Environ. Sci. Technol. 2005, 39, 4172-4179
Spatial Variability of Fine Particle Mass, Components, and Source Contributions during the Regional Air Pollution Study in St. Louis E U G E N E K I M , † P H I L I P K . H O P K E , * ,† JOSEPH P. PINTO,‡ AND WILLIAM E. WILSON‡ Department of Chemical Engineering, Clarkson University, Potsdam, New York 13699-5705, and U.S. Environmental Protection Agency, Research Triangle Park, North Carolina 27711
Community time-series epidemiology typically uses either 24-hour integrated particulate matter (PM) concentrations averaged across several monitors in a city or data obtained at a central monitoring site to relate PM concentrations to human health effects. If the day-to-day variations in 24-hour integrated concentrations differ substantially across an urban area (i.e., daily measurements at monitors at different locations are not highly correlated), then there is a significant potential for exposure misclassification in community time-series epidemiology. If the annual average concentration differs across an urban area, then there is a potential for exposure misclassification in epidemiologic studies that use annual averages (or multiyear averages) as an index of exposure across different cities. The spatial variability in PM2.5 (particulate matter e 2.5 µm in aerodynamic diameter), its elemental components, and the contributions from each source category at 10 monitoring sites in St. Louis, Missouri were characterized using the ambient PM2.5 compositional data set of the Regional Air Pollution Study (RAPS) based on the Regional Air Monitoring System (RAMS) conducted between 1975 and 1977. Positive matrix factorization (PMF) was applied to each ambient PM2.5 compositional data set to estimate the contributions from the source categories. The spatial distributions of components and source contributions to PM2.5 at the 10 sites were characterized using Pearson correlation coefficients and coefficients of divergence. Sulfur and PM2.5 are highly correlated elements between all of the site pairs Although the secondary sulfate is the most highly correlated and shows the smallest spatial variability, there is a factor of 1.7 difference in secondary sulfate contributions between the highest and lowest site on average. Motor vehicles represent the next most highly correlated source component. However, there is a factor of 3.6 difference in motor vehicle contributions between the highest and lowest sites. The contributions from point source categories are much more variable. For example, the contributions from incinerators show a difference of a factor of 12.5 between the sites with the lowest and highest contributions. This study demonstrates that the spatial distributions of elemental * Corresponding author phone: (315)268-3861; fax: (315) 2686654; e-mail:
[email protected]. † Clarkson University. ‡ U.S. Environmental Protection Agency. 4172
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 11, 2005
components of PM2.5 and contributions from source categories can be highly heterogeneous within a given airshed and thus, there is the potential for exposure misclassification when a limited number of ambient PM monitors are used to represent population-average ambient exposures.
Introduction The statistical association between particulate matter (PM) and adverse health effects has been shown in many studies (1-3). Since the U.S. Environmental Protection Agency promulgated new national ambient air quality standards for airborne PM (4), many air quality and epidemiology studies have been undertaken (e.g., refs 5-9). Community timeseries epidemiology typically uses either daily 24-hour integrated PM concentrations averaged across several monitors in a city or data obtained at a central monitoring site to relate the daily PM concentrations to daily human health effects. In addition to examining the association of adverse health effects with PM concentrations, such investigations may also relate health outcomes to PM components and source contributions (6, 10). If the daily time series of concentrations at different monitoring sites across an urban area are not highly correlated, then there will be a potential for exposure misclassification. The reduction in the risk estimated by epidemiologic analyses, due to a less than perfect correlation between the true community average and an experimental measure at one site (or an average of measurements at several sites) will be related to the R2 (coefficient of determination) between the two time series. Thus, differences in the correlation between the time series of concentrations of PM mass, components, or source contributions at various monitoring sites in an urban area give some indication of the reduction in estimated risk due to use of a less-than-perfect surrogate for the communityaverage concentrations. Epidemiologic studies of the effects of chronic exposure to pollutants use annual averages (or multi-year averages) as an index of exposure across different cities. If the annual average concentrations differ across an urban area, then there is a potential for exposure misclassification in this type of epidemiologic study. Thus, if site-to-site correlations are low or if annual average concentrations differ substantially across an urban area, then exposure misclassification could be an important consideration when a limited number of ambient PM monitors are used to represent population-average ambient concentrations. Little quantitative information exists on the effect of exposure misclassification on estimated risks. However, one study found considerable variation in the level and significance of the calculated health risk per unit PM when using each of several individual sites as well as various averages of multiple sites (11). A number of studies have examined the spatial variability of PM mass measured in urban areas (12-15). However, there have been very few data sets collected in urban areas from which one could determine the spatial variability of both chemical components and source contributions to PM. The Regional Air Pollution Study/Regional Air Monitoring System (RAPS/RAMS) (16-18) conducted in St. Louis, Missouri between 1975 and 1977 is one such study in which enough data are available to evaluate the spatial heterogeneity of PM2.5 (particulate matter e2.5 µm in aerodynamic diameter), its components, and its source contributions across an urban area. 10.1021/es049824x CCC: $30.25
2005 American Chemical Society Published on Web 04/27/2005
TABLE 1. Summary of PM2.5 Mass Concentrations for 10 Monitoring Sites concentration (µg/m3)
FIGURE 1. Location of the monitoring sites and major point sources during RAPS/RAMS. The objective of this study is to characterize the spatial uniformity in PM2.5 species and source contributions so that the potential for exposure characterization error in health outcome studies can be better understood. In the present study, positive matrix factorization (PMF) was applied to ambient PM2.5 compositional data from ten RAPS/RAMS sites in St. Louis to identify PM2.5 sources and estimate their contributions to PM2.5 mass concentrations in the St. Louis metropolitan area. The spatial variability in PM2.5, its components, and its contributions from source categories are discussed. PMF nominally determines source categories rather than specific sources, even though there is only one source in a given category in the community. In this paper, the word source is used with the understanding that it refers to either a source category or a specific source as appropriate.
Sample Collection and Data The ambient PM2.5 compositional data set used in this study was collected in St. Louis, Missouri between May 1975 and April 1977 as part of the RAPS/RAMS program. In the RAPS/ RAMS program, automated dichotomous samplers (19) were operated over a two year period at 10 monitoring sites across the St. Louis metropolitan area. The sampling network was designed to sample ambient air in different parts of the greater St. Louis urban area to which the general public would have access, not to sample emissions from specific sources. The 10 monitoring sites were located along the direction of the prevailing winds from southwest to northeast. Figure 1 shows the location of the 10 monitoring sites in the RAPS/RAMS program where PM composition measurements were made and the locations of major point sources in the study area (20, 21). A two-stage vertical impactor in the automated dichotomous samplers separated particles into PM2.5 and PMcoarse (particles between 2.5 and 15 µm in aerodynamic diameter). The particles were then collected onto 37-mm-diameter cellulose ester membrane filters. Ambient PM samples were analyzed for total mass concentration by β-gauge measurements and for 27 elements by energy-dispersive X-ray fluorescence (Lawrence Berkeley Laboratory). Integrated 12hour PM2.5 samples were taken at 8 sites and 6-hour samples were collected at the others. Six-hour and 2-hour period
Site no.
no. of samples
arithmetic mean
minimum
maximum
103 105 106 108 112 115 118 120 122 124
582 593 422 489 566 534 513 509 546 364
27.71 25.61 24.71 24.61 22.91 21.10 18.43 21.97 18.73 16.44
4.50 5.25 4.60 4.20 4.00 4.65 2.90 3.90 2.35 3.25
121.51 88.10 87.50 80.50 74.80 77.40 79.50 79.10 61.20 58.60
samples were obtained during an intensive study in the summer of 1975. One-hour averages of wind speed and direction, carbon monoxide (CO), nitrogen oxides (NOx), and gaseous sulfur were measured automatically at each site. A majority of the elemental measurements were beneath their detection limits and were therefore excluded from the analysis. Finally, PM2.5 mass concentration, 13 species, and 3 gaseous species were included in this study: S, Si, Al, Ti, Br, Cu, Cr, Ca, Fe, Zn, Pb, Mn, K, CO, NOx, and SO2. In this study, 24-hour averaged PM2.5 mass and species concentrations were estimated from 1-, 2-, 6-, or 12-hour integrated data: measurements which had an error flag were considered as missing data. Days on which samplers operated for less than 50% of the sampling time were excluded. Daily averaged PM2.5 mass and species concentrations were then estimated by averaging the valid values. There were several values that were either much lower or much higher compared to other measurements. As a data screening method for the 24-hour averaged data, values that were lesser or greater than three times the standard deviation away from the geometric mean values were considered to be unrealistic or data input mistakes and therefore excluded for each elemental species. A summary of the PM2.5 mass concentrations used in this study is shown in Table 1. Detailed summaries of the species concentrations are shown in the Supporting Information.
Data Analyses and Results Source Apportionment. Positive matrix factorization (PMF) (23) was applied to the 10 ambient PM2.5 compositional data sets of the 24-hour averaged samples as well as CO, NOx, and SO2 in this study. PMF has been shown to be a powerful alternative to traditional receptor modeling of airborne PM (23, 24). PMF has been used successfully to assess particle source contributions in the Arctic (25), Phoenix (26), Vermont (27), three northeastern U.S. cities (28), a northwestern U.S. city (29), and Atlanta (30). The general receptor modeling problem can be stated in terms of the contribution from p sources to all of the chemical species in a given sample as follows (31) p
xij )
∑g
ik fkj
+ eij
(1)
k)1
where xij is the jth species concentration measured in the ith sample, gik is the particulate mass concentration from the kth source contributing to the ith sample, fkj is the jth species mass fraction from the kth source, eij is the residual associated with the jth species concentration measured in the ith sample, and p is the total number of sources. PMF provides a solution that minimizes an object function, Q(E), based upon the uncertainties of each observation (23, 32). This function is VOL. 39, NO. 11, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
4173
TABLE 2. Comparison of Average Source Contribution (µg/m3) to PM2.5 Mass Concentrations Measured at 10 Monitoring Sites average source contribution (standard error) Site no.
sec. sulfate
motor vehicle
metal processing
cement kiln
incinerator
airborne soil
pigment/metal processing
103 105 106 108 112 115 118 120 122 124
13.9 (0.5) 12.2 (0.5) 11.7 (0.5) 13.9 (0.5) 12.0 (0.4) 14.0 (0.5) 14.9 (0.4) 8.8 (0.3) 10.3 (0.4) 10.5 (0.4)
3.6 (0.2) 2.4 (0.1) 3.1 (0.1) 2.3 (0.1) 2.4 (0.1) 2.0 (0.1) 1.3 (0.04) 2.4 (0.1) 3.3 (0.1) 1.0 (0.04)
4.2 (0.1) 6.8 (0.1) 0.8 (0.08) 3.6 (0.1) 0.3 (0.02) 1.7 (0.1) 0.2 (0.02) 9.1 (0.1) 1.7 (0.1) 0.2 (0.01)
0.7 (0.03) 1.4 (0.04)
0.7 (0.03)
1.7 (0.1) 2.0 (0.1) 3.6 (0.2)
2.3 (0.1) 0.8 (0.1) 2.7 (0.2)
1.5 (0.1)
0.4 (0.03)
defined as
[ ] p
n
Q(E) )
m
∑∑ i)1 j)1
xij -
∑
2.9 (0.1) 2.7 (0.1) 2.6 (0.1) 1.1 (0.1) 1.0 (0.1) 0.9 (0.04) 2.6 (0.1)
2
gik fkj
k)1
uij
(2)
where uij is an uncertainty estimate in the jth constituent measured in the ith sample. The application of PMF depends on the estimated uncertainties of each of the data values. The uncertainty estimation provides a useful tool to decrease the weight of missing and below-detection-limit data in the solution. The procedure of Polissar et al. (32) was used to assign the measured data and the associated uncertainties as the input data to the PMF. The concentration values were used for the measured data, and the sum of the analytical uncertainty and 1/3 of the detection limit value were used as the overall uncertainty assigned to each measured value. Values below the detection limit were replaced by half of the detection limit values, and their overall uncertainties were set at 5/6 of the detection limit values. Missing values were replaced by the geometric mean of the measured values, and their accompanying uncertainties were set at four times this geometric mean value. Because the atmospheric process of gaseous species between the source and the monitoring site is different from those of the PM, the gaseous species should not influence the model fit. The estimated uncertainties of the gaseous species were set at four times their measured concentrations so that the large uncertainties decreased their weight in the model fit. Therefore, gaseous species are only served to determine the resolved sources. The results of PMF were then normalized by a scaling constant, sk, so that the quantitative source contributions as well as profiles for each source were obtained. In the same way, the standard deviation associated with the source contributions and profiles were normalized. Specifically
()
p
xij )
∑(s
k gik)
k)1
fkj sk
+ eij
(3)
where sk is determined by regressing the total PM2.5 mass concentrations in the ith sample, mi, against the estimated source contribution values. p
mi )
∑
sk gik
(4)
k)1
It is necessary to test different numbers of sources and find the optimal fit with the most physically reasonable results in order to determine the number of sources. The robust mode was used to reduce the influence of extreme values on the PMF solution. A measured data point was classified as 4174
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 11, 2005
1.6 (0.1) 1.6 (0.1) 2.5 (0.1) 0.7 (0.04) 0.3 (0.02) 0.2 (0.01) 1.8 (0.1) 1.9 (0.1)
0.8 (0.03)
an extreme value if the residual exceeded four times the error estimate. The estimated uncertainties of those extreme values were then increased so that the weight of the extreme values in the solution was decreased. As pointed out by Henry (33), there are an infinite number of possible solutions to the factor analysis problem due to the free rotation of matrices. PMF uses non-negativity constraints on the factors to decrease the rotational freedom. The FPEAK parameter and a FKEY matrix are used to control the rotations (34, 35). The routine is forced to add one gik vector to another and subtract the corresponding fkj factors from each other by setting FPEAK equal to a nonzero value. PMF was run with different FPEAK values to determine the range within which the objective function Q(E) value in eq 2 remains relatively constant (35). The final PMF solution should lie in this FPEAK range. In this way, subjective bias was avoided to some extent. The external information can be imposed on the solution to control the rotations. If specific species in the source profiles are known to be zero, then it is possible to pull down those values toward lower concentration through appropriate settings of FKEY resulting in the most interpretable source profiles. Each element of the FKEY matrix controls the pulling-down of the corresponding element in the fkj matrix by setting a nonzero integer value in the FKEY matrix (34). The final PMF solutions were determined by experiments with a different number of sources, different FPEAK values, and different FKEY matrices, with the final choice based on the evaluation of the resulting source profiles as well as the quality of the species fits. PMF identified between 5 and 10 sources of PM2.5 at the 10 monitoring sites, as shown in detail in the Supporting Information. Estimates of the source contributions obtained at each site for the seven major source types are shown in Table 2. The results of the analyses agreed well with the existing information about the location and nature of local point sources (Figure 1). The influence of a number of specific point sources can be observed with varying strength across the sampling sites. The comparisons of the daily reconstructed PM2.5 mass concentrations (sum of the contributions from PMF resolved sources) with measured PM2.5 mass concentrations, the identified source profiles (value ( standard deviation resolved by PMF), and time series plots of estimated daily contributions to PM2.5 mass concentrations from each source are presented in the Supporting Information. The high concentrations of S are attributed to secondary sulfate aerosol. This source was also identified in the previous RAPS/RAMS studies (18, 20, 21, 36, 37). Secondary sulfate aerosol was the major contributor to PM2.5 (40% - 80%) in St. Louis. In the previous RAPS/RAMS studies, secondary sulfate aerosol accounted for 61% of the average PM2.5 mass concentrations at Site 108 (18). In the target-transformation factor analysis of 55 samples collected at Sites 105 and 106, Chang et al. (21) apportioned 67% of the average PM2.5 mass
concentrations to the secondary sulfate. Alpert and Hopke (36) reported a 65% contribution of secondary sulfate to samples collected during July and August 1976 at Site 112. Secondary sulfate shows strong seasonal variation with higher concentrations in the summer when photochemical activity is highest. Motor-vehicle-related PM are characterized by high concentrations of Pb and Br. During the RAPS, lead concentrations in motor vehicle emissions were variable and depended on the ratio of leaded gasoline to unleaded gasoline and the proportion of diesels in the vehicle fleet. Chang et al. (37) reported a value of 11.5% Pb in motor vehicle emissions and a ratio of Br/Pb of 0.24 in the previous RAPS/ RAMS study for five monitoring sites (Sites 105, 106, 112, 118, and 120). A value of 10.4 ( 1.3% Pb and a Br/Pb ratio of 0.38 ( 0.04 are obtained in this study for the 10 monitoring sites. Motor-vehicle-related PM accounts for between 6 and 18% of the PM2.5 mass concentrations. This result is consistent with the results from previous RAPS/RAMS studies in which motor vehicle emissions accounted for 15% of the PM2.5 mass concentrations at Site 112 (36), 17% at Site 108 (18), and 11% at Sites 105, 106, 112, 118, and 120 (37). The motor-vehiclerelated PM source profiles also had high loadings of CO and NOx at the 10 monitoring sites. Motor-vehicle-related PM contributed more to the PM2.5 mass in the winter. The observed seasonal variations are probably due mainly to reduced mixing and dilution in the planetary boundary layer during the winter and possibly also to increases in emissions. The winter maxima in contributions from this source are consistent with the results from source apportionment studies of other cities (28-30). Metal processing, including steel, copper, lead, zinc, and aluminum smelters/plants, are indicated by source profiles with high loadings of Fe, Cu, Pb, Zn, and Al, respectively. The contributions of this source to PM2.5 mass vary between 1 and 41%. Liu et al. (18), Chang et al. (37), and Gatz (38) reported several metal-processing sources in St. Louis, as shown in Figure 1. A cement kiln was identified by high loadings of Ca and Si (39). This source contributed between 3 and 16% of the PM2.5 mass concentration. Incinerators are identified by high loadings of Zn, Pb, and K (28) and contributed between 1 and 11% of the PM2.5 mass concentrations. A previous analysis of Sites 105 and 106 (21) identified an incinerator source that had high concentrations of Pb, Cl, K, and Zn in its emissions. In this study, PMF did not extract an incinerator source at Site 105. The incinerator profile in Site 105 is thought to be split among the metal processing profiles of other sources such as the zinc and lead smelter. SO2 was inconsistently high in incinerator, cement kiln, and metal processing profiles. Airborne soil, represented by Si, Al, Fe, Ca, and K (40, 41), contributes from 6 to 14% of the PM2.5 mass concentration. For Sites 106 and 112, PMF did not clearly separate the soil contributions from the steel processing sources located northwest of the sites from other airborne soil contributions. Therefore, airborne soil contributions to Sites 106 and 112, shown in Table 2, include contributions from steel processing. The previous Washington, DC (28) and Atlanta aerosol studies (30) showed a 3% contribution of airborne soil to the PM2.5 mass concentration. Crystal particles could be contributed by unpaved roads, parking areas, construction sites, and wind-blown soil dust. Fly ash originating from the combustion of coal could also contribute to this source. Airborne soil was identified by previous RAPS/RAMS studies contributing 11 (36), 9 (18), 7 (37), and 3% (21) to the PM2.5 mass concentration. The airborne soil shows seasonal variation with higher concentrations in the dry summer season.
A paint pigment plant is represented by high loadings of Ti (21). At several monitoring sites, the pigment plant contributions were mixed with those from metal-processing emissions. They were not separated because they were located in the same direction relative to the monitoring sites or their daily integrated contribution patterns were similar. Pigment/metal processing contributed between 2 and 11% of the PM2.5 mass concentration. The paint pigment source was identified in a previous study as contributing 1.5% to the PM2.5 mass concentration at Site 112 (36). Analysis of Spatial Variability. The spatial characteristics of concentrations of PM2.5, its components, and its source contributions were determined using several metrics including the following: geometric means (µg), standard deviations (σg) (42), Pearson correlation coefficients (r), and coefficients of divergence (COD) (13). The geometric standard deviations and COD are defined as
(x
∑(ln x
σgf ) exp
CODfn )
)
n
if
- ln µgf)2
i)1
n-1
x ( ) 1
n
xif - xih
∑x n i)1
if
(5)
2
(6)
+ xih
where xif is the ith concentration measured at the f th site. f and h represent two monitoring sites, and n is the number of observations. The geometric standard deviation is the slope of the log-probability plot showing the frequency distribution of the data over time for a log-normal distribution of concentrations. If the measured values at a monitoring site are similar during the study period, then σg approaches zero. The Pearson correlation coefficient provides information on how well the concentrations at different monitoring sites go up and down together. The COD is a coefficient used extensively in numerical analyses of ecological data to determine the resemblance either between objects under study or the variables describing them (43). The COD provides information on the degree of uniformity between monitoring sites. For the spatial distribution, the COD approaches zero if the measured values at two monitoring sites are similar. In contrast, if the measured values are very different, then the COD approaches unity. As shown in Table 3, the highest values for all of the chemical species, except for sulfur, are found at sites located close to sources. This result demonstrates the importance of discrete point sources for determining the spatial distribution of species other than S. Fe and Mn show spatial characteristics of geometric mean concentrations that are similar to each other, indicating that there could be a source that releases both species. A similar situation exists with regard to Br and Pb. The arithmetic average of geometric standard deviations of PM2.5 mass concentration of Si, S, K, Ca, Cr, Fe, Br, and Pb is 2.06 ( 0.21, which shows reasonable agreement with 1.85 ( 0.14 from a previous study for California aerosols (42). The geometric standard deviations of Ti, Mn, Cu, and Zn are higher than the others, ensuring the presence of multiple sources that give rise to the observed broad or multimodal distributions over time. The arithmetic average of the Pearson correlation coefficients and COD between all of the site pairs for PM2.5 and trace elements are shown in Figure 2a and b. Although there were sources of primary S in operation during the RAPS/ RAMS program period, S is the most highly correlated element (mean r ) 0.92, mean COD ) 0.17) and has the lowest spatial variation among all of the sites, as can be seen from Figure 2a,b, reflecting in large part its secondary nature. PM2.5 VOL. 39, NO. 11, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
4175
TABLE 3. Comparison of Geometric Mean Concentrations (ng/m3) and Associated Geometric Standard Deviations of PM2.5 and Its Elemental Components geometric mean concentration (geometric standard deviation) species
103
105
106
108
112
115
118
120
122
124
PM2.5
23494 (1.8) 2330 (2.0) 222 (2.2) 144 (1.8) 18 (2.5) 96 (2.3) 24 (3.7) 7.1 (1.8) 121 (2.3) 166 (2.0) 84 (2.4) 441 (2.0) 13 (2.1) 159 (1.7)
21930 (1.8) 2252 (2.0) 216 (1.9) 152 (1.6) 24 (3.0) 121 (2.0) 19 (4.7) 7.6 (1.9) 117 (2.0) 193 (2.1) 82 (2.7) 555 (1.8) 19 (2.6) 152 (1.7)
21642 (1.7) 2458 (2.0) 275 (2.1) 120 (1.7) 34 (4.7) 152 (2.0) 14 (3.6) 7.1 (2.3) 102 (2.1) 207 (2.2) 69 (2.5) 647 (1.7) 22 (3.7) 145 (1.8)
21267 (1.8) 2389 (2.0) 193 (2.2) 104 (2.0) 13 (2.9) 100 (2.3) 15 (3.8) 5.5 (2.2) 111 (2.5) 163 (2.6) 77 (3.2) 451 (2.2) 15 (2.8) 162 (2.2)
19653 (1.8) 2139 (2.1) 200 (2.1) 122 (1.9) 19 (4.1) 169 (1.9) 8.4 (3.6) 4.4 (2.4) 82 (2.3) 123 (2.2) 47 (2.6) 650 (1.9) 12 (3.2) 121 (1.9)
18424 (1.7) 2322 (1.9) 165 (2.0) 76 (2.0) 9.4 (2.6) 52 (2.0) 12 (4.5) 3.9 (2.1) 64 (2.5) 95 (1.8) 67 (2.7) 296 (1.9) 6.5 (2.3) 102 (1.8)
15609 (1.8) 1967 (2.0) 152 (2.1) 67 (2.0) 15 (4.0) 36 (2.0) 4.8 (4.0) 3.0 (2.3) 99 (2.6) 89 (2.0) 34 (2.7) 230 (1.9) 5.4 (2.7) 103 (1.7)
18859 (1.8) 1915 (2.1) 147 (2.0) 89 (2.0) 11 (3.3) 133 (2.0) 5.0 (3.4) 3.8 (2.6) 64 (2.7) 86 (2.2) 32 (2.5) 504 (1.8) 6.2 (2.9) 99 (1.8)
15375 (1.9) 1961 (2.2) 133 (2.3) 56 (2.1) 6.8 (3.0) 21 (2.2) 6.9 (3.6) 2.5 (2.6) 60 (2.9) 74 (2.3) 35 (3.3) 136 (2.3) 5.2 (2.7) 85 (2.0)
14195 (1.7) 1942 (2.0) 126 (2.2) 51 (1.9) 7.5 (2.6) 21 (2.0) 2.9 (3.1) 3.8 (2.5) 74 (2.5) 74 (2.2) 22 (2.3) 136 (1.9) 4.4 (2.9) 80 (1.7)
S Si Al Ti Br Cu Cr Ca Fe Zn Pb Mn K
FIGURE 2. (a) Pearson correlation coefficients and (b) coefficients of divergence for PM2.5 and trace elements obtained at the 10 monitoring sites. concentrations are the next most highly correlated element between all of the site pairs (mean r ) 0.79, mean COD ) 4176
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 11, 2005
FIGURE 3. (a) Pearson correlation coefficients and (b) coefficients of divergence for source contributions to PM2.5 obtained at the 10 monitoring sites. 0.22). However, there are large differences in the behavior of component species (mean r e 0.59, mean COD g 0.30).
FIGURE 5. Ratio of calculated source contributions at different sites to those at Site 105.
FIGURE 4. Spatial distributions of (a) secondary sulfate, (b) motor vehicle, and (c) incinerator contributions (µg/m3) derived for the 10 monitoring sites.
The arithmetic average of the Pearson correlation coefficients and the COD for major source categories deduced by PMF are shown in Figure 3a,b. The spatial characteristics of metal processing and airborne soil could not be analyzed because they could not be clearly separated from each other. It can be seen that secondary sulfate, produced mainly by coal-fired power plants, is the most highly correlated source component and shows the smallest spatial variability. Motor vehicles represent the next most highly correlated source component, and finally, contributions from point sources are the least well correlated among the source types. The spatial distributions for the average contributions of PMF-deduced secondary sulfate, motor vehicle, and incinerator are shown in Figure 4a-c, respectively. Although the sulfate sources are the most highly correlated, it can be seen from Figure 4a that there is still a factor of 1.7 between the highest and lowest sites (difference: 6.1 µg/m3). The contributions from motor vehicles are more variable, as shown in Figure 4b, and there is a factor of 3.6 between the sites with the lowest and the highest contributions (difference: 2.6 µg/m3). As shown in Figure 4c, the contributions from the incinerator are much more variable, showing a difference of a factor of 12.5 between the sites with the lowest and highest contributions (difference: 2.3 µg/m3). One of the RAPS/RAMS monitoring sites, Site 105, was located close to the St. Louis site used in the Harvard sixcities study (3). Compositional data from this site were later used by Laden et al. (7) to relate human mortality to source contributions. They used lead as the tracer for motor vehicle exhaust in the specific rotation factor analysis and reported statistically significant association between motor vehicle sources and daily mortality in St. Louis. However, in this study, the lead contribution from a lead smelter (Figure S2b in the Supporting Information), which was not identified in the study of Laden et al. (7), was 54% higher than that from the motor vehicle source at Site 105. This indicates that there were multiple sources of lead and, therefore, the use of lead as a tracer for the motor vehicle source could cause problems in their specific rotation factor analysis. Similar problems in Laden et al.’s source apportionments were indicated by Grahame and Hidy (44), who reported that sulfur was also contributed from local residual oil combustion in Boston, MA, and therefore the coal combustion source was overestimated. Figure 5 shows the ratios of estimated source contributions in this study at different sites to those at Site 105. The source contributions to Site 105 can differ by over a factor of 3 from those at other sites. Because these results were derived from a 10-site network, they also imply that the distributions of VOL. 39, NO. 11, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
4177
many trace elements and source contributions derived mainly from point sources can be highly heterogeneous within a given air shed. This study demonstrated that the spatial distributions of PM2.5 elemental components and source contributions can be highly heterogeneous within a given air shed. The results suggest that there can be significant potential for exposure misclassification in time-series epidemiologic studies when regressing health outcomes against source contributions estimated at a monitoring site if there is heterogeneity in the distribution of local sources. Exposure misclassification errors resulting from the neglect of spatial variability may contribute to uncertainty in the relative risk estimates resulting from epidemiologic investigations. In future air quality and epidemiology studies, it is important that the spatial variation in ambient PM2.5 species and source contributions within a study area be taken into consideration in order to reduce their source of exposure missclassification.
Acknowledgments This paper has been reviewed in accordance with the U.S. Environmental Protection Agency’s (EPA) peer and administrative review policies and approved for publication. This study was supported by the U.S. EPA under contract ICR268-NTEX. Although the research described in this article has been funded by the U.S. EPA, the views expressed herein are solely those of the authors and do not represent the official policies or positions of the U.S. EPA.
Supporting Information Available Additional figures and tables. This material is available free of charge via the Internet at http://pubs.acs.org.
Literature Cited (1) Dockery, D. W.; Pope, C. A.; Xu, X. P.; Spengler, J. D.; Ware, J. H.; Fay, M. E.; Ferris, B. G. An association between air-pollution and mortality in 6 United-States cities. N. Engl. J. Med. 1993, 329, 1753-1759. (2) Pope, C. A.; Thun, M. J.; Namboodiri, M. M.; Dockery, D. W.; Evans, J. S.; Speizer, F. E.; Heath, C. W. Particulate air pollution as a predictor of mortality in a prospective study of U.S. adults. Am. J. Respir. Crit. Care Med. 1995, 151, 669-674. (3) Schwartz, J.; Dockery, D. W.; Neas, L. M. Is daily mortality associated specifically with fine particles? J. Air Waste Manage. Assoc. 1996, 46, 927-939. (4) U.S. EPA. National Ambient Air Quality Standards for Particulate Matter. Code of Federal Regulations, Title 40, Part 50, 1997; Fed. Regist. 1997, 62. (5) Clarke, R. W.; Coull, B. A.; Reinisch, U.; Catalano, P. J.; Killingsworth, C. R.; Koutrakis, P.; Kavouras, I.; Lawrence, J.; Lovett, E. G.; Wolfson, J. M.; Verrier, R. L.; Godleski, J. J. Inhaled concentrated ambient particles are associated with hematologic and bronchoalveolar lavage changes in canines. Environ. Health Perspect. 2000, 108, 1179-1187. (6) Mar, T. F.; Norris, G. A.; Koenig, J. Q.; Larson, T. V. Associations between air pollution and mortality in Phoenix. Environ. Health Perspect. 2000, 108, 347-353. (7) Sarnat, J. A.; Koutrakis, P.; Suh, H. Assessing the relationship between personal particulate and gaseous exposures of senior citizens living in Baltimore, MD. J. Air Waste Manage. Assoc. 2000, 50, 1184-1198. (8) Schwartz, J.; Laden, F. The concentration-response relationship between PM2.5 and daily deaths. Environ. Health Perspect. 2002, 110, 1025-1029. (9) Dominici, F.; Zanoetti, A.; Zeger, S. L.; Schwartz, J.; Samet, J. M. Hierarchical bivariate time series models: a combined analysis of the effects of particulate matter on morbidity and mortality. Biostatistics 2004, 5, 341-360. (10) Laden, F.; Neas, L. M.; Dockery, D. W.; Schwartz, J. Association of fine particulate matter from different sources with daily mortality in six US cities. Environ. Health Perspect. 2000, 108, 941-947. (11) Ito, K.; Kinney, P. L.; Thurston, G. D. Variations in PM10 concentrations within two metropolitan areas and their implications for health effects analyses. Inhalation Toxicol. 1995, 7, 735-748. 4178
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 39, NO. 11, 2005
(12) Wilson, W. E.; Shu, H. H. Fine particles and coarse particles: concentration relationships relevant to epidemiologic studies. J. Air Waste Manage. Assoc. 1997, 47, 1238-1249. (13) Wongphatarakul, V.; Friedlander, S. K.; Pinto, J. P. A comparative study of PM2.5 ambient aerosol chemical databases. Environ. Sci. Technol. 1998, 32, 3926-3934. (14) Leaderer, B. P.; Naeher, L.; Jankun, T.; Balenger, K.; Holford, T. R.; Toth, C.; Sullivan, J.; Wolfson, J. M.; Koutrakis, P. Indoor, outdoor, and regional summer and winter concentrations of PM10, PM2.5, SO42-, H+, NH4+, NO3-, NH3, and nitrous acid in homes with and without kerosene space heaters. Environ. Health Perspect. 1999, 107, 223-231. (15) Kinney, P. L.; Aggarwal, M.; Northridge, M. E.; Janssen, N. A. H.; Shepard, P. Airborne concentrations of PM2.5 and diesel exhaust particles on Harlem sidewalks: a community-based pilot study. Environ. Health Perspect. 2000, 108, 213-218. (16) Pilotte, J. O.; Winchester, J. W.; Nelson, J. W. Components of lead in the atmosphere of St. Louis, Missouri. J. Appl. Meteorol. 1978, 17, 627-635. (17) U.S. EPA. Aerosol Analysis for the Regional Air Pollution Study; EPA 600/4-81-006; Research Triangle Park, NC, 1981. (18) Liu, C. K.; Roscoe, B. A.; Severin, K. G.; Hopke, P. K. The application of factor analysis to source apportionment of aerosol mass. Am. Ind. Hyg. Assoc. J. 1982, 43, 314-318. (19) U.S. EPA. Regional Air Pollution Study: Dichotomous Aerosol Sampling System; EPA 297310; Research Triangle Park, NC, 1979. (20) Dzubay, T. G. Chemical element balance method applied to dichotomous sampler data. Ann. N.Y. Acad. Sci. 1980, 338, 126144. (21) Chang, S. N.; Hopke, P. K.; Gordon, G. E.; Rheingrover, S. W. Target-transformation factor analysis of airborne particle samples selected by wind-trajectory analysis. Aerosol Sci. Technol. 1988, 8, 63-80. (22) Paatero, P. Least-squares formulation of robust nonnegative factor analysis. Chemom. Intell. Lab. Syst. 1997, 37, 23-35. (23) Huang, S.; Rahn, K. A.; Arimoto, R. Testing and optimization two factor-analysis techniques on aerosol at Narragansett, Rhode Island. Atmos. Environ. 1999, 33, 2169-2185. (24) Willis, R. D. Workshop on UNMIX and PMF as Applied to PM2.5; EPA 600-A-00-048; U.S. EPA: Research Triangle Park, NC, 2000. (25) Xie, Y. L.; Hopke, P. K.; Paatero, P.; Barrie, L. A.; Li, S. M. Identification of source nature and seasonal variations of Arctic aerosol by positive matrix factorization. J. Atmos. Sci. 1999, 56, 249-260. (26) Ramadan, Z.; Song, X. H.; Hopke, P. K. Identification of sources of Phoenix aerosol by positive matrix factorization. J. Air Waste Manage. Assoc. 2000, 50, 1308-1320. (27) Polissar, A. V.; Hopke, P. K.; Poirot, R. L. Atmospheric aerosol over Vermont: Chemical composition and sources. Environ. Sci. Technol. 2001, 35, 4604-4621. (28) Song, X. H.; Polissar, A. V.; Hopke, P. K. Source of fine particle composition in the northeastern US. Atmos. Environ. 2001, 35, 5277-5286. (29) Kim, E.; Larson, T. V.; Hopke, P. K.; Slaughter, C.; Sheppard, L. E.; Claiborn, C. Source identification of PM2.5 in an arid northwest U.S. city by positive matrix factorization. Atmos. Res. 2003a, 66, 291-305. (30) Kim, E.; Hopke, P. K.; Edgerton, E. S. Source identification of Atlanta aerosol by positive matrix factorization. J. Air Waste Manage. Assoc. 2003b, 53, 731-739. (31) Miller, M. S.; Friedlander, S. K.; Hidy, G. M. A chemical element balance for the Pasadena aerosol. J. Colloid Interface Sci. 1972, 39, 165-176. (32) Polissar, A. V.; Hopke, P. K.; Paatero, P.; Malm, W. C.; Sisler, J. F. Atmospheric aerosol over Alaska 1. Elemental composition and sources. J. Geophys. Res. 1998, 103, 19045-19057. (33) Henry, R. C. Current factor analysis models are ill-posed. Atmos. Environ. 1987, 21, 1815-1820. (34) Paatero, P. User’s Guide for Positive Matrix Factorization Programs PMF2 and PMF3, Part 1: tutorial; 2000. (35) Paatero, P.; Hopke, P. K.; Song, X. H.; Ramadan, Z. Understanding and controlling ratations in factor analytic models. Chemom. Intell. Lab. Syst. 2002, 60, 253-264. (36) Alpert, D. J.; Hopke, P. K. A determination of the sources of airborne particles collected during the regional air pollution study. Atmos. Environ. 1981, 15, 675-687. (37) Chang, S. N.; Hopke, P. K.; Rheingrover, S. W.; Gordon, G. E. Target-transformation factor analysis of wind-trajectory selected samples; The 75th Annual Meeting of the Air Pollution Control Association, New Orleans, LA, 1982. (38) Gatz, D. F. Identification of aerosol sources in the St. Louis area using factor analysis. J. Appl. Meteorol. 1978, 17, 600-608.
(39) U.S. EPA. US EPA SPECIATE, version 3.2; U.S. EPA: Research Triangle Park, NC, 2002. (40) Watson, J. G.; Chow, J. C. Source characterization of major emission sources in the Imperial and Mexicali Valleys along the US/Mexico border. Sci. Total Environ. 2001a, 276, 33-47. (41) Watson, J. G.; Chow, J. C.; Houck, J. E. PM2.5 chemical source profiles for vehicle exhaust, vegetative burning, geological material, and coal burning in northwestern Colorado during 1995. Chemosphere 2001b, 43, 1141-1151. (42) Kao, A. S.; Friedlander, S. K. Frequency distributions of PM10 chemical components and their sources. Environ. Sci. Technol. 1995, 29, 19-28.
(43) Legendre, P.; Legendre, L. Numerical Ecology, 2nd English ed., Developments in Environmental Modeling 20; Elsevier: Amsterdam, 853, 1998. (44) Grahame, T.; Hidy, G. Using factor analysis to attribute health impacts to particulate pollution sources. Inhalation Toxicol. 2004, 16, 143-152.
Received for review February 2, 2004. Revised manuscript received February 26, 2005. Accepted March 25, 2005. ES049824X
VOL. 39, NO. 11, 2005 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
4179
