Atmospheric aerosols over two sites in a Southeastern region of Texas.

Chiou, Paul ; Tang, Wei ; Lin, Che-Jen 等

INTRODUCTION

It is well known that high concentrations of particulate matter (PM) are a main source of air pollution, and the air pollution Iis an important issue in the United States (Vedal, 1997; Rudell et al., 1999). Particles in the air can originate from a variety of natural or anthropogenic sources. It has been well documented that high PM concentrations can lead to serious health effects such as morbidity and mortality (Dockery et al., 1993; Schwartz et al., 1993; Lipfert and Wyzga, 1995), particularly, the PM that is in the fine mode with aerodynamic diameter less than 2.5 [micro]m (Gilliland et al., 2001; Peters et al., 2001; Pope et al., 2002).

Houston, TX (29[degrees]46'N, 95[degrees]23'W) located on the bank of the Buffalo Bayou, 80 km northwest of the Gulf of Mexico, is the largest city in the state of Texas and the fourth-largest in the United States. The energy industry in Houston is recognized worldwide, particularly, for oil and biomedical research. Aeronautics and the ship channel are also large parts of its economic base. The area is the world's leading centre for building oilfield equipment. Much of Houston's success as a petrochemical complex is due to its busy man-made ship channel, the Port of Houston. The port ranks first in the United States in international commerce, and is the 10thlargest port in the world. The high oil and gasoline prices are generally seen as beneficial to the economy of Houston; however, the oil industries are a major cause of the city's air pollution. Beaumont, TX (30[degrees]O5'N, 94[degrees]06'W) located on the west bank of the Neches River, 130 km east of Houston and 45 km north of the Gulf of Mexico, is a medium size urban area in Southeast Texas. With two smaller neighbouring cities, Port Arthur and Orange, it constitutes the so-called Golden Triangle in Texas, a major industrial area on the Texas Gulf Coast. Shipbuilding, livestock raising, and rice farming spread in the surrounding area. Several major chemical, petrochemical, and paper plants, refineries, rice mills, and waste management sites are located in the area of Golden Triangle. The Bayland Park monitoring site in Houston and Orange monitoring site in Golden Triangle operated by US EPA and maintained by TCEQ (Texas Commission on Environmental Quality) are both located in the southeastern Texas (Figure 1).

In an effort to better characterize the ambient air quality in Southeastern Texas, it is important to identify the possible sources of P[M.sub.2.5] in the region. To understand the source/receptor relationship, multivariate receptor models have been applied to the observed speciated PM over the years. The multivariate approach is based on the fundamental principle that mass conservation can be assumed, and a mass balance analysis can be used to identify and apportion sources of airborne particulate matter in the atmosphere (Hopke, 1985, 1991). Under this principle, the time dependence of a chemical species at the same receptor site will remain the same for species from the same source. Concentrations of chemical species are measured in a large number of samples gathered at a single receptor (monitoring) site over the time. Species of similar variability are grouped in a minimum of factors that explain the variability of the data. It is assumed that each factor is associated with a source or source type. Among the multivariate receptor models, positive matrix factorization (PMF) is a relatively new technique developed by Paatero (1997) and Paatero and Tapper (1993, 1994). It has been successfully applied to several source attribution studies (e.g. Juntto and Paatero,1994; Anttila et al., 1995; Polissar et al., 1996, 1999, 1998, 2001; Lee et al., 1999; Paterson et al., 1999; Xie et al., 1999; Chueinta et al., 2000; Kim et al., 2003, 2004).

The objectives of this study are to (1) identify the sources of particulate pollutants at the two sites, (2) estimate the source contributions as well as source composition of each possible source (e.g. Ramadan et al., 2000, 2003; Song et al., 2001; Zheng et al., 2002; Hansen et al., 2003; Liu et al., 2005; Chiou et al., in press), and (3) investigate the regional-local source contrast using estimated source contributions of each common factor for the two sites. Such comparison between the two sites in this southeastern region of Texas has not been reported in earlier literature.

SAMPLING AND CHEMICAL COMPOSITION MEASUREMENTS

The P[M.sub.2.5] composition sample data analyzed in this study was downloaded from the EPA website at http://www.epa. gov/ttn/airs/airsags/, and processed to conform to the PMF data format. The original 24-h integrated samples were collected at the Bayland Park monitoring site (29[degrees]41'45" N, 95[degrees]29'57" W) and Orange site (30[degrees]11'39" N, 93[degrees]52'01" W) using a fine particle sequential sampler (Rupprecht/Patashnick Model 2025). The Bayland Park, 3 km east of Highway 59 and 4 km west of I610, is located 15 km southwest of Houston downtown, and the Orange site, 6 km north of I-10, is located 30 km northeast of Beaumont.

Integrated 24-h P[M.sub.2.5] particle samples were collected on Teflon filters. Most of PM samples were collected every third day and some were collected daily during the time period between July 2003 and August 2005. A total of 256 and 293 samples were separately obtained at the Bayland Park and Orange sites. Both mass concentration and elemental chemical speciation were determined using an energy dispersive X-ray fluorescence (XRF). An ion chromatography (IC) was used to analyze sulphate (S[O.sup.2.sub.4-]), ammonium (N[H.sup.+.sub.4]), and nitrate (N[O.sup-.sub.3]) concentrations. The thermal optical transmission technique was used to measure both organic carbon (OC) and elemental carbon (EC). A total of 52 chemical elements was analyzed, including: Ag, Al, As, An, Ba, Br, Ca, Cd, Ce, CI, Co, Cr, Cs, Cu, Eu, Fe, Ga, Hf, Hg, In, Ir, K, La, Mg, Mn, Mo, Na, Nb, Ni, P, Pb, Rb, S, Sb, Se, Si, Sm, Sn, Sr, Ta, Tb, Ti, V, Y, Zn, Zr, W, OC, EC, S[O.sup.2-.sub.4] , N[H.sup.+.sub.4] , N[O.sup-.sub.3].

In the data, the concentration of XRF S and S[O.sup.2.sub.4] were highly correlated (slope = 2.77, [r.sup.2] = 0.96 for the Bayland Park data; slope = 2.72, [r.sup.2] = 0.94 for the Orange data), thus it is reasonable to exclude XRF S from the analysis (Kim et al., 2004). On the other hand, there are 22 chemical species from Bayland Park and Orange, respectively, with a lower signal-to-noise ratio because of too many below-detection-limit measurements. As a result, these species were excluded in the PMF analysis as well. Among the species excluded, 19 are common species to the two sites. The analysis of the compositional data, however, still revealed a mass closure violation after excluding these species. The comparison of measured PM mass to the sum of PM compositional data indicates that 12.5 % of the measured PM mass concentrations for the Bayland Park and 7.5 % for the Orange site, respectively, were less than the sum of species concentrations. In the data matrices, there were missing and below-detection-limit values. The analytical uncertainty estimates associated with each measured concentration and the detection limits for instruments were also reported. Tables 1 and 2 summarize the P[M.sub.2.5] speciation data used in this study.

METHODOLOGY

In this study, PMF was used with the data collected at the Bayland Park and Orange site as discussed previously. PMF is an approach of factor analysis, and it is described in detail by Paatero (1997). Only a brief description of this approach is provided here.

[FIGURE 1 OMITTED]

PMF and Data Handling

PMF uses the method of weighted least-squares to solve a general receptor modelling problem. The general model assumes there are p sources, source types or source regions (termed factors) impacting a receptor and the observed concentrations of various species at the receptor are linear combinations of the impacts from the p factors. The factor analysis model (PMF) can be written as:

X=GF+E (1)

where X is a known n x m concentration matrix of the m measured chemical species in n samples, G is an n x p matrix of source (or factor) contributions to the samples (time variations), F is a p x m matrix of source compositions (source or factor profiles), and E is an n x m residual matrix. It is assumed that only the concentration matrix X is known, and both G and F are unknown matrix to be determined. Furthermore, all of the elements of G and F are non-negative which means the samples cannot have any negative source contribution, and sources cannot have any negative species concentration. E represents the portion of the data variance unexplained by the p-factor model, and it is the difference between the measurement of X and the model Y=GF. The concentration x1j in Equation (1) can then be written as:

[MATHEMATICAL EXPRESSIONS NOT REPRODUCIBLE ASCII.]

where [x.sub.ij] is the jth species concentration measured in the ith sample, [g.sub.ik] is the particulate mass concentration from the kth source (or factor) contributing to the ith sample, [f.sub.kj] is the jth species mass fraction from the kth source (or factor), [e.sub.e.j] is the residual associated with the jth species concentration measured in the ith sample, and p is the total number of sources (or factors).

The objective of PMF is to estimate the mass contributions [g.sub.ik] and the mass fractions (profiles) fkj in Equation (2) by the weighted least-squares. The task of PMF is thus to minimize the sum of the squares of the residuals weighted inversely with error estimates (estimated uncertainties) of the data points. In other words, the data analysis by PMF can be described as to minimize the objective function Q:

[MATHEMATICAL EXPRESSIONS NOT REPRODUCIBLE ASCII.]

under constraints [g.sub.i.k] > 0, [f.sub.k.j] > 0, and with [s.sub.i.j] as the error estimate (estimated uncertainty) for [x.sub.i.j]. The estimates of source contributions and source profiles are obtained by a unique algorithm in which both matrices G and F are adjusted in each iteration step. The process continues until convergence occurs (Paatero, 1997; Polissar et al., 1998).

The application of PMF requires the estimated uncertainty [s.sub.i.j] for each data value [x.sub.i.j] to be carefully selected so that it reflects the quality and reliability of each data point. This important feature of PMF enables us to properly handle any below detection limit and missing data values. The uncertainty estimate provides a useful tool to decrease the weight of any below detection limit and missing data values when searching for the minimum of Q in Equation (3). In this study, the procedure of Polissar et al. (1998) was adopted as follows: (i) the concentration value [x.sub.i.j] was the actually measured concentration, and the sum of the analytical uncertainty and one third of the detection limit value was used as the estimated uncertainty [s.sub.i.j] if [x.sub.i.j] was a determined value; (ii) the concentration value [x.sub.i.j] was replaced by half of the detection limit value, and 5 sixths of the detection limit value was used as the estimated uncertainty [s.sub.i.j] if [x.sub.i.j] was below detection limit; (iii) the concentration value [x.sub.i.j] was set equal to the geometric mean of all the measured values of [x.sub.i.j] for element j, and its corresponding uncertainty szj was set equal to four times of this geometric mean value if [x.sub.i.j] was a missing data value. Half of the average detection limits were used for below detection limits values in the calculation of the geometric means. Furthermore, the estimated uncertainties of OC and [S0.sup.2.sub.4] were increased by a factor of three because of its magnitude compared to the lower concentration species.

Robust Mode

It is well known that extreme data values as well as true outliers can distort the least-squares estimation profoundly. A delicate handling of these data values is important, and PMF offers a robust mode to properly weigh these data points in the process of searching for the minimum of Q. The robust factorization based on the Huber influence function (Huber, 1981) is a technique of iterative reweighing of the individual data values. The least squares approach with the robust factor analysis leads now to n limit v.

[MATHEMATICAL EXPRESSIONS NOT REPRODUCIBLE ASCII.]

where

[MATHEMATICAL EXPRESSIONS NOT REPRODUCIBLE ASCII.]

and [aLpha] is the outlier threshold distance. The value of [alpha]=4.0 was chosen in this study

Mass Apportionment

The results of the PMF analysis reproduce the data and ensure that the source profiles and mass contributions are non-negative. However, it has not yet taken into account the measured mass. In addition, the results are uncertain relative to a multiplicative scaling factor.

Assuming that all of the sources contributing mass to the particulate matter samples have been identified, the sun of the mass contributions s hould have been identified, should be equal to the measured PM mass.With the measured PM mass in each samples as the response, a multiple linear regression can be performed to regress the mass concentration against the estimated factor (source contribution values) obtained from PMF that is:

[MATHEMATICAL EXPRESSIONS NOT REPRODUCIBLE ASCII.]

This regression provides several useful indicators of the quality of the solution. Obviously each of the regression coefficients must be non-negative. If there is any negative coefficient, it suggests that too many factors have been used. The regression coefficients are used to scale the factor profiles into those with physically meaningful units. Once the profiles are scaled and summed, it can be determined if the sum of a source profile exceeds 100 %. In this case, it suggests that too few factors may have been chosen (Hopke et al., 1980).

Because of the mass closure violation noted previously, the measured particle mass concentration was included as an independent variable in the PMF modelling to directly obtain the mass apportionment instead of using a regression analysis (Kim et al., 2003, 2004). The estimated uncertainties of the P[M.sub.2.5] mass concentrations were set at four times of their values to reduce their weight in the model fit so that the magnitude of PM mass will not skew the analysis. When the measured particle mass concentration is included as an independent variable, the PMF apportions a mass concentration for each source based on its temporal variation without using a multiple linear regression. The results of PMF modelling are then normalized by the apportioned particle mass concentrations so that the quantitative source contributions are obtained. Specifically:

[MATHEMATICAL EXPRESSIONS NOT REPRODUCIBLE ASCII.]

where [c.sub.k] denotes directly apportioned mass concentration by PMF for the kth factor.

Discrete Fourier Transform

Frequency separation in a pollutant time series is extremely important as the dynamic processes operate on different frequencies (Pasquill, 1974; Eskridge et al., 1997; Rao et al., 1997; Hies et al., 2000). The discrete Fourier transform was employed to investigate the time-frequency relationship of the source contributions estimated by PMF. The discrete Fourier transform of a time series x (t), X (k), is defined as:

[MATHEMATICAL EXPRESSIONS NOT REPRODUCIBLE ASCII.]

where n is the number of samples, x (t) is a time series, i = I is the imaginary unit, and vk = k/n. The periodogram at frequency [v.sub.k] equals the squared magnitude of X(k) in Equation (7), and it is an estimate for the spectral density function of a finite time series. The spectral density function of a time series indicates the strength of the signal in frequency domain.

As discussed in the previous studies of air pollutant concentrations (Hies et al., 2000; Liu et al., 2005), the logarithmic transformation was commonly used to stabilize the variance of meteorological data. The average was subtracted from all values to obtain a zero mean for the series. Discrete Fourier transform using a fast Fourier transform algorithm of the log-transformed source contributions was calculated to construct the periodogram which estimates the spectral density function of the source contributions. The spectral density function can detect periodic components in noisy time series such as the source contributions by splitting up the variance to the underlying periodicities (Rotach, 1995; Sun and Wang, 1996; Schlink et al., 1997). Thus, the periodogram was employed to investigate the frequency variations of source contributions in each of the identified factors. The regional factors show predominantly low frequency variations due to the lack of local impacts, however, the area-related and local factors show both high and low frequency variations.

RESULTS AND DISCUSSION

Determination of the Number of Sources An essential step in PMF analysis is to determine the number of factors and source apportionment. For determination of the number of factors, the basic consideration is to obtain a good fit of the model to the original data, and the model can well explain the physical meaning of the data. If there is goodness of fit, the theoretical Q value in Equation (4) should be approximately equal to the number of degrees of freedom or approximately equal to the number of entries in the data array provided that correct values of [s.sub.i.j] have been used (Yakovleva et al., 1999). However, there is actually no totally reliable information for selecting correct values of [s.sub.i.j] in practice, and the potential outlier presence complicates the situation to search for goodness of fit. As a result, determining the number of factors from the Q value bears uncertainty. The other two indicators for a correct determination of the number of factors are the regression coefficients bk in Equation (5) if available and the distribution of scaled residuals ([e.sub.i.j],/s.sub.i.j]). In a well-fit model, the residuals [e.sub.i.j] and the error estimates [s.sub.i.j] should not be too much different in size, and the ratio ([e.sub.i.j],/[s.sub.i.j.]) should fluctuate between [+ or -]3. Jumto and Paatero (1994) recommended values of [+ or-]2 for the ratio.

Sources and Source Profiles

Based on the criterion of obtaining the most physically meaningful solution with the calculated Q value (Q = 8486 and 9915 for the Bayland Park and Orange site, respectively) close to the theoretical Q value (Q = 7680 and 8790 for the Bayland Park and Orange sites, respectively), the PMF identified ten common source types by trial and error with different numbers of factors. We termed these factors as sulphate-rich secondary aerosol I, sulphate-rich secondary aerosol II, cement/carbon-rich, wood smoke, motor vehicle/road dust, nitrate-rich secondary aerosol, metal processing, soil, sea salt, chloride-depleted marine aerosol. The contributions of these factors towards the PM mass from [c.sub.k] in Equation (6) at the two sites are summarized in Table 3. To study the spatial variations contributed by different factors between the two sites, the square of correlation coefficient ([r.sub.2]) was calculated from the estimated source contributions with respect to the com mon factor for the two sites. Table 4 presents the summary of the squared correlation coefficients for the factors. The other detailed results are displayed in Figures 2 to 15.

Both Figures 2 and 3 show a relationship between the reconstructed P[M.sub.2.5] mass contributions from all sources and the measured P[M.sub.2.5] mass concentrations. It is clear that the resolved sources effectively reproduce the measured values and account for most of the variation in the P[M.sub.2.5] mass concentrations (slope = 0.87 and [r.sup.2] = 0.90 for the Bayland Park site; slope = 0.75 and [r.sup.2] = 0.81 for the Orange site). The predicted mass concentrations slightly underestimate for the higher measured mass concentrations possibly due to a mass closure violation. As indicated, 12.5% of measured PM mass concentrations for the Bayland Park and 7.5% for the Orange sites, respectively, were less than the sum of species concentrations. Figures 4 to 11, Figures 14 and 15 present the time series plots of estimated source contributions to P[M.sub.2.5] mass concentrations for each source, the identified source profile at the two sites, and the periodogram plot for each source at the Orange site. The seasonal variations in the time series plots may be explained by variation in source strength, atmospheric transport, and possible chemical reactions in the atmosphere, or a combination of the three. Figures 12 and 13 show the source contributions with wind direction in polar coordinates for soil, sea salt, and chloride-depleted marine factor at the two sites, respectively.

[FIGURE 2 OMITTED]

[FIGURE 3 OMITTED]

Among the ten factors identified, the first and second common factors are two sulphate-rich secondary aerosols. Secondary sulphate aerosols are formed mainly due to the presence of sulphur dioxide in the atmosphere. Figures 4 and 5 show the contribution and factor profile resolved by PMF for the two factors, respectively. Both sources have a high concentration of carbon, [S[O.sup.2.sub.4] , and N[H.sup.+.sub.4]. OC and EC were associated with these factors. The OC association was consistent with several previous studies (Ramadan et al., 2000; Kim et al., 2004). The mixed EC concentration probably reflects that the resolved factor by PMF may not merely represent one source. The amount of ammonium found in these sources accounts for about 93% and 94% of the total ammonium concentration at the Bayland Park and Orange sites, respectively. It indicates a strong association between ammonium and sulphate. Molar ratios of ammonium to sulphate for the two factors were 1.8 and 2.4 at the Bayland Park site, and that for the two factors were 1.7 and 1.9 at the Orange site. Because of the possible evaporation of ammonium during sample analysis and/or the uncertainty of the PMF estimate, sulphate is likely present mainly as ammonium sulphate with molar ratios not equal to 2.0 at the two receptor sites. Sulphate-rich secondary aerosol I has the highest source contribution to P[M.sub.2.5] mass concentration with 40% and 36.5%, and sulphate-rich secondary aerosol II has the second highest with 19.5% and 17.4% at the two sites, respectively. Carbon and trace elements usually become associated with the secondary sulphate aerosol in the atmosphere (Kim et al., 2004). However, sulphate rich secondary aerosol II has higher loadings of trace metals at the sites. The middle panel in Figures 4 and 5 indicate that the major difference between the two types of sulphate-rich aerosol is in silicon and some elements such as Ba, P, and Fe or Na. The top panel in Figures 4 and 5 clearly indicate that these two types showed up at two different time frames. The sulphate I factor was mostly "on" at both sites until about November 1, 2004 and then turned "off", while sulphate II was "off" until about November 1, 2004 and then coincidentally turned "on" at both sites. A possible explanation is that the XRF analysis of PM2.5 speciation filters at sites in Texas was conducted by the Research Triangle Institute. After October 31, 2004, the XRF analysis of filters from all except three of Texas PM2.5 speciation sites was switched to the Desert Research Institute Laboratory. The two sites used in this PMF analysis are among those switched to. The sulphate-rich secondary aerosol I sources show slightly higher concentrations in late summer and early fall when the photochemical activity is high in the region (Polissar et al., 2001; Song et al., 2001). The two sulphate-rich secondary aerosol sources account for almost 59% and 54% of the PM2.5 mass concentration at the two sites, respectively. This is quite similar to the study of three northeastern US cities which identified its contributions of 47%, 55%, and 51 % to the PM2.5 mass concentration (Song et al., 2001). The bottom panel in Figures 4 and 5 show the periodogram for the two factors, respectively, at the Orange site. There is a large peak at high frequency for annual cycle and almost no significant peak at high frequency. It indicates the seasonal dependence of sulphate formation with limited local impact on these factors at the Orange site. The top panel in Figures 4 and 5 show highly similar seasonal variations at the two sites. This highly similar seasonal variations at the two sites and a significant squared correlation coefficient of [r.sub.2] = 0.56 and 0.67 for the two factors, respectively, between the two sites imply that these factors are regional factors, and due to the laboratory change about November 1, 2004, it is likely that the two sulphate-rich secondary aerosols were separately identified by PMF when in fact there is only one sulphate source.

[FIGURE 4 OMITTED]

[FIGURE 5 OMITTED]

The third common factor is related to a cement/carbon-rich source characterized by Ca, OC, and EC (US EPA, 2002; Kim et al., 2004). The middle panel in Figure 6 shows the factor profiles at the two sites. Fe and a small amount of K and Zn were associated with this factor at the Bayland Park site. A small amount of S[0.sup.2-.sub.4], N[H.sub.+.sub.4] , Fe, and Si were associated with this factor at the Orange site. It contributes 13.7% and 11.7% to the P[M.sub.2.5] mass concentration at the Bayland Park and Orange sites, respectively, and likely includes contributions from the construction sites and an unknown carbon-rich source possibly from chemical plants in the Houston and Golden Triangle areas, respectively. The high carbon concentration of this source indicates that the cement and a carbon-rich source are co-located and daily emission patterns are similar (Kim et al., 2004). The bottom panel in Figure 6 shows the periodogram for this factor at the Orange site. The peaks at low frequency corresponded to the annual and semiannual cycle, and the peaks at high frequency were related to a monthly and weekly cycle. The high weekly peak suggested this factor was dominated by weekday-weekend local activity such as reduced activity at the building/highway construction sites over weekends. The top panel in Figure 6 shows no similar temporal variability between the two sites. The factor contribution peak did not match between the two sites with [r.sup.2] = 0. This indicates that the two sites may be influenced by some distinct local sources under certain meteorological conditions.

[FIGURE 6 OMITTED]

The fourth common factor was identified as wood smoke source which is characterized by K, OC, and EC (Watson et al., 2001; Kim et al., 2004; Liu et al., 2005). It contributes 3.2% and 11.1% to the P[M.sub.2.5] mass concentration at the two sites, respectively. The middle panel in Figure 7 shows the factor profiles at the two sites. Sulphate and a small amount of nitrate were associated with this factor at the Bayland Park site. The wood smoke probably comes from residential wood burning, local agricultural biomass burning, and occasional forest fires. For the two sites, this factor has a slightly higher trend in winter season possibly related to residential wood burning for heating. The short-term peaks in spring and summer were probably due to forest fires, and/or biomass burning from Central America. However, the squared correlation coefficient of this factor between the two sites was [r.sup.2] = 0.03 which is considered substantially low. The top panel in Figure 7 shows little similar temporal variability between the two sites. It indicates that the two sites are influenced by different types of local sources. The periodogram for this factor at the Orange site is shown at the bottom panel of Figure 7. The large peaks at low frequency reflected different annually and semiannual wood burning activity. The local characteristic of this factor in the area was represented by a large peak at high frequency such as monthly or weekly.

The fifth common factor was not as readily interpreted as the other factors; however, it was identified as motor vehicle/road dust source characterized by higher concentration of Si and OC along with Ca, Fe, and K (Chueinta et al., 2000). The middle panel in Figure 8 shows the quite similar factor profiles at the two sites. A small amount of Ti, ammonium, and nitrate were mixed in this factor during the formation and transport. A small amount of EC was also associated with this factor at the Orange site. This source might be accounted for the mixing of sources such as vehicles on highway, road dust, summer soil, and emission from vegetation or wood smoke. The busy highway I-610 and SH 59 intersect in the proximity of Bayland Park, and the monitoring site is located about 3 km east of SH 59 and 4 km west of I-610. The Orange site is located about 6 km north of I-10 and 0.5 km west of SH 62. It has short-term peaks in June and July, and shows a summer-high seasonal trend possibly due to the higher concentration of soil dust during the period. This source accounts for 6.8 % of the PM2.s mass concentration at the two sites, respectively. The bottom panel in Figure 8 shows the periodogram for this factor at the Orange site. The large peak at low frequency for the annual cycle indicated the seasonal dependence of formation of this source possibly from the transported dust. The peaks at high frequency suggest that the urban area traffic has significant impact on this factor at the Orange site. The top panel in Figure 8 shows significantly similar seasonal variations at the two sites due to the summer dust and the monitoring sites being in the area of Southeastern Texas. The apparently similar temporal variability and a squared correlation coefficient of [r.sup.2] = 0.55 between the two sites imply that this source is highly influenced by the summer soil dust and area traffic sources.

[FIGURE 7 OMITTED]

[FIGURE 8 OMITTED]

The sixth common factor resolved at the two sites mainly consists of ammonium and nitrate. The nitrate-rich secondary aerosol is identified by its high concentration of N[O.sup.-.sub.3] and N[H.sup.+.sub.4]. Figure 9 shows the factor contribution and profile results for this source. OC and EC were associated with this factor at the Bayland Park site. EC and a small amount of trace metals such as Ca, K, and Na were also associated with this factor at the Orange site. This source includes N[H.sup.+.sub.4] that becomes associated with the secondary nitrate aerosol in the atmosphere. Molar ratios of ammonium to nitrate were 1.1 and 0.4 for the Bayland Park and Orange sites, respectively. Because of the possible evaporation of ammonium during sample analysis and/or the uncertainty of the PMF estimate, nitrate is probably present mainly as ammonium nitrate with molar ratios not equal to 1.0 at the two receptor sites. Nitrate is formed in the atmosphere mostly through the oxidation of NO, depending on ambient temperature, relative humidity, and the presence of ammonia (Liu et al., 2005). It has short-term peaks and higher trend in cool seasons possibly indicating that low temperature and high humidity foster the formation of nitrate aerosol in the region as discussed in the study for Atlanta (Kim et al., 2004) and three northeastern US cities (Song et al., 2001). The bottom panel in Figure 9 shows the periodogram for this factor at the Orange site. The seasonal dependence of nitrate formation is reflected by a high peak at low frequency. The top panel in Figure 9 shows the similar seasonal variations of ammonium nitrate at the two sites. It reflects the regional characteristic of ammonium nitrate formation and transport. The local characteristic of this source in the area was reflected by the small monthly or weekly peak, and apparently the local impact was limited. The [r.sup.2] value of 0.18 between the two sites is not as high as those of sulphate factors possibly due to the shorter lifetime of NO, than SO4. The source accounts for 4.8% and 4.4% of the PMz s mass concentration at the Bayland Park and Orange sites, respectively.

[FIGURE 9 OMITTED]

The seventh common factor suggested a source of metal processing because of the profile characterized by its high concentration of Zn along with OC, EC, and S[sub.O.sup.2.sub.4] at the Bayland Park site and high concentration of Fe associated with EC and N[H.sup.+.sub.4] at the Orange site (US EPA, 2002; Kim et al., 2004). The middle panel in Figure 10 shows the factor profiles at the two sites. This factor at the Bayland Park site is highly likely to include contributions from metal processing facilities in Houston area. One of the facilities, 12 km north of the Bayland Park receptor, uses the process of galvanization to coat steel and iron with zinc. A major steel mill, 1 km south of I-10 and 10 km east of Beaumont, is also likely to contribute to this factor at the Orange site. The source showed a slightly reduced seasonal trend at the receptors in the late spring and summer possibly due to the locations of monitoring sites and southerly winds. About 82 % and 61 % of the data from June to August with southerly wind direction were reported for the Bayland Park and Orange sites, respectively. This source accounts for 5.5% and 4.0% of the P[M.sub.2.5] mass concentration at the two sites, respectively. The bottom panel in Figure 10 shows the periodogram for this factor at the Orange site. The peaks at low frequency corresponded to the annual and semiannual cycle possibly reflecting the seasonal variation, and the peak at high frequency was related to a weekly cycle. The weekly high peak suggested this factor was dominated by weekday-weekend local activity. The top panel in Figure 10 shows little similar temporal variability between the two sites. The much different time variations at the two sites and a squared correlation coefficient of rz = 0 between the two sites emphasize this factor may come from different local metal sources or common sources with different impacts on the two sites.

[FIGURE 10 OMITTED]

The eighth common factor was identified as soil source represented by high concentration of Al and Si along with Fe, K, Mg, Na, and Ti (Watson et al., 2001; Kim et al., 2004). It contributes 2.2% and 3.0% to P[M.sub.2.5] mass concentration at the two sites, respectively. The crustal particles could be contributed by unpaved roads, construction sites, and soil dust. The contribution and factor profile in Figure 11 show the highly similar results of this factor between the two sites. The airborne soil shows seasonal variation with higher concentrations in the summer. The short-term peaks in summer of 2004 and 2005 likely reflect the intercontinental dust transport as indicated in several analyses across the eastern US (Liu et al., 2005). Prospero (2001) showed that the summer trade winds carry African dusts into US from the direction of southeast which is consistent with what the top panel in Figures 12 and 13 indicate. The mixed OC, EC, and [S.sup.2-.sub.4] concentration in this factor imply that this source was mixed with some other sources during the long-range transport. The bottom panel in Figure 11 shows the periodogram for this factor at the Orange site. The large peaks at low frequency for the annual and semiannual cycle indicated the seasonal variations of this factor. The small peaks at high frequency suggest that local dust has limited impact on this factor at the Orange site. The top panel in Figure 11 shows highly similar seasonal trends at the two sites. This highly similar seasonal variations at the two sites and a significantly large squared correlation coefficient of [r.sup.2] = 0.79 between the two sites imply that this factor is a regional factor.

[FIGURE 11 OMITTED]

The ninth common factor at the two sites has high concentration of Cl and Na. It is clearly from the marine or sea salt aerosol source (Lee et al., 1999). As the Bayland Park and Orange monitoring sites are located 80 km northwest and 55 km north of the Gulf of Mexico, respectively (Figure 1), the presence of marinerelated aerosol is expected. Figure 14 shows the contribution and factor profile resolved by PMF for this factor. The middle panel in Figure 14 shows the comparable factor profiles at the two sites. Both nitrate and sulphate were associated with this factor at the Orange site probably due to scavenging of nitrate and sulphate during the transport from the coast, however, OC and a small amount of EC were also associated with this factor at the two sites. It has slightly higher concentrations in summer possibly due to the southerly winds. This source accounts for 1.1 % and 2.6% of the P[M.sub.2.5] mass concentration at the Bayland Park and Orange sites, respectively. The bottom panel in Figure 14 shows the periodogram for this factor at the Orange site. The peaks at low frequency for the annual and semiannual cycle indicated the seasonal variations of this factor. The peaks at high frequency suggest that the Gulf of Mexico has substantial impact on this factor at the Orange site. The middle panel in Figures 12 and 13 clearly show the relationship of this factor with wind direction from the Gulf of Mexico. The top panel in Figure 14 shows highly similar seasonal variations at the two sites due to the proximity of the monitoring sites to the Gulf of Mexico. This highly similar seasonal variations at the two sites and a squared correlation coefficient of [sup.r.2] = 0.33 between the two sites imply that this source is highly influenced by the monitoring site being in the proximity to the Gulf of Mexico.

[FIGURE 12 OMITTED]

[FIGURE 13 OMIITED]

The tenth common factor was identified as chloride-depleted marine aerosol that is related to the sea salt factor (Lee et al., 1999). The middle panel in Figure 15 shows the comparable factor profiles at the two sites. It has high concentration of Na and S[O.sup.2-sub.4]. OC and a small amount of Ca, Mg, and nitrate were associ ated with this factor at the two sites, however, a small amount of K and EC were also associated with this factor at the Orange site. It is originated from sea salt aerosol which has undergone the chloride loss reactions through acid substitution and yielded a higher loading of S[O.sup.2-.sub.4] in the source than sea salt aerosol (Lee et al., 1999). This chemical reaction usually occurs in the coastal areas with high sulphur loading. The composition of chloride-depleted marine aerosol depends on air quality and meteorological conditions, and therefore it was separately identified from sea salt. The higher sulphate loading in chloride-depleted marine aerosol compared to sea salt has led almost no chloride associated with this factor identified by PME However, the lower sulphate loading in sea salt compared to chloride-depleted marine aerosol has led a high chloride loading in sea salt identified by PME At the bottom panel of Figures 12 and 13, there are indications of higher concentrations at the two sites from the direction of south which is consistent with the source direction of sea salt aerosol. The chloride-depleted marine aerosol source has slightly higher concentrations in spring and summer. This source accounts for 3.2 and 2.5 % of the P[M.sub.2.5] mass concentration at the two sites, respectively. The bottom panel in Figure 15 shows the periodogram for this factor at the Orange site. The large peak at low frequency for the annual cycle indicated the seasonal dependence of formation of this source. The peaks at high frequency suggest that the Gulf of Mexico has substantial impact on this factor at the Orange site. The top panel in Figure 15 shows similar seasonal variations at the two sites due to the proximity of the monitoring sites to the Gulf of Mexico. This similar seasonal variations at the two sites and the squared correlation coefficient of [r.sup.2] = 0.18 between the two sites imply that this source is likely influenced by the monitoring site being in the proximity to the Gulf of Mexico.

[FIGURE 14 OMITTED]

[FIGURE 15 OMITTED]

Uncertainty of Profile Estimates

The profile estimates discussed in the preceding section do not have the associated error of estimates. They are merely the point estimates, not how much reproducibility there is in those point estimates. To estimate the uncertainties of source profiles obtained from PMF, a bootstrapping technique (Efron and Tibshirani,1993) combined with a method to account for the rotational freedom in the solution was used. The middle panels of Figures 4 to 11, Figures 14 and 15 display the lower and upper limit of a 90% confidence interval for the mean profiles as well as the profile estimates. The smaller magnitude the whisker, the more consistent the estimate is and the larger magnitude the whisker, the less consistent the result is. In other words, the smaller magnitude the whisker, the smaller associated error the estimate has and the larger magnitude the whisker, the larger associated error the estimate has.

CONCLUSIONS

An air quality study has been carried out to identify and compare the sources of particulate pollutants at two EPA monitoring sites in Texas, namely, the Bayland Park and the Orange monitoring sites located in Southeastern Texas with about 175 km separation along Interstate Highway 10. The two sites have average annual PM concentrations of 10.58 and 12.05 [micro]g/[m.sup.3], respectively, which are below the National Ambient Air Quality Standards of 15 [micro]g/[m.sup.3] for PM.

In the study, aerosol composition data of P[M.sub.2.5] derived from samples collected at two monitoring sites in Southeastern Texas were analyzed by positive matrix factorization. The PMF effectively identified ten possible common source-related factors for P[M.sub.2.5]. The estimated source contributions for the common factor between the two sites were used to analyze spatial differences and correlations. Fourier transform was employed to investigate the frequency variations of the identified factors. The results showed the possible source types and factor contributions (%) for the two sites are quite comparable. The factors are classified as regional, area-related, and local sources.

Two different sulphate-rich secondary aerosols were extracted by PMF, which, respectively, had the first and second highest contribution to the P[M.sub.2.5] mass in the region accounting for almost 59% and 54% of the total concentration at the two sites, respectively. Sulphate and nitrate mainly exist as ammonium sulphate and ammonium nitrate at the receptor sites. Sulphate, nitrate, and soil show regional characteristics with similar seasonal variation patterns and low frequency variations at the two sites. The soil factor has high source contribution peaks during the summer likely reflecting the intercontinental dust transport. The regional factors account for about 61-66% of the P[M.sub.2.5] mass concentration. The sea salt factor is clearly seen at the sites from the Gulf of Mexico. The chloride-depleted marine aerosol was originated from sea salt aerosol; however, it was separately identified because of the chloride loss during chemical reactions in the atmosphere.

The correlations between the two sites are slightly lower than moderate for sea salt and chloride-depleted marine aerosol, and moderate for motor vehicle/road dust. The periodogram from Fourier transform for the motor vehicle/road dust factor shows high variations at both low and high frequency. It implies that this source is likely influenced by both the summer soil dust and area traffic sources. The periodograms for sea salt and chloride-depleted marine aerosol show peaks at high frequency reflecting the impact of the area being in the proximity to the Gulf of Mexico. The correlations between the two sites are poor for cement/carbon-rich, wood smoke, and metal processing. The periodograms for these factors show large high-frequency variations. It indicates these factors are mainly dominated by local sources. The metal processing facilities and steel mills in Houston and a steel mill in the Golden Triangle, respectively, are clearly suggested of being related to the source of metal processing. The local factors on the average contribute about 22-27 % to the P[M.sub.2.5] mass concentration for the two sites, respectively.

ACKNOWLEDGEMENTS

This study was supported in part by the US EPA through project R07-0159. The authors wish to thank Professor Hopke of Clarkson University for helpful e-mail communications and the reviewers for comments. The result of this research represents only the authors' assessments and does not reflect the funding agency's views on the air quality issues in this region.

NOMENCLATURE

X a known n x m concentration matrix
G an n x p matrix of source (or factor) contributions
 to the samples
F a p x m matrix of source compositions (source or
 factor profiles)
E an n x m residual matrix
[x.sub.i.j] the jth species concentration measured in the ith sample
[g.sub.i.k] the mass concentration from the kth source contributing
 to the ith sample
[f.sub.k.j] the jth species mass fraction from the kth source
[e.sub.i.j] the residual associated with the jth species
 concentration measured in the ith sample
Q objective function
[s.i.j] estimated uncertainty
x (t) source contributions
X (k) discrete Fourier transform of x (t)

REFERENCES

Anttila, P., P. Paatero, U. Tapper and O. Jarvinen, "Source Identification of Bulk Wet Deposition in Finland by Positive Matrix Factorization," Atmos. Environ. 29, 1705-1718 (1995).

Chiou, P., W Tan, C. J. Lin, H. W Chu and T. C. Ho, "Atmospheric Aerosol over a Southeastern Region of Texas: Chemical Composition and Possible Sources," Environ. Model. Assess. DOI: 10.1007/S10666-007-9120-8.

Chueinta, W, P. K. Hopke and P. Paatero, "Investigation of Sources of Atmospheric Aerosol at Urban and Suburban Residential Areas in Thailand by Positive Matrix Factorization," Atmos. Environ. 34, 3319-3329 (2000).

Dockery, D. W, C. A. Pope, III, X. P. Xu, J. D. Spengler, J. H. Ware, M. E. Fay, B. G. Ferris and F. E. Speizer, "An Association between Air Pollution and Mortality in Six United States Cities," N. Engl. J. Med. 329, 1753-1759 (1993).

Efron, B. and R. L. Tibshirani, "An Introduction to the Bootstrap," Chapman and Hall, London (1993).

Eskridge, R. E., J. Y. Ku, S. T. Rao, P. S. Porter and I. G. Zurbenko, "Separating Different Scales of Motion in Time Series of Meteorological Variables," Bull. Am. Meteorol. Soc. 78, 1473-1483 (1997).

Gilliland, F. D., K. Berhane, E. B. Rappaport, D. C. Thomas, E. Avol, W J. Gauderman, S. J. London, H. G. Margolis, R. McConnell, K. T Islam and J. M. Peters, "The Effects of Ambient Air Pollution on School Absenteeism due to Respiratory Illnesses," Epidemiology 12, 43-54 (2001).

Hansen, D. A., E. S. Edgerton, B. E. Hartsell, J. J. Jansen, N. Kandasamy, G. M. Hidy and C. L. Blanchard, "The Southeastern Aerosol Research and Characterization Study. Part 1. Overview," J. Air Waste Manage. Assoc. 53, 1460-1471 (2003).

Hies, T., R. Treffeisen, L. Sebald and E. Reimer, "Spectral Analysis of Air Pollutants. Part 1. Elemental Carbon Time Series," Atmos. Environ. 34, 3495-3502 (2000).

Hopke, P. K., "Receptor Modeling in Environmental Chemistry," John Wiley & Sons, New York (1985).

Hopke, P. K., "Receptor Modeling for Air Quality Management," Elsevier Science, Amsterdam (1991).

Hopke, P. K., R. E. Lamb and D. F. S. Natusch, "Multielemental Characterization of Urban Roadway Dust," Environ. Sci. Technol. 14, 164-172 (1980).

Huber, P. J., "Robust Statistics," John Wiley, New York (1981).

Juntto, S. and P. Paatero, "Analysis of Daily Precipitation Data by Positive Matrix Factorization," Environmetrics 5, 127-144 (1994).

Kim, E., P. K. Hopke and E. S. Edgerton, "Source Identification of Atlanta Aerosol by Positive Matrix Factorization," J. Air Waste Manage. Assoc. 53, 731-739 (2003).

Kim, E., P. K. Hopke and E. S. Edgerton, "Improving Source Identification of Atlanta Aerosol Using Temperature Resolved Carbon Fractions in Positive Matrix Factorization," Atmos. Environ. 38, 3349-3362 (2004).

Lee, E., C. K. Chan and P. Paatero, "Application of Positive Matrix Factorization in Source Apportionment of Particulate Pollutants in Hong Kong," Atmos. Environ. 33, 3201-3212 (1999).

Lipfert, F. W and R. E. Wyzga, "Air Pollution and Mortality: Issues and Uncertainties," J. Air Waste Manage. Assoc. 45, 949-966 (1995).

Liu, W, Y. H. Wang, R. Armistead and E. S. Edgerton, "Atmospheric Aerosol over Two Urban-Rural Pairs in the Southeastern United States: Chemical Composition and Possible Sources," Atmos. Environ. 39, 4453-4470 (2005).

Paatero, P., "Least Squares Formulation of Robust, Non-Negative Factor Analysis," Chemomet. Intell. Lab. Syst. 37, 23-35 (1997).

Paatero, P. and U. Tapper, "Analysis of Different Modes of Factor Analysis as Least Squares Fit Problems," Chemomet. Intell. Lab. Syst. 18, 183-194 (1993). Paatero, P. and U. Tapper, "Positive Matrix Factorization: A Non-Negative Factor Models with Optimal Utilization of Error Estimates of Data Values," Environmetrics 5, 111-126 (1994).

Pasquill, F., "Atmospheric Diffusion," Wiley, Chichester (1974).

Paterson, K. G., J. L. Sagady, D. L. Hooper, S. B. Bertman, M. A. Carroll and P. B. Shepson, "Analysis of Air Quality Data Using Positive Matrix Factorization," Environ. Sci. Technol. 33, 635-641 (1999).

Peters, A., D. W Dockery, J. E. Muller and M. A. Mittleman, "Increased Particulate Air Pollution and the Triggering of Myocardial Infraction," Circulation 103, 2810-2815 (2001).

Polissar, A. V., P. K. Hopke, W C. Malm and J. F. Sisler, "The Ratio of Aerosol Optical Absorption Coefficients to Sulfur Concentrations, as an Indicator of Smoke from Forest Fires When Sampling in Polar Regions," Atmos. Environ. 30, 1147-1157 (1996).

Polissar, A. V, P. K. Hopke, P. Paatero, Y. J. Kaufman, D. K. Hall, B. A. Bodhaine, E. G. Dutton and J. M. Harris, "The Aerosol at Barrow, Alaska: Long-Term Trends and Source Locations," Atmos. Environ. 33, 2441-2458 (1999).

Polissar, A. V., P. K. Hopke, P. Paatero, W C. Malm and J. F. Sisler, "Atmospheric Aerosol over Alaska. 2. Elemental Composition and Sources," J. Geophys. Res. 103, 19045-19057 (1998).

Polissar, A. V., P. K. Hopke and R. L. Poirot, "Atmospheric Aerosol over Vermont: Chemical Composition and Sources," Environ. Sci. Technol. 35, 4604-4621 (2001).

Pope, C. A. III, R. T Burnett, M. J. Thun, E. E. Calle, D. Krewski, K. Ito and G. D. Thurston, "Lung Cancer, Cardio-Pulmonary Mortality, and Long-Term Exposure to Fine Particulate Air Pollution," J. Am. Med. Assoc. 287, 1123-1141 (2002).

Prospero, J. M., "African Dust in America," Geotimes (November 24-27, 2001).

Ramadan, Z., B. Eickhout, X. H. Song, L. M. C. Buydens and P. K. Hopke, "Comparison of Positive Matrix Factorization and Multilinear Engine for the Source Apportionment of Particulate Pollutants," Chemomet. Intell. Lab. Syst. 66, 15-28 (2003).

Ramadan, Z., X. H. Song and P. K. Hopke, "Identification of Sources of Phoenix Aerosol by Positive Matrix Factorization," J. Air Waste Manage. Assoc. 50, 1308-1320 (2000).

Rao, S. T., I. G. Zurbenko, R. Neagu, P. S. Porter, J. Y. Ku and R. F. Henry, "Space and Time Scales in Ambient Ozone Data," Bull. Am. Meteorol. Soc. 78, 2153-2166 (1997).

Rotach, M. W., "Profiles of Turbulence Statistics and Above an Urban Street Canyon," Atmos. Environ. 29, 1473-1486 (1995).

Rudell, B., A. Blomberg, R. Helleday, M. C. Ledin, B. Lundback, N. Sternjerb, P. Horstedt and T. Sandstorm, "Bronchoalveolar Inflammation after Exposure to Diesel Exhaust: Comparison between Unfiltered and Particle Trap Filtered Exhaust," Occup. Environ. Med. 56, 527-534 (1999).

Schlink, U., O. Herbarth and G. Tetzlaff, "A Component Time-Series Model for SOz Data: Forecasting, Interpretation, and Modification," Atmos. Environ. 31, 1285-1295 (1997).

Schwartz, J., D. Slater, T. V Larson, W E. Pierson and J. Z. Koenig, "Particulate Air Pollution and Hospital Emergency Room Visits for Asthma in Seattle," Am. Rev. Respir. Dis. 47,826-831 (1993).

Song, X. H., A. V Polissar and P. K. Hopke, "Sources of Fine Particle Composition in the Northeastern US," Atmos. Environ. 35, 5277-5286 (2001).

Sun, L. and M. Wang, "Global Warming and Global Dioxide Emission: An Empirical Study," J. Environ. Manage. 46, 327-343 (1996).

US EPA, "SPECIATE version 3.2," US Environmental Protection Agency, Research Triangle Park, NC (2002).

Vedal, S., "Critical Review: Ambient Particles and Health-Lines that Divide," J. Am. Med. Assoc. 47, 551-581 (1997).

Watson, J. G., J. C. Chow and J. E. Houck, "PMZ.5 Chemical Source Profiles for Vehicle Exhaust, Vegetative Burning, Geological Material, and Coal Burning in Northwestern Colorado During 1995," Chemosphere 43, 1141-1151 (2001).

Xie, Y. L., P. K. Hopke P. Paatero, L. A. Barrie and S. M. Li, "Identification of Source Nature and Seasonal Variations of Arctic Aerosol by Positive Matrix Factorization," J. Atmos. Sci. 56, 249-260 (1999).

Yakovleva, E., P. K. Hopke and L. Wallace, "Receptor Modeling Assessment of PTEAM Data," Environ. Sci. Technol. 33, 3645-3652 (1999).

Zheng, M., G. R. Cass, J. J. Schauer and E. S. Edgerton, "Source Apportionment of PMZ.5 in the Southeastern United States Using Solvent-Extractable Organic Compounds as Tracers," Environ. Sci. Technol. 36, 2361-2371 (2002).

Manuscript received January 14, 2008; revised manuscript received February 15, 2008; accepted for publication February 18, 2008.

Paul Chiou, [1] Wei Tang, [2] Che-Jen Lin,[3] Hsing-Wei Chu ,[4] Rafael Tadmor [2] and T. C. Ho [2]

[1.] Department of Mathematics, Lamar University, Box 10047, Beaumont, TX 77710, U.S.A.

[2.] Department of Chemical Engineering, Lamar University, Box 10053, Beaumont, TX 77710, U.S.A.

[3.] Department of Civil Engineering, Lamar University, Box 10024, Beaumont, TX 77710, U.S.A.

[4.] Department of Mechanical Engineering, Lamar University, Box 10028, Beaumont, TX 77710, U.S.A.

Can. J. Chem. Eng. 86:421-435, 2008 2008 Canadian Society for Chemical Engineering DOI 10.1002/cjce.20047

Table 1. Summary of [PM.sub.2.5] and 29 species mass concentrations at
Bayland Park used for PMF analysis

Species Concentration (ng [m.sup.-3])

 Geometric Arithmetic Minimum Maximum
 mean (b) mean

[PM.sub.2.5] 9232 10574 2459 29800
Al 16.2 54.2 4.0 156.4
As 0.87 1.54 0.1 16.2
Ba 6.7 8.5 1.85 144.0
Br 2.0 2.8 0.23 9.7
Ca 36.1 47.2 1.5 297.2
CI 6.5 34.0 0.75 617.0
Cr 0.75 1.0 0.35 11.5
Cs 5.3 6.0 1.8 91.9
Cu 1.7 2.53 0.5 22.5
Fe 49.4 75.2 1.4 869.7
K 57.6 70.5 11.9 477.0
Mg 14.9 27.7 3.25 395.4
Mn 1.2 1.6 0.28 9.78
Na 62.0 115.8 9.5 855.2
Ni 0.7 0.89 0.34 7.6
P 6.9 23.5 1.9 192.2
Pb 1.8 2.4 0.55 10.7
Si 75.9 164.2 3.05 2715
Sr 1.0 1.38 0.31 12.4
Ti 3.0 6.0 0.55 95.9
V 1.4 2.1 0.1 13.1
Y 0.7 0.79 0.39 4.89
Zn 6.7 11.7 0.5 105.0
Zr 1.25 1.44 0.5 10.8
OC 2490 3097 49.0 8500
EC 372 466 37.9 1920
[SO.sup.2-.sub.4] 2770 3394 324 11658
[NH.sup.+.sub.4] 916 1150 73.1 4385
[NO.sup.-.sub.3] 123 210 8.3 2740

Species Number of BDL (a) Number of missing
 values (%) values (%)

[PM.sub.2.5] 0 (0) 2 (0.8)
Al 138 (55.0) 1 (0.4)
As 107 (42.6) 1 (0.4)
Ba 188 (74.9) 1 (0.4)
Br 54 (21.5) 1 (0.4)
Ca 2 (0.8) 1 (0.4)
CI 98 (39.0) 1 (0.4)
Cr 151 (60.2) 1 (0.4)
Cs 217 (86.5) 1 (0.4)
Cu 76 (30.3) 1 (0.4)
Fe 1 (0.39) 1 (0.4)
K 0 (0) 1 (0.4)
Mg 168 (66.9) 1 (0.4)
Mn 134 (53.4) 1 (0.4)
Na 112 (44.6) 1 (0.4)
Ni 139 (55.4) 1 (0.4)
P 147 (58.6) 1 (0.4)
Pb 134 (53.4) 1 (0.4)
Si 9 (3.6) 1 (0.4)
Sr 159 (63.4) 1 (0.4)
Ti 88 (35.1) 1 (0.4)
V 74 (29.5) 1 (0.4)
Y 222 (88.5) 1 (0.4)
Zn 7 (2.8) 1 (0.4)
Zr 207 (82.5) 1 (0.4)
OC 0 (0) 11 (4.38)
EC 1 (0.39) 11 (4.38)
[SO.sup.2-.sub.4] 0 (0) 1 (0.4)
[NH.sup.+.sub.4] 0 (0) 1 (0.4)
[NO.sup.-.sub.3] 2 (0.79) 1 (0.4)

(a) Below detection limit.

(b) Data below the limit of detection were replaced by half of the
reported detection limit values for the geometric mean calculations.

Table 2. Summary of [PM.sub.2.5] and 29 species mass concentrations at
Orange used for PMF analysis

Species Concentration (ng [m.sup.-3])

 Geometric Arithmetic Minimum Maximum
 mean (b) mean

[PM.sub.2.5] 10719 12050 150 60400
Al 16 54 4.6 1420
As 0.59 0.82 0.1 4.2
Ba 7.1 9 2 84.6
Br 1.9 2.9 0.2 49.3
Ca 47 61 1.5 824
Cl 8.0 63 0.75 1580
Co 0.4 0.45 0.28 6.9
Cr 0.78 1.2 0.35 23.2
Cu 1.6 2.9 0.50 40.8
Fe 73 116 0.43 861
K 72 87 2.2 656
Mg 14 24 3.7 343
Mn 1.5 2.6 0.31 24.5
Na 67 129 11 1280
Ni 0.72 0.9 0.34 11
P 6.8 22 1.8 183
Pb 1.8 2.3 0.55 26
Se 0.5 0.6 0.1 21
Si 87.6 177 3 2477
Sr 1.2 1.6 0.35 15.3
Ti 3.1 6.4 0.55 90
V 1.8 2.6 0.1 11
W 3.9 4.5 1.3 120
Zn 6.2 7.9 0.5 72
OC 3269 3750 49 36800
EC 350 396 38 1820
[SO.sup.2-.sub.4] 2788 3392 2.5 11135
[NH.sup.+.sub.4] 838 1046 3.3 3730
[NO.sup.-.sub.3] 103 167 1.8 1363

Species Number of BDL (a) Number of missing
 values (%) values (%)

[PM.sub.2.5] 1 (0.34) 29 (10)
Al 146 (49.8) 29 (10)
As 146 (49.8) 29 (10)
Ba 199 (67.9) 29 (10)
Br 68 (23.2) 29 (10)
Ca 2 (0.68) 29 (10)
Cl 106 (36.2) 29 (10)
Co 244 (83.3) 29 (10)
Cr 162 (55.3) 29 (10)
Cu 112 (38.2) 29 (10)
Fe 2 (0.68) 29 (10)
K 1 (0.34) 29 (10)
Mg 180 (61.4) 29 (10)
Mn 121 (41.3) 29 (10)
Na 118 (40.3) 29 (10)
Ni 138 (47.1) 29 (10)
P 161 (55.0) 29 (10)
Pb 147 (50.2) 29 (10)
Se 238 (81.2) 29 (10)
Si 5 (1.7) 29 (10)
Sr 145 (49.5) 29 (10)
Ti 96 (32.8) 29 (10)
V 77 (26.3) 29 (10)
W 239 (81.6) 29 (10)
Zn 2 (0.68) 29 (10)
OC 7 (2.39) 16 (5.5)
EC 12 (4.1) 16 (5.5)
[SO.sup.2-.sub.4] 1 (0.34) 29 (10)
[NH.sup.+.sub.4] 3 (1.0) 29 (10)
[NO.sup.-.sub.3] 6 (2.0) 29 (10)

(a) Below detection limit.

(b) Data below the limit of detection were replaced by half of the
reported detection limit values for the geometric mean calculations.

Table 3. Possible source types and factor contributions (%) obtained
by PMF.

Source type Bayland Park site Orange site

Sulphate-rich I 40.0 36.5
Sulphate-rich II 19.5 17.4
Cement/carbon-rich 13.7 11.7
Wood smoke 3.2 11.1
Motor vehicle/road dust 6.8 6.8
Nitrate-rich 4.8 4.4
Metal processing 5.5 4.0
Soil 2.2 3.0
Sea salt 1.1 2.6
CI-depleted marine aerosol 3.2 2.5

Table 4. Bayland Park versus Orange [r.sup.2] for the factors by PMF
at the two sites.

Source type Squared correlation coefficient

Sulphate-rich I 0.56
Sulphate-rich II 0.67
Cement/carbon-rich 0
Wood smoke 0.03
Motor vehicle/road dust 0.55
Nitrate-rich 0.18
Metal processing 0
Soil 0.79
Sea salt 0.33
CI-depleted marine aerosol 0.18