Andrés Ramírez
Restrepo
ab,
Stephen J.
Hayward
ac,
James M.
Armitage
*a and
Frank
Wania
a
aDepartment of Physical and Environmental Sciences, University of Toronto Scarborough, 1265 Military Trail, Toronto, Ontario M1C 1A4, Canada. E-mail: james.armitage@utoronto.ca; Tel: +1 416-287-7225
bGrupo GDCON, Facultad de Ingeniería, Sede de Investigaciones Universitarias (SIU), Universidad de Antioquia, Calle 70 No 52 -21, Medellín, Colombia
cDepartment of Chemical Engineering and Applied Chemistry, University of Toronto, 200 College St W, Toronto, Ontario M5S 3E5, Canada
First published on 9th June 2015
The main objective of this study was to evaluate the performance of a model for simulating the uptake of various pesticides on passive air samplers (PAS). From 2006–2007 a series of PAS using XAD-resin were deployed at Egbert, a rural agricultural site in southern Ontario, Canada, to measure the uptake of pesticides for time periods ranging from two months to one year. A continuous increase in sequestered amounts was observed for most pesticides, except for trifluralin and pendimethalin, which could conceivably be subject to substantial degradation inside the sampler. Continuous low-volume active air samples taken during the same period, along with data on weather conditions, allowed for the simulation of the uptake of the pesticides using the model (PAS-SIM). The modelled accumulation of pesticides on the PAS over the deployment period was in good agreement with the experimental data in most cases (i.e., within a factor of two) providing insight into the uptake kinetics of this type of sampler in the field. Passive sampling rates (PSR, m3 d−1) were determined from the empirical data generated for this study using three different methods and compared with the PSRs generated by the model. Overall, the PAS-SIM model, which is capable of accounting for the influence of temperature and wind variations on PSRs, provided reasonable results that range between the three empirical approaches employed and well-established literature values. Further evaluation and application of the PAS-SIM model to explore the potential spatial and temporal variability in PAS uptake kinetics is warranted, particularly for established monitoring sites where detailed meteorological data are more likely to be available.
Environmental impactPassive air samplers are frequently deployed in the field in order to monitor ambient concentrations of various contaminants in the atmosphere. Although the basic principles underlying the accumulation of organic chemicals on passive air samplers are well-established, interpretation of monitoring data is complicated by varying ambient concentrations and meteorological conditions over time. This study reports on the performance of a modeling tool (PAS-SIM) for simulating the accumulation of organic chemicals on XAD-2 passive air samplers using a calibration study for pesticides. The modelled accumulation of pesticides on the PAS was in good agreement with the experimental data in most cases (i.e., within a factor of two) providing insight into the uptake kinetics of this type of sampler in the field. |
Recently, the PAS-SIM model was presented as a tool for simulating the behaviour of organic chemicals on PAS using divinylbenzene-styrene-co-polymeric resin (XAD-2) as sorbent under different exposure scenarios.16,17 One potential use of the PAS-SIM model is to estimate PSRs prior to actual deployments, based only on the meteorological conditions (i.e., temperature, wind speed) at the sampling sites and the chemical properties of the target analytes. The model's performance has been evaluated for PCBs and PAHs but not for pesticides. Accordingly, the main objective of this study is to assess the performance of the PAS-SIM model for simulating the uptake kinetics of various pesticides on PAS. PAS using XAD-2 as sorbent (hereinafter referred to as XAD-PAS) were deployed for up to one year, alongside a continuous active air sampler. The active air sampling data in combination with the PAS-SIM model were used to simulate the uptake of pesticides in the XAD-PAS. Sampling rates for a range of pesticides were derived by using the PAS-SIM model and compared with those obtained by direct data calibration methods. Three empirical methods for estimating PSRs were considered. A secondary objective of this study is therefore to provide guidance on the appropriateness and applicability of these methods for deriving empirical sampling rates from calibration studies using XAD-PAS.
(1) |
m = PSR(CLV-AAS·t) | (2) |
PSR is then derived as the slope of the linear regression of the sequestered amount m against the product of CLV-AAS and time (Method 2).
Methods 1 and 2 assume the PSR to be constant during deployment, but previous research demonstrated that PSRs can vary with temperature and wind speed.7,13–15,18 To address this concern, a third method was used to account for seasonal variations. Sampling rates during each two months interval between retrievals were derived from the increase in the amount captured by subsequently retrieved XAD-PAS and the average air concentration (CLV-AAS)i, during the interval (Method 3).
(3) |
The use of depuration compounds has also been proposed as a method to estimate PSRs from PAS data (in the absence of AAS), based on the loss of the spiked compound over the deployment period.9,19,20 This approach is predicated on the assumption that uptake of target analytes and loss of the depuration compounds are subject to the same transport resistances, where typically it is assumed that transport across the air-side boundary layer is rate limiting. As discussed previously, transport through the porous medium on the sampler-side is an important consideration and hence these assumptions may not be valid.21,22 Additional research is required to better understand the use of depuration compounds for estimating PSRs as a function of chemical properties and environmental conditions. Regardless, because the XAD-PAS deployed in this study were not spiked with depuration compounds, this method cannot be applied to the current data.
The LV-AAS data was used as an input to the model, and the output (i.e., the amount m sequestered in the PAS) was compared against the empirical data obtained from the deployed XAD-PAS. The normalized residuals error (NRE) in the model estimation was calculated using the following equation:31
(4) |
On the other hand, whereas the LV-AAS air concentrations of pendimethalin and trifluralin remain elevated throughout the summer months (≥91 pg m−3 and 88 pg m−3 respectively), the empirical XAD-PAS data do not show continued uptake even though the amounts accumulated on the samplers (4.4 ng and 3.2 ng respectively by Sept 1) do not appear to reflect equilibrium partitioning at any time. For example, the expected amount of trifluralin on the XAD-PAS samplers deployed for this study at equilibrium with an air concentration of 20 pg m−3 at 30 °C is 13.2 ng. However, the measured amount of trifluralin on the XAD-PAS was ≤4.2 ng throughout the summer (i.e., well below the amount corresponding to thermodynamic equilibrium with the ambient air). Accordingly, the apparent loss of pendimethalin and trifluralin from the passive samplers cannot be explained by fluctuations in ambient air concentrations or enhanced volatilization in the summer caused by warmer temperatures because the sampler is below the expected equilibrium. Although both of these compounds exhibit relatively large 2nd-order OH radical reaction rate constants (kOH, as estimated using the EPISUITE AOPWIN v1.92 module), so do some of the other compounds sampled here (e.g., atrazine, disulfoton). As the experimental PAS data for these other compounds is broadly consistent with expectations, it does not seem likely that reaction with OH radicals in the pore space of the sampler alone can explain the apparent discrepancies for pendimethalin and trifluralin. Moreover, the mass fraction of the compounds in the pore air of the sampler is negligible in comparison to the sorbed fraction. An alternative explanation for these observations is that degradation of the compounds within or sorbed to the passive sampler is facilitated by the sampler medium itself. For example, it was recently reported that XAD-2 ‘artificially transformed’ chlorpyrifos to its oxygenated analogue chlorpyrifos-oxon to a substantial extent whereas PUF used in the same study did not.32 Because pendimethalin and trifluralin have a similar dinitro-aniline structure, they both may be prone to the same type of reaction process(es). Additional studies are required to explore this hypothesis experimentally; PAS-SIM model simulations incorporating this process are presented below. Because of the discrepancy identified above, care must be taken when interpreting the air concentrations of pendimethalin and trifluralin, as the estimated air concentrations using the sampler data may be inaccurate.
Of the 17 pesticides detected, acceptable agreement was found for eight compounds, seven pesticides were systematically underestimated, and there was no agreement for two pesticides. The compounds with acceptable agreement between model output and the empirical data were alachlor, atrazine, cis-chlordane, trans-chlordane, DCPA, disulfoton, metolachlor, and trans-nonachlor (Fig. 1). The emission profiles of these compounds are diverse, from pesticides with a strong seasonal variability to others with seemingly random fluctuations over time. For some pesticides in this group (DCPA, disulfoton, trans-nonachlor), not all solute descriptors had been reported in the literature so estimated values were used (i.e., Absolv output). In the case of trans-nonachlor, all the descriptors were estimated, but its NRE absolute value is less than 0.1. These results suggest that altogether the solute descriptor estimation, the pp-LFERs and the PAS-SIM model are a good and rugged assembly, able to make accurate predictions even for compounds with little experimental property data available.
When the modeled and measured shape of the uptake curve was similar, but the NRE was systematically greater than 2σ, model results, shown in Fig. 2, were judged systematically biased. This was the case for HCB, α- and γ-HCH, the endosulfans and chlorothalonil. For the pesticides in this study, the bias was always positive, suggesting that the model is prone to underestimating the residues on XAD-PAS. The extent of bias can be expressed by the number of standard deviations n. For example, an n of approximately 3 for HCB means that the experimental values usually were three standard deviations above the predicted values. The common range for n was between 2 and 5, with the noticeable exception of chlorothalonil (n ≈ 16). Compounds with the lowest systematic bias were the endosulfans, whose structure is quite similar to the chlordanes, for which acceptable agreement was found (Fig. 1). The model bias also can be expressed by a factor of agreement (FoA), which was a factor of 2 for almost all compounds, indicating that modeled amounts (m) are approximately half of the experimental values. For chlorothalonil the modeled amounts are only a fifth of the empirical values obtained and hence the FoA is 5. The blue lines in Fig. 2 show a fitted estimation using the FoA as a correction factor for the values obtained by the model using 10 mm stagnant boundary layer. Accordingly, the values of the fitted estimates double the values obtained by the PAS-SIM model for all the compounds except for chlorothalonil, for which the fitted values are five times higher.
Uncertainty in the estimated sampler-air partition coefficients can be an important consideration for chemicals with relatively small values (i.e., logKSA ≤ 7.5 at 25 °C). For example, model output for HCB and γ-HCH approaches the lower bound of the empirical XAD-PAS data if the logKSAs are increased by 1 order of magnitude. The improved model performance reflects increased net uptake of the chemical (i.e., decreased volatilization) over the simulation period due to higher sorption capacity. logKSA values greater than 9 did not substantially improve model performance. Although the ppLFER method used to derive all KSA values is well-validated, the potential for errors remains. As it may not be possible to know when the estimated KSAs are in fact biased low, this consideration could be taken into account as an uncertainty factor in the interpretation of PAS-SIM outputs for more volatile compounds. Bias in the estimation of the aerosol-air partition coefficient can also influence model performance if this property value is relatively large (i.e., estimated logKQA > 9), as the PAS-SIM model for XAD-2 assumes that the fraction of chemical bound to particulates is completely unavailable for uptake.17 Overestimation of log KQA can therefore lead to underestimation of the amount of chemical accumulated on the sampler in the model calculations because the available fraction is inaccurately quantified. This aspect may partly explain the model performance for endosulfan II and endosulfan sulfate, given that the estimated logKQAs for these compounds are greater than nine (ESI, Section S2†). For the other pesticides, this consideration is not expected to be relevant. Note that the PAS-SIM model assumptions are based on empirical XAD-PAS data for PAHs such as benzo(b)fluoranthene, benzo(a)pyrene and indeno(1,2,3-c,d)pyrene, compounds known to be predominantly particle-bound under typical atmospheric conditions. Low sampling efficiencies of particle-bound PAHs on PUF-PAS were also reported for a recent calibration study33 and the reliability of PAS for particle-bound compounds in general remains unclear.34
Two compounds, trifluralin and pendimethalin, showed a completely different behavior in the model (Fig. 3). While the overall NRE indicates significant agreement, the NRE itself has a tendency to have significant changes from extreme positive to extreme negative values (i.e., the low overall NRE is due to error cancellation). Consistent with the equilibrium-based calculations discussed above, the calculated fugacities35 in the sampler and ambient air over the course of the simulation (data not shown) indicate that the XAD-PAS should depurate these chemicals only towards the end of the simulation (days 283–365), when the concentration of these chemicals in ambient air concentration becomes negligible. To explore the hypothesis of degradation within the sampler,32 a 1st-order degradation rate constant applied to the sorbed phase (i.e., kdeg-PAS, d−1) was introduced and fitted until the simulation shape resembled the experimental data obtained. In both cases, a rate constant of 0.0125 d−1 produced acceptable results with respect to the temporal trends in the empirical uptake curves.
Compound | PSRE (m3 d−1) | PSRW (m3 d−1) | ||
---|---|---|---|---|
Method 1 | Method 2 | Method 3 | ||
Alachlor | 0.66 ± 0.07 | 0.63 ± 0.02 | 0.61 ± 0.09 | 0.54 |
Atrazine | 0.72 ± 0.06 | 0.70 ± 0.03 | 0.66 ± 0.10 | 0.61 |
Chlorothalonil | 2.27 ± 0.30 | 2.29 ± 0.09 | 1.90 ± 0.62 | 0.67 |
cis-Chlordane | 0.48 ± 0.06 | 0.43 ± 0.02 | 0.42 ± 0.10 | 0.50 |
trans-Chlordane | 0.56 ± 0.03 | 0.55 ± 0.01 | 0.54 ± 0.07 | 0.51 |
DCPA | 0.47 ± 0.09 | 0.42 ± 0.03 | 0.35 ± 0.13 | 0.57 |
Disulfoton | 0.68 ± 0.05 | 0.67 ± 0.03 | 0.65 ± 0.19 | 0.53 |
Endosulfan I | 0.89 ± 0.09 | 0.84 ± 0.04 | 0.78 ± 0.19 | 0.54 |
Endosulfan II | 0.67 ± 0.06 | 0.65 ± 0.03 | 0.62 ± 0.15 | 0.53 |
Endosulfan sulfate | 0.31 ± 0.01 | 0.31 ± 0.01 | 0.34 ± 0.06 | 0.55 |
HCB | 0.88 ± 0.10 | 0.86 ± 0.04 | 0.77 ± 0.33 | 0.67 |
α-HCH | 1.15 ± 0.18 | 1.06 ± 0.07 | 0.99 ± 0.32 | 0.64 |
γ-HCH | 0.93 ± 0.16 | 0.91 ± 0.04 | 0.88 ± 0.31 | 0.63 |
Metolachlor | 0.74 ± 0.04 | 0.72 ± 0.02 | 0.68 ± 0.10 | 0.50 |
trans-Nonachlor | 0.43 ± 0.04 | 0.41 ± 0.02 | 0.39 ± 0.08 | 0.52 |
The empirical PSR values (PSRE) for HCB, α-HCH and γ-HCH agree very well with values that were determined previously in the same region (Burnt Island and Point Petre, in central and southern Ontario, respectively).16 In contrast, the values are higher than those obtained under Arctic conditions (Alert),16 and lower than those from Costa Rica,7 confirming the need for temperature-dependent calibration. These results support the use of Method 3 as an accurate way to obtain PSRE, despite its associated variability which is higher than the uncertainty obtained for the other two methods, largely because it accounts for the temperature-dependent variation in PSRE. The ratio of the average PSRE during the Spring–Summer (end of April–August) and Fall–Winter (March–April, September–February) periods, PSRsummer and PSRwinter, respectively (data shown in the ESI, Section S9†), indicates that the sampling rates are an average of 40% higher during the warmest deployment periods. Thus Method 3 is likely the best estimate of the XAD-PAS sampling rates over the entire year, as it most accurately accounts for the seasonal variations throughout the deployment. However, as can be seen in Table 1, differences in the estimated PSREs via Methods 1–3 are often less than the associated uncertainties in the estimates and therefore, in practical terms, data availability (i.e., sampling intervals over deployment period) is the key factor.
The wind speed adjusted sampling rates estimated by the PAS-SIM model for a 10 mm stagnant boundary layer (PSRW) also are presented in Table 1. As shown, the PSRW agrees with the empirical sampling rates derived from the field deployment in Egbert for the majority of chemicals, following from the model performance illustrated in Fig. 1–3. Note that the modeled PSRW is an inherent sampling rate that is independent of the concentration of the chemical in the air and amount of chemical on the sampler and only the sampler dimensions, meteorological conditions and the physicochemical properties are taken into account to calculate it.17 Because the information needed to calculate PSRW over time is often available, it can be estimated for any site prior to and during the deployment period to inform the interpretation of empirical XAD-PAS data. Such calculations could be particularly useful to probe site-to-site and year-to-year variations in passive air monitoring data. Parameterization of the model with the most detailed meteorological records available (e.g., temperature and wind speed at daily resolution) is recommended for this purpose, especially for sites experiencing substantial weather variability.
A secondary objective was to further assess the appropriateness and applicability of three methods for deriving empirical sampling rates from calibration studies. The selection of the most appropriate approach to quantify PSR depends on the availability of the data and the accuracy needed. Simple approaches (i.e., Method 1 or 2) may be sufficient to estimate the empirical sampling rate (PSRE) if the weather conditions or ambient concentrations are expected to be relatively stable over the deployment period. When a site is expected to have strong seasonal or meteorological variability, Method 3 will likely yield a better estimate and is recommended if PAS were deployed and collected at appropriate intervals. While empirical PSREs for target compounds from a given site could be assumed to be valid for other sites with similar meteorological conditions, a more rigorous approach would be to use all available literature data to make an estimate by linear regression of the experimental PSR against a site-specific characteristic, e.g., temperature.36 Using multiple literature values has a real advantage over the extrapolation of a single empirical sampling rate. The main limitation for this approach may be the lack of data needed and/or the lack of congruence across that data. As noted in the Methods section, the use of depuration compounds has been promoted as a method to estimate PSRs in the absence of concurrent AAS data but is subject to some uncertainty.21,22 Although outside the scope of the current study, the PAS-SIM model can also be used to simulate the behaviour of depuration compounds under different meteorological conditions. Simulated PSRs based on uptake scenarios and derived from depuration scenarios could then be compared and used to gain insight into potential error associated with the use of depuration compounds to estimate PSR using the current approach. Such simulations are considered a priority for future applications of the PAS-SIM model. Development of a PAS-SIM parameterization set for simulating the uptake of organic compounds on PUF-PAS is also desirable, given the widespread use of this type of PAS for field deployments.
Footnote |
† Electronic supplementary information (ESI) available. See DOI: 10.1039/c5em00122f |
This journal is © The Royal Society of Chemistry 2015 |