Environ. Sci. Technol. 2004, 38, 1471-1479
Simulating Organic Aerosol Formation during the Photooxidation of Toluene/NOx Mixtures: Comparing the Equilibrium and Kinetic Assumption C R A I G A . S T R O U D , * ,†,‡ P A U L A . M A K A R , § DIANE V. MICHELANGELI,† MICHAEL MOZURKEWICH,| DONALD R. HASTIE,| ANDREEA BARBU,| AND JANYA HUMBLE† Departments of Earth and Atmospheric Science and Chemistry, York University, 4700 Keele Street, North York, Ontario, Canada M3J 1P3, and Air Quality Research Branch, Meteorological Service of Canada, 4905 Dufferin Street, Downsview, Ontario, Canada M3H 5T4
Organic compounds contribute an appreciable mass to particulate matter and thus impact the hygroscopic and radiative properties of an aerosol distribution. Being able to predict the chemical and physical properties of aerosols based on their size and composition is critical to assessing their impact on air quality, visibility, and climate change. In this study, a comparison was performed between an equilibrium and a kinetic model for simulating organic aerosol formation during the photooxidation of toluene/NO/ isopropyl nitrite mixtures. Both models used an explicit gasphase toluene scheme (University of Leeds Master Chemical Mechanism version 3.0) and provided a prediction of individual products partitioned to the aerosol phase. After incorporating a heterogeneous wall reaction scheme regenerating NOx from HNO3 and HNO2, the gas-phase scheme was able to simulate the observed toluene decay within 5% and NO decay within 30% for all of the chamber experiments. The models reproduced the general magnitude of the aerosol yields but suggest a weaker trend dependence on aerosol mass loading. A few nonvolatile compounds were predicted to compose the majority of the aerosol-phase mass with multifunctional organic nitrates being the dominant organic aerosol functional group. The hygroscopic diameter growth factor for the organic phase was predicted to be 1.1 at a relative humidity of 79%. We conclude with a list of recommended laboratory experiments to help constrain and validate aerosol process models.
1. Introduction Enhanced fine particulate matter has been observed in environments with substantial photochemical ozone (O3) * Corresponding author phone: (303) 497-1449; fax: (303) 4971477; e-mail:
[email protected]. † Department of Earth and Atmospheric Science, York University. ‡ Present address: Atmospheric Chemistry Division, National Center for Atmospheric Research, 1850 Table Mesa Dr., Boulder, CO 80303. § Meteorological Service of Canada. | Department of Chemistry, York University. 10.1021/es030546w CCC: $27.50 Published on Web 01/22/2004
2004 American Chemical Society
production and, along with O3, has been identified as a component contributing to poor air quality and adverse human health (1). An important contributor to fine particulate matter, especially in urban environments, is secondary organic aerosol (SOA) (2, 3). SOA is defined as organic particular matter formed in situ due to the atmospheric oxidation of organic compounds. In general, precursor reactive hydrocarbons (RH) can react with any of several possible oxidants (OH, O3, and NO3) to form partially oxidized volatile organic compounds (OVOCs) and condensable organic compounds (COCs). The OVOCs can react further to form additional COCs and OVOCs. Thus, several oxidation steps may be involved in forming COCs. In turn, the COCs can be lost by further oxidation, surface deposition, and/or partitioning to the aerosol phase. If existing organic aerosol is present in the atmosphere, the gas-to-particle partitioning process will likely be absorptive with the equilibrium partitioning being strongly dependent on the volatility of the absorbate (4). Hydrocarbons with six or fewer carbons generally do not act as precursors to SOA, because the resultant oxidation products have high-saturation vapor pressures (5). However, studies by Pankow have shown that the interstitial vapor pressures of the COCs do not have to increase above their saturation vapor pressures for condensation to occur (6). This can be explained by the impurity of the condensed phase lowering the surface vapor pressures. A recent study by Jang and Kamens (7) also suggested that heterogeneous reactions effectively lower the surface vapor pressures of some of the products by allowing them to polymerize on the surface of the aerosol. Of the anthropogenic hydrocarbons, aromatics and heavy alkanes have been observed to generate significant SOA in chamber studies (8, 9). Of the 32 most prevalent nonmethane hydrocarbons observed in urban air, seven are aromatics, of which toluene is the most abundant, accounting for 6% of the observed nonmethane carbon concentration (10). Due to the importance of toluene in urban SOA formation, there have been several recent chamber experiments that have focused on quantifying SOA formation following toluene oxidation (11, 12). The goal of this study was 2-fold: (1) to develop and compare results from two explicit aerosol models in simulating organic aerosol yields during the chamber photooxidation of toluene/NOx mixtures and (2) to provide a list of target aerosol compounds to guide future experimental studies. Both models utilized the same gas-phase toluene mechanism; however, they differed in the assumptions used to partition products between the gas and aerosol phases (e.g., equilibrium- vs mass-transfer-limited). Equilibrium partitioning theory, originally developed by Pankow, has been used successfully in lumped chemical models where each reactive hydrocarbon forms two surrogate aerosol products with stoichiometric yields and partitioning coefficients derived from fits to chamber experiments (13-17). Masstransfer-limited partitioning was originally implemented within an explicit chemical model by Kamens et al. (18, 19) in predicting SOA aerosol yields from the reaction of R-pinene with O3. Chemically mechanistic models, such as those examined here, incorporate a detailed aerosol- and gas-phase chemical speciation. Predictions of detailed organic aerosol composition are useful for several reasons: (1) the properties (e.g., refractive index, water uptake) of the aerosol may be calculated on the basis of their explicit composition; (2) a wide range of different atmospheric conditions may be simulated (e.g., varying RH/NOx ratio, relative humidity, VOL. 38, NO. 5, 2004 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
1471
TABLE 1. Initial Conditions for Chamber Experiments (Initiation of Lamps)a exp
toluene (ppmv)
NO (ppmv)
isopropyl nitrite (ppmv)
particle no. (cm-3)
median particle diam (nm)
σb
10 Dec 2001 19 Dec 2001 21 Dec 2001 4 Jan 2002 14 Jan 2002 16 Jan 2002
0.497 0.152 0.240 0.389 0.375 0.382
2.0 1.0 1.4 3.1 1.7 1.8
4.97 1.52 2.40 3.89 3.75 3.82
27 600 28 000 27 800 26 600 35 900 31 600
47 47 47 47 49 46
1.58 1.59 1.60 1.58 1.60 1.59
a Experiments were performed at 298 K, 1.0 × 105 Pa, and relative humidity < 10%. The initial aerosol seed was composed of (NH ) SO . 4 2 4 ) the log-normal geometric standard deviation.
TABLE 2. Final Conditions for Chamber Experiments (After Irradiation Time of 3 h)a
expt
toluene (ppmv)
NO (ppmv)
particle no. (cm-3)
median particle diam (nm)
σb
10 Dec 2001 19 Dec 2001 21 Dec 2001 4 Jan 2002 14 Jan 2002 16 Jan 2002
0.223 0.169 0.122 0.217 0.160 0.168
0.96 0.65 0.76 N/A 1.1 1.0
22 700 15 600 19 600 18 000 24 000 23 900
180 78 113 116 142 148
1.20 1.47 1.32 1.32 1.29 1.25
a Final concentrations were corrected for chamber dilution flow rate of 1-3.5 L min-1. Particles were composed of an internal mixture of inorganic seed and secondary organic. b σ ) the log-normal geometric standard deviation.
temperature), allowing predictions for nonlaboratory conditions; and (3) predicted organic aerosol speciation may be used to suggest candidate compounds for future laboratory analysis. In the future, chemically mechanistic models will also be used to evaluate proposed heterogeneous chemical pathways.
2. Experimental Description A series of toluene/NO/isopropyl nitrite irradiations was performed in a 9.6 m3 Teflon-walled indoor smog chamber (20). Tables 1 and 2 summarize initial and final observations. Isopropyl nitrite photolysis was used to initiate the production of OH radicals. Toluene decay was monitored with gas chromatography/flame ionization detection. NO decay was monitored by reacting O3 with NO in a chemiluminescent detector. Particle size distributions were measured with a TSI Model 3071 DMA and a TSI Model 3010 CPC operated in fast-scanning mode. Interested readers can find further experimental details (chamber operation protocol, observed size distributions) in the Supporting Information. The observations used here were not corrected for wall losses since wall losses were accounted for in the model. Observed organic aerosol mass concentrations were determined by first subtracting initial seed aerosol volumes from particle volumes (volumes calculated from observed number size distributions assuming spherical particles) and then applying a conversion factor to mass concentration using an organic particle density of 1.4 g cm-3, as measured by Martin-Reviejo and Wirtz (21) for organic aerosol derived from aromatic oxidation. 2.1. Modeling Description. (2.1.1) Gas-Phase Toluene/ NOx Photochemistry. An explicit gas-phase oxidation scheme for the precursor hydrocarbon was required to model the gas-particle partitioning of specific COCs. In this study, we used the University of Leeds Master Chemical Mechanism version 3.0 (MCM3) to model toluene oxidation. The dominant processes for OH-initiated toluene oxidation are addition of OH to the aromatic ring and hydrogen abstraction from the toluene methyl side chain, leading to five initial 1472
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 38, NO. 5, 2004
b
σ
pathways. Scheme S1 in the Supporting Information summarizes the initial stages of toluene oxidation in the MCM3 and shows the updates to the toluene branching ratios used here, as suggested by Wagner et al. (22). Another recent compilation by Calvert et al. (23) shows a similar oxidation scheme (Figure II-E-2 in Calvert) with channels also forming dicarbonyls, an unsaturated epoxide, cresol, and benzaldehyde. Other toluene reactions initiated by O3, NO3, and NO2 were considered but determined to be of negligible importance. The J(NO2) photolysis rate coefficient in the chamber was measured in prior experiments to be 7 × 10-4 s-1 by quantifying the decay of NO2 in N2 within a quartz tube placed in the chamber. Modeled photolysis rates were calculated for atmospheric conditions representative of the mid-day, summertime, mid-latitude boundary layer. All J-values were then scaled by a common factor to ensure the modeled and measured J(NO2) were equivalent. The calculation of J-values is a source of uncertainty in this study due to potential differences between the solar spectrum assumed in the MCM3 and the black lamp spectrum in the chamber. It was necessary to include a chamber wall conversion of HNO2 to NO and HNO3 to NO2, each with a first-order rate coefficient of 5.6 × 10-4 s-1, to model the observed decay of toluene and NO for all experiments. Recent studies have suggested possible renoxification mechanisms (24-26). It should be noted that chamber renoxification rates were optimized for the 19 Dec 2001 experiment and then applied to the model simulations for all other chamber experiments. This appeared to be a robust value for our chamber, independent of the range of initial concentrations. This is a common confounding factor in chamber studies as the magnitude of wall loss processes needs to be optimized for each chamber and reaction system (27, 28). For example, Carter and Lurmann (29) include wall renoxification in a series of five reactions with rate constant magnitudes that range over an order of magnitude depending on the chamber. Wang et al. (30) describe detailed mathematical procedures for determining the best-fit values for wall processes. For further details on chamber wall processes, the reader is referred to the Supporting Information. For the purposes of identifying the COCs that partition to the aerosol phase, a gas-phase-only box model simulation was performed on the 19 Dec 2001 experiment. The saturation ratios for all of the organic compounds were determined after 1, 2, and 3 h. The COCs were ranked on the basis of the saturation ratio and the top 20 COCs defined the list of species considered to significantly contribute to the aerosol-phase composition (Tables 3 and 4). Although the ranking order varied, the list of top 20 COCs did not change between 1 and 3 h. 2.1.2. Kinetic Aerosol Model. The kinetic based modeling approach developed by Kamens et al. (18) was applied to the explicit toluene scheme. In this approach, it was necessary to calculate forward and backward rate coefficients representing COC absorption and desorption. The ratio of the forward rate coefficient (units of volume per mass per time) to the backward rate coefficient (units of per time) is equal
TABLE 3. Ranking of Individual COCs Based on Saturation Ratio Using Results from Gas-Phase-Only Photochemical Simulations of the 19 Dec 2001 Experiment sat. ratio species
saturation vapor pressure (Torr) 10-13
6.8 × 1.0 × 10-10 4.7 × 10-9 3.4 × 10-6 6.4 × 10-6 2.0 × 10-4 3.5 × 10-4 1.3 × 10-3 3.9 × 10-3 7.3 × 10-3 4.1 × 10-3 1.9 × 10-2 3.4 × 10-3 4.1 × 10-2 1.7 × 10-1 2.0 × 10-2 7.1 × 10-3 4.9 × 10-2 1.2 2.7
TLEMUCNO3 C6CO2OHPAN TLEMUCCO TOL1OHNO2 C5134CO2OH C5COO2NO2 TLOBIPEROH C54CO C5CO14OH CO2H3CHO TLBIPERNO3 PTLQONE TLEMUCPAN MMALANHY MALANHY C6H5CH2NO3 TL3O2OH MC3CODBPAN BENZAL EPXC4DIAL
MW (g mol-1) 219 233 172 153 130 175 156 128 114 102 203 122 217 112 98 153 124 175 106 102
to the partitioning coefficient (units of volume per mass) for a given compound i:
Ki ) ikon/ikoff
(E1)
Recent studies have shown that SOA formation can be accurately modeled as a multicomponent absorption of species between a gas and liquid phase (4, 8). We calculated Ki values for individual products using the expression derived by Pankow (4, 6) from thermodynamic principles:
Ki )
RT 106MWomζipio
(E2)
where R is the ideal gas constant, T is the temperature, MWom is the average molecular weight of the absorbing medium, ζi is the activity coefficient of species i, and Pi0 is the saturation vapor pressure over the pure liquid of species i. Recent modeling studies have suggested that it may be a useful first approximation to assume that ζi are close to unity for SOA generated in chamber experiments (16). In the kinetic model, we assumed unity ζi and tested this assumption with our equilibrium model (where ζi were calculated explicitly using UNIFAC; see the following section). An initial MWom for the particle phase was estimated at 200, with subsequent MWom calculated at each time step on the basis of the modeled aerosol composition. Our results were insensitive to the initial MWom estimate. The Pi0 values were calculated using group contribution methods as summarized in Reid et al. (31). Pi0 values were calculated utilizing the Lee and Kesler formulation as a function of species boiling point temperature and acentric factor (7-2.6 to 7-2.8 in Reid). The table of Joback’s properties (Table 2-2 in Reid) for individual functional groups was used to calculate species boiling point temperature and acentric factor (2-2.4, 2-2.5, 2-3.4 in Reid). The estimation procedure was verified by comparing calculated saturation vapor pressures for toluene and benzaldehyde with experimental values (agreement within 10%). However, it is anticipated that applying this methodology to complex multifunctional COCs will introduce greater uncertainty. We found it necessary to calculate our own Joback critical properties for the -ONO2 and -C(O)OONO2 groups, as our vapor pressure estimates from grouping combinations of -O-, -NO2, and -C(O)O- resulted in a significant under-
gas diffusivity (cm2 s-1)
1h
2h
3h
0.061 0.061 0.066 0.072 0.077 0.071 0.071 0.078 0.080 0.086 0.066 0.077 0.062 0.089 0.099 0.074 0.076 0.071 0.080 0.088
3.1 × 7.1 × 102 2.7 × 10 9.4 × 10-2 3.8 × 10-2 1.4 × 10-3 9.4 × 10-4 3.7 × 10-4 4.2 × 10-4 1.4 × 10-4 8.6 × 10-5 4.9 × 10-5 4.1 × 10-5 2.1 × 10-5 1.2 × 10-5 1.2 × 10-5 1.4 × 10-5 1.5 × 10-6 1.3 × 10-6 4.1 × 10-7
3.0 × 9.4 × 102 3.0 × 10 1.3 × 10-1 4.0 × 10-2 1.7 × 10-3 8.1 × 10-4 4.9 × 10-4 3.9 × 10-4 1.4 × 10-4 7.2 × 10-5 4.7 × 10-5 4.6 × 10-5 2.7 × 10-5 1.6 × 10-5 1.4 × 10-5 1.2 × 10-5 1.8 × 10-6 1.4 × 10-6 3.5 × 10-7
2.8 × 105 9.2 × 102 2.8 × 10 1.4 × 10-1 3.9 × 10-2 1.5 × 10-3 7.4 × 10-4 4.8 × 10-4 3.7 × 10-4 1.4 × 10-4 6.7 × 10-5 4.5 × 10-5 4.1 × 10-5 2.9 × 10-5 1.8 × 10-5 1.4 × 10-5 1.1 × 10-5 1.7 × 10-6 1.4 × 10-6 3.2 × 10-7
105
105
estimation for the organic nitrate and peroxyacyl nitrate species (see Supporting Information for critical property values of the new groups and further details). A comparison between our vapor pressure calculation for pinonaldehyde and a group contribution calculation by Yu et al. (32) yielded a difference of a factor of 3. Vapor pressures calculated using group contribution methods are usually within a factor of 2-3 of measured values (33). Measurement-based multigroup methods can have higher accuracy but require very compound-class specific vapor pressure measurements that are not available for many COCs (34). The free molecular kinetic limit, ikkin ) 3Rici/(4Fr), and continuum diffusion limit, ikdif ) 3D/(Fr2) were used to formulate ikon ) ikkinikdif/(ikkin + ikdif), where Ri is the accommodation coefficient of species i on the organic aerosol, ci is the mean gas-phase molecular speed of species i, r is the particle radius, D is a gas-phase diffusion coefficient, and F is the particle density (35). Gas-phase diffusion coefficients were calculated using the Fuller et al. group contribution method (11-4.4 and Table 11-1 in Reid) with values listed in Table 3. Our kinetic model does not explicitly treat the evolution of the particle size distribution. Here, we use the mean particle size from the observed aerosol size distribution to continually update the calculation of ikon within the kinetic model. Sensitivity tests were performed assuming a constant particle diameter for two limiting cases: a lower limit diameter of 30 nm (median D - 1σD from Dec 10th in Table 1) and an upper limit diameter of 216 nm (median D + 1σD from Dec 10th in Table 2). Final aerosol yields only varied by 12% between these limits. Thus, we feel the particle size distributions were sufficiently narrow so as to not introduce significant uncertainty in our analysis. Accommodation coefficients for COCs equal to 0.2 resulted in the most reasonable fit to the observed aerosol mass concentrations. This value was also used by Bowman et al. (14) and is similar to the range of values listed by Jacob (36) for organic species on liquid water. With Ri ) 0.2, mass transfer was dominated by the kinetic limit. A sensitivity simulation removing the diffusion limit from our formulation introduced errors of less than 5% compared to the full masstransfer formulation. The rate of COC condensation for a given COC depends on the rate coefficient, ikon, along with the gas-phase COC concentration and aerosol mass concentration. The presence VOL. 38, NO. 5, 2004 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
1473
TABLE 4. Chemical Structures of COCs Considered in MCM3
and allowed to decay. Estimated first-order gas-phase species wall loss rates were in the range 2 × 10-6 to 3 × 10-6 s-1. Mass transfer to the wall is driven by the difference between a species gas-phase concentration and its saturation concentration. Initial concentrations in these wall loss experiments were much larger than their saturation concentrations so mass transfer should largely be species independent, as observed by the narrow range of first-order loss rates. Using these experiments as a guideline, a first-order upper-limit gas-phase wall loss rate coefficient of 3 × 10-6 s-1 was assumed for all the toluene oxidation products. Given a typical value for ikon ∼ 2 × 107 cm3 g-1 s-1 and an ∼10 µg m-3 aerosol loading, we calculate an effective gas-to-particle rate coefficient of 2 × 10-4 s-1, 2 orders of magnitude larger than observed gas wall loss rates. The weak model dependence on gas-phase species wall loss was also confirmed in sensitivity tests reported in section 3.3.1. Our gas wall loss rate for toluene products is in the range reported for other multifunctional organics in the literature (e.g., 8 × 10-6 s-1 for pinonaldehyde (38)). For interested readers, we have compiled a list of wall loss rates in different chambers in the Supporting Information. Wall loss rates in the literature do not necessarily scale with chamber size (19, 28, 29, 37). All of the above-mentioned processes (gas-phase photochemistry, gas-condensed-phase equilibrium partitioning, wall loss) were parametrized in terms of reactions and integrated with a Gear-type solver of ordinary differential equations (Facsimile 3.0). (2.1.3) Equilibrium Aerosol Model. The equilibrium model included gas-phase reaction and aerosol equilibrium modules, operator-split with a time step of 30 s. The gasphase module was solved using the Gear-type solver with NCAR’s chemistry box model (39). The aerosol module iteratively solved the Raoult’s law equilibrium partitioning for a mixture of condensable organic gases and condensedphase liquids, following Pankow (6):
Fij ) τi
[( ) 1
Fkj-1
∑ MW k
of an increasing organic aerosol mass coupled with the production of low-volatile gas-phase COCs therefore drives the growth of organic aerosol in the model (the absorption rate is proportional to the ratio of aerosol mass concentration and radius, which is proportional to the product of radius squared and particle number concentration). In the model, the initial inorganic mass concentration provides the seed onto which organic species absorb. ikoff values for desorption in the kinetic model were calculated through the substitution of (E2) into (E1). Particle wall loss rates were measured experimentally in York’s chamber as a function of particle size. The measured rates decreased from 2.9 × 10-5 s-1 for a diameter of 26 nm to 1.3 × 10-5 s-1 at 140 nm and then increased to 2.4 × 10-5 s-1 at 720 nm (see Supporting Information for further details). Because of this small variation and the fact that the wall loss lifetimes were 5-10 times the duration of the experiment, a constant wall loss rate coefficient (2.0 × 10-5 s-1) was used in the model. Our particle wall loss rate can be compared to values reported for other chambers (1.3 × 10-5 (18) and 9.4 × 10-4 s-1 (37)). Several experiments have been performed in York’s chamber characterizing gas-phase species wall loss rates. Samples of long-chain alkyl nitrates (e.g., 2-ethylhexyl nitrate) and citraconic anhydride were injected into the chamber 1474
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 38, NO. 5, 2004
k
(
)
109ζij-1pL,i0 760RT
]
-1
+1
(E3)
where Fij is the jth iterative value of the condensed-phase mass concentration of the ith organic compound (ng of organic compound/(m3 of air)), τi is the total mass concentrations (particle + gas-phase) of the ith organic compound (ng of organic compound/(m3 of air in the grid square)), MWk is the molecular weight of the kth compound (g mol-1), ζij is the jth iterative value of the activity coefficient of the ith organic compound (mole fraction scale), pL,i0 is the vapor pressure of the ith organic compound with respect to its own liquid (Torr), R is the universal gas constant (8.2 × 10-5 m3 atm mol-1 K-1), and T is the temperature (K). Activity coefficients were calculated using a recent version of the UNIFAC group method algorithm (40, 41). The initial value of the condensed-phase mass in the iteration was taken to be the total condensable organic mass (Fi0 )τi). Gas-phase concentrations were then calculated on the basis of mass conservation and returned to the gas-phase reaction module. The equilibrium and gas reactions were treated as consecutive operators in an operator-split system; a sensitivity study using successively smaller time steps showed that an operator step of less than 1 min was required to prevent the introduction of splitting-associated errors. Wall loss rates as described in the kinetic model were also incorporated into the gas-phase reaction module of the equilibrium model.
3. Results 3.1. Summary of Gas-Phase Model-Measurement Comparison. Figure 1 summarizes the model-measurement
FIGURE 2. Comparison between the measured and modeled aerosol yield dependence on aerosol mass concentration. The line represents a two-product fit to observed data (not corrected for wall losses) using (E4) with the following coefficients: a1 ) 12.4; K1 ) 0.0182; a2 ) 17.9; K2 ) 0.00236. This is appropriate for comparison with the models. For comparison with other experimental results, when the York observations are corrected for particle wall loss, the fitted parameters are as follows: a1 ) 0.0406; K1 ) 385; a2 ) 0.652; K2 ) 7.31. A particle density of 1.4 g cm-3 was used to convert volume to mass.
FIGURE 1. Comparison between measured and modeled toluene and NO mixing ratio time series for three chamber experiments (panel A, 19 Dec 2001; panel B, 14 Jan 2002; panel C, 16 Jan 2002). comparison for the two measured gas-phase species, toluene and NO. Model-measurement toluene percent differences were less than 5% for all of the experiments. Both the kinetic and equilibrium models yielded nearly identical toluene decays. Both models overestimated NO observed decays with a percentage difference after 3 h in the range of 10-30%. The kinetic model yielded slightly faster NO decays than the equilibrium model; however, modeled NO mixing ratios were within 10% of each other after the 3 h experiments. 3.2. Summary of Aerosol-Phase Model-Measurement Comparison. Figure 2 summarizes the comparison between the models and measurements for the dependence of aerosol yield on aerosol mass concentration. Here, aerosol yield is defined as the aerosol mass concentration divided by the mass of toluene reacted as a percentage after a 3 h experiment. The circular points are the results from individual York chamber experiments. The line is a two-product best fit to the observations minimizing weighted squared residuals using the following expression from Odum et al. (8): 2
Y ) Mo
RiKi
∑1 + K M i)1
i
(E4) o
where Ki is a partitioning coefficient and Ri is a gas-phase stoichiometric yield. The square and triangular points are
the equilibrium and kinetic model results, respectively. The models reproduced the general magnitude of the aerosol yields but suggest a weaker trend dependence on aerosol mass loading (overpredicted aerosol yields for the lower aerosol loading experiments by up to 272% and underpredicted aerosol yields for the higher aerosol loading experiments by up to 43%). It is reasonable that uncertainties in the gas-phase mechanism and partitioning would not impact the high- and low-aerosol-loading experiments equally. To illustrate this point, we need to consider the high and low aerosol limits of (E4). At high aerosol loadings, (KiMo . 1) and Y ) ∑Ri. Thus, aerosol yields at high aerosol loadings depend critically on stoichiometric coefficients (Ri) and less on partitioning parameters (Ki). At lower aerosol yields, KiMo ,1 and Y )Mo∑RiKi. The lower aerosol loading experiments therefore depend critically on both gas-phase stoichiometric coefficients and the partitioning parameters. The accuracy of the partitioning parametrization might therefore be expected to have an important impact at low aerosol loadings. Results from these two limits imply that the modeled underprediction at high aerosol loadings stems from uncertainties in gas-phase photochemistry and modeled overprediction at low aerosol loading stems from an overpartitioning of COC products to the aerosol phase. In the next section, we discuss uncertainties in wall losses, gas-phase photochemistry, and partitioning parameters in an attempt to better understand the model-measurement trend. 3.3. Analysis of Model Sensitivities. (3.3.1) Wall Losses. Modeled organic gas-phase species wall loss processes were of negligible importance in our standard simulations (1% difference from simulation with no gas wall losses). Increases in gas-phase species wall loss rates by a factor of 10 resulted in only a 5.1% decrease in modeled aerosol mass concentration after a 3 h simulation. Modeled aerosol wall loss processes also played only a small role in decreasing aerosol mass concentration for our standard simulations (11% difference from simulation with no aerosol wall losses). However, increases in particle wall losses by a factor of 10 resulted in VOL. 38, NO. 5, 2004 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
1475
a 65% decrease in aerosol mass concentration. These results suggest that modeled aerosol concentrations in our chamber are not sensitive to possible uncertainties in the measured gas-phase organic wall loss rates within a factor of 10 but could be sensitive if measured particle wall loss rates are severely underestimated. 3.3.2. Organic Chemistry. As outlined in the model description, the initial toluene branching ratios in the chemical mechanism were from a recent recommendation by Wagner et al. (22). Using the initial toluene branching ratios in the original MCM3, as reported in Jenkin et al. (42), resulted in a 70% increase in aerosol yields. This stems from the fact that the epoxide and quinone branches are the most effective at producing COCs and the Jenkin et al. recommendation increases both of these branches at the expense of the dicarbonyl and cresol branches. Selecting initial toluene branching ratios between the Wagner et al. and Jenkin et al. recommendations would increase the modeled aerosol yields for all the simulations and yield better model-measurement consistency for the high-aerosol-loading experiments. Jenkin et al. also point out that the photolysis of many intermediates in the chemical mechanism is highly uncertain. In the MCM3 formulation, the R,β-unsaturated γ-dicarbonyls (e.g., TLEPOXMUC) are a class of intermediates for which photolysis rates were estimated by applying a 0.028 factor to the J(NO2) photolysis rate. Measured cross-sections and quantum yields by Klotz et al. (43) for muconaldehyde species imply a factor of 0.014-0.042 should be applied to the J(NO2) photolysis rate. To evaluate the sensitivity of modeled aerosol yields to TLEPOXMUC photolysis, we increased its rate of photolysis by a factor of 5. Increasing this one photolysis rate alone resulted in a 13% decrease in modeled aerosol yields. These results support further laboratory studies to continue improving explicit oxidation schemes, as the details are important. 3.3.3. Gas-to-Particle Conversion. In the equilibrium model, the key uncertain factors impacting the partitioning coefficients are saturation vapor pressures and activity coefficients. In the kinetic model, the key factors are the saturation vapor pressures, activity coefficients, accommodation coefficients, and the initial adsorption process. A factor of 10 increase in saturation vapor pressures resulted in a 10% decrease in modeled aerosol yields. While this percentage change improves the model’s overprediction for the low-aerosol-loading experiment, it also decreased the aerosol yields for the higher aerosol loading experiments by a similar percentage (Figure 3A). From (E3) it can be seen that the fraction of the COCs in the condensed phase is roughly proportional to the inverse of the saturation vapor pressureserrors in vapor pressure estimates will therefore impact all simulations in a similar fashion. Sensitivity studies showed that the kinetic model is in a regime where partitioning is highly dependent on the assumed accommodation coefficients for all the experiments. A factor of 10 decrease in the accommodation coefficients resulted in a 28% decrease in aerosol mass concentration nearly independent of aerosol loading. Thus, variations in accommodation coefficient also could not reconcile the model-measurement differences for aerosol yields over the entire range of aerosol experiments. We have not been able to identify one single variable that would cause the strong observed aerosol yield dependence on aerosol mass concentration. A combination of altered branching ratios (raising aerosol yields for all experiments) and an accommodation coefficient dependence on organic layer film thickness (decreasing aerosol yields preferentially for the lower aerosol loading experiments) could provide the combination needed to reproduce the observed aerosol yield trend. A calculation of gas-aerosol transport characteristic times, as described in Bowman et al. (14), results in τGP ) 7 min and 0.9 h for 1476
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 38, NO. 5, 2004
FIGURE 3. Sensitivity results for aerosol mass concentration time series. Panel A compares model to measurements for base case simulation and a simulation with all COC saturation vapor pressures increased by a factor of 10. Panel B illustrates the impact of decreasing all COC accommodation coefficients by a factor of 10. All simulations were performed with the kinetic model. the highest and lowest aerosol loading experiments, respectively. Thus, mass transfer that is limited kinetically may be one explanation for the equilibrium model’s overestimation at lower aerosol loading. However, the kinetic model should take this size-dependent transfer into account, and thus we hypothesize size-dependent accommodation coefficients as a possible mechanism for the differences at either end of the aerosol mass range. 3.4. Modeled Product Distribution and Comparison to Other Chamber Experiments. Figure 4 shows the product distribution obtained from both the equilibrium and kinetic models’ simulation of the 14 Jan 2002 experiment. Similar distributions of products were observed from both models. Three species tended to dominate the aerosol phase (C6CO2OHPAN, TLEMUCNO3, TOL10HNO2) comprising more than 97% of the aerosol mass after the 3 h simulation. The predominant species were nitroaromatics and multifunctional organic nitrates (both alkyl nitrates and peroxyacyl nitrates). Generally, species with saturation vapor pressures below 10-3 Torr partitioned to the aerosol phase (see Supporting Information for illustration). Forstner et al. (44) determined the molecular composition of organic aerosol produced from the oxidation of toluene-NOx mixtures. Unsaturated anhydrides were determined to be the predominant aerosol components in the extractable/elutable portion of the total organic mass. However, only 15-30% of the material which could be extracted from filters and which eluted through a gas chromatograph could be detected. Thus, the majority of the aerosol mass remained unidentified and presumably is composed of more nonvolatile material that remained on the filters or was lost with the chromatographic column. The two predominant unsaturated anhydrides identified were 2,5-furandione and 3-methyl-2,5-furandione. These species were also modeled aerosol components in our study and correspond to MALANHY and MMALANHY, respectively. These unsaturated anhydrides were only minor species in our modeled particles with lower vapor pressure
FIGURE 4. Comparison of aerosol composition (% of total mass) derived from the kinetic and equilibrium model for the 14 Jan 2002 experiment. Species names correspond to structures in Table 4. species dominating the aerosol phase. This is qualitatively consistent with the majority of the aerosol mass in the Forstner et al. experiments not being extracted/eluted due to their low volatility. More recently, Jang and Kamens (7) made a comprehensive identification of aerosol particles derived from toluene-NOx photooxidation experiments, listing species as major, minor, or trace products. They report several ringretaining products also found in our modeled particles, namely, a methylnitrocresol (TOL1OHNO2) and a quinone (PTLQONE), as well as ring-opening products such as the unsaturated carboxylic acid CO2H3CHO. Jang and Kamens did not report measurements for organic nitrates. The formation of organic nitrates may be enhanced in our chamber experiments compared to others due to the relatively low initial toluene/NOx ratio for our experiments (0.13-0.25). The Forstner et al. and Jang and Kamens experiments are both initiated with toluene/NOx ratios between 2.8 and 4.0. We have performed a sensitivity simulation (14 Jan 2002) with an initial toluene/NOx ratio of 2.2. As expected, the importance of hydroxy-dicarbonyls increased (TLEMUCCO, 0.32 to 15%; C5134CO2OH, 0.0017 to 0.16%) with multifunctional alkyl nitrates decreasing (TLEMUCNO3, 32 to 13%), although nitrated species were still very important species in the aerosol phase (see Supporting Information for detailed speciation). 3.5. Activity Coefficients in Toluene-Derived Particles and Their Water Uptake Properties. The equilibrium model uses UNIFAC to continuously calculate the activity coef-
ficients of the organic species in the aerosol phase. A sensitivity simulation was performed to evaluate the importance of nonideal interactions where the activity coefficients were all constrained to unity. The impact of a unity activity coefficient assumption on modeled aerosol yields was negligible. This stems from the fact that the calculation of activity coefficients varied over a limited range from 0.43 to 2.8 so that they will only matter for semivolatile products; most of the products in the model were either highly volatile or nonvolatile. Thus, the assumption of unity activity coefficient in the kinetic model’s calculation of partitioning coefficients, Ki, was a reasonable approximation, assuming that the calculated range of activity coefficients in the equilibrium model was accurate. Recent work by Demou et al. (45) suggests that the standard UNIFAC formulation may not capture all of the nonlinearities associated with activity coefficient calculations for a large range of condensed-phase mole fractionssone possible cause for the deviation between model and measurement results is that the actual mixture may be less ideal than that implied by the activity coefficients calculated here. The water uptake properties of organic particles are currently a major area of uncertainty in Earth system models. To look at the hygroscopicity of organic particles produced from toluene oxidation, we performed a simulation at 79% relative humidity and included water as a partitioning species in the equilibrium model (no inorganic module was employed; did not consider inorganic/water interactions). The modeled aerosol yields increased by a factor of 1.4 (diameter VOL. 38, NO. 5, 2004 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
1477
growth factor, G(79%) ) 1.1) with a 69% increase due to water itself and 31% increase due to enhanced partitioning of organics. The upper limit for activity coefficients increased considerably with a new range from 0.44 to 30. Thus, organic particle components derived from toluene are slightly hygroscopic. Our G(79%) ) 1.1 is comparable to laboratory studies for organic aerosol water uptake. Cruz and Pandis (46) observed an G(85%) ) 1.1 and 1.0 for pure glutaric acid and pinonic acid particles, respectively. Virkkula et al. (47) report an G(84%) ) 1.1 for particles in irradiated mixtures of pinene/NOx. The results of Aklilu and Mozurkewich (48) give G(79%) ) 1.05-1.12 for atmospheric organic particles. These results provide indirect support for the equilibrium model and its predicted aerosol composition.
4. Recommendations for Future Experiments A major goal of this exploratory study was to develop a target list of potential aerosol species to guide future experimental studies. Throughout the course of developing and assessing our conceptual models and through a literature review of chamber studies, we have also identified a list of future laboratory measurements needed to help constrain and validate explicit models: (1) characterization of J-values and heterogeneous wall processes within chambers; (2) measurements of saturation vapor pressure and activity coefficients for our target list species, especially nitrated species; (3) gas-phase kinetic and product distribution studies for toluene; (4) accommodation coefficient measurements of our target list species on organic films of varying thickness with an inorganic core; and (5) better characterization of organic gas- and aerosol-phase speciation during experiments. We also recommend expanding the database of chamber and modeling studies at different initial toluene mixing ratios and varying toluene/NOx ratios. Our preliminary findings suggest that toluene-precursor particles are mainly composed of a few nonvolatile components. The model dependence for several individual aerosol yields on total mass concentration from a series of model simulations at varying initial toluene is shown in the Supporting Information. Nonvolatile species should be largely independent of the partitioning process (less reliance on saturation vapor pressure calculations and existing particle mass and composition), and thus their aerosol yield measurement should provide a better validation of the proposed gas-phase chemical pathways in the model. Chamber experiments with a heated transfer line to the SMPS would confirm the relative importance of nonvolatiles compared to semivolatiles predicted by the model. Our preliminary results also suggest that the particles are slightly hygroscopic with a G(79%) ) 1.1. We recommend future experiments with a humidified transfer line between the chamber and SMPS to confirm the modeled diameter growth factors.
Acknowledgments The authors are thankful to Sam Saunders, Mike Pilling, Mike Jenkin, Dick Derwent, and William Carter for supplying the authors with downloadable versions of the MCM v3.0. The authors are grateful to York University’s Centre for Atmospheric Chemistry Research (CAC) for funding and administrative support throughout the study and the Air Quality Research Branch of Environment Canada for in-kind support. Financial support from the Canadian Foundation for Climate and Atmospheric Science (CFCAS) is acknowledged. The authors are appreciative of Bhavina Patel, Vinoba Indran, and Ilona Kletskin for their contributions during summer internships. C.A.S. would also like to thank Environment Canada’s Science Horizons program for funding young scientists performing environment-related research. 1478
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 38, NO. 5, 2004
Supporting Information Available Extensive details on experiment protocols, vapor pressure calculations, a critical review of chamber wall processes, and model output not included in the manuscript (sensitivity tests at lower NOx, aerosol fractions, and aerosol yields for individual species). This information is available free of charge via the Internet at http://pubs.acs.org.
Literature Cited (1) Schwartz, J.; Dockery, D. W.; Neas, L. M. J. Air Waste Manage. Assoc. 1996, 46, 927. (2) Turpin, B. J.; Saxena, P.; Andrews, E. Atmos. Environ. 2000, 34, 2983. (3) Jacobson, M. C.; Hansson, H. C.; Noone, K. J.; Charlson, R. J. Rev. Geophys. 2000, 38 (2), 267. (4) Pankow, J. F. Atmos. Environ. 1994, 28A, 185. (5) Grosjean, D. Atmos. Environ. 1992, 26A, 953. (6) Pankow, J. F. Atmos. Environ. 1994, 28A, 189. (7) Jang, M.; Kamens, R. M. Environ. Sci. Technol. 2001, 35, 4758. (8) Odum, J. R.; Jungkanmp, P. W.; Griffin, R. J.; Flagen, R. C.; Seinfeld, J. H. Science 1997, 276, 96. (9) Kourtidis, K.; Ziomas, I. Global Nest:Int. J. 1999, 1, 33. (10) Jeffries, H. E. Photochemical Air Pollution. In Composition, Chemistry, and Climate of the Atmosphere; Singh, H. B., Ed.; Van Nostrand Reinhold: New York, 1995. (11) Kleindienst, T. E.; Smith, D. F.; Li., W.; Edney, E. O.; Driscoll, D. J.; Speer, R. E.; Weathers, W. S. Atmos. Environ. 1999, 33, 3669. (12) Hurley, M. D.; Sokolov, O.; Wallington, T. J.; Takekawa, H.; Karasawa, M.; Klotz, B.; Barnes, I.; Becker, K. H. Environ. Sci. Technol. 2001, 35, 1358. (13) Odum J. R.; Hoffmann, T.; Bowman, F.; Collins, D.; Flagan, R. C.; Seinfeld, J. H. Environ. Sci. Technol. 1996, 30, 2580. (14) Bowman, F. M.; Odum, J. R.; Seinfeld, J. H.; Pandis, S. N. Atmos. Environ. 1997, 31, 3921. (15) Strader, R.; Lurmann, F.; Pandis, S. N. Atmos. Environ. 1999, 33, 4849. (16) Seinfeld, J. H.; Erdakos, G. B.; Asher, W. E.; Pankow, J. F. Environ. Sci. Technol. 2001, 35, 1806. (17) Pankow, J. F.; Seinfeld, J. H.; Asher, W. E.; Erdakos, G. B. Environ. Sci. Technol. 2001, 35, 1164. (18) Kamens, R.; Jang, M.; Chien, C. J.; Keach, K. Environ. Sci. Technol. 1999, 33, 1430. (19) Kamens, R.; Jaoui, M. Environ. Sci. Technol. 2001, 35, 1394. (20) Barbu, A. M.Sc. Dissertation, York University, 2003. (21) Martin-Reviejo, M.; Wirtz, K. Presented at the IGAC Conference, Greece, 2002. (22) Wagner, V.; Jenkin, M. E.; Saunders, S. M.; Stanton, J.; Wirtz, K.; Pilling, M. J. Atmos. Chem. Phys. Discuss. 2002, 2, 1217. (23) Calvert, J. G.; Atkinson, R.; Becker, K. H.; Kamens, R. M.; Seinfeld, J. H.; Wallington, T. J.; Yarwood, G. The Mechanisms of Atmospheric Oxidation of Aromatic Hydrocarbons; Oxford University Press: New York, 2002. (24) Ammann, M.; Kalberer, M.; Jost, D. T.; Tobler, L.; Ro¨ssler, E.; Piguet, D.; Ga¨ggeler, H. W.; Baltensperger, U. Nature 1998, 395, 157. (25) Chang, T. Y.; Nance, B. I.; Kelly, N. A. Atmos. Environ. 1999, 33, 4695. (26) Rivera-Figueroa, A. M.; Sumner, A. L.; Finlayson-Pitts, B. J. Environ. Sci. Technol. 2003, 37, 548. (27) Wang, L.; Milford, J. B.; Carter, W. P. L. Atmos. Environ. 2000, 34, 4337. (28) Chang, T. Y.; Nance, B. I.; Kelly, N. A. Atmos. Environ. 1999, 33, 4695. (29) Carter, W. P. L.; Lurmann, F. W. Atmos. Environ. 1991, 25A, 2771. (30) Wang, L.; Milford, J. B.; Carter, W. P. L. Atmos. Environ. 2000, 34, 4337. (31) Reid, R. C.; Prausnitz, J. M.; Poling, B. E. The Properties of Gases and Liquids, 4th ed.; McGraw-Hill: New York, 1987. (32) Yu, J.; Cocker, D. R.; Griffin, R. J.; Flagan, R. C.; Seinfeld, J. H. J. Phys. Chem. 1930, 34, 207. (33) Asher, W. E.; Pankow, J. F.; Erdakos, G. B.; Seinfeld, J. H. Atmos. Environ. 2002, 36, 1483. (34) Makar, P. A. Atmos. Environ. 2001, 35, 961. (35) Seinfeld, J. H.; Pandis, S. N. Atmospheric Chemistry and Physics; Wiley: New York, 1997. (36) Jacob, D. J. Atmos. Environ. 2000, 34, 2131. (37) Sarwar, G.; Corsi, R.; Allen, D.; Weschler, C. Atmos. Environ. 2003, 37, 1365.
(38) Jaoui, M.; Kamens, R. M. J. Atmos. Chem. 2003, 44, 259. (39) Stroud, C. A.; Madronich, M.; Atlas, E.; Ridley, B.; Flocke, F.; Weinheimer, A.; Talbot, R.; Fried, A.; Wert, B.; Shetter, R.; Lefer, B.; Coffey, M.; Heikes, B.; Blake, D. Atmos. Environ. 2003, 37, 3351. (40) Fredenslund, A.; Gmehling, J.; Rasmussen, P. Vapor-Liquid Equilibrium Using UNIFAC; Elsevier: Amsterdam, 1997. (41) Sandler, S. I. Chemical and Engineering Thermodynamics; Wiley: New York, 1999. (42) Jenkin, M. E.; Saunders, S. M.; Derwent, R. G.; Pilling, M. J. Atmos. Environ. 2002, 36, 4725. (43) Klotz, B.; Bierbach, A.; Barnes, I.; Becker, K. H. Environ. Sci. Technol. 1995, 29, 2322.
(44) Forstner, H. J. L.; Flagan, R. C.; Seinfeld, J. H. Environ. Sci. Technol. 1997, 31, 1345. (45) Demou, E.; Visram, H.; Donaldson, D. J.; Makar, P. A. Atmos. Environ., in press. (46) Cruz, C. N.; Pandis, S. N. Environ. Sci. Technol. 2000, 34, 4313. (47) Virkhula, A.; Dingenen, R. V.; Raes, F.; Hjorth, J. J. Geophys. Res. 1999, 104 (D3), 3569. (48) Aklilu, Y. A.; Mozurkewich, M. Aerosol Sci. Technol., in press.
Received for review July 5, 2003. Revised manuscript received December 5, 2003. Accepted December 12, 2003. ES030546W
VOL. 38, NO. 5, 2004 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
1479