Precision of Atmospheric Persistent Organic Pollutant Concentration

Dec 5, 2016 - atmospheric concentration measurements of persistent organic pollutants (POPs) near the North American Great Lakes measure- ments that ...
0 downloads 0 Views 2MB Size
Download PDF
Article pubs.acs.org/est

Precision of Atmospheric Persistent Organic Pollutant Concentration Measurements Daniel C. Lehman, James C. Bays, and Ronald A. Hites* School of Public and Environmental Affairs, Indiana University, Bloomington, Indiana 47405, United States S Supporting Information *

ABSTRACT: Environmental measurement programs are often undertaken with the assumption that measurements at a given location will be comparable to others that would be observed at the same time in the immediate vicinity, but this assumption has seldom been tested. This paper does so. We discuss here the precision of atmospheric concentration measurements of persistent organic pollutants (POPs) near the North American Great Lakesmeasurements that we have been conducting since 1994. We report the relative percent differences between the measured values for 100−200 duplicate samples, and through our use of surrogate (recovery) standards, we have separated the analytical error from the sampling error for the target compounds. The error contributions we calculated were on the order of 5% for the analytical error and 20% for the sampling error, suggesting that the latter is the greatest hindrance to increased precision. In a comparison of relative percent differences for measurements among different atmospheric phases, we observed the highest errors for precipitation samples, with an average median of 35 ± 3, which is more than for vapor-phase samples (27 ± 3) or particle-phase samples (27 ± 2). We suggest that sampling errors are principally the result of inaccuracies in measuring the sample volume and possibly the result of spatial heterogeneity of the atmosphere.

■

INTRODUCTION The measurement of persistent organic pollutant (POP) concentrations in the environment is accompanied by some error. Determining the accuracy of these measurements is difficult and is best accomplished by interlaboratory studies of the same samples, particularly samples for which there is a known (or at least agreed upon) answer. These samples usually come from national standardization laboratories, such as the United States’ National Institute of Standards and Technology, but they can also be generated by so-called “round-robin” studies. With the exception of these sorts of experiments, it is usually difficult to really know the accuracy of a given measurement. Precision is another matter. The precision of environmental measurements can usually be determined by the repeated measurement of the same sample or of samples taken at the same time and place. It is common to report the mean (or better the geometric mean1) of several replicate measurements plus or minus the standard deviation (or the standard error) of that mean. If these replicates are samples taken at the same time and place, the reported precision has two components: first, the variability resulting from trying to take two or more samples that are actually the same (“sampling error”), and second, the precision or variability of the analytical method used to generate the numbers (“analytical error”). The resulting overall precision of the measurements is then a combination of these two sources of variability. © XXXX American Chemical Society

In our laboratory, we have been making measurements of the atmospheric concentrations of 100−200 POPs over the last 20 years. These samples are all collected on the shores of the North American Great Lakes as part of the U.S. Environmental Protection Agency’s Integrated Atmospheric Deposition Network (IADN). We sample both the vapor and particle phases in the atmosphere and precipitation at six sites. We have repeatedly reported these concentrations as a function of both sampling date and location,2,3 but we have not directly addressed the precision of these measurements, nor have we separated the precision into the two sources of error (sampling vs. analytical). In fact, it is possible to address these issues because we have collected duplicate samples at four of these sites and because we have added surrogate standards to the extracts just before analysis. The duplicate samples will give us information on the precision of the sampling plus analytical errors, and the surrogate standards will give us information on the precision of the analytical measurements alone.

■

EXPERIMENTAL SECTION Sampling. The details of the sample collection and chemical analyses have been published previously;4−6 thus, Received: August 31, 2016 Revised: November 10, 2016 Accepted: November 16, 2016

A

DOI: 10.1021/acs.est.6b04428 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

Figure 1. Available duplicate measurements by year (yellow bars) for four IADN sites, for four compound groups, and for three phases. The yellow bars represent usable duplicate measurements in that year, for that phase, for that site, and for that compound group.

sites can be found on the IADN Web site.7 A summary of sampling years and data availability is given in Figure 1. The duplicate air samples were collected for 24 h two or three times per year. They were collected by high-volume samplers located within 3 m of one another at a flow rate such that 815 m3 is

only a summary is given here. The duplicate samples were collected at Chicago, Illinois (41.8343° N, 87.6238° E), Sturgeon Point, New York (42.6928° N, 79.0389° E), Sleeping Bear Dunes, Michigan, (44.7611° N, 86.0586° E), and Eagle Harbor, Michigan (47.4595° N, 88.1491° E). Details of these B

DOI: 10.1021/acs.est.6b04428 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology sampled over a 24 h period. The flow rate was set at the beginning of the sampling period and checked at the end. The air is first pumped through a 0.2 μm filter to collect the particles and then through a bed of XAD-2 resin to collect the vaporphase components. Precipitation samples were integrated over an entire month, and one or two duplicate samples per year were collected. Duplicate precipitation samples were collected using automated wet-only samplers located within 3 m of one another at each of the four sites. Each sampler consists of a shallow funnel connected to a glass column packed with XAD-2 resin. The sampler is normally covered, but it is opened automatically during a precipitation event, which is sensed by a conductivity grid. The precipitation flows by gravity from the funnel through the XAD-2 resin column and into a large carboy, which is used to measure the total precipitation volume associated with a given sample. At the end of each sampling period, the sampler is rinsed with 300−500 mL of reagent water, which also flows through the XAD-2 resin column. Analyses. Once returned to the laboratory, the particle- and vapor-phase and precipitation samples were extracted separately. The polychlorinated biphenyls (PCBs) and organochlorine pesticides were measured by electron capture gas chromatography with a 60 m column. The polycyclic aromatic hydrocarbons (PAHs) were measured by isotope-dilution gas chromatographic mass spectrometry (GC−MS) in electron impact ionization mode. The polybrominated diphenyl ether (PBDE) flame retardants were measured by isotope-dilution GC−MS in the electron capture negative ionization mode. All analyses are based on internal calibration compounds. Extensive QA/QC procedures have been implemented.8 None of the concentrations were blank- or recovery-corrected. For the purposes of this paper, it is important to point out that all of the samples were spiked with known amounts of surrogate standards just before the sample extraction began. These standards allowed us to verify the effectiveness of the analytical method, and these results have been reported in the past as percent recoveries. The following surrogate standards have been used throughout the project: For the PCBs, we have used PCB-14, PCB-65, and PCB-166, none of which are present in the commercial Aroclor mixtures. For the pesticides, we have used δ- and ε-hexachlorocyclohexane (δ- and ε-HCH) and dibutyl chlorendate (DBC). The latter is synthesized from the Diels−Alder condensation of hexachlorocyclopentadiene and maleic anhydride followed by esterification with n-butyl alcohol; it is structurally similar to chlordane and endosulfan. For the PAHs, we have used phenanthrene-d10 and pyrene-d10. For the PBDEs, we have used BDE-77 and BDE-166, neither of which is present in the commercial flame retardant mixtures, and 13C12-BDE-209. Because these surrogates were added to the Soxhlet extractor just before the extraction of the sample started, the precision of their measurements gives us a way to evaluate the analytical error. The target analytes are reported as concentrations, which were calculated by dividing the mass of the analyte measured relative to the internal quantitation standards (in ng) by the volume of air sampled (in m3) or precipitation collected (in L). Thus, these concentrations include two errors: the analytical error and the sampling error. The latter includes the error caused by errors in measuring the sample volume (in either m3 or in L) and the error caused by taking duplicate samples with separate samplers located 3 m apart. The surrogates are reported here as masses measured relative to the internal quantitation standards (in ng). Thus, the

surrogates include only the analytical error. These will be discussed separately below. Data Analyses. The data set for the following discussion consists of values of the relative percent difference (RPD) between the duplicate measurements, which for the analyte concentrations is defined as RPDC =

200|C1 − C2| C1 + C2

(1) 3

where C1 and C2 are the concentrations (in pg/m or ng/L) measured in each of the duplicate samples. For the surrogate mass measurements, the RPD is RPDM =

200|M1 − M 2| M1 + M 2

(2)

where M1 and M2 are the surrogate masses (in ng) measured in each of the duplicate samples. If either or both of these concentrations or masses were nondetects, the RPD was not calculated, and the spreadsheet cell was left empty. RPD data for which there were 19 or fewer measurements were omitted because they would not give reliable percentiles.

■

RESULTS AND DISCUSSION Neither of the two air or precipitation samplers gave consistently higher results than the other, as determined using the sign test; thus, we calculated the absolute values of the differences between the duplicates, as given in eqs 1 and 2. A preliminary ANOVA indicated that there were no significant (P < 0.05) differences in the RPD values among sampling sites or among sampling years; thus, the data from all sites and all years were pooled. This lack of significant differences among the sampling sites suggests that the RPD values are not related to the absolute concentrations of a given compound. For example, the concentration of chlordane at Chicago is about 15 times higher than that at Eagle Harbor, but the RPDs for chlordane at these two sites are similar. This observation indicates that measurements at the low end of the concentration scale (at Eagle Harbor in this case) are more or less as precise as those at the higher end of the concentration scale (Chicago in this case). All of the RPD data are given in the Supporting Information, which presents the number of duplicate measurements and the 10th percentile, the median, and the 90th percentile for each compound in each of the three phases. For example, for PCB52 there were 254 duplicate pairs in the vapor phase, and the middle 80% of these RPD values ranged from 3.5% to 73.2% with a median of 22.9%. To make the following discussion tractable, we have elected to focus on 26 compounds and their 11 surrogates, and these data are given in Table 1. These data include 10 PCB congeners, 12 organochlorine pesticides, eight polycyclic aromatic hydrocarbons, and seven polybrominated diphenyl ethers. We selected these compounds because they are typical of their chemical classes and because they gave RPDs for virtually all duplicate samples measured. Figure 2 shows the median RPD for each of these compounds in each phase, with the surrogate results marked in red. The median RPD values for the samples are in the 20− 40% range for all compounds and all phases, with a few exceptions. The first exception is BDE-209 in the vapor phase, which shows a median RPD of 72%. This is puzzling given that BDE-209 has such a low vapor pressure that it should not even be measurable in the vapor phase. The second exception is that C

DOI: 10.1021/acs.est.6b04428 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

samples, which tend to show higher RPDs compared with the vapor and particle phases. Put simply, the errors predominately lay with the denominator (sample volume) and not with the numerator (measured mass) in our concentration calculations. Let us examine in more detail how we actually measure the volume of the air sample. The high-volume air samplers are from Tisch Environmental, and they consist of a blower motor that pulls air through a quartz fiber filter and then through a bed of XAD resin. The flow rate is calibrated once every 4 months with a standard orifice calibrator device that itself has been calibrated once a year by Tisch Environmental. The total pressure differential between the sampler exhaust and the ambient pressure is measured by a magnehelic gauge in inches of water. The primary calibration by Tisch and our seasonal calibrations provide a linear relationship between the magnehelic gauge reading and the flow through the sampler in cubic feet per minute. During calibration, the sampler motor speed is adjusted using the linear calibration to give a flow rate through the sampler corresponding to 0.566 m3/min. At this flow rate, for a 24 h sample, we obtain a total sample volume of 815 m3. During operation, the flow rate through the sampler is regulated by a mass-flow controller equipped with a hot-wire anemometer inserted into the air stream just below the filter. This probe is monitored by the flow controller, which adjusts the motor speed as the filter begins to collect particles. The sampler blower is turned on by a mechanical or electronic timer and off 24 h later by the same timer. During setup, the blower is turned on temporarily for 2 min before the timer is set, and a magnehelic gauge reading is taken. This procedure is repeated after the sampling is completed. The two magnehelic gauge readings are recorded and reported to our laboratory. These magnehelic readings are compared to those taken during the most recent calibration. The magnehelic reading fluctuates with the local temperature and barometric pressure, factors that the set-point calculation takes into account. However, if the magnehelic gauge reading before or after the sampling differs by more than 50% from the calibration reading, the samples are usually not analyzed. There are several sources of possible error associated with measuring the sampled air volumes. First, there could be errors in the flow rate calibration caused by drift in the pump speed between the seasonal calibrations. We guard against this possibility by noting the magnehelic gauge flow rates at the start and end of sampling. Second, there could be errors in the timer start and stop times caused by malfunctions in the timer itself. An add-on timer, which operates only in the presence of the electromagnetic field generated during blower motor operation, protects against ambiguity should a timer fail. In this way, a sampler that ran properly would not be disqualified because the controller timer did not count the hours of operation. Of course, we note the times on these devices at the start and end of sampling. The high-volume air samplers are also equipped with a mechanical retracting hood within the shelter, which prevents passive particle deposition while the sampler is not operating. Failure of this hood to retract could result in a noticeable difference between simultaneously run samplers. There are also three O-rings in our modified high-volume air samplers that are necessary to seal the XAD cartridge to the blower motor and to maintain the flow through the system. If any of these are missing, the flow will be erratic. It should also be noted that even with all of the O-rings in place and with the filter holder fastened tightly, unless the XAD cartridge tube is tightly

Table 1. Median Relative Percent Differences between Duplicate Samples Taken at Four Sites on the Shores of the Great Lakesa vapor phase compound

N

median

PCB-14 (surr) PCB-18 PCB-28 PCB-52 PCB-65 (surr) PCB-83 PCB-101 PCB-118 PCB-166 (surr) PCB-180 α-HCH δ-HCH (surr) ε-HCH (surr) γ-HCH α-chlordane γ-chlordane DBC (surr) endosulfan I endosulfan II p,p′-DDD p,p′-DDE p,p′-DDT phenanthrene phenanthrene (surr) fluoranthene pyrene pyrene (surr) benz[a]anthracene benzo[a]pyrene coronene BDE-47 BDE-77 (surr) BDE-99 BDE-100 BDE-166 (surr) BDE-209 BDE-209 (surr)

255 255 254 254 254 174 255 254 255 190 263 218 83 251 248 235 261 241 153 104 256 218 266 252 265 228 200 43

4.6 23.6 29.1 22.9 4.2 29.1 22.6 29.3 3.1 26.7 15.2 6.4 6.2 17.9 16.4 22.2 7.7 19.8 28.7 45.4 15.1 26.8 17.4 5.7 17.9 26.1 6.0 38.9

99 98 100 86 99 75 99

26.6 6.7 34.3 26.2 4.7 71.5 13.1

particle phase N

median

56 108

20.5 6.2

112 127 124 141 138 101 60 27 51 241 251 256 200 205 154 180 139 99 104 105 92 105 97 105

28.1 24.4 24.4 6.3 17.5 23.9 44.5 26.7 45.7 20.6 6.5 21.1 22.6 5.7 21.7 23.8 22.3 24.6 4.9 22.1 28.5 5.0 43.9 14.6

precipitation N

median

108 107 101 104 108 55 108 108 108 91 141

5.2 36.6 33.2 21.9 5.3 51.6 29.2 33.1 4.2 34.8 15.9

54 134 118 114 151 143 134 46 144 123 148 143 148 138 109 115 126 97 63 61 63 63 63 63 61

8.9 20.5 37.2 36.6 9.4 22.7 16.4 43.4 26.7 42.5 22.5 7.8 20.9 25.7 6.2 23.9 23.4 29.6 91.6 11.2 62.8 48.0 9.5 52.7 18.2

N is the number of duplicate pairs used to find the median. The full data set, including the 10th and 90th percentiles for all of the compounds in all three phases, is given in the Supporting Information. The surrogates (“surr”) are indicated by italics.

a

all of the PBDEs show relatively high RPDs in precipitation, ranging from 48 to 92%. This is expected since we have relatively shorter data sets for these compounds, starting only in 2005 as opposed to 1994 for most of the other compounds. The averages of the medians shown in Figure 2 are summarized in Table 2. The median RPD is 27−35% for all of the compounds taken together, but the RPD for the surrogate measurements is 6−9% for the same set of measurements. The difference between these two sets of errors is the sampling error, which in this case is on the order of 20− 25%. In other words, the measurement of the mass of the analyte in the sample is much more precise than is the measurement of the volume of the sample, or taking two duplicate atmospheric samples with side-by-side samplers is problematic. This is particularly noticeable for precipitation D

DOI: 10.1021/acs.est.6b04428 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

Figure 2. Median relative percent differences for duplicate samples taken at four sites around the Great Lakes organized by compound and phase sampled. The red bars represent the relative percent differences for the surrogate compounds that were added to the samples just before extraction.

falling snow and to open the unit when it might otherwise remain closed during such an event. All water from the funnel flows through a glass column filled with wetted XAD resin to absorb the organic analytes and empties into a large glass carboy that is used to measure the volume of precipitation for a particular sample. The precipitation samples are pooled together over an entire month. At the end of the month, or whenever the site operator sees that the carboy is greater than one-half full, the water in the carboy is poured into a 1 L graduated cylinder in 1 L increments. The number of these increments times 1000 mL and the volume of the last partial increment are summed, and the precipitation volume is reported to the nearest 10 mL. This is not a precise process. There can be some spillage as water in the large carboy is poured into the graduated cylinder, and there is a chance for miscounting the number of 1 L increments. On the basis of this experience, we have replaced all of the carboys with ones that have a spigot at the bottom to facilitate measuring the volume of collected precipitation. An additional source of error for the precipitation samples could be differences in the sensitivity to precipitation events of the sensors that activate the opening and closing of the hoods that cover the collection basins. It has been occasionally noted by a site operator that one hood will be open while the other is closed. It is possible that the first minutes of rainfall will carry with it the highest concentration of atmospheric contaminantswhat might be called the “initial scrub” of the atmospheric scavenging process. In this case, a less sensitive precipitation sensor might not open the hood quickly enough to collect the dirtiest portion of rainfall. If the two precipitation samplers do not open and close at the same time, in essence different precipitation events are being sampled, and the differences in the measured concentrations could be real. Occasionally, the flow through the Teflon tube connecting the bottom of the XAD resin column to the carboy gets

Table 2. Average Median RPDs for All Compounds Measured in Each Phasea compound PCBs

pesticides

PAHs

PBDEs

all compounds combined

phase vapor particle precipitation vapor particle precipitation vapor particle precipitation vapor particle precipitation vapor particle precipitation

samples 26.2 NA 34.3 23.1 28.4 29.1 25.1 22.0 24.3 39.7 29.8 63.8 27.1 26.7 34.8

± 1.2 ± ± ± ± ± ± ± ± ± ± ± ± ±

3.4 3.2 3.3 3.7 5.0 0.5 1.2 10.8 4.9 9.8 2.5 1.9 3.3

surrogates 4.0 NA 4.9 6.8 6.2 9.1 5.9 6.1 7.0 8.2 8.2 13.0 6.2 7.0 8.6

± 0.5 ± ± ± ± ± ± ± ± ± ± ± ± ±

0.4 0.5 0.1 0.3 0.2 0.4 0.8 2.5 3.2 2.7 0.8 1.3 1.3

a The errors are given as standard errors of the mean. “NA” means duplicates were not measured for that compound in that phase.

installed, the blower flow rate will be affected. Our local site operators guard against this possibility. Insects and snow or ice can cause increased flow rates if the blower exhaust tube becomes partially or wholly blocked. Human activity (mowing, construction, operating vehicles, etc.) in the immediate vicinity of one sampler versus the other could also contribute to differences in the duplicate samples. The precipitation samplers consist of a shallow 46 cm × 46 cm funnel that is normally closed and is opened by a sensor that detects precipitation (rain or snow). The funnel and the sampler enclosure are heated to about 5−7 °C to melt snow. The underside of the sensor is also heated in order to evaporate morning condensation and to prevent the unit from opening for a nonprecipitation event. This heating also serves to melt E

DOI: 10.1021/acs.est.6b04428 Environ. Sci. Technol. XXXX, XXX, XXX−XXX

Article

Environmental Science & Technology

(4) Venier, M.; Hung, H.; Tych, W.; Hites, R. A. Temporal trends of persistent organic compounds: A comparison of different time series models. Environ. Sci. Technol. 2012, 46, 3928−3934. (5) Venier, M.; Hites, R. A. Time trend analysis of atmospheric POPs concentrations near the Great Lakes since 1990. Environ. Sci. Technol. 2010, 44, 8050−8055. (6) 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. (7) Great Lakes Integrated Atmospheric Deposition Network. https://www.epa.gov/great-lakes-monitoring/great-lakes-integratedatmospheric-deposition-network (accessed Aug 31 2016). (8) Wu, R.; Backus, S.; Basu, I.; Blanchard, P.; Brice, K. A.; DryfhoutClark, H.; Fowlie, P.; Hulting, M. L.; Hites, R. A. Findings from quality assurance activities of the Integrated Atmospheric Deposition Network. J. Environ. Monit. 2009, 11, 277−296.

restricted. This problem is usually obvious because of the standing water left in the collection funnel at the top of the sampler. If this happens, the site operators clear the obstruction with a metal probe and allow the accumulated precipitation to flow through the XAD column and into the carboy. In any case, the volume of precipitation measured in the carboy is usually the volume that has passed through the XAD column unless the funnel has overflowed, in which case, the duplicates would not agree. In addition to sample volume errors, it is also possible that the concentrations of these compounds in the atmosphere are actually variable on a small spatial scalewhat might be called spatial heterogeneity. Thus, even though the samplers are located within 3 m of one another, it is possible that the concentrations in the two different volumes of air that are sampled are in fact different from one another. If this is the case, no improvements in the measurement of the sample volumes will reduce the apparent error between the duplicate samples. One way to investigate this possible effect would be to systematically vary the distance between the two samplers, but the number of replications needed to uncover this effect, if it exists, may make this approach impractical.

■

ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.est.6b04428. Numbers of duplicate samples and RPDs for duplicate measurements of various compounds in the atmospheric vapor and particle phases and in precipitation collected at three sites around the North American Great Lakes; the 10th, 50th (median), and 90th percentiles of the RPDs are given for each compound in each phase (PDF)

■

AUTHOR INFORMATION

Corresponding Author

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

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

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS We thank the U.S. Environmental Protection Agency’s Great Lakes National Program Office for funding through agreement GL00E01422 (Todd Nettesheim, Project Officer), Marta Venier and Amina Salamova for helpful discussions, and all of the Integrated Atmospheric Deposition Network (IADN) team members for sampling and analyses.

■

REFERENCES

(1) Hites, R. A. A statistical approach for left-censored data: Distributions of atmospheric PCB concentrations near the Great Lakes as a case study. Environ. Sci. Technol. Lett. 2015, 2, 250−254. (2) 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. (3) Venier, M.; Hites, R. A. DDT and HCH, two discontinued organochlorine insecticides in the Great Lakes region: Isomer trends and sources. Environ. Int. 2014, 69, 159−165. F

DOI: 10.1021/acs.est.6b04428 Environ. Sci. Technol. XXXX, XXX, XXX−XXX