Source Apportionment of Urban Particulate ... - ACS Publications

Apr 28, 2001 - (Ilias G. Kavouras) and phone: +30 81 393628; fax: +30 81 393678; e-mail: [email protected] (Euripides G. Stephanou). † Harv...
0 downloads 0 Views 83KB Size
Environ. Sci. Technol. 2001, 35, 2288-2294

Source Apportionment of Urban Particulate Aliphatic and Polynuclear Aromatic Hydrocarbons (PAHs) Using Multivariate Methods I L I A S G . K A V O U R A S , * ,† PETROS KOUTRAKIS,† MANOLIS TSAPAKIS,‡ EVAGGELIA LAGOUDAKI,‡ E U R I P I D E S G . S T E P H A N O U , * ,‡ DIETRICH VON BAER,§ AND PEDRO OYOLA# Environmental Science and Engineering Program, Harvard School of Public Health, 665 Huntington Avenue, Boston, Massachusetts 02115, Environmental Chemical Processes Laboratory, Department of Chemistry, University of Crete, 71409 Heraclion, Greece, Department of Instrumental Analysis, University of Concepcion, Casilla 160-C, Barrio University s/n, Concepcion, Chile, and Commission Nacional del Medio Ambiente (CONAMA), Valentin Letelier 13, Santiago, Chile

Samples of organic aerosol were collected in Santiago de Chile. An activated-charcoal diffusion denuder was used to strip out organic vapors prior to particle collection. Both polynuclear aromatic hydrocarbons (PAHs) and aliphatic hydrocarbons were determined using gas chromatography/mass spectrometry (GC/MS). Organic particle sources were resolved using both concentration diagnostic ratios and multivariate methods such as hierarchical cluster analysis (HCA) and factor analysis (FA). Four factors were identified based on the loadings of PAHs and n-alkanes and were attributed to the following sources: (1) high-temperature combustion of fuels; (2) fugitive emissions from oil residues; (3) biogenic sources; and (4) unburned fuels. Multilinear regression (MLR) analysis was used to determine emission profiles and contributions of the sources. The reconstructed concentrations of particle phase aliphatic and polynuclear aromatic hydrocarbons were in good agreement (R2 > 0.70) with those measured in Santiago de Chile.

Introduction Polynuclear aromatic hydrocarbons (PAHs) have been of scientific interest for several decades due to their carcinogenic and mutagenic properties. They are ubiquitous pollutants of the atmosphere and originate from diesel, vehicle exhausts, and other combustion sources (1-6). While the molecular markers approach has been widely applied to source reconciliation of PAHs in various environments (7-10), its use has been very limited due to the lack of suitable “source * Corresponding authors e-mail: [email protected] (Ilias G. Kavouras) and phone: +30 81 393628; fax: +30 81 393678; e-mail: [email protected] (Euripides G. Stephanou). † Harvard School of Public Health. ‡ University of Crete. § University of Conception. # Commission Nacional del Medio Ambiente (CONAMA). 2288

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 35, NO. 11, 2001

signatures” (11). Receptor modeling using both the chemical mass balance (CMB) model and factor analysis (FA) has also been used to identify the sources of organic compounds (1223). However, there are present certain limitations. CMB models require the input of source emission profiles to calculate source contributions. These models have been used in previous studies to develop to identify and quantify the sources of organic material in California (12-14). Factor analysis (FA) can provide information on source contributions based on the time-variation of the particle composition. Several models have been developed to apportion the sources of PM10 (particles with aerodynamic diameter less than 10 µm) and PM2.5 (particles with aerodynamic diameter less than 2.5 µm) in the United States (18-20). Factor analysis, in conjunction with multilinear regression, has been applied to determine the contribution of particulate PAH sources using ambient concentrations of PAHs, inorganic species, and metals (22, 23). A previous source apportionment study, which used particle elemental analysis, identified automobiles, buses, and trucks as the major sources of particulate material in Santiago de Chile (24). The contributions of oil combustion and sulfur-, arsenic-, and chloride-associated sources in the fine particle mass were also found to be important (24). The objective of this work is to identify and quantify the sources of particle-phase PAHs and n-alkanes in Santiago de Chile using a combination of statistical methods. In this current study, hierarchical cluster analysis (HCA) is used to investigate associations between the organic compounds. In addition, factor analysis (FA) is employed to identify the most important sources of organic aerosol. Subsequently, the contribution of the identified sources to PAH and n-alkane concentrations will be determined by regressing the factor scores on compound concentrations using multilinear regression (MLR) analysis.

Methodology Sampling and Analysis. Aerosol samples were collected during the winters of 1997 (from June 1997 to September 1997) and 1998 (from August 1998 to October 1998). Santiago de Chile is the major urban and industrial area of Chile with a population of approximately 5 000 000 inhabitants. It is located in a narrow valley at the foot of Andes Mountains, about 150 km distant from the Pacific Ocean. As a result, temperature inversion layers are frequently formed. Air pollution levels during the winter season are often high due to the presence of vehicles, especially buses, and industrial sources such as power plants and copper smelters. Forty samples were collected using a PM2.5 virtual impactor at a sampling flow of 93.5 L/min. The major flow passed through a parallel plate organic diffusion denuder (10, 25). After the denuder, the air stream passed through a transitional section to an adaptor plate with four sampling ports. A filter pack containing a 47 mm (pore size, 1 µm) Teflon filter (TFs) (Gelman Laboratories, Zefluor) was attached to each sampling port. Polyurethane foam (PUF) sampling tubes (Supelco) were placed downstream of the filter packs to trap volatilized SVOCs (10). It has been shown that, for the tested organic compounds (from naphthalene to pyrene), the denuder has high collection efficiency and adequate capacity at ambient (50-60%) and high (85-92%) relative humidity (25). A detailed description of the analytical procedure used for extraction, separation, and analysis of the main lipid fractions has been published elsewhere (26). All samples were analyzed using a Finnigan GCQ ion trap gas chromatogra10.1021/es001540z CCC: $20.00

 2001 American Chemical Society Published on Web 04/28/2001

TABLE 1. Concentration Diagnostic Ratios of PAHs and Aliphatic Hydrocarbons Measured in Santiago de Chile and of PAHs from Specific Source Emissionsd

a

diagnostic ratios

Santiago de Chile

diesel enginesa

gasoline enginesb

wood combustionc

CPI1 CPI2 CPI3 UCM/NA %WNA MPh/Ph Fl/(Fl+Py) BaA/(BaA+CT) B[e]P/B[e]P+B[a]]P) IP/(IP+BP)

0.79 ( 0.11 0.76 ( 0.07 1.33 ( 0.32 3.12 ( 1.50 19.90 ( 3.55 0.36 ( 0.11 0.43 ( 0.06 0.24 ( 0.07 0.65 ( 0.15 0.21 ( 0.12

0.60-0.70 0.38-0.64 0.29-0.40 0.35-0.70

0.40 0.43 0.60-0.80 0.18

0.74 0.56 0.48

Literature cited: (1, 8).

b

Literature cited: (1-5). c Literature cited: (4).

phy-mass spectrometer (at the University of Crete) in the electron and methane-chemical ionization mode, equipped with a HP-5MS capillary column (30 m × 0.25 mm i.d. × 0.25 µm film thickness). The aliphatic fraction of each sample was also analyzed using gas chromatography equipped with a flame ionization detector (FID) to quantify the mixture of unresolved hydrocarbons. The identification of n-alkanes (C15-C40) and PAHs was performed using reference standards (Dr. Ehrenstorfer GmbH, Germany). Data Analysis. Diagnostic ratios were used to identify the sources of PAHs and aliphatic hydrocarbons. In addition, a more comprehensive data analysis including hierarchical cluster and factor analysis was conducted. These methods made possible to (a) obtain more information about the structure of the data (organic compounds concentrations); (b) separate and identify the sources of organic compounds; and (c) quantify the source contributions to PAHs and n-alkanes concentrations. Also as a part of this analysis, the source emission profiles were estimated. This information is very valuable in our efforts to apportion sources of atmospheric organic aerosol. (I) Diagnostic Ratios. Because the distributions of homologues are strongly associated with formation mechanisms of carbonaceous aerosol, the following concentration diagnostic ratios (along with similar characteristics of organic species) can be used to qualitatively reconcile sources of organic species. (a) The carbon preference index (CPI) for n-alkanes (ratio odd-to-even) is calculated as follows (27, 28):

C13 + C15 + ....... + C31 + C33 + C35 C12 + C14 + ....... + C30 + C32 + C34

(1)

CPI2 )

C13 + C15 + ....... + C23 + C25 C12 + C14 + ....... + C22 + C24

(2)

CPI3 )

C25 + C27 + ....... + C33 + C35 C24 + C26 + ....... + C32 + C34

(3)

CPI1 )

n-Alkanes originate from epicuticular waxes of terrestrial plants and exhibit high values of CPI (CPI .1), whereas CPI values for vehicular emissions and other anthropogenic activities are close to unit (CPI ≈ 1). (b) The biogenic “wax” concentration of n-alkanes is calculated as follows (28):

%wax Cn )

(

)

Cn - 0.5‚[Cn-1 + Cn+1] ‚100 Cn

(4)

This equation calculates the “wax” concentration of each n-alkane and thus allows the estimation of biogenic sources.

d

For abbreviations see Methodology.

(c) The ratio of unresolved complex mixture (UCM) to the total n-alkanes (NA) concentration (UCM/NA): UCM is composed of cyclic, unsaturated, and branched aliphatic hydrocarbons, which is depicted as a broad envelope in chromatograms. Gough and Rowland (29) identified UCM in both petroleum and its refined products. Subsequently, UCM can be produced as a result of the incomplete combustion of oil, wood, and coal (1-8). Wood and coal combustion exhibit UCM/NA ratio values from 2.3 to 3.9, whereas vehicular emissions result values higher than 4.0 (1-8, 29). (d) PAH concentration diagnostic ratios that are characteristic of the anthropogenic emissions (see Table 1 (1-5)) are calculated as follows: (i) methylphenanthrene to phenanthrene, (MPh/Ph); (ii) fluoranthene to (fluoranthene+pyrene), [Fl/Fl+Py]; (iii) benzo[a]anthracene to (benzo[a]anthracene + chrysene/triphenylene), BaA/(BaA+CT); (iv) benzo[e]pyrene to (benzo[e]pyrene+benzo[a]pyrene), [BeP/BeP+BaP]; and (v) indeno[1,2,3-cd]pyrene to (indeno[1,2,3-cd]pyrene + benzo[ghi]perylene), [IP/IP+BgP]. (II) Statistical Methods. Using the results of the chemical analysis, the concentration matrix (X(nxm)) with n-rows (the number of analyzed species) and m-columns (the number of samples analyzed) was constructed. For both FA and HCA, the concentration matrix is standardized using the Z-score as follows

Zij )

hi Xij - X σi

(5)

where Xij is the concentration of the i-element on the j-day, and Xi and σi are the mean and standard deviation of the i-element, respectively. (a) Hierarchical Cluster Analysis. Cluster analysis is an exploratory multivariate method that can be used to describe the relationships among variables (30). Classification of variables into groups using cluster analysis does not require a priori information on the number and the properties of the groups. Several mathematical criteria can be used to examine the similarity (or difference) between variables and cases. The Euclidean distance defined as the length of the straight line between two points, the square of the Euclidean distance, and the Pearson correlation coefficient (R2) can be used in the classification of data. The Euclidean distance is used to explore similarities between cases, whereas Pearson correlation coefficient measures the similarities between variables. The initial outcome of hierarchical cluster analysis is a number of clusters that is equal to the number of variables. Afterward, a new cluster is formed based on the similarities between variables. This process is repeated as many times as required to form a single cluster. The similarities among VOL. 35, NO. 11, 2001 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

2289

equal to 90°, has been applied to overcome this problem (21). Oblique rotation was also used to split up correlated factors because the right angle between the factors can be varied and therefore they can take up any position (20, 21). Because the initial conclusions from factor loadings are not easily interpretable, these factors are orthogonaly transformed using the VARIMAX method. This transformation (rotation) results in high loadings of certain variables on a given factor and relatively low loadings of the others. Therefore, this rotation produces a better correlation of factors with the measured species, allowing them to be attributed to specific sources of organic aerosol. The factor scores (FS) are the product of the component score coefficient matrix (W(pxn)) and the Z-score matrix (Z(nxm)) (Figure 1). Since, the Z-score is calculated using the mean concentrations, these initially calculated FS values do not correspond to the “real” factor scores. To determine the “real” factor scores (source contributions), the “absolute zero factor score” must be calculated. This was achieved by separately scoring an extra day where ambient concentrations of elements are set at zero (31). The difference between the FS and the “absolute zero” FS is the actual factor score (AFS). Finally, by regressing daily concentrations of organic compounds on these AFS values, the source contributions are estimated as follows p

FIGURE 1. Flow diagram of the factor analysis method.

M j ) ao +

the clusters decrease as clusters are merged into a single group. The average method is used to link clusters (30). Using different methods to measure similarities and link clusters and/or adding small perturbations can examine the sensitivity and accuracy of cluster analysis. For this application, the Pearson correlation coefficient (R2) was used as a measure of the similarities between the variables, initially, and then between the clusters. The number and the structure of clusters formed using these methods are not affected with either the addition of small numbers of “bad” points or the use of a limited number of measurements. (b) Factor Analysis. The basic assumption for receptor models is that the concentration of a pollutant at the receptor (Xi) for a given sample is the linear sum of the products of the emission profile (ai) and contribution (Si) of n-sources, see eq 6. n

Xi )

∑a ‚S i

i

(6)

i)1

Note that ai remains constant and Si varies with time. Using the Z-score matrix described above, the correlation matrix C(nxn) is calculated. Factor analysis proceeds through the determination of the eigenvector (factor) matrix which is a matrix of weights each applicable to the variables, B(nxn), and the corresponding transpose, B-1(nxn), matrix, which are used to calculate a diagonal table Λ(nxn) using the following equation (Figure 1):

B(nxm)‚C(nxn)‚B-1 (nxn) ) Λ(nxn)

(7)

The matrix Λ(nxn) contains the eigenvalues of the corresponding eigenvectors, which are equal to the sum of the squares of the corresponding factor loadings. The number of extracted factors is equal to the number of variables. However, typically a smaller number p, of eigenvectors (factors) can explain a large fraction of the total variance. Therefore, the dimensionality of the system can be reduced from n to p. The virtually infinite number of mathematically equivalent solutions introduces an important limitation to the use of factor analysis. The orthogonal rotation of eigenvectors, where the right angle between the factors is 2290

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 35, NO. 11, 2001

∑ a (AFS) k

kj

(8)

k)1

where Mj is the concentration of particle mass or element (in ng/m3) for the j-sample; AFSkj* is the rotated absolute factor score for k-source on sample j; ak is the regression coefficient of the AFS to organic compounds concentrations; and ao is the intercept. The contribution of each factor-source to the total concentration of PAHs and aliphatic hydrocarbons can be determined by multiplying the regression coefficients by the corresponding absolute factor score (Figure 1). A similar methodology was previously applied to apportion the sources of particulate material in Watertown, MA (31). The SPSS 8.0 (SPSS, Chicago, IL) was used to perform both the cluster and factor analysis. An EXCEL-based algorithm used to apportion sources of PM10 and PM2.5 in five Chilean cities (32) was applied to calculate factor scores and estimate source contributions on organic compounds daily concentrations.

Results and Discussion Diagnostic Ratios. The aliphatic fraction was composed of n-alkanes (from C14 to C36), isoprenoid hydrocarbons (pristane and phytane), and an unresolved complex mixture (UCM), consisting of cyclic, unsaturated, and branched alkanes. The calculated CPI1 values (0.79 ( 0.11) (Table 1) were comparable to those observed for other urban areas (10, 33, 34). In addition, the CPI2 (for C14-C25) were equal to 0.76 ( 0.07 (Table 1), which indicates the importance of anthropogenic activities. This is further supported by the values of the UCM/ NA ratio (3.12 ( 1.50; Table 1), which are within the range measured for unburned fossil fuels emitted by vehicles (18). These observations suggested the importance of the petroleum and diesel residues and gasoline emissions contributions. However, the CPI3 values >1 (1.33 ( 0.32; Table 1) indicated also a certain contribution of other nonanthropogenic sources for this range of n-alkanes. These n-alkanes are mainly emitted from leaf epicuticular waxes (35). The relative contribution of biogenic wax n-alkanes to the total n-alkanes concentration (%WNA) (28) in Santiago de Chile was only 19.90 ( 3.55%. This percentage is comparable or lower than that measured to other urban areas (7), probably due to the lower emissions from both deciduous and coniferous trees in the winter.

TABLE 2. Eigenvalues, Variance Explained, and Loadings of Factors factor

1

2

3

4

eigenvalue explained variance phenanthrene methylphenanthrene chrysene/triphenylene benzo[e]pyrene perylene indeno[1,2,3-cd]pyrene coronene C21 C27 C29 C31 pristane phytane UCM

4.83 34.52 0.18 -0.26 0.83 0.90 0.50 0.77 0.86 -0.33 0.09 0.11 0.27 -0.17 -0.13 0.50

3.73 26.65 -0.30 0.17 -0.07 0.01 0.35 -0.45 -0.19 0.79 0.13 0.06 -0.20 0.96 0.93 0.72

2.01 14.35 -0.07 -0.16 0.33 0.37 0.55 -0.16 0.15 0.12 0.97 0.97 0.82 -0.09 0.19 -0.20

1.57 11.23 0.89 0.89 -0.09 0.02 0.11 -0.14 0.04 -0.12 -0.10 -0.10 -0.12 0.10 -0.10 -0.06

In addition to aliphatic hydrocarbons, polynuclear aromatic hydrocarbons were also identified in ambient samples. These samples contained mostly semivolatile nonsubstituted 4- and 5-ring compounds. High molecular weight PAHs, including chrysene, benzofluoranthenes, and benzopyrenes, were abundant. The mean methylphenanthrene-to-phenanthrene ratio (MPh/Ph) during the sampling period was 0.36 ( 0.11 (Table 2). This value is within the typical range for combustion-derived products (7). The mean [Fl/Fl+Py] ratio was 0.43 ( 0.06 (Table 1). This ratio is similar to that reported for catalytic and noncatalytic automobiles emissions (2, 3) and has also been observed in other urban areas (7). The ratio BaA/(BaA+CT) (0.24 ( 0.07; Table 1) indicates, as a source, oil combustion (1-5, 36). The mean [BeP/BeP+BaP] ratio measured was 0.65 ( 0.15 (Table 1) similar to that reported for gasoline combustion emissions (1, 3-5, 35). The [IP/IP+BgP] ratios obtained in our study were 0.21 ( 0.12 (Table 1) which is similar to that for diesel emissions (1, 8). Although, photochemical reactions of PAHs can introduce a significant error to source reconciliation studies, the values of these observed ratios in this study were similar to those obtained from other investigations (2,3, 7, 8, 35). This suggests that the use of molecular diagnostic ratios for PAHs was appropriate for the identification of their sources. Overall, our results suggest that the most important sources of organic aerosol in Santiago during the study period were fossil fuel combustion from automobiles and unburned petroleum residues. Because diagnostic ratios provide only qualitative information about the contribution of the sources of organic species, hierarchical cluster analysis (HCA) and factor analysis (FA) were used to classify the organic compounds into groups and quantify their sources. Hierarchical Cluster Analysis (HCA). Cluster analysis was applied to the entire standardized concentration matrix (Z(nxm)) to explore the structure of the concentration data. The correlation coefficient (R2) was used to initially measure the similarities between variables and clusters. The average method was applied to link the clusters. The results of HCA did not change significantly when different linkage methods were used. Figure 2 depicts the HCA results presented in the form of a dendrogram. Four distinguished clusters were classified with correlation coefficients higher than 0.70. The first group was composed of n-alkanes from C23 to C35. These aliphatic hydrocarbons have a dual origin (both anthropogenic and biogenic) (2, 3, 7, 8, 34, 35). However, the absence of PAHs with pyrolytic origin in this group suggests that the impact of biogenic sources to n-alkanes from C25 to C35 concentrations might be important. Nonsubstituted PAHs such as benzofluoranthenes and benzopyrenes were included in the second cluster. These PAHs are produced from the

FIGURE 2. Agglomerative hierarchical dendrogram of polynuclear aromatic hydrocarbons (PAHs) and aliphatic hydrocarbons in Santiago de Chile. high-temperature combustion of gasoline, diesel wood, and other fuels (1-8). In addition, perylene was correlated with this cluster (R2 ) 0.48). Semivolatile PAHs from fluorene to pyrene constituted the third cluster, which was attributed to unburned fossil fuels. The fourth group included low molecular weight n-alkanes, pristane, phytane, and UCM. The presence of isoprenoid hydrocarbons in this group suggested that oil residues were an important source of aliphatic and aromatic hydrocarbons (37). Factor Analysis (FA). Because of the limited number of samples, only a subset of 14 variables was selected for the factor analysis. These variables (seven PAHs, four n-alkanes, pristane, phytane, and UCM) were selected based on the results of diagnostic ratios and HCA presented above. Using this subset of variables, the factor loadings and scores were determined. The rest of the variables (not included in the factor analysis) were introduced to the analysis subsequently. More precisely, these variables were regressed against the absolute factor scores to determine the respective source emission profiles and contributions. The majority of the variance (86.7%) of this subset of data was explained by four eigenvectors-factors (Table 2). Factor 1 was mostly associated with high molecular weight PAHs (MW > 202) and the UCM and accounted for 34.5% of the total variance (Table 2). The loadings of benzo[a]anthracene and chrysene, usually emitted through both diesel and natural gas combustion (5, 38), were also high in this factor. Benzo[a]pyrene and coronene also had high loadings in this factor (Table 2). These compounds are emitted from catalyst and noncatalyst automobiles (2, 3). In addition, indeno[1,2,3cd]pyrene, which is present in petroleum and it is a tracer of diesel combustion (39), was strongly correlated with this factor. Phenanthrane and its methylated derivatives had very low loadings, suggesting that factor 1 was not associated with unburned and residual fossil fuels (38). Consequently, this factor was attributed to the emissions from pyrolysis and combustion of fuels. As expected, this factor was the major source of heavier PAHs. The entire concentration levels of chrysene/triphenylene, benzofluoranthenes, benzo[a]pyrene, benzo[e]pyrene, anthrathrene, indeno[1,2,3-cd]pyrene, and coronene were explained by this source (Table VOL. 35, NO. 11, 2001 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

2291

TABLE 3. Sources Contribution on PAH Concentrationa mean ( S.E. combustion fugitive natural unburned emissions (µg/m3) emissions (µg/m3) sources (µg/m3) fossil fuels (µg/m3)

compound fluorene phenanthrene anthracene methylphenanthrenes dimethylphenanthrenes fluoranthene pyrene benzo[a]antrhacene chrysene/triphenylene benzo[bjk]fluoranthene benzo[e]pyrene benzo[a]pyrene perylene anthranthrene indeno[1,2,3-cd]pyrene benzo[ghi]perylene coronene MPh/Ph Fl/(Fl+Py) BaA/(BaA+CT) BeP/(BeP+BaP) IP/(IP+BP) a

5.0 ( 1.4 6.0 ( 1.3 0.6 ( 0.1 2.6 ( 0.8 2.6 ( 0.7 3.1 ( 0.9 4.1 ( 1.3 16.2 ( 5.1 23.6 ( 7.1 13.1 ( 3.8 5.9 ( 1.8 2.2 ( 0.5 2.2 ( 0.6 4.3 ( 1.4 24.4 ( 6.7 4.1 ( 1.1 0.46 0.20 0.69 0.15

0.7 ( 0.2 3.1 ( 0.9

0.8 ( 0.2 0.4 ( 0.1

6.7 ( 1.9 31.0 ( 9.0 4.7 (1.3 8.4 ( 2.3 0.7 ( 0.2 2.8 ( 0.9 5.2 ( 1.6

0.4 ( 0.1 2.3 (1.0 0.1 ( 0.1

7.5 ( 1.0

0.2 ( 0.1

1.2 ( 0.2 0.2 ( 0.1 0.27 0.35

calculated (µg/m3)

measured (µg/m3)

11.6 ( 2.4 37.1 ( 9.0 5.9 ( 1.3 11.6 ( 2.2 3.3 ( 0.8 5.4 ( 1.1 8.3 ( 1.9 4.9 ( 0.7 16.2 ( 5.1 23.6 ( 7.1 13.8 ( 3.8 6.3 ( 1.9 4.5 ( 1.1 2.3 ( 0.7 4.3 ( 1.4 33.0 ( 6.8 4.3 ( 1.1 0.31 0.39 0.23 0.69 0.12

14.0 ( 3.3 31.0 ( 9.8 4.0 ( 1.5 11.3 ( 2.7 5.9 ( 1.1 5.8 ( 1.2 7.7 ( 2.0 5.6 ( 1.4 18.1 ( 5.8 23.4 ( 7.7 13.5 ( 4.3 7.2 ( 1.9 4.4 ( 2.8 3.0 ( 0.7 6.7 ( 1.7 25.0 ( 8.5 3.8 ( 1.4 0.36 ( 0.11 0.43 ( 0.06 0.24 ( 0.07 0.65 ( 0.15 0.21 ( 0.12

R2 0.87 0.97 0.90 0.96 0.78 0.96 0.93 0.93 0.89 0.97 0.97 0.96 0.83 0.96 0.92 0.95 0.90

For abbreviations see Methodology.

TABLE 4. Sources Contribution on Aliphatic Hydrocarbons Concentration mean ( S.E. combustion emissions (µg/m3) C14 C15 C16 C17 C18 C19 C20 C21 C22 C23 C24 C25 C26 C27 C28 C29 C30 C31 C32 C33 C34 C35 Pristane Phytane UCM CPI1 CPI2 CPI3 UCM/NA

1.2 ( 0.2 0.6 ( 0.1

0.3 ( 0.1 0.1 ( 0.1 0.1 ( 0.1 1.0 ( 0.2 0.9 ( 0.2 1.1 ( 0.2 1.3 ( 0.3 1.0 ( 0.3 1.0 ( 0.2 0.7 ( 0.1 0.5 ( 0.1

306.0 ( 5.5 0.63 0.05 1.00 46.31

fugitive emissions (µg/m3)

biogenic (µg/m3)

10.2 ( 1.5 9.0 ( 2.4 44.3 ( 6.6 32.7 ( 5.1 22.8 ( 4.9 11.5 ( 1.6 5.0 ( 1.3 5.1 ( 0.4 7.2 ( 1.1 6.3 ( 0.6 5.4 ( 1.0 3.3 ( 0.9 3.0 ( 1.2 2.7 ( 1.0 0.1 ( 0.1 0.1 ( 0.1

0.7 ( 0.1

27.4 ( 2.8 22.0 ( 2.4 858.9 ( 13.3 0.72 0.72 0.73 4.32

5.4 ( 1.0 6.5 ( 1.6 0.3 ( 0.1 3.2 ( 1.0 4.0 ( 1.4 5.1 ( 1.5 7.7 ( 3.2 6.4 ( 2.3 8.5 ( 2.9 6.9 ( 2.4 7.4 ( 3.0 3.8 ( 1.4 4.7 ( 2.5 2.4 ( 0.9 1.8 ( 0.7 1.2 ( 0.4 0.5 ( 0.2

0.85 0.58 1.49 -

3). Perylene was also associated with oil residues, since it is a product of the diagenetic processes of degradation of organic material to produce petroleum (40). Pyrene, fluoranthene, and phenathrene are also components of fossil fuels and a portion of them was associated with their combustion. The potential contribution of other sources such as biomass burning and waste incinerators cannot be resolved because pyrene, fluoranthene, and phenathrene were not identified 2292

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 35, NO. 11, 2001

unburned fossil fuels (µg/m3)

0.4 ( 0.1 2.5 ( 0.8 0.3 ( 0.1

-

calculated mass (µg/m3)

measured mass (µg/m3)

10.9 ( 1.6 9.0 ( 2.4 51.4 ( 6.2 35.3 ( 5.1 29.8 ( 5.1 11.9 ( 1.6 5.0 ( 1.3 5.4 ( 0.4 10.8 ( 1.5 10.4 ( 1.5 10.5 ( 1.8 11.1 ( 3.3 10.4 ( 2.5 12.2 ( 3.0 8.0 ( 2.3 8.8 ( 2.9 4.8 ( 1.4 5.7 ( 2.4 3.0 (0.9 2.3 ( 0.7 1.2 ( 0.2 0.5 ( 0.2 27.4 ( 2.6 22.0 ( 2.6 1165.0 ( 260.9 0.77 0.70 1.48 10.86

8.6 ( 2.9 13.0 ( 4.3 41.3 ( 4.8 32.1 ( 5.7 31.2 ( 6.1 13.2 ( 2.1 14.5 ( 1.8 10.8 ( 1.2 14.4 ( 1.9 12.7 ( 1.8 14.6 ( 2.7 12.3 ( 2.9 10.6 ( 3.1 10.2 ( 3.0 9.6 ( 2.8 8.5 ( 2.0 5.6 ( 1.5 5.9 ( 1.4 3.1 ( 1.0 2.4 ( 0.9 1.0 ( 0.5 0.5 ( 0.3 23.5 ( 4.0 20.9 ( 3.9 1002.0 ( 192.5 0.79 ( 0.11 0.76 ( 0.07 1.33 ( 0.32 3.12 ( 1.50

R2 0.56 0.37 0.94 0.98 0.79 0.97 0.67 0.87 0.94 0.89 0.70 0.96 0.97 0.99 0.88 0.98 0.92 0.90 0.84 0.82 0.82 0.73 0.99 0.97 0.90

in the same cluster or correlated with the first factor. The diagnostic concentration ratios, calculated using the estimated concentration of PAHs for each source, indicated that combustion of gasoline, diesel, and other fuels were the major source of these PAHs. Indeed, the mean ratios of Fl/(Fl+Py) (0.46), BaA/(BaA+CT) (0.20), BeP/(BeP+BaP) (0.69), and IP/ (IP+BP) (0.15) ratios (Table 3) were comparable to those measured for emissions from automobiles, trucks (diesel

combustion), and natural gas combustion (1-6, 38). Only a small fraction of the UCM and C23, C24, and C25 were associated with the first factor. Thus, the combustion of fossil fuels at high temperatures favored the formation of aromatic than aliphatic hydrocarbons. The emissions from mobile combustion sources exhibit high values of UCM/NA ratios (46.31; Table 4), which is in agreement with previous studies. This suggested that PAHs are more favorable products of the hightemperature combustion of fossil fuels, which is in agreement with previous studies (2-4). Factor 2 accounted for 26.6% of the total variance (Table 2). This factor was dominated by C16, C18, pristane, phytane, and perylene and to a lesser extent by C14, C15, C17, C19, and C20 (Table 2). The presence of isoprenoid hydrocarbons, pristane, and phytane is characteristic of petroleum and oil residues (1, 4, 5) (Table 2). The moderate loadings of benzo[ghi]perylene and UCM suggested the impact of oil residues (Table 2). As a result, this factor was attributed to oil residues. The majority of low molecular weight n-alkanes from C14 to C18 and isoprenoid hydrocarbons were associated with this factor (Table 4). In addition, minor quantities of n-alkanes up to C29, were coming from oil residues (Table 4). Finally, oil contamination was responsible for the majority of UCM (858.97 ( 133.32 ng/m3 (Table 2). The CPI1 (0.72) and CPI2 (0.72) ratios of n-alkanes originated from oil residues were low (Table 4). Oil residues exhibit relatively high values of the UCM/NA (4.32) (Table 4). This was in agreement with previous studies, which suggested that UCM is a component of petroleum and its refined products (29); subsequently it can be emitted during the incomplete combustion of fossil fuels. Only anthracene, methylphenanthrenes, perylene, and benzo[ghi]perylene were associated with oil residues. Although, heavier PAHs are also indicators of oil residues, they were not identified here. Factor 3 explained 14.3% of the total variance (Table 2). The loadings of n-alkanes from C27 to C31 in this factor were high (Table 2). In addition, the loadings of pyrolytic PAHs produced from the combustion of fossil fuels were low. The isoprenoid hydrocarbons and UCM, tracers of oil residues, had also relatively low loadings (Table 2). This suggested that indeed biogenic sources such as the mechanical abrasion of terrestrial plants leaf waxes, vegetation debris, and soil erosion are responsible for the presence of nonvolatile n-alkanes (>C27) (35). Thus, this factor was attributed to biogenic sources. The calculated CPI(C25-C35) (1.49; Table 4) ratios were in agreement with the characteristics of biogenic sources. Perylene was the only PAH which was associated with this source. Factor 4 accounted for 11.2% of the total variance (Table 2). This factor was characterized by high loadings of phenanthrene and methylphenanthrene. These compounds may originate from both coal combustion and unburned fossil fuels. However, unburned fossil fuels contribute to elevated levels of concentration of alkylated phenanthrenes and pyrenes (6, 7, 9). Therefore, this factor was attributed to unburned fossil fuels. The larger fractions of phenanthrene, methylphenanthrenes, fluoranthene, and pyrene were explained by this factor. Trace concentrations of coronene, a tracer of petroleum, were detected in this factor. The emission profile of this source yielded low values of MPh/Ph (0.27; Table 3) and Fl/(Fl+Py) (0.35; Table 3) ratios, which are in agreement with those observed for unburned petroleum and oil (9) in other urban areas (1-8, 10, 32). Overall, the results from both exploratory methods (HCA and FA) were in good agreement and suggested the classification of the compounds into four clusters (and factors). The four identified factors were attributed to specific sources of PAHs and aliphatic hydrocarbons; namely, combustion processes, fugitive emissions from oil, biogenic emissions, and unburned fossil fuels. The results from this study indicate

that combustion of fossil fuels was the major source for high molecular weight nonsubstituted PAHs, while emissions from unburned fossil fuels were responsible for the presence of the more volatile PAHs. In addition, n-alkanes and isoprenoid hydrocarbons were originating from both biogenic emissions, fugitive emissions from oil residues, and unburned fossil fuels.

Acknowledgments The Commission Nacional del Medio Ambiente (CONAMA) of Chile supported this study. The development and laboratory performance evaluation of the sampler was supported by U.S. EPA STAR grant No. R825 270-01-0. The Research Account of the University of Crete (EPEAEK program) is also acknowledged for financial support.

Literature Cited (1) Grimmer, G.; Jacob, J.; Naujack, K. W. Fresenius Z. Anal. Chem. 1983, 316, 29-36. (2) Rogge, W. F.; Hildemann, L.; Mazurek, M. A.; Cass, G. R.; Simoneit, B. R. T. Environ. Sci. Technol. 1993, 27, 636-651. (3) Rogge, W. F.; Hildemann, L.; Mazurek, M. A.; Cass, G. R.; Simoneit, B. R. T. Environ. Sci. Technol. 1993, 27, 1892-1904. (4) Pyyssalo, H.; Tuominen, J.; Wickstrom, K.; Skytta, E.; Tikkanen, L.; Salomaa, S.; Sorsa, M.; Pohjola, V. Atmos. Environ. 1987, 21, 1167-1180. (5) Khalili, N. R.; Scheff, P. A.; Holsen, T. M. Atmos. Environ. 1995, 29, 533-542. (6) Masclet, P.; Bresson, M. A.; Mouvier, G. Fuel 1987, 66, 556-562. (7) Gogou, A. I.; Stratigakis, N.; Kanakidou, M.; Stephanou, E. G. Org. Geochem. 1996, 25, 79-96. (8) Sicre, M. A.; Marty, J. C.; Saliot, A.; Aparicio, X.; Grimalt, J.; Albaiges, J. Atmos. Environ. 1987, 21, 2247-2259. (9) Simoneit, B. R. T. Sci. Tot. Environ. 1984, 36, 61-72. (10) Kavouras, I. G.; Lawrence, J.; Koutrakis, P.; Stephanou, E. G.; Oyola, P. Atmos. Environ. 1999, 33, 4977-4986. (11) Sporstol, S.; Gjos, N.; Licternhaler, R. G.; Gustavsen, K. O.; Urdal, K.; Orels, F., Skel, J. Environ. Sci. Technol. 1983, 17, 282-286. (12) Schauer, J. J.; Cass, G. R. Environ. Sci. Technol. 2000, in press. (13) Schauer, J. J.; Rogge, W. F.; Hildemann, L. M.; Mazurek, M.; Simoneit, B. R. T. Atmos. Environ. 1996, 20, 3837-3855. (14) Fraser, M. P.; Kleeman, M. J.; Schauer, J. J.; Cass, G. R. Environ. Sci. Technol. 2000, 34 1302-1312. (15) Baldasano, J. M.; Delgado, R.; Calbo, J. Environ. Sci. Technol. 1998, 32, 405-412. (16) Wadden, R. A.; Scheff, P. A.; Uno, I. Atmos. Environ. 1994, 28, 2507-2521. (17) Pirrone, N.; Keeler, G. J.; Holsen, T. M. Environ. Sci. Technol. 1995, 29, 2123-2132. (18) Dzubay, T. G.; Stevens, R. K.; Gordon, G. E.; Olmez, I.; Sheffieldm, A. E.; Courtney, W. J. A. Environ. Sci. Technol. 1988, 22, 46-52. (19) Pratsinis, S. E.; Zeldin, M. D.; Ellis, E. C. Environ. Sci. Technol. 1988, 22, 212-216. (20) Koutrakis, P.; Spengler, J. D Atmos. Environ. 1985, 21, 15111517. (21) Hopke, P. K. Receptor Modeling in Environmental Chemistry; Wiley: New York, 1985. (22) Harrison, R. M.; Smith, D. J. T.; Luhana, L. Environ. Sci. Technol. 1996, 30, 825-832. (23) Simcik, M. F.; Eisenreich, S. J.; Lioy, P. J. Atmos. Environ. 1999, 33, 5071-5079. (24) Artaxo, P.; Oyola, P.; Martinez, R. Nucl. Instrum. Methods Phys. Res. B 1999, 150, 409-416. (25) Koutrakis, P.; Sioutas, C.; Lawrence, J. Design and evaluation of a novel SVOC sampler: 1. Gas-phase diffusion denuder to trap organics; EPA Progress Report; Research Triangle Park, NC, 1998. (26) Gogou, A. I.; Apostolaki, M.; Stephanou, E. G. J. Chromatogr. 1998, 799, 215-231. (27) Bray, E. E.; Evans, E. D. Geo. Cosmoch. Acta. 1961, 22, 2-15. (28) Simoneit, B. R. T.; Cardoso, J. N.; Robinson, N. Chemosphere 1990, 21, 1285-1301. (29) Johnson, R. A.; Wichern, D. W. Applied Multivariate Statistical Analysis; Prentice Hall; 1998. (30) Gough and Rowland, Nature. (31) Thurston, G. D.; Spengler, J. D. Atmos. Environ. 1985, 22, 9-25. (32) Kavouras, I. G.; Koutrakis, P.; Cereceda-Balic, F.; Oyola, P. J. Air Wastes Manag. Assoc. 2000, in press. VOL. 35, NO. 11, 2001 / ENVIRONMENTAL SCIENCE & TECHNOLOGY

9

2293

(33) Kavouras, I. G.; Stratigakis, N.; Stephanou, E. G. Environ. Sci. Technol. 1998, 32, 1369-1377. (34) Nielsen, T. Atmos. Environ. 1996, 30, 3481-3490. (35) Rogge, W. F.; Hildemann, L.; Mazurek, M. A.; Cass, G. R.; Simoneit, B. R. T. Environ. Sci. Technol. 1993, 27, 2700-2711. (36) Smith, D. J. T.; Harrison, R. M. Atmos. Environ. 1996, 30, 25132525. (37) Volkman, J. K.; Maxwell, J. R. Biological markers in the Sendimentary Record. In Methods, Geochemistry Geophysics; Johns, R. B., Ed.; Elsevier: 1987.

2294

9

ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 35, NO. 11, 2001

(38) Rogge, W. F.; Hildemann, L.; Mazurek, M. A.; Cass, G. R.; Simoneit, B. R. T. Environ. Sci. Technol. 1993, 27, 2700-2711. (39) Li, C. K.; Kamens, R. M. Atmos. Environ. 1993, 27A, 523-532. (40) Aizenshtat, Z. Geochim. Cosmochim. Acta 1973, 37, 559-567.

Received for review July 31, 2000. Revised manuscript received March 7, 2001. Accepted March 9, 2001. ES001540Z