Comparison of methods for measuring gas-particle partitioning of semivolatile compounds

Comparison of methods for measuring gas-particle partitioning of semivolatile compounds

ARTICLE IN PRESS Atmospheric Environment 37 (2003) 3177–3188 Comparison of methods for measuring gas-particle partitioning of semivolatile compounds...

219KB Sizes 0 Downloads 14 Views

ARTICLE IN PRESS

Atmospheric Environment 37 (2003) 3177–3188

Comparison of methods for measuring gas-particle partitioning of semivolatile compounds John Volckens*, David Leith Department of Environmental Sciences and Engineering, University of North Carolina at Chapel Hill, CB #7431 Rosenau Hall, Chapel Hill, NC 27599-7431, USA Received 17 November 2002; accepted 30 April 2003

Abstract This work demonstrates several difficulties associated with sampling semivolatile aerosols. Gas-particle partition coefficients (Kp ¼ ½F =TSP=A) of semivolatile PAHs and alkanes were estimated from chamber tests using four different sampling methods: filter-adsorbent (FA), filter–filter-adsorbent (FFA), Denuder–filter-adsorbent (DFA), and electrostatic precipitator-Adsorbent (EA). The FA, FFA, and EA sampling methods estimated Kp accurately for high concentrations of PAHs in diesel engine exhaust aerosol. The FA sampler, which used a Teflon Zeflour filter, was biased by filter adsorption of alkane gases sampled at lower concentrations, whereas the FFA sampler, which used a Teflon coated glass fiber filter, was biased by particle evaporation during the same tests. The EA sampler introduced small biases at low concentrations due to chemical artifacts from the corona, non-ideal particle collection, and evaporative loss of collected particles. The DFA sampler was biased by particle losses in the denuder for PAHs in diesel exhaust aerosol, but gave accurate estimates of Kp for alkanes measured at lower concentrations. Results from this research show that none of the methods tested was able to measure gas-particle partition coefficients accurately for each series of compounds under all sampling situations. r 2003 Elsevier Ltd. All rights reserved. Keywords: Kp ; Sampling; Filter; ESP; Denuder; Semivolatile

1. Introduction Semivolatile compounds exist in both gas and particle phases in the atmosphere. At equilibrium, the distribution of mass between phases can be described by a partitioning coefficient, Kp (m3 mg1): Kp ¼

F ; A  TSP

ð1Þ

where F is the particle phase concentration of the compound of interest (ng m3), A is the gas phase concentration (ng m3), and TSP is the amount of total *Corresponding author. Present address: National Exposure Research Laboratory, U.S. EPA, MD E205-3, Research Triangle Park, NC 27711, USA. Tel.: +1-919-541-2323; fax: +1-919-541-0960. E-mail address: [email protected] (J. Volckens).

suspended particulate matter (mg m3) (Pankow, 1994). Gas-particle partition coefficients are used to model the formation, transport, and fate of semivolatile compounds in the environment (Bidleman, 1988). Multiplying Kp by TSP gives a ratio of the particle to gas-phase concentrations: F Kp  TSP ¼ : A

ð2Þ

When Kp  TSP is greater than 1, the compound partitions primarily into the particle phase; Kp  TSP values less than 1 indicate partitioning primarily to the gas phase. For example, a compound with Kp  TSP=100 will partition 100 molecules to the particle phase for every one molecule to the gas phase. Time-integrated measurements of gas and particlephase concentrations are commonly used to determine Kp : Unfortunately, unbiased measurements of Kp are

1352-2310/03/$ - see front matter r 2003 Elsevier Ltd. All rights reserved. doi:10.1016/S1352-2310(03)00352-2

ARTICLE IN PRESS 3178

J. Volckens, D. Leith / Atmospheric Environment 37 (2003) 3177–3188

difficult to make. Several factors can contribute to measurement bias: (1) the sampling method disturbs the gas-particle equilibrium, causing mass transfer between phases during sampling, (2) the conditions governing gas-particle equilibrium change while sampling (i.e. temperature shift), causing previously collected particles and gases to migrate between phases, (3) the capture efficiency for particles and/or gases is less than 100% for a given method, and (4) some of the collected mass reacts during sampling (i.e. reactions with ozone). As a result, gas-phase compounds are incorrectly measured as particle-phase, or vice versa. When gas-phase compounds are measured erroneously as particle-phase, F is artificially increased, A is artificially decreased, and the sampling method overestimates Kp ; the inverse happens when particle-phase compounds are measured as being in the gas-phase. The objective of this research is to compare sampling methods used to measure Kp for semivolatile compounds. Four sampling methods were evaluated through chamber tests with alkane and diesel engine exhaust aerosols: filter-adsorbent (FA), filter–filter-adsorbent (FFA), denuder–filter-adsorbent (DFA), and electrostatic precipitator-adsorbent (EA). Diesel exhaust aerosol was chosen because it is a frequently studied atmospheric pollutant containing many semivolatile constituents (Kamens et al., 1995; Jang et al., 1997). Alkanes were tested because their physical and chemical behavior are well characterized. Diesel exhaust aerosol was sampled at a high, constant concentration whereas alkanes were sampled at lower concentrations with four pre-determined concentration profiles. A brief description of each sampling method follows. The FA system is a simple, traditional method for measuring Kp : Mass collected on the filter represents the particle phase concentration, F ; and mass collected on the downstream adsorbent represents the gas phase concentration, A: Thus, Kp is estimated from the ratio of the filter catch to the adsorbent catch, normalized to TSP. FA systems are prone to biases from particle evaporation (i.e. particle-phase compounds that evaporate during sampling and are subsequently measured as gases) and gas adsorption (i.e. gas-phase semivolatiles that adsorb to the filter during sampling and are subsequently measured as particle-phase) (Turpin et al., 2000). The FFA system is similar to the FA system except that a second filter (FF2) is placed behind the first (FF1) to correct for the adsorption artifact. Assuming equal amounts of gas-phase semivolatiles adsorb to each filter, mass on FF2 is subtracted from FF1 and the residual represents the particle phase concentration. Partitioning measurements with FFA samplers are biased when unequal amounts of gas-phase compounds adsorb to each filter or when particle-bound compounds collected on the front filter evaporate (Turpin et al.,

2000; Mader and Pankow, 2001; Volckens and Leith, 2002a). Denuders are designed to remove gases from an airstream while allowing particles to pass through unhindered (Gundel et al., 1995). In the DFA system, gas-phase semivolatiles are trapped on an upstream denuder while particles (and the semivolatile compounds bound to them) are captured on a downstream filter. Since air that passes through the filter is devoid of gasphase semivolatiles, particles on the filter may evaporate in an attempt to re-establish gas-particle equilibrium. Therefore, an adsorbent is placed behind the filter to collect any particle-phase mass that evaporates during sampling. The two most common biases associated with DFAs are particle deposition in the denuder and gas penetration of the denuder, also known as breakthrough. Particle losses in the denuder occur from impaction, sedimentation, diffusion, and evaporation (Ye et al., 1991). Gas breakthrough occurs whenever the denuder captures semivolatile gases with less than 100% efficiency (Turpin et al., 2000). Electrostatic precipitators (ESPs) charge particles using a corona and collect them on a conducting surface using an applied electric field. The EA system works much like a filter in that particles collect in the ESP while gas-phase compounds collect on an adsorbent downstream. Adsorption and evaporation artifacts are substantially reduced in an ESP compared to a filter because the particle collection surface area, a small sheet of aluminum foil, is substantially smaller than the effective surface area of a filter (Volckens and Leith, 2002a). The disadvantage of the EA sampler is that the corona discharge process can be destructive to semivolatile aerosols and cause chemical artifacts (Volckens and Leith, 2002b).

2. Experimental 2.1. Diesel engine exhaust aerosol Exhaust from a 1980 Mercedes Benz sedan (Model 300SD) was passed through a Kr85 charge neutralizer and into a 25 m3 chamber lined with 5 mil Teflon film. Details of the sampling chamber are described elsewhere (Kamens et al., 1995; Jang et al., 1997). Aerosol from the chamber was drawn through a 15 mm diameter glass manifold and into a cyclone designed to remove particles smaller than 2.5 mm in diameter (VSCC-A, BGI Inc., Waltham, MA). From the cyclone, aerosol passed through a four-way flow splitter to ensure equivalent aerosol into each sampling train (FA, FFA, DFA, EA). A schematic of the sampling array is shown in Fig. 1. Two of these systems were employed to replicate the results from each sampler, for a total of 8 samplers.

ARTICLE IN PRESS J. Volckens, D. Leith / Atmospheric Environment 37 (2003) 3177–3188

Key TFE Z TFE GF

PM2.5 Cyclone

Zeflour PTFE Filter Fiberfilm PTFE Coated Glass Fiber Filter Electrostatic Sampler

Flow Splitter

DFA

FFA

XAD-4 Coated Annular Denuder EA

TFE GF

2.2. Alkane aerosol

FA TFE Z

TFE GF TFE GF

To Flowmeters, Pumps

Fig. 1. Schematic of the sampling train for the diesel exhaust aerosol and alkane aerosol experiments.

Sampling was performed for 30 min at a flow of 4.2 l m1 per sampler. The following substrates were used: FA: 47 mm Teflon (Pallflex Zeflourt, Pall Gelman Sciences, Ann Arbor, MI), FFA and DFA: 47 mm TFE coated glass Fiber (Pallflex Fiberfilmt T60A20, Pall Gelman Sciences, Ann Arbor, MI), EA: 60  80 mm aluminum foil (Reynolds Metals Co., Richmond, VA). Annular glass denuders (URG, Chapel Hill, NC), coated with XAD-4, were used for all gas-phase collections. A second denuder was placed behind the FA sampling train to measure any gas breakthrough that may have occurred during a test. Aerosol residence time in each denuder was approximately 0.5 s. Particle size distribution was measured with a scanning mobility particle sizer (SMPS, TSI Inc., St. Paul, MN). Immediately after each experiment, substrates were spiked with a suite of deuterated internal standards, extracted with high purity dichloromethane, and stored in pre-cleaned glass jars prior to soxhlet extraction. Denuders were spiked with the same internal standard solution and rinsed three times with dichloromethane, with the final rinse acting as a blank. Details of the experimental workup and analysis have been previously described (Kamens et al., 1995; Jang et al., 1997). High TSP concentrations, on the order of 1–10 mg m3, were established in the chamber to ensure a sufficient collection of sample mass, given the short sampling time and low flow rate. The goal was to have a measurable amount of mass present in both gas and particle phases so that estimates of Kp would be free of uncertainty from detection limits and sample contamination. Six PAHs were selected for analysis: acenaphthene (ACE), fluorene (FLU), phenanthrene (PHN), fluoranthene (FLA), pyrene (PYR), and benz[a]anthracene (BAA).

Alkane aerosol was nebulized from a solution of C12– C24 n-alkanes into a 1.0 m3 chamber made from acrylic plastic. This chamber was designed for the generation of aerosol concentrations that can vary in time; its operating characteristics are published elsewhere (Volckens and Leith, 2002b). Four, time-dependent concentration profiles, shown in Fig. 2, were generated to simulate conditions favoring: (1) neither adsorption nor evaporation artifacts, (2) gas-phase adsorption, (3) particle evaporation, and (4) both adsorption and evaporation. Although each profile is unique, the chamber characteristics were set so that the timeweighted, average TSP was constant at 50 mg m3 for all tests. Sampling parameters were identical to those used in the diesel tests, except that the duration was 120 min. Details of the sample workup and analysis are published elsewhere (Volckens and Leith, 2002a). Particle size distribution was measured with a timeof-flight particle spectrometer (APS 3321, TSI Inc., St. Paul, MN). 2.3. Data analysis Pankow developed predictive equations for gasparticle partition coefficients (Pankow, 1987, 1994). When particles are predominantly liquid-like, as is expected with alkane and diesel exhaust aerosols, absorptive partitioning dominates and Kp may be calculated as Kp ¼

fom RT ; Mom gi p0l 106

ð3Þ

where fom is the mole fraction of organic matter in the particle, R is the gas constant (8.2  105 m3 atm K1 mol1), T is the ambient temperature (K), Mom is the average molecular weight of the absorbing liquid phase (g mol1), gi is the activity coefficient of the compound absorbed in the particle, p0l is the sub-cooled, liquid vapor pressure of the Target Conc. , µg/m³

Aerosol Inlet

3179

100

Alkane 4 (Ads + Evap)

75

Alkane 2 (Adsorption)

50

Alkane 1 (No Ads or Evap)

25

Alkane 3 (Evaporation)

0

1 Time, hr

2

Fig. 2. Target profiles of concentration vs. time for alkane aerosol tests: (1) constant concentration, no adsorption or evaporation, (2) adsorption favored, (3) evaporation favored, (4) both adsorption and evaporation favored. Time-weighted average concentration was set at 50 mg m3 per test.

ARTICLE IN PRESS J. Volckens, D. Leith / Atmospheric Environment 37 (2003) 3177–3188

log Kp ¼

mr log p0L

þ br ;

where the intercept   fom RT : br ¼ log Mom gi 106

ð4Þ

ð5Þ

A plot of log Kp vs. log p0L for a series of compounds gives a slope of mr ¼ 1 if br remains constant for the compounds of interest. Since the concentration profiles in Fig. 2 were designed to keep the average TSP equal between tests and the remaining terms in (5) are constant for alkanes, br should also remain constant. Therefore, the effects of adsorption and evaporation biases on each method may be quantified by observing the deviations in the predicted vs. measured values of Kp; mr and br taken from each profile. For PAHs in diesel engine exhaust, values of Kp can be normalized to gi to produce a linear plot of log Kp;g vs. log p0L ; as recommended by Jang et al. (1997). A quantitative estimate of measurement accuracy was determined by calculating a sum-of-squares error term ðSSEÞ for each sampler based on the logarithms of the measured partition coefficient (Kp;m ) and the expected partition coefficient (Kp ): Pn 2 i i i¼1 ðlog Kp  log Kp;m Þ ; ð6Þ SSE ¼ n where the superscript i represents each compound and n equals the total number of compounds measured. Data were analyzed with Stata software (Intercooled Stata 6.0, Stata Corp., College Station, TX) using one-way ANOVA, Tukey mean comparison tests, and simple linear regression. Statistical significance was determined at a five percent level (a ¼ 0:05).

3. Results and discussion

120

Concentration,µg/m³

compound (atm), and the final term in the denominator is a conversion factor (mg g1) (Pankow, 1994). In this work, empirical measurements of Kp are made from Eq. (1) and theoretical estimates are made from Eq. (3). Taking the logarithm of (3) results in the linear relationship shown in Eq. (4) (Pankow and Bidleman, 1992):

∑ PAH Particle

∑Alkane Particle

∑ PAH Gas

∑Alkane Gas

100 80 60 40 20 0

FA

FFA

DFA

EA

Fig. 3. Particle and gas mass balance by sampler for diesel exhaust aerosol and alkane aerosol tests. Values for PAHs represent the sum of ACE, FLU, PHN, FLA, PYR, BAA. Values for alkanes represent the sum of n-alkanes from C15– C24. Error bars represent 1SD.

1

Fraction of Total Mass Lost in ESP

3180

0.8

0.6

0.4

0.2

0

C 16

C20

C24

0.05 0.5 5 30 60 90 99 99.9 Percent of Mass in Particle Phase

3.1. Chemical mass balance

Fig. 4. Chemical bias from the ESP corona for n-alkanes C15– C24 present in varying quantities in the particle phase. Error bars indicate 1SD.

A mass balance of particle and gas-phase compounds measured by each sampler is shown in Fig. 3 for PAHs and alkanes. No significant difference was detected between the total mass of PAH (Sparticle+Sgas) measured by each sampler with diesel exhaust aerosol, indicating that the aspiration efficiency was equal between samplers. For the alkane tests, the EA measured on average 25% less mass than the FA, FFA, and DFA systems. This difference is probably due to a chemical artifact from the ESP corona, an effect

that has been previously observed when sampling alkanes at low concentrations (Volckens and Leith, 2002b). However, for these tests, the chemical bias appears to affect particles and gases equally. A plot of the ESP chemical bias vs. percentage of mass in the particle phase for alkanes C15–C24 is shown in Fig. 4. A linear regression of the bias vs. percent particle-phase resulted in a slope estimate of 0.000270.0002, which is not significantly different from zero, indicating that the

ARTICLE IN PRESS J. Volckens, D. Leith / Atmospheric Environment 37 (2003) 3177–3188

3181

chemical bias does not change with increasing alkane number. Therefore, for these alkane tests, the ESP chemical bias should not substantially affect Kp ; as particle and gas-phase compounds were lost in equal proportions. The ESP corona should not affect aerosols sampled by the ESP at high concentrations (i.e. diesel exhaust aerosol) (Volckens and Leith, 2002b). Gas breakthrough was less than 1% for all compounds tested (both diesel exhaust and alkanes) except for ACE (3.5%) and FLU (2.5%). 3.2. Diesel exhaust measurements A log–log plot of Kp;g vs. p0L for PAHs in diesel exhaust aerosol is shown for each sampler in Fig. 5. The expected values of Kp ; calculated from (4) and (5) and normalized to compound activity coefficient, gi ; are plotted for comparison. Values of gi for PAHs in diesel exhaust aerosol were adopted from Jang et al. (1997) who used both UNIFAC and Hansen structure-activity models. Regression data for these results are provided in Table 1. Concentration data for all tests are provided in Table 2. Temperature, relative humidity, and TSP in the chamber were 5 C, 80%, and 4100 mg m3, respectively. The constant br was estimated as 10.4 from these data, assuming fom ¼ 0:5 and Mom ¼ 300 g mol1. Some caution should be taken when comparing predicted vs. measured values of br here because of the uncertainty associated with the fom and Mom parameters. However, such variability is greatly reduced through the use of a log transform. Surprisingly, the FA sampler resulted in the most accurate estimates of Kp and hence, the best estimates of mr and br for PAHs in diesel exhaust aerosol, as assessed by the SSE in Table 1. Adsorption artifacts, usually associated with FA sampling, may have been minimized here by the large amount of particulate collected (B500 mg) on the filter. Furthermore, evaporation artifacts may have been reduced by the constancy of the sampled aerosol concentration and the short sampling duration. The FA and EA samplers also gave reasonable estimates of Kp for all PAHs except BAA. The discrepancy with BAA can be explained if the FFA and EA samplers collected particles with less than 100% efficiency. BAA had a Kp;g TSP near 1000 under the conditions tested here, indicating that each nanogram of BAA in the gas phase must have approximately 1000 counterparts in the particle phase. If 1% of the particlephase BAA were measured incorrectly as gas-phase (e.g. 10 ng) then the measured partition ratio (Kp;m TSP=990/11=90) would be in error by a an order of magnitude (Kp =Kp;m ¼ 11). The diesel exhaust particles measured here have a mass median diameter of 0.15 mm and were captured with an efficiency between 95% and 99% in the ESP sampler (Cardello et al., 2002).

Fig. 5. log Kp;g vs. log p0L for PAHs measured in diesel exhaust aerosol: ACE (1), FLU (2), PHN (3), FLA (4), PYR (5), BAA (6). Error bars are shown when the standard deviation of the measurement is larger than the data point.

The Fiberfilmt filters used in the FFA may also have collection efficiencies slightly less than 100% (Lee and Ramamurthi, 1993). Thus, a small, but non-zero percentage of particle phase mass was probably measured incorrectly as gas-phase in the FA and EA samplers. More important to note, however, is that even small sampling inefficiencies can result in large measurement biases to Kp under some conditions. The DFA method reported significantly lower values of Kp than the other methods and exhibited substantial deviations in the predicted Kp for the lower volatility compounds, as seen in Fig. 5. These deviations are primarily attributable to particle losses in the denuder, as the denuder walls were visibly darkened following the experiment. Particle evaporation during transmission through the denuder may also have contributed to these loses. A substantial amount of non-volatile PAHs, such as benzo[ghi]perylene and indeno[123cd]pyrene, were detected in the DFA denuder extracts after further analyses. These compounds were not detected in the gasphase filter extracts, nor the blanks, giving further evidence to support a bias from particle deposition in the denuder. The bias to Kp from particle loss to the denuder is analogous to the EA and FA errors discussed previously, since these losses result in particle-phase compounds being measured erroneously as gas-phase. For BAA, the deviation in expected vs. measured Kp;g is 2.1 log units, which is in error by a factor of 100. Nearly 10% of the particle mass must have deposited in the denuder to produce such an error. This number is higher

ARTICLE IN PRESS J. Volckens, D. Leith / Atmospheric Environment 37 (2003) 3177–3188

3182

Table 1 Sample regression statistics (71SD) for log Kp vs. log p0L for alkane and diesel exhaust aerosols log Kp vs. log p0L Slope (mr )a

Intercept (br )b

SSE

R2

1 2 3 4

0.9070.03 0.9070.03 0.9070.04 a 0.9370.03a 1.0670.04a

8.8270.20 8.6470.28 8.8270.31 9.2270.28 10.470.30b

0.2570.10 0.4570.14 0.2370.10 0.1170.09 0.1970.19

0.995 0.996 0.997 0.995 0.987

Alkane Alkane Alkane Alkane Diesel

1 2 3 4

2.0570.18 1.7270.14 1.9470.53 2.0270.22 0.6870.07

20.5473.20 17.171.31 19.6474.87 20.0472.00 8.1270.54

0.9071.01 0.2770.34 1.0971.18 0.6270.78 0.3070.64

0.972 0.993 0.930 0.988 0.904

DFA Fiberfilm

Alkane Alkane Alkane Alkane Diesel

1 2 3 4

1.1070.04a 0.8870.09a 1.0570.08a 0.8770.11a 0.6770.10

10.9270.34b 8.9870.79b 10.1670.68b 8.8770.92b 8.7270.76

0.0370.04 0.0470.05 0.1470.09 0.1070.04 1.1971.60

0.932 0.948 0.978 0.946 0.832

EA Al foil

Alkane Alkane Alkane Alkane Diesel

1 2 3 4

1.2470.08 1.2170.02 1.3170.09 1.2270.09a 0.8270.07

12.5870.74 12.0870.21 11.5070.88 13.3171.28 8.9770.57

0.0870.07 0.0370.04 0.2870.25 0.0970.12 0.1870.28

0.996 0.999 0.976 0.970 0.926

FFA (U) Fiberfilm

Alkane Alkane Alkane Alkane Diesel

1 2 3 4

1.1770.05 1.2370.03 1.1770.07 1.2370.06 0.7670.08

11.7870.40 12.0970.28 11.9870.64 12.2770.50 4.9670.59

0.1370.13 0.0470.05 0.1870.15 0.1470.12 0.2270.36

0.989 0.993 0.987 0.985 0.908

Sampler

Test

FA Zeflour

Alkane Alkane Alkane Alkane Diesel

FFA Fiberfilm

Numbers in bold indicate where estimates are not statistically different from expected values. Particle collection substrates are given below each sampler type. a Slope, mr ; is not significantly different from the expected value of 1. b Intercept, br ; is not significantly different from the expected value of 10.1 for alkanes, or 10.4 for diesel exhaust.

than expected for denuder samplers, which usually exhibit particle losses on the order of 1–5% by mass (Kamens et al., 1995). Since very fine particles, dp o0:1 mm, deposit primarily through diffusive mechanisms, decreasing the aerosol residence time in the denuder would reduce these losses. However, a shorter residence time may diminish the gas-phase collection efficiency and lead to increased gas breakthrough for higher volatility compounds, resulting in more gas-phase mass being measured as particle-phase. 3.3. Alkane aerosol Plots of log Kp vs. log p0L for the alkane tests are arranged by sampler type in Fig. 6 and by concentration profile in Fig. 7. The expected values of log Kp vs. log p0L ; calculated from (4) and (5), are plotted for comparison.

Regression statistics for these plots are provided in Table 1. Time-weighted average particle concentrations were 46, 44, 45, and 48 mg m3 for alkane tests 1–4, respectively. Chamber temperature and relative humidity were constant between tests averaging 2971 C and 5572.5%, respectively. The constant br was estimated as 10.1, assuming fom ¼ 1; g ¼ 1; and Mom ¼ 325 g mol1. Substrate blanks were low and accounted for less than 5% of any particle phase concentration and less than 10% of any gas phase concentration for all samplers. Slopes from the FA plots, shown in Fig. 6a, were close to the expected value of 1 (FA mr D  0:9). However, the FA sampler consistently overestimated Kp ; resulting in an overestimation of br by about 1.2. Gas adsorption to the Teflon filter is probably responsible for this error. Adsorption tests, similar to those conducted elsewhere

ARTICLE IN PRESS J. Volckens, D. Leith / Atmospheric Environment 37 (2003) 3177–3188

3183

Fig. 6. log Kp vs. log p0L organized by sampler type for alkane aerosol tests: (a) FA, (b) FFA, (c) DFA, (d) EA. Error bars are shown when the standard deviation of the measurement is larger than the data point.

(Volckens and Leith, 2002a), showed that the adsorption capacity of Zeflour Teflon filters used in the FA sampler was more than two times higher than the adsorption capacity of the Teflon Fiberfilm filters used in the FFA and DFA samplers. Furthermore, alkanes appear to have a high affinity for the Zeflour filter, as measurable quantities of every gas-phase species were detected on this filter type, down to C12. The FFA sampler, Fig. 6b, consistently underestimated mr, with an average value near 2. As a result, values of br were overestimated by a factor of 2 compared to the expected value of 10.1. The FFA sampler produced the largest error terms (SSE), as seen in Table 1. These errors can be explained if particles that collected on the front filter, FF1, evaporated and adsorbed to the backup, FF2, during sampling. Indeed, Kp could not be determined for the C18–C20 alkanes because more mass was detected on FF2 than on FF1 for these compounds. In such cases, the FF2 correction would produce a negative particle-phase concentration, which is impossible. The log Kp vs. log p0L relationships for the FFA sampler improve significantly when the FF2

correction is ignored (i.e. no adsorption correction is made and mass on the FF2 is treated as if it were captured on the adsorbent). The uncorrected data, termed FFA(U), are provided at the bottom of Table 1, and produce substantially better estimates of mr and br : Interestingly, evaporation artifacts appears to have biased the TFE-coated glass fiber filters, whereas adsorption artifacts biased the TFE–Zeflour filter. However, these filters are distinctly different, although both utilize Teflon. The Zeflour filter in the FA sampler consists of a Teflon membrane on top of a Teflon support pad and weighs approximately 300 mg, whereas the Fiberfilm filter in the FFA sampler is made from borosilicate glass fibers coated with Teflon and weighs approximately 60 mg. The DFA sampler, Fig. 6c, produced good estimates of Kp ; mr ; and br for tests with alkanes. As a result, the DFA sampler reported the lowest error estimates, as seen in Table 1. The mass median aerodynamic diameter of the alkane aerosol was approximately 1.5 mm; particles of this size easily pass through the denuder. Thus, the errors from particle losses in the denuder seen

ARTICLE IN PRESS 3184

J. Volckens, D. Leith / Atmospheric Environment 37 (2003) 3177–3188

Fig. 7. log Kp vs. log p0L for alkanes measured by FA, FFA, DFA, and EA samplers for test conditions 1–4 (see Fig. 2 for details). Error bars are shown when the standard deviation of the measurement is larger than the data point.

with diesel exhaust were negligible here. Particles that collected on the filter behind the denuder readily evaporated, as judged by the high percentage of mass collected on the backup adsorbent. The percentage of evaporated particle mass varied from 2% of the C24 to 100% of the C19, which highlights the importance of a backup adsorbent when sampling semivolatile compounds with a DFA sampler. The EA sampler, Fig. 6d, produced mr estimates of approximately 1.25, resulting in estimates of br that were about 25% lower than expected. This discrepancy is probably caused by evaporation of the C19 and C20 alkanes (Kp TSPo 1); these compounds showed the largest deviation in Kp ; as seen in Fig. 6, which is consistent with the FFA errors. Despite this discrepancy, the EA sampler produced reasonably good estimates of Kp with SSE values only slightly larger than the DFA sampler. Examination of Figs. 6a, b, d shows that the test favoring adsorption, Alkane 2, resulted in higher estimates of Kp than the test favoring evaporation,

Alkane 3, for every compound measured by the FA, FFA, and EA samplers. The adsorption test began with a TSP of 25 mg m3 and ended with a TSP of 75 mg m3; the evaporation test used the same concentration profile but in reverse, as seen in Fig. 2. Although the timeweighted average of log Kp should be the same for these tests, the difference between log (½F=TSP=A) during the adsorptive and evaporative portions of the concentration profiles is about 0.47 log units (i.e. log 75  log 25). The average difference in measured log Kp between alkane tests 2 and 3 was 0.21, 0.41, and 0.37 for the FA, FFA, and EA samplers, respectively. This observation indicates that aerosol collected during the latter half of the sampling duration affects the measured Kp more than aerosol collected at the beginning. Such behavior is expected from time-integrated methods when the sample remains in contact with the flowing airstream, as collected particles will constantly attempt to re-establish equilibrium with aerosol that flows into the sampler (Pankow and Bidleman, 1992). Therefore, these samplers do not report an average measure of Kp ; but rather

ARTICLE IN PRESS J. Volckens, D. Leith / Atmospheric Environment 37 (2003) 3177–3188

3185

Table 2 Particle- and gas-phase concentration data for alkane and diesel exhaust aerosol experiments (blank-corrected) Measured concentration (mg m3) C12

C13

C14

C15

C16

C17

C18

C19

C20

C21

C22

C23

C24

Alkane 1 Particle-phase EA 1 EA 2 FFA 1A FFA 2A FFA 1B FFA 2B FA 1 FA 2 DFA 1 DFA 2

NDa ND ND ND ND ND ND ND ND ND

ND ND ND ND ND ND ND ND ND ND

ND ND ND ND ND ND 0.11 LOD ND ND

ND ND ND LOD ND ND 0.13 0.16 ND ND

LODb ND LOD LOD ND LOD 0.27 0.22 ND ND

LOD LOD ND 0.25 LOD LOD 0.42 0.40 ND ND

LOD LOD 0.14 0.25 0.22 0.46 1.47 1.97 ND ND

LOD LOD 0.28 0.25 0.46 0.45 2.05 2.61 ND ND

0.23 0.17 0.73 1.11 1.13 2.03 2.67 4.72 0.19 0.34

2.72 2.72 3.55 4.49 3.22 3.90 6.41 8.37 2.59 2.58

8.67 8.78 11.26 11.55 2.71 2.96 12.09 13.07 10.62 10.25

10.30 11.37 12.83 16.05 0.69 0.75 15.49 16.99 15.44 14.96

11.74 13.31 14.13 17.56 0.16 0.17 16.52 17.96 17.59 16.92

Gas-phase EA 1 EA 2 FFA 1 FFA 2 FA 1 FA 2 DFA 1A DFA 2A DFA 1B DFA 2B

66.72 76.19 81.99 73.55 75.61 77.60 81.90 74.11 ND ND

14.39 16.13 17.76 16.25 15.59 27.61 17.03 16.04 LOD LOD

10.76 9.07 11.11 10.81 11.26 10.09 10.80 10.79 LOD 0.13

6.29 5.64 6.89 6.96 6.64 6.84 6.72 6.72 0.15 0.25

9.13 7.65 9.39 9.69 9.29 8.72 9.65 9.33 0.29 0.24

7.07 5.98 7.38 7.42 6.71 6.77 7.40 7.31 0.35 0.38

11.35 10.13 10.84 13.85 10.89 12.48 15.88 11.31 0.73 0.57

5.48 7.70 5.41 7.92 4.14 6.10 5.50 8.42 0.70 0.74

4.20 6.08 2.59 6.79 1.85 3.61 3.58 6.59 1.84 1.87

3.36 4.58 1.08 1.83 1.60 2.17 1.93 4.33 3.60 3.10

3.36 2.87 ND LOD 1.00 0.66 1.06 1.83 2.04 1.80

0.66 0.67 ND ND 0.23 ND 0.48 0.73 LOD LOD

ND ND ND ND ND ND LOD 0.14 ND ND

Alkane 2 Particle-phase EA 1 EA 2 FFA 1A FFA 2A FFA 1B FFA 2B FA 1 FA 2 DFA 1 DFA 2

ND ND ND ND ND ND ND ND ND ND

ND ND ND ND ND ND LOD LOD ND ND

ND ND ND ND ND ND LOD LOD ND ND

ND ND ND ND ND ND 0.17 0.14 ND ND

ND ND LOD LOD LOD 0.22 0.32 0.46 ND ND

ND ND ND ND LOD 0.50 0.57 0.52 ND ND

LOD 0.10 LOD 0.22 0.28 0.87 2.00 1.81 ND ND

ND ND 0.33 0.30 0.58 0.73 2.08 1.95 LOD LOD

0.64 0.59 1.21 1.31 1.50 1.88 3.22 3.45 0.13 0.19

3.01 2.94 5.12 5.69 3.24 3.96 6.99 7.87 3.93 3.91

7.62 8.12 11.49 12.50 1.62 2.06 12.06 12.32 10.87 10.72

9.94 10.34 15.00 17.87 0.53 0.37 15.50 15.26 14.02 13.99

11.03 11.25 15.74 18.60 LOD LOD 15.92 16.30 15.23 15.11

Gas-phase EA 1 EA 2 FFA 1 FFA 2 FA 1 FA 2 DFA 1A DFA 2A DFA 1B DFA 2B

40.95 47.89 38.94 37.64 52.92 56.37 54.94 48.37 ND ND

13.81 18.11 14.59 13.52 16.90 18.89 18.91 17.27 ND ND

9.30 8.64 11.07 11.44 11.60 12.32 11.42 12.07 ND ND

5.07 4.51 6.17 6.63 6.66 6.70 6.32 6.72 LOD LOD

6.65 6.22 8.11 8.42 8.42 8.31 8.56 8.61 0.17 0.27

5.04 5.04 6.23 6.63 6.52 6.39 6.51 6.89 0.16 LOD

7.00 6.57 9.49 9.91 8.47 8.31 8.99 9.72 ND ND

3.97 3.67 3.82 4.16 3.32 3.45 4.51 5.15 0.72 0.72

2.48 3.09 1.62 2.76 1.39 2.10 2.29 3.65 1.66 1.53

2.40 2.81 0.67 0.82 1.25 1.46 2.08 2.84 2.69 2.66

1.39 1.58 ND ND 0.62 0.46 2.01 1.62 1.13 1.50

0.47 0.43 ND ND 0.20 ND 1.79 0.81 ND 0.55

ND ND ND ND LOD ND 1.21 0.23 ND 0.25

Alkane 3 Particle-phase EA 1 EA 2 FFA 1A FFA 2A FFA 1B FFA 2B FA 1 FA 2 DFA 1 DFA 2

ND ND ND ND ND ND LOD LOD ND ND

ND ND ND ND ND ND LOD LOD ND ND

ND ND ND ND ND ND 0.11 LOD ND ND

ND ND ND ND ND ND 0.11 LOD ND ND

ND ND ND ND ND ND LOD LOD ND ND

LOD LOD LOD LOD LOD ND 0.30 0.24 ND ND

ND LOD LOD 0.18 0.17 0.17 0.94 0.86 ND ND

ND LOD 0.19 0.14 0.32 0.25 1.53 1.30 ND ND

0.23 0.33 0.39 0.49 0.77 0.82 2.59 2.63 0.12 ND

1.41 1.65 3.37 3.31 3.01 3.07 6.40 6.41 2.66 2.84

6.31 7.06 10.29 10.16 2.33 2.70 11.55 10.90 11.32 10.41

8.80 9.62 15.62 14.54 0.89 1.47 13.83 15.08 16.19 14.13

10.05 10.89 16.88 16.35 0.35 1.53 14.65 15.43 18.77 16.21

Gas-phase EA 1 EA 2

55.84 56.55

9.27 16.64

12.24 9.67

21.99 5.09

1.80 0.14

2.75 3.99

5.61 5.50

3.83 3.48

3.68 4.01

3.35 3.56

1.93 1.73

1.28 0.83

0.79 0.41

ARTICLE IN PRESS J. Volckens, D. Leith / Atmospheric Environment 37 (2003) 3177–3188

3186 Table 2 (continued)

Measured concentration (mg m3) C12

C13

C14

C15

C16

C17

C18

C19

C20

C21

FFA 1 FFA 2 FA 1 FA 2 DFA 1A DFA 2A DFA 1B DFA 2B

63.95 66.25 77.64 67.48 64.33 67.12 ND LOD

21.08 19.49 22.82 20.43 19.08 20.54 ND LOD

12.33 11.70 12.42 11.50 12.54 11.65 ND LOD

6.64 6.33 6.62 6.02 6.64 6.25 ND LOD

0.22 0.16 0.29 0.11 0.13 0.15 ND LOD

5.41 5.18 5.46 4.43 5.65 4.96 ND 0.11

7.50 8.14 6.69 5.74 7.88 7.15 ND ND

4.18 4.10 3.32 3.06 4.24 3.79 0.73 0.72

3.22 3.65 2.37 2.38 2.45 2.35 2.84 2.52

1.42 1.30 1.60 1.75 1.37 1.50 4.76 3.62

Alkane 4 Particle-phase EA 1 EA 2 FFA 1A FFA 2A FFA 1B FFA 2B FA 1 FA 2 DFA 1 DFA 2

ND ND ND ND ND ND LOD LOD ND ND

ND ND ND ND ND ND LOD LOD ND ND

ND ND ND ND ND ND LOD LOD ND ND

ND ND ND ND ND ND 0.12 0.14 ND ND

ND ND ND ND ND ND 0.29 0.25 ND ND

LOD LOD ND LOD LOD LOD 0.46 0.41 ND ND

0.22 0.18 0.13 0.26 0.22 0.36 1.36 1.46 ND ND

LOD LOD 0.25 0.26 0.48 0.43 2.08 1.93 ND ND

0.54 0.56 0.91 0.91 1.74 1.80 2.83 2.69 0.24 0.28

Gas-phase EA 1 EA 2 FFA 1 FFA 2 FA 1 FA 2 DFA 1A DFA 2A DFA 1B DFA 2B

44.48 48.29 63.39 56.06 54.33 59.64 61.62 58.12 0.11 LOD

16.48 16.09 22.28 19.84 19.06 19.68 21.15 20.13 LOD LOD

12.76 13.66 17.58 16.25 16.70 16.45 17.97 16.37 LOD LOD

7.47 8.05 10.68 9.65 10.52 10.64 10.50 10.12 LOD LOD

7.61 8.07 10.87 9.70 10.25 10.22 10.46 10.40 0.17 0.17

5.23 5.97 7.91 6.97 7.44 7.23 7.68 7.45 0.24 0.26

7.88 9.02 12.91 12.47 20.17 10.73 12.24 15.05 5.79 6.46

5.12 5.03 6.64 5.92 5.46 5.51 6.45 6.19 1.22 1.36

3.62 4.33 4.03 4.51 3.61 3.48 3.47 3.37 2.59 2.89

ACEc

FLU

PHN

FLA

PYR

BAA

Diesel Particle-phase EA 1 EA 2 FFA 1A FFA 2A FFA 1B FFA 2B FA 1A FA 2A DFA 1 DFA 2

0.42 0.47 0.55 0.48 0.02 0.02 0.41 0.45 0.03 0.03

2.65 3.00 3.30 2.84 0.23 0.27 2.94 2.79 0.12 0.11

14.05 15.71 15.76 13.91 1.29 1.54 14.77 13.31 4.83 3.95

4.91 5.62 5.59 5.14 0.22 0.27 5.28 4.88 4.20 3.55

7.01 8.09 8.16 7.41 0.30 0.35 7.68 8.17 6.21 5.28

1.56 1.76 1.80 1.69 0.05 0.06 1.66 1.63 1.47 1.15

Gas-phase EA 1 EA 2 FFA 1 FFA 2 FA 1A FA 2A DFA 1A DFA 2A DFA 1B DFA 2B

2.29 2.97 2.94 2.46 2.98 2.82 4.05 2.78 0.10 0.03

5.61 7.61 6.97 6.99 7.33 6.82 10.72 7.64 0.27 0.16

6.83 7.39 6.04 6.34 6.97 6.48 21.90 16.25 3.08 2.08

0.49 0.23 0.06 0.05 0.17 0.21 2.26 2.15 0.26 0.16

0.63 0.27 0.06 0.06 0.19 0.25 2.92 2.84 0.31 0.20

0.10 0.02 0.011 0.009 0.003 0.004 0.45 0.47 0.03 0.02

a

C22

C23

C24

0.29 0.13 0.75 0.75 0.92 0.85 2.07 1.51

0.48 ND LOD 0.35 0.31 0.49 0.30 0.18

0.69 ND ND 0.21 LOD 0.15 ND ND

3.18 3.12 4.47 4.46 3.51 3.79 7.14 7.01 3.62 3.60

8.14 7.39 11.93 11.53 2.14 2.62 12.16 12.48 11.94 11.30

9.54 9.69 15.61 16.06 0.39 0.54 16.19 15.58 15.97 15.47

10.61 10.46 16.95 17.28 ND LOD 16.63 16.19 17.90 17.15

2.81 3.15 1.16 1.43 2.05 2.48 1.94 2.03 3.35 3.42

1.44 1.29 LOD 0.21 0.89 1.09 1.32 1.20 1.85 1.74

0.86 0.61 ND 0.19 0.32 0.66 1.69 0.76 0.80 0.35

0.59 0.15 ND 0.20 LOD 0.54 1.87 0.43 0.73 LOD

‘ND’ not detected by GC-MS. ‘LOD’ GC-MS peak observed, but value is below the limit of detection. c Abbreviations for PAHs in diesel exhaust: acenaphthene (ACE), fluorene (FLU), phenanthrene (PHN), fluoranthene (FLA), pyrene (PYR), and benz[a]anthracene (BAA). b

ARTICLE IN PRESS J. Volckens, D. Leith / Atmospheric Environment 37 (2003) 3177–3188

a value that is weighted towards concentrations collected near the end of the sampling period. Samples collected in a DFA sampler are constantly in a state of non-equilibrium, because gas-phase compounds are quickly stripped from the airstream and collected particles are continually evaporating. Consequently, the DFA sampler is not affected by transient changes in concentration. Comparison of Figs. 6 and 7 shows that the variability in measured Kp is larger between methods than within any given method (Po0:001). In some cases the measured Kp varies over an order of magnitude between methods, whereas, the errors bars for each Kp measured within a given methods are usually within the size of the data point, as seen Figs. 5–7. Such large variations between methods emphasize the difficulties associated with sampling semivolatile aerosols, as well as the importance of selecting an appropriate method for a given sampling situation.

4. Conclusions For the conditions tested here, no sampler outperformed the others in every case. The FA sampler produced accurate estimates of Kp for PAHs in diesel exhaust when considerable mass collected on the filter and sampling time was short, but was biased by adsorption artifacts during tests with alkane aerosol when concentrations were more dilute. The FFA sampler also performed reasonably well in tests with diesel exhaust but was considerably biased by particle evaporation during tests with alkane aerosol. The DFA sampler appears to produce reasonable estimates of Kp when gas breakthrough is avoided and when particle losses to the denuder are negligible. However, particle losses to denuders vary with aerodynamic diameter and may be difficult to resolve as the organic carbon content of atmospheric aerosols varies with particle size (Subramanyam et al., 1994; Kaupp and McLachlan, 2000). Electrostatic precipitators are less prone to adsorption biases but not evaporation bias. Although chemical artifacts from the ESP corona did not bias the Kp coefficients measured here, such a bias may be substantial when a measurement of absolute concentration is desired.

Acknowledgements The authors would like to thank Dr. Michael Tolocka of the University of Delaware for his help in developing the test procedures with diesel exhaust. This work was supported in part by grant T32ES07018 from the National Institute for Environmental Health Sciences and a US EPA-NNEMS fellowship (U-91567401-

3187

0).Appendix AThe concentration data for alkane and diesel exhaust aerosols (blank corrected) are presented in Table 2.

References Bidleman, T.F., 1988. Wet and dry deposition of organic compounds are controlled by their vapor-particle partitioning. Environmental Science and Technology 22 (4), 361–367. Cardello, N., et al., 2002. Technical note: performance of a personal electrostatic precipitator particle sampler. Aerosol Science and Technology 36 (2), 162–165. Gundel, L.A., et al., 1995. Direct determination of the phase distributions of semi-volatile polycyclic aromatic hydrocarbons using annular denuders. Atmospheric Environment 29, 1719–1733. Jang, M., et al., 1997. A thermodynamic approach using group contribution methods to model the partitioning of semivolatile organic compounds on atmospheric particulate matter. Environmental Science and Technology 31, 2805–2811. Kamens, R.M., Odom, J.R., Fan, Z., 1995. Some observations on times to equilibrium for semivolatile polycyclic aromatic hydrocarbons. Environmental Science and Technology 29 (1), 43–50. Kaupp, H., McLachlan, M.S., 2000. Distributuion of dibenzop-dioxins and polycyclic aromatic hydrocarbins (PAHs) within the full size range of atmospheric particles. Atmospheric Environment 34, 78–83. Lee, K.W., Ramamurthi, M., 1993. Filter collection. In: Willeke, K., Baron, P. (Eds.), Aerosol Measurement: Principles, Techniques, and Applications. Wiley, New York. Mader, B.T., Pankow, J.F., 2001. Gas/solid partitioning of semivolatile organic compounds (SOCs) to air filters 3. An analysis of gas adsorption artifacts in measurements of atmospheric SOCs and organic carbon (OC) when using Teflon membrane filters and quartz fiber filters. Environmental Science and Technology 35 (17), 3422–3432. Pankow, J.F., 1987. Review and comparative analysis of the theories on partitioning between the gas and aerosol particulate phases in the atmosphere. Atmospheric Environment 21 (11), 2275–2283. Pankow, J.F., 1994. An absorption model of the gas/particle partitioning of organic compounds in the atmosphere. Atmospheric Environment 28 (2), 185–188. Pankow, J.F., Bidleman, T.F., 1992. Interdependence of the slopes and intercepts from log–log correlations of measured gas-particle partitioning and vapor pressure—I. Theory and analysis of available data. Atmospheric Environment 26A (6), 1071–1080. Subramanyam, V., et al., 1994. Gas-to-particle partitioning of polycyclic aromatic hydrocarbons in an urban atmosphere. Atmospheric Environment 28 (9), 3083–3091. Turpin, B.J., Saxena, P., Andrews, E., 2000. Measuring and simulating particulate organics in the atmosphere: problems and prospects. Atmospheric Environment 34 (18), 2983–3013. Volckens, J., Leith, D., 2002a. Filter and electrostatic samplers for semivolatile aerosols: physical artifacts. Environmental Science and Technology 36 (21), 4613–4617.

ARTICLE IN PRESS 3188

J. Volckens, D. Leith / Atmospheric Environment 37 (2003) 3177–3188

Volckens, J., Leith, D., 2002b. Electrostatic sampler for semivolatile aerosols: chemical artifacts. Environmental Science and Technology 36 (21), 4608–4612.

Ye, Y., et al., 1991. Particle transmission characteristics of an annular denuder ambient sampling system. Journal of Aerosol Science 14, 102–111.