, although a relatively signi"cant evolution is observed in ethene/ethyne ratios in Paris (1994, 1996 and 1998), the corresponding m-xylene/p-xylene ratios remain statistically constant. Finally, one can see from Fig. 3 that the ethene/ethyne ratio does not signi"cantly in#uence the m-xylene/p-xylene ratio, indicating that the use of catalytic converters does not in#uence the m-xylene/p-xylene ratio which remains statistically constant.

Comparisons between the C 2 -alkylbenzenes and source emission markers
It is obvious from above that the C -alkylbenzenes are emitted by the same sources. To distinguish between combustion and solvent evaporation sources, two comparisons have been made: one between the C -alkylbenzenes and ethene, which is a marker for combustion, and one between the C -alkylbenzenes and n-pentane which is a marker for solvent evaporation (even if this component is also emitted in combustion processes). Ethylbenzene has been chosen representative of the C alkylbenzenes because its atmospheric lifetime towards OH-oxidation is comparable to that of ethene and npentane (Table 5). The comparisons between ethene and ethylbenzene and between n-pentane and ethylbenzene show a very low correlation throughout all the cities investigated ( Table 6). The correlation coe$cients are particularly low in tra$c samples (R"0.31 and 0.22, respectively). Investigating each city individually, one "nds good correlation between ethylbenzene and ethene and between ethylbenzene and n-pentane, with slopes varying widely (from 0.035 to 0.26, and from 0.13 to 0.98, respectively). This explains the poor correlation found when all locations were grouped together. In most cities, urban aira samples show both ethylbenzene/ethene and ethylbenzene/n-pentane ratios consistent with those in the tra$c samples (Table 7). This indicates that the unique urban source for C -alkylbenzenes is automotive emissions, and the comparisons between toluene and ethylbenzene indicates that combustion may be the major source for these compounds (see above). However, this does not apply for three cities: Santiago (November), Krakow, and Prague.
In Santiago (November), the ratio between ethylbenzene and ethene is signi"cantly higher in`urban aira samples than in tra$c samples, indicating that extra sources (other than automotive) of ethylbenzene exist in the urban area. Because the ratio between ethylbenzene and n-pentane is statistically the same in`urban aira samples and in tra$c samples, the extra sources may be solvent evaporation from factories, or chemical feedstock. Table 7 Comparison between one C alkylbenzene and (i) one combustion marker (ethene) and (ii) one evaporation marker (n-pentane) for each city in urban air and tra$c air in this study In Krakow, the ratio between ethylbenzene and ethene is signi"cantly lower in`urban aira samples than in tra$c samples, however, only one tra$c sample was collected in this city, and the mixing ratios may not represent the mean values for automotive emissions in Krakow.
In Prague, as in Krakow, only one tra$c sample was collected. However, an unknown source of ethylbenzene was determined. In`urban aira samples the actual ethylbenzene versus n-pentane correlation is: [ethylbenzene]"0.13[n-pentane]#238 (pptV), with R"0.82. Other good correlation (R'0.80) with a Y intercept signi"cantly di!erent from zero was also observed in Prague for ethene versus n-pentane, o-xylene versus n-pentane, m-xylene versus n-pentane, p-xylene versus npentane, benzene versus n-pentane, toluene versus n-pentane, n-butane versus n-pentane, i-butane versus n-pentane, ethyne versus n-pentane and propene versus n-pentane. Therefore, there is an unknown source containing ehtylbenzene, o-, m-, p-xylene, benzene, toluene, n-butane, i-butane, ethene, ethyne, propene but no npentane (nor i-pentane), but further studies need to be done in this city to identify this source.

Conclusions and implications for atmospheric chemistry
The persistent ratios obtained between m-, p-, o-xylene and ethylbenzene concentrations in all the locations investigated indicate that these compounds have the same sources everywhere in the world. Comparing them to ethene (which is a combustion marker) and to n-pentane (which is a solvent evaporation marker) show that in all cities except in Santiago and Prague, C alkylbenzenes are emitted only by automotive sources. In Santiago, extra C alkylbenzene emissions may come from solvent evaporation from factories or chemical feedstock, and in Prague, there is an unknown source containing ehtylbenzene, o-, m-, p-xylene, benzene, toluene, n-butane, i-butane, ethene, ethyne, propene but no n-pentane (nor i-pentane), but further studies need to be done in this city to identify this source. Comparing C alkylbenzenes to toluene and benzene indicate that combustion (and not fuel evaporation) is the major common source of these compounds in areas dominated by automotive emissions. The xylenes and ethylbenzene ratios remain persistent close to their urban sources no matter of the type of combustion, of the type of fuel used, etc. Moreover, the three xylenes ratios remain similar in very di!erent kinds of environments: urban and close to biomass combustion. This information should be very useful for ambient air measurement networks that routinely measure the sum of the three xylenes as well as for modeling studies. These persistent ratios should also be useful in analytical testing of gas chromatography or any other aromatic hydrocarbons' analytical device and for the use of previous data sets in which p-and m-xylenes were reported as one.
Finally, the constancy of the sources of ethylbenzene and m-xylene and the di!erence in atmospheric lifetimes with respect to OH removal could be a useful tool to estimate the amount of photochemical processing in an advecting urban plume.