Environ. Sci. Technol. 2002, 36, 4613-4617
Filter and Electrostatic Samplers for Semivolatile Aerosols: Physical Artifacts JOHN VOLCKENS* AND DAVID LEITH Department of Environmental Sciences and Engineering, University of North Carolina at Chapel Hill, CB 7431 Rosenau Hall, Chapel Hill, North Carolina 27599-7431
Adsorptive and evaporative artifacts often bias measurements of semivolatile aerosols. Adsorption occurs when the sampling method disrupts the gas-particle partitioning equilibrium. Evaporation occurs because concentrations of semivolatiles are rarely constant over time. Filtration is subject to both adsorptive and evaporative artifacts. By comparison, electrostatic precipitation reduces these artifacts by minimizing the surface area of collected particles without substantially disrupting the gasparticle equilibrium. The extent of these artifacts was determined for filter samplers and electrostatic precipitator samplers for semivolatile alkane aerosols in the laboratory. Adsorption of gas-phase semivolatiles was lower in electrostatic precipitators by factors of 5-100 compared to the filter method. Particle evaporation from the electrostatic sampler was 2.3 times lower than that from TFE-coated glassfiber filters. Use of a backup filter to correct for compoundspecific adsorption artifacts can introduce positive or negative errors to the measured particle-phase concentration due to competition among the adsorbates for available adsorption sites. Adsorption of evaporated particles from the front filter onto the backup filter increased the measured evaporative artifact by a factor of 1.5-2.
Introduction Semivolatile organic compounds (SOCs) in air exist simultaneously in both gas and particle phases. At equilibrium, SOCs distribute mass between the two phases according to a dimensionless partitioning ratio, Kp
Kp )
cp cg
(1)
where cp is the concentration of SOC in the particle phase (ng/m3) and cg is the concentration in the gas phase (ng/m3) (1). The ratio Kp depends largely on a compound’s subcooled liquid vapor pressure, which is strongly related to temperature, and to a lesser extent on available surface area, particle composition, and relative humidity (2-6). Any departure from equilibrium results in mass transfer between the phases until equilibrium is re-established. Mass exchange occurs via absorption, adsorption, and evaporation. The rate of transfer between the phases can be rapid or slow depending on the individual SOC and the transfer mechanism (7-9). Consequently, a sampling method to quantify Kp accurately must * Corresponding author phone: (919)966-7337; fax: (919)966-7911; e-mail:
[email protected]. 10.1021/es020711s CCC: $22.00 Published on Web 09/28/2002
2002 American Chemical Society
not substantially disturb the conditions governing gas-particle equilibrium. Traditionally, Kp is determined from simultaneous, timeintegrated measurements of cp and cg. The most common sampling apparatus is a filter used in conjunction with an adsorbent downstream. The concentration of particles is determined from the catch on the filter, whereas the gasphase concentration is determined from the catch on the adsorbent. Filters are prone to evaporative artifacts because particles that are trapped within a filter present a large contact area to the flowing airstream. This condition enhances particle volatilization when corresponding gas-phase concentrations fall below equilibrium levels. This negative bias causes underestimation of particle-phase concentration and overestimation of gas-phase concentration. Diurnal variations in temperature and aerosol concentration will promote such losses (10). A positive bias may occur during sampling because the high specific surface area of filters allows gas-phase semivolatiles to adsorb onto the filter surface. In this situation, adsorbed SOC will be measured incorrectly as particle-phase mass. The magnitude of the adsorption artifact may vary with face velocity, filter type, and the influent concentration of gas-phase organic carbon and is reported to account for 10-30% of the organic carbon measured in the particulate phase (11, 12). A second filter is often placed behind the first to form a filter-filter system to correct the adsorption artifact (13). Particle-phase SOCs collect only on the front filter, while gas-phase SOCs can adsorb to both filters. Assuming that equal amounts of gas-phase SOCs adsorb to each filter, the mass measured on the second filter can be subtracted from that on the front filter. However, if the gas to filter partitioning is efficient, then only after the front filter (FF1) has equilibrated will the backup filter (FF2) begin to adsorb gas-phase SOCs (1). Although filter-filter sampling trains attempt to correct for an adsorption artifact, they cannot account for particle-phase mass that evaporates from the front filter during sampling. Furthermore, if particles evaporate from the front filter and readsorb to the second filter, then the evaporative artifact would be increased by a factor of 2 when the adsorption correction is made. Electrostatic precipitation has been proposed as an alternative to filter-based sampling. Electrostatic precipitators (ESPs) charge particles using corona discharge and separate them from the airstream using electric force. Particles collect at the flow boundary on a conductive foil substrate where they aggregate. This aggregation reduces exposed particle surface area and thus should lower the rate of particle evaporation when the gas-phase concentrations of SOC drop below equilibrium. The adsorption artifact should also be reduced because the substrate surface area, approximately 35 cm3, is substantially less than the surface of the fibers in TFE and glass fiber filters. Unfortunately, the extent to which these artifacts bias an ESP measurement is unknown. Although the ESP collection substrate has a significantly lower surface area than a filter, gas-phase semivolatiles may still adsorb there. Adsorbates may also have a selective affinity for the type of surface, this is true for different filter types (12) and with the amount of semivolatile material already collected on the substrate. In addition to adsorption, collected particles may evaporate from the substrate during sampling. Therefore, the degree to which these artifacts bias ESP measurements of semivolatile aerosols merits examination. VOL. 36, NO. 21, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
4613
TABLE 1. Composition of Adsorption/Evaporation Mixtures substrate cocktail
mole fraction
vapor cocktail
mole fraction
dodecane, C12 tetradecane, C14 hexadecane, C16 octadecane, C18 eicosane, C20 docosane, C22 tetracosanec24
1/7 1/7 1/7 1/7 1/7 1/7 1/7
pentadecane, C15 heptadecane, C17 nonadecane, C19 heneicosnae, C21 tricosane, C23
1/5 1/5 1/5 1/5 1/5
FIGURE 1. Schematic for physical artifact experiments. The objective of the present manuscript is to investigate the physical artifacts associated with electrostatic precipitation and with filters (i.e., adsorption and evaporation). Laboratory tests were conducted to determine the magnitude of adsorptive and evaporative artifacts for a homologous series of SOCs whose vapor pressures span the semivolatile range. Filter-filter samplers were tested alongside ESPs to evaluate and compare the biases associated with each (14).
Experimental Section Adsorptive and evaporative artifacts were evaluated for filterfilter and ESP samplers under three conditions: (i) adsorption to clean, unused substrates, (ii) adsorption to substrates that had previously sampled SOC, and (iii) evaporation from substrates that had previously been used to sample SOC aerosol. The first condition was evaluated separately, while the latter two were tested simultaneously. Four sets of experiments were conducted in random order for each condition with durations of 20, 80, 160, and 260 min, for a total of 16 samples per condition. This procedure was followed to evaluate artifacts as a function of time. Two alkane mixtures of dissimilar composition were created; the compositions of these mixtures are listed in Table 1. Chemicals had a minimum purity of 99% (Sigma-Aldrich Chemical Co., St. Louis, MO). The first, or “substrate mixture”, consisted only of alkanes with an even number of carbon atoms. The second, or “vapor mixture”, consisted only of odd numbered alkanes. The substrate mixture was used to evaluate evaporative artifacts, whereas the vapor mixture was used to evaluate adsorption artifacts. Data were analyzed with Stata software (Intercooled Stata 6.0, Stata Corp., College Station, TX) using linear regression and paired t-tests. Significance was determined at R ) 0.05, and p-values (P) are reported where appropriate. (i) Adsorption to Clean Substrates. In these experiments, the adsorption artifact was tested by passing gas-phase SOCs over the ESP and filter-filter samplers and measuring the amount of adsorbed mass. A schematic of the adsorption apparatus is shown in Figure 1. SOC vapors were generated by passing purified, compressed air (UNC Scientific Supply, Chapel Hill, NC) through a 250 mL column packed with 5 mm glass beads (Fisher Scientific, Pittsburgh, PA) that were 4614
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 36, NO. 21, 2002
previously coated with the vapor mixture. The packed column was immersed in a constant temperature water bath maintained at 22 °C to sustain a constant concentration of SOC vapor in the airstream. Vapor laden air was passed into a mixing chamber and diluted by a factor of 10 to ensure that the vapor concentrations were well below saturation. A final filter was placed immediately upstream of the sampling manifold to remove any particles from the airstream. The vapor stream was then passed through two ESP and two filter-filter samplers operated in parallel for the four time periods mentioned above. The ESP samplers contained clean, unused aluminum foil substrates (Reynolds Metals Co, Richmond, VA) and measured 60 × 80 mm; the filter substrates were 47 mm TFE-coated glass fiber filters (Pallflex Fiberfilm T60A20, Pall Gelman Sciences, Ann Arbor, MI). Vapor that did not adsorb to the substrate was collected on 40 cm, XAD-4 coated annular denuders (URG, Chapel Hill, NC) located immediately behind each sampler. Substrates were analyzed via GC-MS as described elsewhere (14) with the exception that samples were concentrated prior to analysis with a nitrogen blowdown method (8) and no gravimetric analysis was performed. Under these conditions, the limit of quantification was approximately 10 ng per sample. Power to the ESP was turned off during these tests to avoid any bias associated with chemical artifacts (14). (ii) Adsorption to Used Substrates with (iii) Simultaneous Evaporation. In these experiments, adsorptive and evaporative artifacts were tested simultaneously by repeating the adsorption artifact tests (i) described above with substrates that had previously been used to sample SOC particles. Even-numbered alkanes from the substrate mixture were nebulized in a 1.0 m3 chamber and sampled by four ESP samplers and four filter-filter samplers operated in parallel for 1 h at a concentration of approximately 100 µg/m3. These conditions were repeated for each test to ensure that all samplers were coated with approximately the same amount of mass from the substrate mixture. Details of the aerosol generation and sampling system are presented elsewhere (14). Immediately after sampling, two ESP and two filterfilter samplers were transferred to 15 mL glass centrifuge tubes, spiked with an internal standard, and desorbed with pentane (Fisher Scientific, Pittsburgh, PA) for later analysis. Results from these samplers were used as controls to determine the mass of particles comprised of even-numbered alkanes that collected on both filters and ESPs. The remaining samplers (two ESP and two filter-filter) were placed in the adsorption apparatus to undergo a period of combined particle evaporation and vapor adsorption. While odd numbered alkanes from the vapor mixture could adsorb to the samplers, the previously collected droplets of even-numbered alkanes from the substrate mixture could evaporate. The adsorption artifact was determined by the odd numbered alkanes detected on each substrate, because odd numbered alkanes could originate only from the vapor mixture. Similarly, the evaporative artifact was determined from the difference between the even numbered alkanes detected on substrates that underwent adsorption/evaporation vs the controls that did not.
Results and Discussion (i) Adsorption to Clean Substrates. A substantial difference in the adsorption artifact between the front filter (FF1) of a filter-filter sampler versus the aluminum foil substrate of an ESP can be seen in Figure 2. The adsorption of gas-phase alkanes to the FF1 appears to follow a Langmuiran isotherm with rapid initial adsorption that tapered off over time, as shown in Figure 2. In contrast, only small amounts of C17, C19, and C23 adsorbed to the ESP substrate. The total mass adsorbed to the ESP substrate, 0.035 ( 0.015 µg, did not significantly increase or decrease over time (P ) 0.893). Blank
FIGURE 2. Mass adsorbed to clean, unused ESP and filter-filter (Data for the front filter shown.) substrates as a function of time and alkane composition. Error bars represent one standard deviation.
FIGURE 3. Mass adsorbed to ESP and filter-filter (Data shown for front filter.) substrates that had previously sampled alkane aerosol. Error bars represent one standard deviation. values were low for all tests, with an average of 0.013 and 0.006 µg and a maximum of 0.051 and 0.022 µg detected for all compounds on filter and ESP substrates, respectively. Alkane concentrations exiting the vapor column were relatively consistent between replicate tests: 650 ( 120 µg/ m3 for C15, 73 ( 5 µg/m3 for C17, 6.3 ( 1.4 µg/m3 for C19, 0.62 ( 0.17 µg/m3 for C21, and 0.37 ( 0.09 µg/m3 for C23. (ii) Adsorption to Used Substrates. The coating process consistently applied approximately 23.5 ( 2.2 µg of particle mass from the substrate mixture of even numbered alkanes to each substrate. Over 99.5% of the mass on each substrate was accounted for by the higher molecular weight alkanes C20 (6.5% ( 0.9), C22 (46.2% ( 0.8), and C24 (46.8% ( 0.8). The adsorption artifact of odd-numbered alkanes to the coated substrates is shown in Figure 3. Several important differences can be seen between the uncoated and coated tests, Figures 2 and 3, respectively. First, the total mass adsorbed to the coated FF1 substrate did not increase over time (P ) 0.899), indicating that the coated FF1 was probably saturated with
FIGURE 4. Mass evaporated from ESP and filter-filter (Data shown for front filter.) substrates as a function of time and alkane composition. Error bars represent one standard deviation. SOC. Second, the amount of odd-numbered alkanes that adsorbed to the coated FF1 substrate was less than that for the uncoated adsorption test. However, this finding was expected since some even numbered particle-phase and adsorbed gas-phase SOCs likely reduced the number of available adsorption sites. Third, considerably more mass adsorbed to the coated ESP substrate than the uncoated one. Fourth, the mass adsorbed to the ESP substrate increased linearly over time, as seen in Figure 3. The latter two points can be explained if different mechanisms governed the gasto-substrate partitioning for the coated and the uncoated tests. Absorption to a coated ESP substrate is consistent with the collection of liquid particles on the ESP substrate and their coalescence into a thin film. On the other hand, adsorption is likely to govern gas-to-substrate partitioning on an uncoated metal surface. (iii) Evaporation from Used Substrates. The particlephase mass evaporated from the FF1 and ESP substrates, as a function of time and alkane component, are shown in Figure 4. For all tests, the amount of mass evaporated from the ESP is significantly lower than from the filter (P ) 0.0039). The evaporation rates for both samplers appear linear within the time range shown. A simple linear regression gave estimates of the evaporation rates from the FF1 and ESP substrates of 0.0143 and 0.0062 µg/min, respectively. Thus, the rate of evaporation from the filter is 2.3 times as high as the rate of evaporation from the ESP substrate for the conditions tested. This result is consistent with findings from previous sampling studies of semivolatile metalworking fluids in occupational environments where SOC volatilization from filters is expected (15, 16). The percentage of mass evaporated from each substrate as a function of sampler type, alkane number, and time is provided in Table S1 (Supporting Information). Filter-Filter Correction. Backup filters produced serious errors when correcting for compound specific adsorption artifacts, as seen in Figure 5. The adsorption correction error, Ea, was calculated by dividing the difference in mass adsorbed to the front and back filters by the mass adsorbed to the front filter
Ea,i )
FF1i - FF2i FF1i
(2)
where i represents the SOC compound of interest. A positive VOL. 36, NO. 21, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
4615
FIGURE 5. Percent difference in adsorbed mass, Ea, between backup and front filter for filter-filter sampler after coated adsorption tests. Error bars represent one standard deviation. Ea means the backup filter underestimated the adsorption artifact; a negative E refers to an overestimation. Mader et al. report that backup filter corrections are possible only after both the FF1 and FF2 have become saturated with a given SOC, which is true for single-component adsorption (1). However, the situation becomes more complicated for multicomponent adsorption when adsorbates compete for available adsorption sites. An example of competition between adsorbates can be seen in Figure 6, showing compound specific adsorption to the FF1 over time. When a filter becomes saturated with adsorbed vapor, new adsorbates will adsorb only when existing sites become available. After 20 min, the FF1 is saturated mainly with the more volatile C17 and C19 compounds, whose concentrations are substantially higher in the vapor stream. However, these compounds are slowly displaced from FF1 by the less volatile C21 and C23. These higher molecular weight compounds have appreciably longer adsorption half-lives on the filter surface. Atkins defines the temperature-dependent half-life, t1/2 (s), for physiosorption as
t1/2 ) τ0eEd/RT
(3)
where τ0 is 10-12 s and Ed is the enthalpy of desorption from the surface (kJ/mol) (17). If Ed is assumed to be approximately equal to the enthalpy of vaporization of the pure compound, then the average residence time of SOCs adsorbed to the filter surface can be estimated using eq 3. Using Ed of 86 and 166 kJ/mol (18), the adsorption half-life is on the order of 20 min for C17 and is greater than seven years for C23. At 260 min, the FF1 is enriched with C21 and C23 and nearly depleted of C17 because the adsorption half-life of C17 is considerably shorter than that for C21 or C23. This phenomenon occurs despite the fact that the air concentration of C17 remained over 100 times higher than the C21 or C23 compounds throughout the test. At this point, more C17 and C19 are on the FF2 than the FF1; thus, as shown in Figure 5, the backup filter is overcorrecting the adsorption artifact for these compounds. A FF2 adsorption correction with multicomponent adsorption is possible only if one adsorbate can displace all other types and thereby occupy every adsorption site on both the FF1 and FF2. For these tests, the most strongly adsorbing compound was C23. If the volume of air required to equilibrate both 4616
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 36, NO. 21, 2002
FIGURE 6. Competitive adsorption of gas-phase SOCs to a saturated filter substrate over time. Error bars represent one standard deviation. front and backup filters, 2Vmin (1), is expressed as a function of the filter adsorption capacity, CF (ng), the gas-phase concentration, cg (ng/m3), and the gas to filter adsorption efficiency, η, is taken as unity here
Vmin )
CF ηcg
(4)
then the time required to reach equilibrium, t (min), is the ratio of 2Vmin to the sampling flowrate, Q (m3/min)
t)
2CF ηcgQ
(5)
For the conditions tested here, the time for C23 to saturate both filters is approximately 18 days. In ambient or workplace environments where mixtures of semivolatile compounds are also present, the time to saturate both filters could also be quite long relative to the sample duration. The use of a backup filter can also lead to errors during periods of particle evaporation. As particles evaporate from the FF1, they may readsorb to the FF2, thereby increasing the evaporation artifact when the adsorption correction is performed. Such was the case in these tests, with docosane. If F is the mass originally on the FF1 (ng), then the evaporation error Ee can be expressed as
Ee ) -
(e +F ηe)
(6)
where e is the amount of mass evaporated from the FF1 (ng) and η is the gas to filter adsorption efficiency to the FF2 for the compound of interest. Here, the first term in the numerator of the right side represents the mass that actually evaporated from the FF1. The second term in the numerator represents the correction to F intended to account for vapor adsorption. Figure 7 shows Ee for C22 expressed as a percentage, with and without the FF2 correction. Use of the FF2 correction greatly increased the evaporation artifact for C22 due to the high efficiency with which evaporated C22 readsorbed to the backup filter, FF2. This efficiency ranged from 93% at 20 min to 77% at 260 min. Gas-to-filter adsorption efficiencies for gas-phase compounds are provided in Table S2 (Supporting Information). Although the compound-specific adsorption artifacts mentioned here were serious, the total amounts of SOC adsorbed
Acknowledgments The authors would like to thank Russell Wiener and Gary Norris of the U.S. EPA for their support of this research and Maryanne Boundy and the Baity lab of the University of North Carolina for their comments and suggestions. This work was supported in part by grant T32ES07018 from the National Institute for Environmental Health Sciences and a U.S. EPA - NNEMS fellowship (U-91567401-0).
Supporting Information Available Percent of mass evaporated from substrate as a function of compound, sampler type, and time (Table S1) and percent of influent vapor adsorbed to substrate as a function of sampler type, alkane number, and time (Table S2). This material is available free of charge via the Internet at http:// pubs.acs.org.
Literature Cited FIGURE 7. Evaporation error, Ee, for docosane when using a backup filter to correct for adsorption artifacts vs time. Error bars represent one standard deviation. to the front and back filters were not substantially or statistically different during the pure adsorption tests (P > 0.4). Therefore, filter-filter sampling is more appropriate for total organic carbon measurements because the relative amounts of adsorbed organic carbon on the FF1 and the FF2 were equal for these adsorption tests. However, the filterfilter technique cannot account for particle evaporation. Filter-filter sampling trains are not well suited for measuring semivolatile partitioning ratios when the sampled air contains a mixture of SOC. Furthermore, the FF2 correction can introduce either positive or negative errors to the observed particle-phase concentration. The relative importance of these errors will depend on the sampling time, flow rate, SOC concentration, SOC chemical composition, and filter type. Front and backup filter equilibration is unlikely to occur when sampling SOC mixtures because the time required to reach equilibrium is prohibitively long. Evaporative artifacts that occur during long sampling periods will further complicate the measurement as compounds volatilized from particles will compete for available adsorption sites on the FF2. Thus, higher flowrates and long sampling times may not alleviate the adsorption problem. Future work will compare the ESP method to denuderbased sampling methods and evaluate the foil substrate for determination of total organic carbon with thermal optical reflectance.
(1) Mader, B. T.; Pankow, J. F. Environ. Sci. Technol. 2001, 35(17), 3422-3432. (2) Yamasaki, M.; Kuwata, K.; Miyamoto, H. Environ. Sci. Technol. 1982, 16(2), 189-194. (3) Pankow, J. F. Atmos. Environ. 1994, 28(2), 185-188. (4) Jang, M.; Kamens, R. M. Environ. Sci. Technol. 1998, 32(9), 12371243. (5) Pankow, J. F. Atmos. Environ. 1987, 21(11), 2275-2283. (6) Jang, M., et al. Environ. Sci. Technol. 1997, 31, 2805-2811. (7) Kamens, R. M.; Coe, D. L. Environ. Sci. Technol. 1997, 31(6), 1830-1833. (8) Kamens, R. M.; Odom, J. R.; Fan, Z. Environ. Sci. Technol. 1995, 29(1), 43-50. (9) Rounds, S. A.; Pankow, J. F. Environ. Sci. Technol. 1990, 24(9), 1378-1386. (10) Wallace, J. C.; Hites, R. A. Environ. Sci. Technol. 1996, 30(2), 444-446. (11) McDow, S. R.; Huntzicker, J. J. Atmos. Environ. 1990, 24A(10), 2563-2571. (12) Turpin, B. J.; Huntzicker, J. J. Atmos. Environ. 1994, 28(19), 30613071. (13) Turpin, B. J.; Saxena, P.; Andrews, E. Atmos. Environ. 2000, 34(18), 2983-3013. (14) Volckens, J.; Leith, D. Environ. Sci. Technol. 2002, 36, 46084612. (15) Raynor, P. C.; Volckens, J.; Leith, D. App. Occ. Environ. Hyg. 2000, 15(1), 90-96. (16) Volckens, J. et al. Am. Ind. Hyg. Assoc. J. 1999, 60(5), 684-689. (17) Atkins, P. Physical Chemistry, 6th ed.; W. H. Freeman and Co.: New York, 1998; p 864. (18) CRC. CRC Handbook of Chemistry and Physics, 78th ed.; Lide, D. R., Ed.; Chemical Rubber Company: Boca Raton, FL, 1997.
Received for review April 26, 2002. Revised manuscript received August 20, 2002. Accepted August 27, 2002. ES020711S
VOL. 36, NO. 21, 2002 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
4617
