Statistical Approach for Assessing the Stockholm Convention's

Jul 22, 2019 - The loss rates of vapor phase polychlorinated biphenyls became more rapid .... filter to collect the particles and then through a bed o...
0 downloads 0 Views 872KB Size
Download PDF
Article Cite This: Environ. Sci. Technol. XXXX, XXX, XXX−XXX

pubs.acs.org/est

Statistical Approach for Assessing the Stockholm Convention’s Effectiveness: Great Lakes Atmospheric Data Ronald A. Hites* O’Neill School of Public and Environmental Affairs, Indiana University, Bloomington, Indiana 47405, United States

Downloaded via UNIV OF SOUTHERN INDIANA on July 23, 2019 at 00:40:13 (UTC). See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

S Supporting Information *

ABSTRACT: The implementation of the Stockholm Convention (SC) in 2004 should become evident in decreases in environmental concentrations of various pollutants even in countries that not have ratified the SC. However, in some cases, there may be no decreases at all. This paper develops a statistical strategy for investigating time-series measurements such that the rate of change of a pollutant’s concentrations at any time can be compared to those at an earlier or later time and thus determine the effectiveness of the SC at any location. The general approach is to modify a first-order regression to include a second order time term: ln(Ct)= a0 + a1 t + a2 t2, where Ct is the concentration at time t. Thus, the rate constant at any time is k(t) = a1 + 2 a2 t. Given that the errors associated with a1 and a2 can be calculated, one can compare the rate constants at different times with statistical rigor to determine if the rates at which the concentrations are changing are significantly different. As examples of this approach, this paper uses vapor and particle phase atmospheric concentrations of several organic pollutants measured at six sites around the North American Great Lakes every 12 days since about 1992. After correcting for the population near the sampling sites, for seasonality, and for the different numbers of samples collected on the same date, up to 830 data were used in this second-order regression. In general, the loss rates of vapor phase chlorinated pesticides have slowed by about a factor of 2 between 1995 and 2015, which is not SC-like behavior. The exceptions are the endosulfans, the vapor and particle phase concentrations of which were both doubling in 1995 but were both decreasing in 2015, probably because of the greatly diminished use of this insecticide in the United States over the last 20−25 years. The loss rates of vapor phase polychlorinated biphenyls became more rapid between 1995 and 2015, which is SC-like behavior.

■

INTRODUCTION The Stockholm Convention (SC) is a United Nations treaty.1 The SC focuses on toxic, persistent organic pollutants (socalled POPs), and its goal is to eliminate the production and use of specified compounds everywhere in the world. It entered into force on 17 May 2004, and it was signed by virtually every country around the globe, with the notable exceptions of the United States, Italy, and Malaysia. The SC initially listed 12 compounds (the “dirty dozen”). These were six compounds produced from hexachlorocyclopentadiene (aldrin, chlordane, dieldrin, endrin, heptachlor, and mirex); toxaphene; polychlorinated biphenyls, dibenzo-p-dioxins and dibenzofurans; hexachlorobenzene; and DDT, although the use of DDT was still allowed in some countries to kill mosquitoes transmitting malaria. Provisions were made in the SC to add compounds to the list of restricted chemicals as new information became available. Thus, since 2009, three conformers of hexachlorocyclohexane (one of which is known as lindane), kepone, pentachlorobenzene, perfluorooctanesulfonic acid and its salts, several brominated flame retardants (such as 2,2′,4,′4tetrabromodiphenyl ether and hexabromocyclododecane), endosulfan, hexachlorobutadiene, pentachlorophenol, polychlorinated naphthalenes, and short-chain chlorinated paraffins have been added to the list of SC chemicals of concern.1 © XXXX American Chemical Society

Given this history, the question is how can the scientific community determine if the restrictions called out in the SC are working or not? In principle, one could measure the concentrations of a restricted compound in the environment or in people as a function of time at a given site and look for a difference in the rate at which these concentrations are changing. If the SC is working, the rate of decrease will become faster after the SC takes effect at the locations where the samples have been taken. An excellent example of this effect are the concentrations of BDE-47 in human milk from Sweden;2 see Figure 1.3 In this case, the concentrations of BDE-47 had been increasing exponentially (with a doubling time of 4.7 ± 0.2 years), before 1996, but after that time, these concentrations decreased exponentially with a halving time of 11 ± 2 years. This sort of statistical investigation is called a breakpoint analysis and is relatively simple to do in Excel. In the case shown in Figure 1, the breakpoint of 1996 for BDE-47 is about when polybromodiphenyl ethers (PBDEs) came to the attention of the Scandinavian public4 and restrictions on their use were put in place. Received: April 11, 2019 Revised: July 2, 2019 Accepted: July 10, 2019

A

DOI: 10.1021/acs.est.9b02190 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Environmental Science & Technology

■

Article

METHODS

The samples were collected at six sites on the shores of the North American Great Lakes; see Figure S1. At each of the sites, an atmospheric sample was collected using a modified high-volume sampler for 24 h once every 12 days (except at Point Petre, where the sampling frequency was once every 24 or 36 days). The air is sampled at a flow rate such that ∼820 m3 is sampled over the 24 h period. The air is first pumped through a 2.2 μm filter to collect the particles and then through a bed of XAD-2 resin to collect the vapor phase components. Once returned to the laboratory, the particle and vapor phase media are extracted separately, and the extracts are cleaned up and analyzed separately. The concentrations of a suite of about 200 POPs were measured. The full list of analytes is given in the IADN Quality Assurance Project Plan.7 The details of the measurement methods have been reported before;8 thus, only a summary is given here. The polychlorinated biphenyls (PCBs) and organochlorine insecticides are measured with electron capture gas chromatography with a 60 m column. The polycyclic aromatic hydrocarbons (PAHs) are measured by isotope dilution gas chromatographic mass spectrometry in the electron impact ionization mode. All analyses are based on internal calibration compounds. Extensive QA/QC procedures have been implemented.9 A summary of sampling years and data availability is given in Table S1 of the Supporting Information. These dates vary because new compounds (for example, endosulfan) and new sampling sites (for example, Cleveland) were added to the project after its initiation. Data before 1 January 2013 have appeared previously,8 but data for 2013−2016 (inclusive) are included here for the first time. Concentrations of hexachlorocyclohexanes (HCHs) and PCBs in the particle phase are not reported at any site because these levels were too low to be measured reliably. Concentrations of the chlordanes, DDTs, and endosulfans were only measured in the particle phase at the Chicago and Cleveland sites because the levels at the other four sites were too low to measure dependably. In general, there are about 3300 data for each compound in the vapor phase, about 2700 data for PAHs in the particle phase, and about 500 data for pesticides in the particle phase. To make this exposition tractable and to be able to include the broadest set of POPs, we have aggregated some of the data as follows: “Total PCBs” are the sum of the concentrations of most of the PCB congeners in the commercial Aroclor mixtures.8 In addition, three individual PCB congeners (18, 52, and 101), which were major components of the commercial Aroclor mixtures, are discussed individually to compare congener-specific results to those for total PCBs. “Total PAHs” are the sum of the concentrations of 22 polycyclic aromatic hydrocarbons. Phenanthrene was included separately as a representative of an abundant vapor and particle phase PAH. “Total DDTs” are the sum of the concentrations of the p,p′ isomers of DDT, DDE, and DDD. “Total chlordanes” are the sum of the concentrations of α- and γ-chlordane and transnonachlor. “Total endosulfans” are the sum of the concentrations of endosulfan-I and -II and endosulfan sulfate. The two hexachlorocyclohexane conformers (α- and γ-HCH) are reported separately.

Figure 1. Concentrations of 2,2′,4′4′-tetrabromodiphenyl ether (BDE-47) in Swedish human milk as a function of sampling year. The data have been replotted from Gyalpo et al.,2 and the breakpoint determination is from Hites.3 The doubling time before 1996 is 4.7 years, and the halving time after 1996 is 11 years. The coefficient of determination and the Student’s t-value are highly significant.

A similar sort of breakpoint analysis of BDE-47 for samples collected in another country may not show a breakpoint until the use of PBDEs was restricted, presumably as a result of the SC being implemented in that specific country. Thus, a breakpoint date will depend on the regulatory policies of a given country or region. In the absence of global temporal trend data, the best one can do is to determine if the SC has had a measurable effect on the concentrations of a given pollutant at a given site. There is another problem with breakpoint analyses: The concentration at the breakpoint itself is a singularity; that is, the slope of the concentration as a function of time at the breakpoint is undefined. In fact, as shown in Figure 1, the presence of a breakpoint suggests that the change in the concentrations at this date (1996) was abrupt, which is not usually realistic. This paper addresses these issues using atmospheric concentrations of several POPs measured over the last 20− 25 years at sites around the North American Great Lakes. The statistical approach used here will demonstrate that such measurements of atmospheric concentrations, if made on a long-term basis, can be used to determine if the SC (or in fact, any usage restriction) has been effective. The data used here are from the Integrated Atmospheric Deposition Network (IADN), which has been measuring the atmospheric concentrations of about 200 POPs at six sites on the shores of the North American Great Lakes since about 1992.5 To date, this project has accumulated about one and a half million measurements, which are available to the public on the Internet.6 Although this paper focuses only on atmospheric measurements around the Great Lakes, the statistical approach presented here can be used for any time series of environmental measurements providing there are enough data to support the regression analyses. The approach discussed here does not presuppose any environmental processes; for example, it could be applied to a time series of biological measurements without making any assumptions about the mechanism of bioaccumulation or its changes over time. This paper addresses the general questions: Do the measured concentrations follow the regression equation discussed here? If so, do the rates of change vary significantly over time? Lastly, do the rates of change increase with the passage of time as expected if the SC is effective?

■

RESULTS AND DISCUSSION Converting a time series of concentration measurements into rates of change as a function of time is, in principle, simple. For B

DOI: 10.1021/acs.est.9b02190 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology any given compound at any given site, one can fit the data using a first-order rate equation ln(Ct ) = a0 + a1t

significance, and comparing rate constants in 1995 to those in 2015 for 11 compounds or compound groups measured at the six IADN sites. These two years were selected to cover a 20 year period spanning the implementation of the SC but to be within the temporal range of the measurements reported here. The complete regression equation is

(1) 3

where Ct is the concentration (in pg/m ) of the compound on day t, a0 is an intercept that rectifies the units, and a1 is a rate constant (in days−1); this is first-order kinetics. In this and in subsequent equations, the ai values (and their standard errors) are all the results of the multiple linear regression analyses. In our earlier papers on these data, we used this approach (eq 1), and in 1997, we discovered that there were large seasonal effects:10 Most vapor phase compound concentrations were much higher in the summer than in the winter, at least around the Great Lakes. Thus, we soon included harmonic terms in the regression to account for these variations ln(Ct ) = a0 + a1t + a 2 sin(zt ) + a3cos(zt )

ln(Ct ) = a0 + a1t + a 2 sin(zt ) + a3cos(zt ) + a4 log 2(pop) + a5t 2 (4)

where the new second-order term is the last one on the right. One can imagine using some other term to give the fit the required curvature; for example, a sinusoidal curve with a very long periodicity might have been a possible choice. But in eq 4, the time-squared term was selected because (a) it fit the data well, (b) it made the subsequent rate constant derivations simple, (c) it made the multiple regression analyses robust, and (d) it did not require any decisions about the date of a breakpoint. Like all such analyses, a fitted curve of this sort should not be used to extrapolate the regression much outside of the range of the measured data. In such cases, a breakpoint analysis may be more suitable. It is also important to point out that eq 4 is not a mechanistic model−it is just statistics. To be able to visualize just the temporal effects and to eliminate the seasonal and population effects, it is convenient to subtract the harmonic and population terms from each measured ln(Ct) value. Thus, let

(2)

where z is 2π/365.25, which converts the date to radians and sets the periodicity to one year; a2 and a3 are unitless and can be used to give the amplitude and maximum date of the seasonal effect.11 IADN’s six sampling locations range from the cities of Chicago and Cleveland to a remote site, Eagle Harbor, on the tip of the Keweenaw Peninsula near Lake Superior with populations ranging from almost 4 million people in Chicago to about 1300 people at the Keweenaw site. Clearly, these large population differences influence the atmospheric concentrations of many compounds;12 for example, there are many more sources of PCBs in Chicago then at the Keweenaw site. To accommodate these variations, we added another term to the regression

ln(Ct )* = ln(Ct ) − a 2 sin(zt ) − a3cos(zt ) − a4 log 2(pop) (5)

The a2, a3, and a4 terms in eq 5 are determined from a regression of all of the ln(Ct) data using eq 4 with a1 and a5 set to zero. The idea is to strip out the seasonal and population variabilities so that one can examine the long-term temporal effects directly. Thus, eq 5 gives

ln(Ct ) = a0 + a1t + a 2 sin(zt ) + a3cos(zt ) + a4 log 2(pop) (3)

where pop is the number of people living and working within a 25 km radius around each sampling site.13 The sign and magnitude of a4 indicates the relative importance of human habitation of the atmospheric concentrations. For example, a small or negative value of a4 suggests that the compound in question does not originate from population centers, suggesting an agricultural source. The squared common logarithm of this term gives some curvature to the function such that the difference between sites with low populations and low concentrations would be smaller than the difference between sites with high populations and high concentrations. With this term, we were able to regress all six sites together and, thus, increase the statistical power of this analysis. Using this approach, we have demonstrated that the atmospheric concentrations of, for example, PCBs around the Great Lakes are decreasing with halving times of 12−19 years and those of lindane with halving times of about 4 years.8 We have also been able to quantitate the rural vs urban effect using the coefficient of the population term (a4); for example, DDT has a surprisingly strong urban source, but γ-HCH (lindane) does not.12,13 These previous results were all based on the linearity of the time term. In other words, the rate of change was assumed to be constant over the entire 20−25 year period of the measurements. This is not necessarily true. For example, one can imagine that the SC has had an effect for some compound(s) at a given site, and thus, the concentrations there would decrease more rapidly after this time than before. This paper addresses this issue by adding a secondorder time term to the regression equation, testing its statistical

ln(Ct )* = a0 + a1t + a5t 2

(6)

In this case, the a0, a1, and a5 terms are determined, along with their standard errors, by a regression of this equation. It is important to do a propagation of errors calculation using these fitted parameters, but to do this calculation, one must know the collinearity between a1 and a5, which is sometimes difficult to obtain. The straightforward way around this problem is to rescale all the time values by subtracting the mean time (t ̅ or tavg) from each time value.14 This sets the collinearity to zero, and thus, it can be ignored. Thus, eq 6 becomes ln(Ct )* = a0 + a1(t − t ̅ ) + a5(t − t ̅ )2

(7)

−1

The rate constant (in days ) at any time t is the derivative of this rate law k(t ) =

d ln(Ct )* = a1 + 2a5(t − t ̅ ) dt

(8)

If a5 is zero, k = a1. The standard error of k(t) is given by the propagation of errors σk(t ) = [σa12 + 4(t − t ̅ )2 σa5 2]1/2

(9)

An example should make these calculations clear. Figure 2 shows all the data for α-HCH in the vapor phase at all six sites. There are a total of 3295 data points in the six graphs. The seasonal periodicity is clear, especially at Sturgeon Point and at Eagle Harbor. The concentrations maximize in the summer, as expected. The range of the ln(C) values is the same in each C

DOI: 10.1021/acs.est.9b02190 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

Figure 3. (Top) Natural logarithms of the concentrations (in pg/m3) of α-HCH in the vapor phase after corrections for the population, sine, and cosine terms (see eq 5). In this case, tavg = 11 Sep 2005 and N = 3295. (Bottom) Natural logarithms of the concentrations in the top panel after those measured on the same day had been averaged. In this case, tavg = 13 Jun 2004. The red line is the second order regression (see eq 6); the regression parameters are given in Table S2. In the bottom panel, the green lines and text indicate the rate constants in 1995 and 2015 and their standard errors; these rate constants have units of 10−4 day−1.

by different average dates; in this case, one average is for 3295 data, and one is for 818 data. Both average dates are given in Table S2. All 818 ln(C)*date values were then regressed with t and t2 as the independent variables using eq 6, and these results were a0 = 3.267 ± 0.020, a1 = −(4.41 ± 0.05) × 10−4, and a5 = (2.45 ± 0.21) × 10−8. These coefficients were used in eq 8 to calculate the rate constants in 1995 and 2015, and these results (in days−1) were k1995 = −(6.10 ± 0.16) × 10−4 and k2015 = −(2.53 ± 0.16) × 10−4. These values are shown in Figure 3, bottom, along with the fitted curve. These values are significantly different (P < 0.001) and clearly indicate that the rate at which α-HCH is clearing from the atmosphere, at least around the Great Lakes, has slowed by about a factor of 2 between 1995 and 2015. These calculations were repeated for each of the compounds and compound groups outlined above and separately for both the vapor and particle phases (where available). The complete regression results are given in Table S2. Comparing Earlier and Later Rates of Change. One can compare the rates of change by simply subtracting k2015 from k1995

Figure 2. Natural logarithms of the concentrations (in pg/m3) of αHCH at the six Integrated Atmospheric Deposition Network sites (see Figure S1 for a map) as a function of sampling date over the period 1992−2016 (inclusive).

graph, indicating that there is relatively little effect of the local population on these α-HCH concentrations and that this compound probably has more agricultural sources than urban sources. All 3295 of these α-HCH data were regressed using eq 4 with the a1 and a5 terms set to zero. The results were a0 = 3.217 ± 0.049, a2 = −0.2567 ± 0.0291, a3 = −0.3573 ± 0.0294, and a4 = −0.0209 ± 0.0018. The negative a4 value confirms that α-HCH does not have many urban sources. These regression coefficients were then used in eq 5 to calculate the 3295 values of ln(Ct)*, and these values as a function of (t − t)̅ are shown in Figure 3, top. There is an obvious problem. As shown in Figure 2 and Table S1, sampling started at the various sites at different times; Eagle Harbor started in 1992, but sampling at Cleveland did not start until 2003. These different start times mean that in the early years, there were only 3 measurements on a given date, but in the later years, there were 6 measurements on a given date. Clearly this difference would incorrectly weigh the regressions toward the later years. Thus, the ln(C)* values were sorted by date regardless of the sampling site, and those values falling on the same date were averaged. This gives a new parameter, which we define here as ln(C)*date. Of course, removing this duplication reduces the data set size; in this case, the number of data points is reduced from 3295 to 818, which is still more than enough for a robust statistical analysis. This date corrected data for α-HCH in the vapor phase are shown in Figure 3, bottom. Note that there is an apparent offset between the data in Figure 3, top, and Figure 3, bottom, which is caused

Δk = k1995 − k 2015

(10)

For scale, it is convenient to change the units of the rate constants to year−1. Given that the errors of the two rate constants are known, the standard error of Δk is σΔk = (σk1995 2 + σk 2015 2)1/2

(11)

The value of Δk for α-HCH as shown in Figure 3 is −0.131 ± 0.009 year−1. In other words, k2015 > k1995. Given that Δk < 0, it is clear that the loss rate for α-HCH from the atmosphere around the Great Lakes has slowed over the last 20 years. In general, if Δk ≤ 0, there has been no improvement in the rates of decrease at the two times, and these data do not show, what D

DOI: 10.1021/acs.est.9b02190 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology can be called, a “SC-like” effect. If Δk > 0, then there has been an improvement in the rate of decrease between 1995 and 2015, and this can be called SC-like behavior. The values of Δk are given in Figure 4 for all of the compounds in the vapor and particle phases. The positive-

significantly between 1995 and 2015; this rate is constant at −0.161 ± 0.025 year−1. The halving time is defined as the time it takes for a concentration to decrease by a factor of 2; it is given by t1/2 = −ln(2)/a1. For lindane, the halving time is 4.3 ± 0.7 years. Perhaps this rate is constant because the production and use of this compound was allowed up until 2009.15 The slowing of the loss rates for the chlordanes, the DDTs, and α-HCH over the last 20 years may be evidence of the “low hanging fruit” effect, in which major sources of these pollutants were controlled early and quickly, leaving minor sources that are still active and yet to be fully controlled. For example, the lakes themselves are now outgassing some of these compounds.16 In both the vapor and particle phases, the endosulfans are the exception. As shown in Figure 4, in the vapor phase, Δk = 0.265 ± 0.027 year−1; similarly, in the particle phase, Δk = 0.366 ± 0.028 year−1. These values are positive and dwarf most of the other Δk values, positive or negative, and clearly indicate SC-like changes. The complete ln(C)*date data for the endosulfans are shown in Figure 5, which clearly shows the

Figure 4. Values of Δk (in years−1) for the 11 compounds or compound groups in the Great Lakes atmosphere; the error bars are standard errors. Positive going bars indicate that the concentrations of the compounds are decreasing more rapidly in 2015 than they did in 1995 (SC-like behavior). Negative going bars indicate that the concentrations of the compounds are decreasing less rapidly in 2015 than they did in 1995 (non-SC-like behavior). Vapor phase and particle phase results are presented separately.

going bars, notably the endosulfans and the PCBs, indicate SClike behavior. The negative-going bars, notably the other pesticides and the PAHs, indicate non-SC-like behavior. Note that (with one exception) none of the error bars intersect zero, indicating that these conclusions are statistically significant. The one exception is γ-HCH (lindane), where Δk is not statistically significant; this indicates that the concentrations of this compound have been decreasing at a constant rate since 1995. Pesticides. In the vapor phase, with one exception, the concentrations of the pesticides were decreasing in both 1995 and 2015; see Figure 4. The average Δk value was −0.035 ± 0.016 year−1 for the chlordanes and the DDTs and −0.131 ± 0.009 year−1 for α-HCH. In the particle phase, Figure 4 shows Δk = −0.230 ± 0.025 year−1 and −0.281 ± 0.038 year−1 for the chlordanes and the DDTs, respectively. (The HCHs were not measured in the particle phase.) The negative sign of all these Δk values indicates that the rates at which the concentrations of these pesticides were decreasing have significantly slowed between 1995 and 2015. It is also clear that the magnitude of Δk for these compounds is about 7 times greater for the particle phase than for the vapor phase. This indicates that particle phase emission control technologies (for example, electrostatic precipitators) may have been especially effective at removing these compounds. In summary, in both the vapor and the particle phases, the rates at which the chlordanes, DDTs, and α-HCH were leaving the atmosphere decreased substantially between 1995 and 2015. Thus, these three compounds show decidedly non-SClike behavior. As previously mentioned, the rate at which γHCH (lindane) is leaving the atmosphere has not changed

Figure 5. (Top) Natural logarithms of the concentrations (in pg/m3) of the total endosulfans in the vapor phase after corrections for the population, sine, and cosine terms (see eq 5) and after ln(C)* data on the same day have been averaged. In this case, tavg = 2 Nov 2005 and N = 723. The red line is the second order regression (see eq 6); the regression results are given in Table S2. (Bottom) The same as the top panel except the endosulfan concentrations are in the particle phase. In this case, tavg = 17 Nov 2006 and N = 575. The red line is the second order regression (see eq 6); the regression results are given in Table S2. In both panels, the green lines and text indicate the doubling (t2) or halving (t1/2) times in 1995 and in 2015.

change from an increasing rate in 1995 to a rapid loss rate in 2015. In both phases, endosulfans’ atmospheric concentrations at these six sites were increasing with doubling times (given by t2 = ln(2)/a1) of 7−15 years in 1995, but by 2015, they were decreasing with halving times of about 3 years. What happened to endosulfan in the last 20 years? The answer is regulation. In the United States, the use of endosulfan decreased from about 1.0 × 106 kg/year in 1995 to virtually zero in 2015.17 In 1995, the uses of endosulfan centered on cotton grown near the Mississippi River as far north as Memphis; in 2015, the few remaining uses of endosulfan were distributed along the northern Michigan shores of Lakes Michigan and Huron, in Texas, and in the central valley of California.17 These changes E

DOI: 10.1021/acs.est.9b02190 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Environmental Science & Technology

■

ACKNOWLEDGMENTS The author thanks Amina Salamova and Marta Venier for helpful comments on the manuscript, Alexander Alexeev for statistical advice, Indiana University’s IADN team for the data, and the U.S. Environmental Protection Agency’s Great Lakes National Protection Office (Derek Ager, project manager) for funding (Agreement No. GL00E01422).

were largely driven by restrictions on endosulfan’s use put in place by the U.S. EPA in 2007.18 Given that the United States is not a signatory of the Stockholm Convention, it should be clear that the restrictions on the use of endosulfan in 2007 were not a direct result of the SC. Nevertheless, this is an example of a strong SC-like temporal trend that one would eventually hope to see for all POPs throughout the globe. PCBs. As shown in Figure 4, the average Δk value for PCBs in the vapor phase was 0.053 ± 0.011 year−1. (Remember that PCBs were not measured in the particle phase because of their low abundances.) These positive Δk values are the opposite those observed for the pesticides, the concentrations of which were decreasing more slowly in 2015 compared to 1995 (except for the endosulfans). The fact that PCBs are decreasing more rapidly in 2015 compared to 1995 (positive Δk values) is another example of the SC-like behavior one would like to observe worldwide. PCBs have been banned in the United States since 1977,19 so the nation is probably well past the “low hanging fruit” effect, but PCB cleanup activities have certainly improved over the last 20 years, which may account for the more rapid recent decrease rates. PAH. For total PAH, Figure 4 shows that Δk = −0.038 ± 0.013 year−1 and −0.132 ± 0.023 year−1 for the vapor and particle phases, respectively. For phenanthrene, Figure 4 shows that Δk = −0.055 ± 0.014 year−1 and −0.118 ± 0.019 year−1 for the vapor and particle phases, respectively. This result is certainly not SC-like behavior, and it suggests that emission controls that were doing a decent job at reducing source strengths of phenanthrene and other PAH in 1995 have lost their effectiveness over the last 20 years. Relative Rates in the Vapor vs Particle Phases. It is clear from Figure 4 that the absolute values of the Δk values of the chlordanes, DDTs, endosulfans, total PAH, and phenanthrene are significantly higher in the particle phase compared to those in the vapor phase. A Student’s t-test comparing these times gives the following results: chlordanes, P < 0.001; DDTs, P < 0.001; endosulfans, P = 0.016, phenanthrene, P = 0.011; and total PAH, P = 0.001. These are significant t-values. Thus, it seems that emission controls in the north American Great Lakes region over the last 20 years have been more effective at removing particle phase pollutants than vapor phase pollutants. In other words, as the United States and Canada have controlled the emissions of atmospheric particles from combustion and other sources, a side benefit has been the control of at least some POPs in the particle phase.

■

■

REFERENCES

(1) United Nations Environment, https://en.wikipedia.org/wiki/ Stockholm_Convention_on_Persistent_Organic_Pollutants, accessed 1 July 2019. (2) Gyalpo, T.; Scheringer, M.; Hungerbuhler, K. Recommendations for evaluating temporal trends of persistent organic pollutants in breast milk. Environ. Health Perspect. 2016, 124, 881−885. (3) Hites, R. A. Break point analysis of human or environmental trends of POPs. Sci. Total Environ. 2019, 664, 518−521. (4) Hites, R. A. Polybrominated diphenyl ethers in the environment and in people: A meta-analysis of concentrations. Environ. Sci. Technol. 2004, 38, 945−956. (5) Buehler, S. S.; Hites, R. A. The Great Lakes’ Integrated Atmospheric Deposition Network. Environ. Sci. Technol. 2002, 36, 354A−359A. (6) Venier, M.; Lehman, D. C.; Salamova, A.; Hites, R. A. The IADN data visualization tool. Sci. Total Environ. 2018, 645, 1617− 1619. (7) Team IADN, Indiana University, https://www.epa.gov/sites/ production/files/2019-02/documents/iadn-qapp-201805-164pp.pdf, Accessed 2 July 2019. (8) Salamova, A.; Venier, M.; Hites, R. A. Revised temporal trends of persistent organic pollutant concentrations in air around the Great Lakes. Environ. Sci. Technol. Lett. 2015, 2, 20−25. (9) Wu, R.; Backus, S.; Basu, I.; Blanchard, P.; Brice, K.; DryfhoutClark, H.; Fowlie, P.; Hulting, M.; Hites, R. Findings from quality assurance activities in the Integrated Atmospheric Deposition. J. Environ. Monit. 2009, 11, 277−296. (10) Hillery, B. R.; Basu, I.; Sweet, C. W.; Hites, R. A. Temporal and spatial trends in a long-term study of gas-phase PCB concentrations near the Great Lakes. Environ. Sci. Technol. 1997, 31, 1811−1816. (11) Venier, M.; Hites, R. A. Time trend analysis of atmospheric POPs concentrations in the Great Lakes region since 1990. Environ. Sci. Technol. 2010, 44, 8050−8055. (12) Venier, M.; Salamova, A.; Hites, R. A. How to distinguish urban vs. agricultural sources of persistent organic pollutants, Current Opinion in Environmental Science & Health, in press. (13) Venier, M.; Hites, R. A. Regression model of partial pressures of PCBs, PAHs, and organochlorine pesticides in the Great Lakes atmosphere. Environ. Sci. Technol. 2010, 44, 618−623. (14) de Levie, R. Collinearity in least-squares analysis. J. Chem. Educ. 2012, 89, 68−78. (15) Wikipedia, Lindane, https://en.wikipedia.org/wiki/Lindane. Accessed: 10 January 2019. (16) Guo, J.; Salamova, A.; Venier, M.; Dryfhout-Clark, H.; Alexandrou, N.; Backus, S.; Bradley, L.; Hung, H.; Hites, R. A. Atmospheric flows of semi-volatile organic pollutants to the Great Lakes estimated by the United States’ Integrated Atmospheric Deposition and Canada’s Great Lakes Basin Monitoring and Surveillance Networks. J. Great Lakes Res. 2018, 44, 665−677. (17) United States Geological Survey, National Water Quality Assessment Project, https://water.usgs.gov/nawqa/pnsp/usage/ maps/show_map.php?year=1995&map=ENDO··· Accessed: 27 December 2018. (18) Wikipedia, Endosulfan, https://en.wikipedia.org/wiki/ Endosulfan. Accessed: 10 January 2019. (19) Hites, R. A.; Raff, J. D. Elements of Environmental Chemistry, 2nd ed.; Wiley: Hoboken, NJ, p 263.

ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.est.9b02190. Map showing the locations of the six sampling sites around the North American Great Lakes; dates of data availability by compound, site, and phase; detailed regression results from eqs 4 and 6 (PDF)

■

Article

AUTHOR INFORMATION

Corresponding Author

* E-mail: [email protected]. Phone: 812-855-0193. ORCID

Ronald A. Hites: 0000-0003-0975-5058 Notes

The author declares no competing financial interest. F

DOI: 10.1021/acs.est.9b02190 Environ. Sci. Technol. XXXX, XXX, XXX−XXX