Environ. Sci. Technol. 2010, 44, 618–623
Regression Model of Partial Pressures of PCBs, PAHs, and Organochlorine Pesticides in the Great Lakes’ Atmosphere MARTA VENIER AND RONALD A. HITES* School of Public and Environmental Affairs, Indiana University, Bloomington, Indiana 47405, 812-855-0193
Received September 16, 2009. Revised manuscript received November 11, 2009. Accepted November 24, 2009.
where P is the partial pressure of the analyte (in atm), ∆HSA is the energy necessary to move a mole of a substance from an environmental surface (soil, water, vegetation, etc.) to the gas-phase (in kJ/mol), T is the atmospheric temperature at the sampling site when the sample was collected (in K), R is the gas constant (0.0083 kJ/K · mol), and a0 is an intercept (2). An example of the application of this equation is given in Figure 1, which shows the atmospheric partial pressures of total PCBs at Sleeping Bear Dunes, Michigan, over the period from January 1992 to December 2007 as a function of reciprocal temperature at the sampling site. The regression is highly significant, giving a ∆HSA value of 40.0 ( 2.3 kJ/mol. To look for temporal trends in this time series data, we have introduced the idea of temperature corrected partial pressures (3), which are given by the following:
( 2881 - T1 )
ln P288 ) ln P - a1
The gas-phase concentrations of polychlorinated biphenyls (PCBs), polycyclic aromatic hydrocarbons (PAHs), and organochlorine pesticides have been measured at six sites around the Great Lakes every 12 days since the early 1990s as part of the Integrated Atmospheric Deposition Network. After converting the concentrations into partial pressures, data from all of the sites were combined and fitted using a multiple linear regression equation that included time (indicating the effect of a chemical’s regulation), atmospheric temperature (indicating seasonality of use or release), the human population within a 25 km radius of the site (indicating the effect of urbanization) and wind speed and wind direction (indicating the source of the chemical). The atmospheric levels of lindane (γ-HCH), DDTs, endosulfans, and chlordanes were largely related to seasonality, with much higher levels in the warm summer months. The levels of ΣPCBs, ΣPAHs, ΣDDTs, and chlordanes were related to urbanization (this was a secondary factor for the latter two), a result that was unexpected for the two pesticides. The levels of only two compounds, R- and γ-HCH, decreased rapidly as a function of time; conversely, most other compounds are declining at much slower rates. Wind speed and wind direction were statistically significant but unimportant variables for most of the compounds.
Introduction
(2)
where a1 is ∆HSA/R and P288 is the partial pressure corrected to an atmospheric temperature of 288 K. A plot of ln P288 vs time allows us to determine a first order rate constant (b1) using ln P288 ) b0 + b1t
(3)
where t is the sampling date in Julian days relative to January 1, 1990. An example of such an approach is given in Figure 2, which shows the data from Figure 1 after temperature correction using eq 2. In this case, the regression is highly significant, and the slope of the regression line (b1) is -0.00028 ( 0.00002 day-1, which converts to a half-life of 6.9 ( 0.4 years. In other words, the atmospheric partial pressures of total PCBs at Sleeping Bear Dunes decrease by a factor of 2 every ∼7 years, which indicates a modest rate of removal of PCBs over the period 1992-2007. In addition to temperature and time, we have also shown that the partial pressures of several POPs strongly depend on the human population living and working near the sampling site, a concept we have parametrized as follows: ln P ) c0 + c1log pop
(4)
where pop is the population within a 25 km radius of the site (4). [Note here that “ln” is the natural logarithm (base e) and “log” is the common logarithm (base 10).] Figure 3 shows a plot of eq 4 for total PCBs at the six IADN sites discussed in
Over the past several decades, the atmospheric concentrations of several persistent organic pollutants (POPs)ssuch as polychlorinated biphenyl’s (PCBs), polycyclic aromatic hydrocarbons (PAHs), and organochlorine pesticidesshave been measured at many locations in an effort to determine the sources and atmospheric residence times of these compounds. One such effort is the Integrated Atmospheric Deposition Network (IADN), which has measured the atmospheric concentrations of several POPs at six United States sampling sites in the Great Lakes region every 12 days for the past 19 years (1). IADN data indicate that gas-phase atmospheric partial pressures of semivolatile organic compounds strongly depend on atmospheric temperature and that this dependence can be modeled by the integrated form of the Clausius-Clapeyron equation ln P ) a0-
∆HSA RT
* Corresponding author e-mail:
[email protected]. 618
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 44, NO. 2, 2010
(1)
FIGURE 1. Relationship between the atmospheric partial pressure of total PCBs (in atm) and 1000/T (K-1) at Sleeping Bear Dunes. The regression equation is ln(P) ) -(4.82 ( 0.28) (1000/T) - (15.6 ( 1.0), N ) 457, r2 ) 0.391, and P < 0.001. 10.1021/es902804s
2010 American Chemical Society
Published on Web 12/11/2009
TABLE 1. Details on the IADN Sites Used in the Regression Equation for PCBs, PAHs, and Pesticides site
populationa
time range
Eagle Harbor Brule River Chicago Sleeping Bear Dunes Cleveland Sturgeon Point
451 2788 3 579 651 16 097 1 301 787 68 361
Nov 1990-Dec 2007 Jan 1996-Aug 2002 Jan 1996-Dec 2007 Jan 1992-Dec 2007 Jan 2003-Dec 2007 Dec 1992-Dec 2007
a Population refers to the 25 km radius value calculated using a GIS technique previously described (see ref (4)).
FIGURE 2. Relationship between the natural logarithm of the total PCBs partial pressures (atm), adjusted to a reference temperature of 288 K, and time, expressed as Julian days after 1/1/1990, for Sleeping Bear Dunes. The equation of the regression line is ln(P288) ) -(31.3 ( 0.1) - (2.76 ( 0.16) × 10-4 t, N ) 456, r2 ) 0.406, and P < 0.001.
FIGURE 3. Relationship between the natural logarithm of the total PCBs concentrations (pg/m3) and the human population within a 25 km radius from the six IADN sites. The equation of the straight regression line is log(PCB conc) ) (0.753 ( 0.189) + (0.351 ( 0.039) log(pop), N ) 6, r2 ) 0.953, P < 0.001; the equation of the curved regression line is log(PCB conc) ) (1.497 ( 0.069) + (0.038 ( 0.026) log2(pop), N ) 6, r2 ) 0.982, P < 0.001. The site abbreviations are as follows: EH, Eagle Harbor; BR, Brule River; SB, Sleeping Bear Dunes; SP, Sturgeon Point; CL, Cleveland; and CH, Chicago. this paper. In this case, the correlation is highly significant. Similarly, high correlations have been found for PAHs from sampling sites around the world (4), for polybrominated diphenyl ethers at the IADN sites (5, 6), and for polychlorinated dibenzo-p-dioxins and dibenzofurans (PCDD/Fs) in the continental United States (7). In previous studies from our laboratory, we have also investigated the effect of directional meteorological parameters, such as wind direction and backward air trajectories, on measured partial pressures. In one study, we determined that PAH concentrations at Sturgeon Point were higher when the air was coming to the site from the south (8). In another study, we used wind directions measured at the site and 4-day average backward trajectories to determine that these directional parameters were relatively unimportant in explaining the variability of POPs atmospheric partial pressures (9). In a third study, we used the potential source contribution function (PSCF) model to determine the sources of several chemicals (PAH, PCBs, and some pesticides) using backward
air trajectories (10). In general, most pollutants were coming to the Great Lakes from southern sources. This approach to determining the effects of atmospheric temperature, sampling date (time), and population as separate factors has been effective, but it has some disadvantages: It is not possible to directly determine which of these parameters (or others that might be subsequently added) are the most important. For example, for PCBs, all of the regressions shown in Figures 1-3 are highly significant, but it is not clear which of the three parameters is the most predictive of the measured partial pressure. Another disadvantage is that these regressions need to be carried out for each chemical at each sampling site, an approach which risks losing the overall picture. To overcome these disadvantages and to gain statistical strength, we have combined these parameters into one multiple linear regression model to predict the atmospheric partial pressures of a particular contaminant based on the sampling time, atmospheric temperature, local human population, wind speed, and wind direction at the sampling site. In addition, data from six IADN sites were combined together to provide an overview of the factors predicting the partial pressures of selected POPs in the Great Lakes region as a whole, rather than on a site by site basis. This new approach was applied to data from the U.S. IADN sites, and the corresponding regression coefficients and the relative importance of each parameter in explaining the observed partial pressures were determined for several POPs.
Experimental Section Sampling and Analytical Methodology. Air samples were collected at the six United States Integrated Atmospheric Deposition Network (IADN) sites over the following time ranges: Brule River, Wisconsin (1996-2002), Eagle Harbor, Michigan (1990-2007), Sleeping Bear Dunes, Michigan (1992-2007), Chicago, Illinois (1996-2007), Cleveland, Ohio (2003-2007), and Sturgeon Point, New York (1992-2007) (see http://www.msc-smc.ec.gc.ca/iadn/index_e.html for more details on the sites and Table 1). Details of the sample collection, extraction, and analysis procedures can be found elsewhere (11), and only a brief description will be presented here. A modified Anderson highvolume air sampler (General Metal Works, model GS2310) was used to collect air samples for 24 h every 12 days at a flow rate giving a total sample volume of ∼820 m3. The gas phase was collected on Amberlite XAD-2 resin (Supelco, 20-60 mesh) held in a stainless steel cartridge. After sampling, recovery standards were spiked into the sample, and the XAD was Soxhlet extracted for 24 h with 1:1 (v/v) acetone: hexane. The extract was reduced in volume by rotary evaporation, the solvent was exchanged to hexane, and this solution was fractionated on a column containing 3.5% (w/ w) water deactivated silica gel. The column was eluted with 25 mL of hexane (fraction 1) and 25 mL of 1:1 (v/v) hexane: dichloromethane (fraction 2). After N2 blow down, the sample was spiked with the internal standards. PCBs and pesticides VOL. 44, NO. 2, 2010 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
619
TABLE 2. Coefficients of the Multiple Linear Regressions with Their Standard Errorsa
ΣPCB ΣPAH R-HCH γ-HCH ΣDDT ΣEndo ΣChlor
N
const. -d0
time -d1 (× 10-4)
t1/2
temp. -d2
popul. +d3 (× 10-3)
WS -d4 (× 10-3)
cos(WD) -d5 (× 10-3)
1886 1844 1888 1847 1870 1788 1779
14.1 ( 0.4 21.7 ( 0.5 20.5 ( 0.3 12.7 ( 0.5 14.6 ( 0.5 -7.53 ( 0.78 14.7 ( 0.4
1.50 ( 0.09 1.02 ( 0.11 5.68 ( 0.07 4.96 ( 0.10 2.45 ( 0.10 1.65 ( 0.20 1.91 ( 0.09
12.6 ( 0.7 18.6 ( 2.0 3.34 ( 0.04 3.82 ( 0.08 7.76 ( 0.33 11.5 ( 1.4 9.96 ( 0.49
5.31 ( 0.11 2.45 ( 0.14 2.98 ( 0.09 5.77 ( 0.10 5.91 ( 0.14 11.8 ( 0.22 5.81 ( 0.12
74.4 ( 1.2 116 ( 2 -9.28 ( 0.96 18.3 ( 1.3 62.5 ( 1.4 19.4 ( 2.1 61.9 ( 1.2
7.46 ( 1.35 32.6 ( 5.9 7.96 ( 3.85 10.6 ( 5.3 NS 32.7 ( 8.9 NS
67.1 ( 21.5 298 ( 26 NS 46.6 ( 23.7 111 ( 25 125 ( 39 149 ( 21
a See eq 5. N is the number of samples included in the regression. The half life t1/2 (in years) was calculated from the coefficient of the time term (d1).
were analyzed by gas chromatography on Hewlett-Packard 5890 and 6890 instruments equipped with 63Ni electron capture detectors and DB-5 and DB-1701 (J&W Scientific) 60 m columns (250 µm i.d. and 0.1 µm film thickness). Quantitation was done using the internal standard method. Surrogate standards were used to estimate recoveries of each compound in each sample. PAHs were analyzed by gas chromatographic mass spectrometry (GC/MS) on an Agilent 6890 GC with a 5973 mass spectrometer, using a DB-5 ms column (J & W Scientific, 30 m long, 250 µm i.d., and 0.25 µm film thickness) (11). Quality control and quality assurance procedures were followed to ensure data accuracy. The detailed procedures are described in the IADN Quality Assurance Program Plan and in the IADN Quality Control Project Plan (12). In this work, ΣPCB represents a suite of 83 congeners that includes the toxicologically relevant ones [see (13)], and ΣPAH represents the sum of 17 compounds (13). The concentrations of several organochlorine pesticides are reported, either as single compounds (R- and γ-HCH) or as groups (ΣDDT ) p,p′-DDT + p,p′-DDD + p,p′-DDE; ΣChlor ) R- + γ-chlordane + trans-nonachlor; and ΣEndo ) endosulfan I + II + endosulfan sulfate). Data Analysis. The gas-phase concentrations (pg/m3) of each target chemical were converted to partial pressures (atm) using the ideal gas law and the average atmospheric temperatures during the 24 h sampling period measured at each site. The logarithms of the atmospheric partial pressures were then fitted using Minitab 15 and the following linear equation: + d log (pop) + d WS + ( 1000 T )
ln P ) d0 + d1t + d2
2
3
4
d5cos WD
(5)
where d0 is an intercept, d1 is a first-order rate constant (in days-1), d2 is the Clausius-Clapeyron slope (in K), d3 (unitless) describes the change of P as a function of population (a surrogate for human activity), and d4 and d5 describe the dependence of the partial pressure on wind speed (in mph) and wind direction, expressed as the cosine of the angle (in radians). In eq 5, the logarithm of the population was replaced by its squared value (compare with eq 4). The rationale behind this choice was the observation that the correlations described by eq 4 (see above) showed systematic residuals; the concentrations at low and high populations were usually above the regression line (see Figure 3), indicating significant nonlinearity in this relationship. The quadratic term for population was introduced to compensate for this nonlinearity; the curve with the improved r2 value is shown in Figure 3. Half-lives were calculated by dividing the d1 coefficient into ln(2). In eq 5, the cosine of wind direction was selected because this function is negative for wind directions from 90° to 270° or from the southsa direction we suspect to be the source of many of the POPs coming to the Great Lakes. 620
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 44, NO. 2, 2010
The importance of each of the regression parameters in explaining the variability of the partial pressures was estimated by calculating the relative contribution of each term to the total variability. This value is reported as the percentage of the sequential sum of squares attributed to each factor relative to the total sum of squares. The sequential sum of squares for each term adds up to the total model sum of squares and allows us to express the importance of each term as a percent of the overall r2 value. The order in which the different factors were entered in the regression was determined using a stepwise regression procedure. This precaution allowed us to use a sequential sum of squares regression calculation, rather than a partial sum of squares calculation.
Results and Discussion The coefficients and their errors from the multiple linear regressions using eq 5 and data from the IADN sites are reported in Table 2, together with the corresponding halflives (t1/2, in years). The quality of the fits is indicated in Figure 4, which shows the observed vs the predicted ln(P) values for ΣPCB, ΣPAH, ΣDDT, ΣChlor, ΣEndo, and γ-HCH at the IADN sites. As expected, the model is precise and accounts for 70-80% of the total variability. Notice that the residuals are evenly distributed around the 1:1 line. The only anomaly is for ΣPAH, which shows two clusters, one at higher values for Chicago and Cleveland and another at lower values for the other four sites. The normality of the residual errors of the regression for all compounds and groups of compounds was assessed using histogram plots, which did not show any evident deviation from the Gaussian distribution. Plots of the residuals versus the predictor variables [JD, 1000/T, log2(pop), WS, and cos(WD)] did not reveal any unusual pattern, confirming the randomness of the errors, and ultimately, the validity of the model. Temporal Trends. All seven of the chemicals or groups of chemicals showed a significant decrease in their partial pressures over time. The longest half-life was for ΣPAH (∼20 years), and the shortest was for R- and γ-HCH (3-4 years). Given that PAH are produced by combustion of every sort, it was expected that ΣPAH would show the slowest rate of decrease, and this was observed (t1/2 ) 18.6 ( 2.0 years). Although the Toxic Release Inventory shows a general decrease of PAH emissions over the past decade in the Great Lakes region (14), the atmospheric concentrations of PAHs are still relatively high (15), probably due to mobile and residential sources. This regression, where all of the sites were combined together, produced results similar to those reported by Sun et al. on a site-by-site basis; these earlier half-lives averaged 14.4 ( 3.6 years (15), confirming that the atmospheric partial pressures of PAHs are slowly decreasing in the Great Lakes region. Despite being banned in the U.S. since the late 1970s, PCBs are still ubiquitous, and their atmospheric levels are declining slowly, as shown by their relatively long half-life
FIGURE 4. Scatter plots of predicted ln(P) versus observed ln(P) values for ΣPCBs, ΣPAHs, ΣDDTs, ΣChlor, ΣEndo, and γ-HCH at the IADN sites with a 1:1 reference line. of 12.6 ( 0.7 years in the Great Lakes region. In 2007, Sun et al. reported an average PCB half-life of 13.8 ( 4.4 years for the same IADN sites [with the exception of Brule River, which was considered unreliable by the authors (16)]. This is a slow rate of decrease for compounds that have not been produced in North America since the mid-1970s; however, there are still large amounts of PCBs that have not been permanently removed from the environment, such as in electrical gear. In addition, “decommissioned” PCBs have not really been removed from the environment either; rather, they have been placed in landfills and in other disposal facilities that may well be leaking into the atmosphere. Among the OC pesticides, the atmospheric partial pressures of R- and γ-HCH are declining most rapidly (half-lives of 3.34 ( 0.04 and 3.82 ( 0.08 years, respectively). The R isomer constituted 60-70% of the technical HCH product, which was banned in the 1970s and replaced by lindane, the purified γ isomer (17). The latter was phased out in Canada in 2004 and in the U.S. in 2009. The half-life of R-HCH observed here is similar to the average half-life of 3.8 ( 0.3 years reported by Sun et al. (18). It is interesting that, despite its ban ∼40 years ago, this insecticide is still being eliminated rapidly from the environment. Conversely, the half-life of γ-HCH reported here is less than that reported by Sun et al. for data covering up to 2003 (an average half-life of 6.1 ( 2.1 years) (18). The more rapid decline observed in this study, which expanded the data set until 2007 and merged all the sites together, could reflect the ban on lindane’s use that took place in Canada starting in 2004. An even more rapid loss rate might be expected in the next 5-10 years, once the ban on this compound in the United States becomes effective. Although the Environmental Protection Agency is reviewing its use, endosulfan is currently used as an agricultural pesticide. Our study showed a half-life of 11.5 ( 1.4 years (calculated as ΣEndo), which is higher than the previously reported average for IADN sites (5.9 ( 2.6 years) (18). The
National Water-Quality Assessment (NAWQA) Program estimated a decrease in endosulfan’s use in the U.S. from 1997 to 2002 (19), but current use estimates are not available. Therefore, we are reluctant in making any connection between use and the observed temporal trend. ΣDDT and ΣChlor, which are all components of banned products, showed half-lives of 7.76 ( 0.33 and 9.96 ( 0.49 years, respectively; these values are in good agreement with those previously reported by Sun et al. (8.0 ( 2.9 and 9.6 ( 2.0 years) (18). These relatively slow rates of decline, despite bans on these products over 20 years ago, may indicate that there are large reservoirs of these compounds in agricultural and urban soils that are only slowly being depleted. Seasonality. In eq 5, the third term describes the relationship between ambient temperature and the gas-phase partial pressure of these semivolatile organic compounds. As reported in Table 2, all of the compounds showed a statistically significant (P < 0.05) negative dependence on reciprocal temperature (-d2), indicating that gas-phase partial pressures increase with increasing temperatures. Ultimately, the absolute value of the d2 coefficient describes the magnitude of seasonality for a specific compound. As such, it is reasonable that the highest value for d2 was found for ΣEndo (-d2 ) 11.8 ( 0.22), a pesticide that is mainly applied in the spring and summer, when atmospheric temperatures are highest. The value of -d2 was lowest for ΣPAH (2.45 ( 0.14), compounds whose atmospheric partial pressures tend to peak in the winter as combustion for space heating increases. Urbanization. The fourth term in eq 5 describes the dependence of the gas-phase partial pressures on the squared logarithm of the local human population. The coefficients for this population term, d3, ranged from 0.12 ( 0.01 for ΣPAH, indicating a strong relationship to urbanization, to 0.018 ( 0.001 for γ-HCH, indicating almost no relationship with urbanization, probably because this compound is used VOL. 44, NO. 2, 2010 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
621
TABLE 3. Contribution (in percent) of Each Term in the eq Regression, Relative to the Total Variabilitya
ΣPCB ΣPAH R-HCH γ-HCH ΣDDT ΣEndo ΣChlor
5
-t
-1000/T
+log2(pop)
WS
cos(WD)
r2
2.86 0.90 69.2 36.6 8.00 0.97 4.77
22.7 4.52 11.1 32.1 42.7 68.3 46.8
54.7 73.9 1.01 2.75 22.3 1.30 27.6
0.31 0.32 0.04 0.05
