Modeling global secondary organic aerosol formation and processing with the volatility basis set: Implications for anthropogenic secondary organic aerosol



[1] The volatility basis set, a computationally efficient framework for the description of organic aerosol partitioning and chemical aging, is implemented in the Goddard Institute for Space Studies General Circulation Model II′ for a coupled global circulation and chemical transport model to simulate secondary organic aerosol (SOA) formation. The latest smog chamber information about the yields of anthropogenic and biogenic precursors is incorporated in the model. SOA formation from monoterpenes, sesquiterpenes, isoprene, and anthropogenic precursors is estimated as 17.2, 3.9, 6.5, and 1.6 Tg yr−1, respectively. Reducing water solubility of secondary organic gas from 105 to 103 mol L−1 atm−1 (1 atm = 1.01325 × 105 N m−2) leads to a 90% increase in SOA production and an increase of over 340% in total atmospheric burden, from 0.54 to 2.4 Tg. Increasing the temperature sensitivity of SOA leads to a 30% increase in production, to 38.2 Tg yr−1. Since the additional SOA is formed at high altitude, where deposition time scales are longer, the average lifetime is doubled from 6.8 to 14.3 days, resulting in a near tripling of atmospheric burden to 1.5 Tg. Chemical aging of anthropogenic SOA by gas-phase reaction of the SOA components with the hydroxyl radical adds an additional 2.7–9.3 Tg yr−1 of anthropogenic SOA to the above production rates and nearly doubles annual average total SOA burdens. The possibility of such high anthropogenic SOA production rates challenges the assumption that anthropogenic volatile organic compounds are not important SOA precursors on a global scale. Model predictions with and without SOA aging are compared with data from two surface observation networks: the Interagency Monitoring of Protected Visual Environments for the United States and the European Monitoring and Evaluation Programme.

1. Introduction

[2] Atmospheric aerosol affects human health [Dockery et al., 1993; Krewski et al., 2003] and visibility [Seinfeld and Pandis, 2006] and can also influence Earth's energy balance either directly [Intergovernmental Panel on Climate Change, 2007] or through the modification of cloud properties [Twomey, 1977; Albrecht, 1989]. A large fraction of the global fine particulate matter (diameter <1 μm) is organic [Putaud et al., 2004; Zhang et al., 2007] and typically is split into the broad categories of primary organic aerosol (POA), emitted in the particle phase, and secondary organic aerosol (SOA), formed by atmospheric oxidation of volatile organic compounds (VOCs). POA has been traditionally treated as nonvolatile and essentially nonreactive [Koch, 2001]; many models convert “hydrophobic” organic aerosol (OA) to “hydrophilic” OA, using an assumed aging time constant [Cooke and Wilson, 1996] to account for, without mechanistic detail, some combination of chemical reaction of OA to more soluble forms and physical mixing with other soluble aerosol components. Recently, this assumption has been challenged [Shrivastava et al., 2006; Robinson et al., 2007]. Early global SOA formation studies, summarized in Table 1, have described SOA in a similar way: nonvolatile particles that are emitted directly into the atmosphere. These studies have assumed that SOA is “emitted” at a fractional constant yield of the parent hydrocarbon [Grosjean and Seinfeld, 1989; Liousse et al., 1996; Penner et al., 1998]. More recent approaches have recognized that SOA behavior more closely resembles a solution, which follows the absorptive partitioning theory of Pankow [1994]. The semivolatile nature of SOA has often been described by the pseudo-ideal solution framework of Odum et al. [1997], which groups the hundreds of semivolatile oxidation products of each VOC precursor into two semivolatile surrogate species (the so-called two-product method). Each of the two surrogates is described by a stoichiometric yield and a partitioning coefficient inversely proportional to the species vapor pressure. Other methods [Derwent et al., 2003; Bonn and Lawrence, 2005] attempt to simulate explicitly the products of oxidation relevant for SOA formation but are restricted to a limited number of VOC precursor species. Recent reviews of the current state of knowledge regarding the chemistry of SOA formation and aging have noted that models should increasingly focus on predicting the volatility changes that accompany these reactions [Kroll and Seinfeld, 2008; Hallquist et al., 2009]. Donahue et al. [2006] presented a new framework for OA modeling, the volatility basis set, which spans a larger range of atmospheric conditions than the two-product model does and conveniently accounts for partitioning, dilution, and chemical aging of organic vapors.

Table 1. Global Secondary Organic Aerosol Modeling Studiesa
Modeling StudySOA ModelBiogenic SOA Production (Tg yr−1)Anthropogenic SOA Production (Tg yr−1)POA Emission (Tg yr−1)
  • a

    Abbreviations are as follows: POA, primary organic aerosol; SOA secondary organic aerosol.

  • b

    Reported in Tg carbon, converted by using organic mass to organic carbon ratio of 1.6.

  • c

    SOA emission prescribed as equal to POA emission.

  • d

    POA concentrations specified as a constant function of altitude.

  • e

    Includes cloud production from glyoxal and methylglyoxal.

  • f

    High estimate assumes partitioning with sulfate.

Liousse et al. [1996]Constant yield7.873.2
Lohmann et al. [1999]Constant yield25.9b142.2b
Cooke et al. [1999]SOA = POAc17.117.1
Raes et al. [2000]Constant yield18.583.1
Kanakidou et al. [2000]Two-product61–7973.1
Chung and Seinfeld [2002]Two-product11.281
Liao et al. [2003]Two-product12.581
Derwent et al. [2003]Explicit products63-0
Tsigaridis and Kanakidou [2003]Two-product2.5–44.50.05–2.6240
Lack et al. [2004]Two-product13.61.762
Reddy and Boucher [2004]Constant yield19.588.9
Stier et al. [2005]Constant yield19.147.2
Tsigaridis et al. [2005]Two-product20.2–21.91.640
Lauer et al. [2005]Two-product16.283.5
Liao and Seinfeld [2005]Two-product18.081
Bonn and Lawrence [2005]Explicit products20–63Specifiedd
Liu et al. [2005]Constant yield14.496.8
Henze et al. [2006]Two-product8.7–16.481
Tsigaridis et al. [2006]Two-product18.11.244.4
Racherla and Adams [2006]Two-product16.381
Hoyle et al. [2007]Two-product52.5–67.52.535.6
Tsigaridis and Kanakidou [2007]Two-product16.81.844.4
Guillaume et al. [2007]Two-product101b9.7b61b
Heald et al. [2008]Two-product36.3b2.7b72b
Henze and Seinfeld [2006]Two-product42.9b5.6b53b
Fu et al. [2008]Two-producte42.7b1.7b
Goto et al. [2008]Two-product6.7488.2
Kim et al. [2008]Two-product49.1b33.8–54.4
Hoyle et al. [2009]Two-productf53.4–68.848.6

[3] Several studies have applied the Odum et al. [1997] approach to estimating global secondary organic aerosol formation. Kanakidou et al. [2000] predicted global SOA production in the range of 61–79 Tg yr−1. Chung and Seinfeld [2002] estimated that the global SOA formation from terpene oxidation is approximately 11.2 Tg yr−1 and OA net anthropogenic direct radiative forcing is on the order of −0.1 to −0.2 W m−2. Investigating uncertainties, Tsigaridis and Kanakidou [2003] attempted to bound annual SOA production from biogenic VOCs (BSOA) to 2.5–44.5 Tg yr−1 and anthropogenic SOA (ASOA) formation to between 0.05 and 2.6 Tg yr−1. Henze and Seinfeld [2006], extending the two-product approach by including isoprene oxidation as a SOA source, concluded that isoprene is as important as monoterpenes and sesquiterpenes on a global scale. Tsigaridis and Kanakidou [2007] estimated that isoprene contributes less than 15% to the SOA burden. Henze et al. [2008] included a treatment of aromatic VOCs from anthropogenic sources, which doubled the previous estimate by Tsigaridis and Kanakidou [2007] of the atmospheric burden of ASOA from 0.04 to 0.08 Tg, which was still less than 10% of the total SOA burden. All the above studies concluded that the contribution of anthropogenic VOCs to global SOA formation was insignificant compared with that of biogenic precursors.

[4] Volkamer et al. [2006] used aerosol mass spectrometer (AMS) measurements to show that the measured surface concentrations of oxidized organic aerosol (OOA) in an urban area dominated by anthropogenic VOC sources cannot be explained by traditional organic aerosol frameworks. Simpson et al. [2007] found that their SOA modeling framework underpredicted SOA concentrations at European sites. Heald et al. [2005] compared free-tropospheric OA measurements from the Asian Pacific Regional Aerosol Characterization Experiment (ACE-Asia) field campaign with predictions from the global chemical transport model GEOS-Chem and concluded that a significant source of SOA was missing from the free troposphere. While this may likely be the case, several other possibilities for underprediction may exist. Fitting the two-product model to smog chamber data tends to bound the data and may not be applicable under more dilute conditions (e.g., in the free troposphere). Henze et al. [2006] illustrated the importance of ΔHvap (the sensitivity of SOA volatility to temperature) on SOA concentrations in the free troposphere, where temperatures are much lower than in typical smog chamber experiments. Henze et al. [2006] also discussed the potential importance of the water solubility of organic vapors on the OA concentrations aloft. Since SOA is semivolatile, changes in gas-phase concentrations will be directly correlated with aerosol concentrations, and the physical properties of both must be accurately described to achieve adequate predictions.

[5] Although there has been a tendency for global models to underpredict OC concentrations, this is by no means universal. Chung and Seinfeld [2002] compared their results with observations from the Interagency Monitoring of Protected Visual Environments (IMPROVE) network and reported that OA concentrations in the United States were consistently underpredicted by a factor of 3 or more. Tsigaridis and Kanakidou [2003] showed no systematic bias in polluted areas but reported more than an order-of-magnitude low bias in marine and remote marine regions. Park et al. [2006] showed little bias (∼20%) in predicted OC concentrations when compared with IMPROVE sites. Model studies typically [Kanakidou et al., 2005] predict a global dominance of POA, in contrast with the findings of Zhang et al. [2006] that ambient OA is mostly oxygenated.

[6] Most global SOA modeling studies neglect the potential increase of SOA yields attributable to continuing chemical reactions of semivolatile VOC products that are initially in the gas phase. Tsigaridis and Kanakidou [2003] estimated model sensitivity to secondary chemistry and found that SOA production can increase by approximately 20%. Due to limitations in the two-product formulation, their approach neglected contributions from VOC oxidation products of high and intermediate volatility. Ng et al. [2006] illustrated the importance of “second-generation” chemistry to BSOA formation, especially for multiply unsaturated precursors. For biogenic systems, the short chemical lifetimes of the first-generation products probably result in most of this behavior being captured directly by the existing smog chamber yields. Lane et al. [2008a] found that SOA is greatly overpredicted in the southeastern United States when secondary biogenic species are modeled to undergo volatility reduction by this mechanism. Shrivastava et al. [2006] applied volatility-reducing aging reactions to organic vapors in the regional chemical transport model PMCAMx to semivolatile POA. Murphy and Pandis [2009] implemented an aging mechanism for both POA and ASOA in PMCAMx and found improved correlation with measurements.

[7] It has been shown that SOA formation is nonlinearly dependent on smog chamber NOx concentrations, changing the calculated SOA yield by a factor of 2 or more [Kroll et al., 2006; Ng et al., 2007a]. Recent modeling studies on the global [Henze et al., 2008] and regional [Lane et al., 2008b] scales have included parameterizations of SOA dependence on NOx.

[8] Recent smog chamber studies investigating the oxidation of anthropogenic VOCs [Ng et al., 2007b; Hildebrandt et al., 2009] have shown that SOA formation from toluene may be significantly higher than previous estimates. Similarly, higher than expected yields have been reported for benzene and m-xylene [Ng et al., 2007b]. The relatively slow photooxidation of anthropogenic VOCs, coupled with a thoroughly inconsistent literature on ASOA yield, suggests that existing smog chamber paradigms may be capturing only the first steps in SOA formation chemistry [Hildebrandt et al., 2009]. This hypothesis is consistent with plausible oxidation mechanisms [Atkinson and Arey, 2003] for aromatic compounds, which suggest that the oxidation of one aromatic molecule by a hydroxyl radical, initiated either through ring opening or by aromatic substitution, may occur over as many as three or more distinct steps.

[9] The present work aims to address some shortcomings of previous models by implementing the volatility basis set for SOA in a global general circulation model (GCM); developing an SOA budget using updated, NOx-dependent yield information for terpenes, isoprene, and anthropogenic species; estimating the effect of SOA aging on the global OA budget; and quantifying the effects of water solubility and heat of vaporization on SOA in the free troposphere.

[10] After a description of the model, the results of two simulations are presented and compared: a reference (“basecase”) simulation and one incorporating an aging mechanism (“aging case”). These model predictions are evaluated against regional measurement data for the United States and Europe. Several sensitivities are then discussed, including scenarios that explore the effects of the rate constant in an aging mechanism, the strength of the SOA temperature dependence, the resistance to gas-phase deposition, and the parameterization of SOA yields with NOx concentration on SOA formation and partitioning.

2. Model Description

2.1. Global Chemical Transport Modeling

[11] The primary tool in this work is the so-called “unified” chemistry-climate-aerosol model [Liao and Seinfeld, 2005], which was developed on the basis of the Goddard Institute for Space Studies General Circulation Model II′ (GISS II′ GCM) [Rind and Lerner, 1996] and includes online tropospheric chemistry [Mickley et al., 1999; Wild et al., 2000] and bulk aerosol thermodynamics modules [Adams et al., 1999; Nenes et al., 1999]. The spatial resolution is 4° longitude × 5° latitude, with nine vertical sigma layers centered at 959, 894, 786, 634, 468, 321, 201, 103, and 26 hPa. The top of the model is at 10 hPa. The GCM is initialized with meteorological conditions reflecting July 1979 (“present day”) and associated monthly variable sea surface temperatures and ocean ice coverage. Although the GCM simulation period is July 1979 through July 1980, the model represents a nonspecific, climatologically accurate twentieth century year. We discard the first month of simulation time as model spin-up. The dynamical GCM time step is 1 h, while chemistry and aerosol modules are invoked every 4 h. Chemical kinetics is solved using the SMVGEAR code [Jacobson and Turco, 1994], which employs an adaptive time step to solve the system of stiff differential equations efficiently in each grid cell. A total of 93 gas and aerosol species are included in simulation. A full description of included species and processes is presented by Liao and Seinfeld [2005, and references therein]. The model has been used to study effects of climate change on ozone and particulate matter globally [Racherla and Adams, 2006] and over the eastern United States [Racherla and Adams, 2008].

2.2. Emissions

[12] Isoprene emissions are parameterized as a function of vegetation type, leaf area index, temperature, and solar radiation following the model of Guenther et al. [1995], as described by Wang et al. [1998]. Isoprene (ISOP) emissions total 550 Tg yr−1. Monoterpenes are grouped into four model species (ALPH, LIMO, TERP, and ALCO) as shown in Table 2. Monthly average emissions of terpenes are scaled by solar zenith angle [Chung and Seinfeld, 2002]. Emissions of monoterpenes and sesquiterpenes (SESQ) total 186 and 15 Tg yr−1, respectively.

Table 2. Precursor Grouping and Emissionsa
Model PrecursorsVOCs IncludedEmissions (Tg yr−1)SOA Speciesb
  • a

    Abbreviations are as follows: SOA, secondary organic aerosol; VOCs, volatile organic compounds.

  • b

    MSOAi has the following basic set of C* values: 1, 10, 100, and 1000 μg m−3, as do the other lumped SOA categories (MSOGi, ISOAi, ISOGi, ASOAi, and ASOGi). Therefore, the model tracks a total of 24 scalar quantities to describe SOA (4 saturation concentrations × 3 source categories × 2 phases).

ALPHα-Pinene50MSOA{1,2,3,4}, MSOG{1,2,3,4}
Sabinene and terpinoid ketones20
TERPα- and γ-Terpinene1.4
Terpenoid alcohols30
ISOPIsoprene550ISOA{1,2,3,4}, ISOG{1,2,3,4}
ALK5C7+ aliphatics17ASOA{1,2,3,4}, ASOG{1,2,3,4}
C4 and C5 olefins3.1
OLE2C6+ olefins3.6
Other C7 aromatics1.6
Trimethyl benzenes0.8
Other C8+ aromatics1.1

[13] Anthropogenic species added to the model to assess SOA formation include aliphatics, olefins, and aromatics. Emissions are derived from the Global Emissions Inventory Activity 1999 ( and are grouped in a manner similar to that of the Statewide Air Pollution Research Center SAPRC99 chemical mechanism [Carter, 2000]. Emitted species, lumped simulated species (ALK5, OLE1, OLE2, ARO1, and ARO2), and emission rates are detailed in Table 2.

[14] Emissions of POA are taken from the work of Liousse et al. [1996], which includes both fossil fuel and biomass combustion organic aerosol sources totaling 81 Tg yr−1. Other, more recent, emissions inventories for POA are available. However, POA emissions in current modeling studies differ by up to a factor of 4, as shown in Table 1. Given this uncertainty, and uncertainties in the volatility of POA, evaluation and updating of POA emissions inventories are reserved for future work. In this work, as in all previous studies, the POA is treated as nonvolatile and nonreactive but it acts as an absorbing phase for SOA condensation, forming one homogeneous, pseudo-ideal organic aerosol phase. Because the effect of POA emissions on the SOA simulation presented here is via partitioning, biases in POA emissions may bias the total amount of absorbing material. Despite the use of a relatively old emissions inventory, comparisons with observations (section 3.3) indicate that the model has a slight low bias (typically by a factor of 2 or less) compared with IMPROVE and European Monitoring and Evaluation Programme (EMEP) observations.

2.3. Secondary Organic Aerosol (SOA) Formation and Partitioning

[15] To describe efficiently the partitioning behavior of SOA, we have used the volatility basis set framework [Donahue et al., 2006]. This approach assumes a wide distribution of semivolatile organic products and then separates them into groups of compounds with logarithmically spaced effective saturation concentrations. The mass yields of these products are fit to laboratory SOA formation experiments and accurately capture the behavior of smog chamber SOA. The partitioning equation is as follows:

equation image

where ξi is the fraction of lumped i in the aerosol phase, COA is the total OA mass concentration, Ci* is the effective saturation concentration, and Ci is the total (gas plus aerosol) phase concentration of species i. In words, the total OA concentration is equal to the total amount of each basis set species that exists in the aerosol phase, where the fraction of each species in the aerosol phase is inversely related to the saturation concentration of that species and directly related to the total OA concentration. C* is the set of effective saturation concentrations specified to cover the range of atmospherically relevant conditions or the experimental conditions on which the empirical fits (yields) are based.

[16] The density assumed in the model is 1.4 g cm−3 for all SOA species [Zhang et al., 2005]. The species-specific basis set yields are presented in Table 3. Basis set yields for limonene oxidation were taken from the work of Zhang et al. [2006]. Yields for the oxidation of lumped species ALK5, OLE1, and OLE2 by hydroxyl radical (OH) and ozone (O3) and all nitrate radical (NOx) yields are identical to those used by Lane et al. [2008a]. Aromatic yields (ARO1 and ARO2) were obtained from Hildebrandt et al. [2009]; as a first approximation, all aromatics are assumed to have yields similar to oxidation of toluene. Additional basis set yields were estimated based on the methods presented by Stanier et al. [2008], using data from the following experiments: Hoffmann et al. [1997], Odum et al. [1997], and Ng et al. [2006, 2007b] for ALPH; Ng et al. [2006] for TERP; Hoffmann et al. [1997], Griffin et al. [1999], and Ng et al. [2006] for ALCO; Griffin et al. [1999] and Ng et al. [2007b] for TERP; Griffin et al. [1999], Kroll et al. [2006], and Ng et al. [2007b] for SESQ; and Kroll et al. [2005, 2006], Dommen et al. [2006], and Kleindienst et al. [2007] for ISOP. Recently observed high yields for the reaction of isoprene with NOx [Ng et al., 2008] were not implemented.

Table 3. Secondary Organic Aerosol Mass Yield Parameters
ReactionSpeciesSaturation Concentration (μg m−3)
VOC + OH and VOC + O3 in high NOxALPH0.
VOC + OH and VOC + O3 in low NOxALPH0.
VOC + NO3Monoterpenes0.

[17] Gas-phase oxidation rates of SOA-forming VOCs are calculated within the model's tropospheric chemical mechanism. Oxidation by O3, OH, and NOx is considered. The SOA is described by a set of 12 particulate-phase species (MSOA{1,2,3,4}, ISOA{1,2,3,4}, and ASOA{1,2,3,4}) and 12 corresponding gas-phase species (see Table 2). These are split into three groups to track the effects of different precursors: monoterpenes and sesquiterpenes, isoprene, and anthropogenics. Oxidation products from each group of precursors are lumped together. The three groups, however, are isolated as separate model basis sets with effective saturation concentrations {1, 10, 100, or 1000} μg m−3 at 298 K. This basis set was chosen because it covers the range of available smog chamber data. The sensitivity of our results to this choice is explored in section 3.4.

2.4. NOx Dependence of SOA Yields

[18] To describe the dependence of SOA yield on NOx concentrations, we have used the approach by Lane et al. [2008a]. Low NOx and high NOx yields are defined, and a linear combination of the two is used:

equation image

where αi,high and αi,low are the mass-based yields of a VOC oxidation product under high NOx and low NOx conditions, respectively, and αi is the effective yield (Table 3). The B factor is a measure of the amount of RO2 radicals that react via the NO + RO2 pathway:

equation image

NO concentrations are used instead of NOx (NO + NO2) due to the mechanistic argument for NOx dependence. The major shortcoming of this approach is the assumption of linearity of SOA yield with respect to the NOx branching ratio, which may not be the case. For the oxidation of isoprene, there is a local maximum in SOA production when initial NOx concentrations are between the NOx-free and “high-NOx” cases [Kroll et al., 2006]. This local maximum is not captured by this framework, which thus may lead to the underprediction of isoprene and perhaps general SOA formation. A sensitivity test will explore the case of B = 0, discussed in section 3.4.2.

2.5. Temperature Dependence of SOA Partitioning

[19] The effective saturation concentration of a species is related to temperature via the Clausius-Clapeyron equation:

equation image

where T0 is a reference temperature and ΔHvap is the effective heat of vaporization. Since the temperature in the troposphere varies greatly, ΔHvap is an important parameter. This work assumes that ΔHvap = 30 kJ mol−1 for the basecase [Chung and Seinfeld, 2002], and sensitivity analysis is performed. This single value of effective ΔHvap is applied to all lumped SOA species (i.e., each combination of SOA precursor and volatility bin).

2.6. Aging of Semivolatiles

[20] In the basecase simulation we are assuming that the existing yields of a smog chamber capture the complete evolution of the SOA, that is, all the generations of chemical aging. In a separate simulation (the “aging case”; see section 3.2), we have investigated the potential importance of SOA volatility reduction by the reaction of secondary organic gases with the hydroxyl radical. The gas-phase molecules that react are assumed to form an equal mass of a species with a lower saturation concentration by a factor of 10. The framework, similar to that presented by Robinson et al. [2007], is illustrated in Figure 1. Based on the work of Shrivastava et al. [2008] and Murphy and Pandis [2009], an OH rate constant of 10 × 10−12 cm3 molecule−1 s−1 is assumed for all gas-phase ASOA species. Since it appears that the smog chamber yields used capture the complete reactions of biogenic precursors, BSOA species are not aged in the aging case [Lane et al., 2008b]. Primary organic aerosol is treated as nonvolatile and thus does not participate in such reactions in this study.

Figure 1.

Organic aerosol aging framework. Solid lines represent reaction with hydroxyl radical; dashed lines represent gas–particle partitioning.

2.7. Dry and Wet Deposition

[21] Wet removal of gas-phase organic semivolatiles is calculated based on the Henry's law constant. For consistency with previous work [Chung and Seinfeld, 2002; Henze et al., 2006], a value of 1 × 105 mol L−1 atm−1 (1 atm =1.01325 × 105 N m−2) is used in the basecase simulation for all gas-phase SOA products. This choice, based on the assumption that the products of VOC oxidation are mostly di-acids, which are estimated to have a high water solubility (H ∼ 106 mol L−1 atm−1), is near the limit of complete solubility and is likely too high, on the basis of Henry's law constants of other VOC oxidation products, including alcohols, ketones, and aldehydes [Sander, 1999]. We will discuss the sensitivity of our results to this parameter in section 3.4. Secondary organic gas- and aerosol-phase dry deposition velocities are calculated by using a resistance-in-series model based on work by Koch et al. [1999]. The surface layer resistance for the gas phase depends on the same Henry's law constant. Aerosol-phase secondary organics are activated into cloud droplets and scavenged by falling raindrops. The efficiency for this process is specified [Chung and Seinfeld, 2002] as 80% for all SOA.

3. Results

3.1. Basecase

[22] Figure 2 shows the basecase annual average total (OA) concentrations in the first model layer. The model suggests a continental background OA concentration of 1–2 μg m−3, the areas of highest OA concentration corresponding to the major biomass-burning regions [Duncan et al., 2003] in South America and Africa, although elevated concentrations also are found over industrialized Europe and Southeast Asia. Predicted annual average OA concentrations in parts of Europe are high, owing to the high emission of POA in the inventory [Liousse et al., 1996].

Figure 2.

Basecase annual average total organic aerosol concentration (μg m−3) at ground level. Cells in black exceed 15 μg m−3 (maximum concentration, 22.8 μg m−3). The average concentration at ground level is 1.36 μg m−3.

[23] The fraction of OA that is predicted to be secondary at ground level is shown in Figure 3. The surface SOA fraction reaches as high as 60% over Brazil, where biogenic precursor emissions are high, and over North America, where primary emissions are relatively low. For the rest of the globe, the predicted contribution of SOA is relatively low. Zhang et al. [2007] used AMS measurements to show that nearly 95% of remote and 63% of urban submicron OA is OOA. Under the simplifying assumption that OOA is synonymous with SOA, it is clear that this model is biased toward POA. While the POA is treated here as nonvolatile and nonreactive, volatilization and subsequent oxidation of POA by hydroxyl radicals are expected to shift this distribution significantly [Robinson et al., 2007].

Figure 3.

Basecase annual average ratio of secondary organic aerosol (SOA) to total organic aerosol (OA) concentration at ground level.

[24] SOA fraction is similarly low for the rest of the troposphere, as shown in Figure 4a. The fraction of SOA is predicted to be greater for higher altitudes due to longer lifetime of some SOA precursors and secondary organic gases (SOGs) compared with aerosol-phase organics. Higher-volatility SOGs that do not contribute to surface SOA may be transported via convection and turbulent diffusion to altitudes where the temperature effect on volatility causes a significant fraction of those gases to condense, even though there is very little preexisting OA. This high-altitude SOA is visible in Figure 5, especially at the equatorial tropopause.

Figure 4.

(a) Basecase and (b) aging case annual average zonal ratio of SOA to total OA.

Figure 5.

Basecase annual average zonal SOA distributions. Concentrations are in micrograms per cubic meter at 1 atm and 298 K.

[25] The basecase surface SOA concentrations, total and resolved by precursor, are shown in Figure 6. At the surface, over 90% of model SOA comes from the oxidation of biogenic precursors. This prediction is consistent with previous model estimates (see Table 1). Primary organic aerosol is present in very high concentrations locally but has the sharpest concentration gradients. This is due to the short (1.5 day) time scale for hydrophobic to hydrophilic conversion and the efficient removal of hydrophilic aerosol through wet deposition. Terpene SOA, that is, SOA formed from the oxidation of monoterpenes and sesquiterpenes, also shows large gradients of concentration due to localized production. Biogenic VOCs have short chemical lifetimes (a few hours) and are not transported significantly away from the source regions. Anthropogenic VOCs, however, have a much longer chemical lifetime than the biogenic species, averaging 1.5 days for ARO1 and 3.8 days for ARO2. This relatively long chemical lifetime allows for transport away from source regions and outside of the boundary layer. This effect is seen in the source- and volatility-resolved annual average atmospheric SOA burdens, production rates, and atmospheric lifetimes shown in Table 4. There are significant differences in SOA lifetime among sources. The oxidation of anthropogenic VOCs outside of the boundary layer creates a comparatively large source of SOA in a region without significant precipitation and without means for dry deposition. Within a given source class, there is usually a general trend of increasing lifetime with increasing volatility. For example, there is approximately a 20% increase in lifetime from low- to high-volatility terpene and isoprene SOA. However, there is almost no change for ASOA. These trends illustrate the slight preference for removal of aerosol-phase organics over gas-phase organics in the basecase model.

Figure 6.

Basecase annual average contributions to OA. “Terpene-SOA” includes SOA from the oxidation of monoterpenes and sesquiterpenes. The x-axes indicate units of annual average surface concentration. Note the different scales of each plot.

Table 4. Basecase Source and Volatility-Resolved Secondary Organic Aerosol Budgeta
 SOA Burdenb (Tg)Productionc (Tg yr−1)Lifetime (days)
  • a

    C* is the saturation concentration.

  • b

    Primary organic aerosol atmospheric burden is constant across all simulations and is 1.1 Tg.

  • c

    Net production rate is assumed to be equal to deposition flux.

Terpene SOA0.3220.85.7
Isoprene SOA0.166.489
Anthropogenic SOA0.061.6213.2
C* = 1 (μg m−3)0.316.36.6
C* = 10 (μg m−3)0.210.57
C* = 100 (μg m−3)0.041.88
C* = 1000 (μg m−3)0.0070.367.3

3.2. Effects of Chemical Aging

[26] To investigate the effect of multiple generations of chemistry on SOA species, we have allowed for anthropogenic SOGs to react with OH, as described in section 2.6 and Figure 1. The surface layer ASOA concentrations are presented in Figure 7. The average surface concentration of ASOA is a factor of 4.6 greater in the aging case than in the basecase. In the aging case, the ASOA is more broadly distributed than BSOA and ASOA in the basecase—in part because of the long chemical lifetime of the anthropogenic VOCs compared with that of the biogenic VOCs, but also because of the additional source of low-volatility material. The total SOA enhancement at the surface is significant, generally by a factor of 1.2–1.6 over North America and Eurasia, near a factor of 3 in northern Africa and the Middle East, and a factor of 6 over midlatitude remote marine areas. The anthropogenic fraction of SOA at the surface, previously considered negligible [Chung and Seinfeld, 2002], increases from 7.6% to 28%. The major effect of this type of volatility evolution, however, is noticed outside of the boundary layer. The annual average total SOA burden for the aging case increased by 82% over the basecase, from 0.54 to 0.98 Tg, and there was an additional 9.4 Tg yr−1 (32%) of SOA formation. The zonal average ratio of total SOA in the aging case to that in the basecase is shown in Figure 8. The most marked SOA enhancement is in the northern midlatitude free troposphere; the enhancement in the free troposphere around 30°N is almost a factor of 3. Heald et al. [2005] illustrated a missing source of SOA in the free troposphere when comparing GEOS-Chem model predictions with data from the ACE-Asia field campaign.

Figure 7.

Aging case annual average anthropogenic SOA at ground level. The average concentration is 0.104 μg m−3. Compare with lower right plot of Figure 6.

Figure 8.

Annual average zonal ratio of total SOA concentrations in the aging case to those in the basecase. Aging of anthropogenic SOA enhances total SOA concentrations in the midlatitude free troposphere by up to a factor of 2.8.

[27] The effect of the aging mechanism on volatility-resolved global SOA burden is illustrated in Figure 9. The high-volatility material in the basecase acts as an SOA precursor and is processed to form the lower-volatility material in the aging case. The additional SOA burden here is due to a redistribution of the products of the initial oxidation.

Figure 9.

Volatility-resolved annual average SOA and secondary organic gas (SOG) burdens in the basecase (SOA = 0.54 Tg) and the anthropogenic SOG aging case (SOA = 0.98 Tg).

3.3. Model Evaluation

[28] Figure 10 shows the comparison of monthly averaged model predicted OA concentrations with measured OA concentrations in the United States and in Europe. Measurement data are usually reported as organic carbon (OC) in micrograms of carbon per cubic meter, while the model predicts organic mass (OM) concentrations. Zhang et al. [2005] reported a range of OM:OC ratios between 1.2 and 2.2 for the Pittsburgh Air Quality Study and suggested an average of 1.6 ± 0.2 for the urban aerosol. Turpin and Lim [2001] suggested the same OM:OC factor for urban OA and 2.1 ± 0.2 for more-oxygenated background aerosols. Polidori et al. [2008] suggested that the conversion factor approaches 1.9–2.1 as the aerosol ages. A conservative OM:OC factor of 1.8 was used for the comparison to represent a moderately oxygenated ambient OA, which is toward the lower limit of expected “global average” OM:OC ratios. The total OA from both the aging case and the basecase is shown. Since the modeled surface OA according to the model is still largely primary, the absolute difference in total OA is not as dramatic as the difference in SOA alone, which was illustrated previously (e.g., Figure 8).

Figure 10.

Total organic aerosol model comparison with the IMPROVE and EMEP observation networks. Dashed lines indicate a factor of 2 difference in concentration. Each point represents a monthly average measurement (x) of organic carbon multiplied by the ratio OM:OC = 1.8, and a model prediction (y) from the corresponding month/location.

[29] More than 160,000 measurements from the IMPROVE network were averaged by location and season and span the year range between 1988 and 2004. The averages were developed from between 5000 and 10,000 measurements per year in the 1990s and from over 10,000 measurements per year from 2000 to 2004. Measurements from the EMEP network were retrieved as monthly averages in 2002 and 2003.

[30] Statistics for the comparisons are shown in Table 5. For the IMPROVE network, consideration of ASOA aging improves model performance, reducing the magnitude of underprediction. Fractional biases of the model when compared against IMPROVE data are −0.41 and −0.26 for the basecase and aging cases, respectively (i.e., the model tends to underpredict by ∼41% and 26%, respectively). The discrepancy between measurements and model predictions is mostly by less than a factor of 2, with 69% and 73% of points falling within this range for the basecase and aging simulations, respectively. For the EMEP sites, however, inclusion of ASOA aging causes increases in the error and further biases already high OA predictions. More than half of the comparisons differ by more than a factor of 2. The average concentrations in this region, however, are much greater than in the IMPROVE data set. As shown in Figure 2, OA is predicted to be mostly primary, while field measurements [Zhang et al., 2007] have shown that even urban OA is mostly oxygenated. This discrepancy is likely due to the inert treatment of POA and is expected to be resolved in future model versions.

Table 5. Model Evaluation Statisticsa
Statistical MetricIMPROVE (N = 721)EMEP (N = 142)
  • a

    Abbreviations are as follows: EMEP, European Monitoring and Evaluation Programme; IMPROVE, Interagency Monitoring of Protected Visual Environments.

Fractional error0.550.490.740.73
Fractional bias−0.41−
Error (μg m−3)0.950.926.26.2
Bias (μg m−3)−0.65−0.361.61.9
Correlation coefficient (R)0.660.660.20.2
Measured mean (μg m−3)
Predicted mean (μg m−3)

3.4. Sensitivity Analysis

[31] Simulation results for all sensitivities are summarized in Table 6. A discussion of these results follows.

Table 6. Summary of Simulations and Resultsa
Simulation NameAdjusted ParametersSOA Burden (Tg)SOA Productionb (Tg yr−1)SOA Lifetime (days)
  • a

    Abbreviations are as follows: SOA, secondary organic aerosol; SOG, secondary organic gases.

  • b

    Net production rate is assumed to be equal to deposition flux.

  • c

    Aging of anthropogenic SOGs.

  • d

    Reduction of gas-phase depositional flux via 100-fold reduction in Henry's law constant.

  • e

    Combination of aging and low-solubility cases.

  • f

    Basecase SOA formation in the limit of [NOx] → 0.

AgingckSOG + OH = 10 × 10−120.9838.19.4
Slow agingkSOG + OH = 1 × 10−120.731.68.1
Low solubilitydH = 103 mol L−1 atm−12.454.816
High ΔHvapΔHvap = 60 kJ mol−11.538.214.3
Low solubility, agingeH, kSOG + OH3.375.111.7
Low NOxfB = 00.7236.67.2

3.4.1. Aging Rate for SOA

[32] To investigate the sensitivity of SOA formation to the chosen rate constant in the SOA aging mechanism, SOA formation was also quantified for the case of k = 1 × 10−12 cm3 molecule−1 s−1, an order of magnitude lower than the default aging case. With this reduction in rate constant, the aged SOA formation is reduced by 71% (from 9.2 to 2.7 Tg yr−1). Compared with the basecase, the additional production from the low-rate aging case results in a 30% increase in atmospheric burden, owing to the increased formation outside the boundary layer (Table 6).

3.4.2. NOx-Dependent SOA Yield Parameterization

[33] To estimate the effect of the NOx-dependent yield parameterization, we calculated an upper bound of SOA formation using the presented yields. The “low-NOx” simulation assumes all SOA is formed under low-NOx conditions (B = 0), where, except for sesquiterpenes, yield is increased. Since the relationship between SOA yield and NOx concentration may be nonlinear and nonmonotonic [Kroll et al., 2006], this assumption is testing the limits of the parameterization, not the physical system.

[34] Under low-NOx conditions, the annual SOA production is estimated at 36.6 Tg yr−1, an additional 7.7 Tg yr−1 (27%) over the basecase. Isoprene SOA formation is most affected by the change, increasing by 50% (6.5 to 9.8 Tg yr−1) from the basecase, whereas terpene SOA and ASOA production each increase by approximately 20% (20.8 to 24.9 and 1.6 to 2.0 Tg yr−1, respectively). Spatial distributions of OA in this case are very similar to those of the basecase, although average concentrations are increased from 0.29 to 0.36 μg m−3 (25%), and atmospheric burden is increased from 0.54 to 0.72 Tg (33%), resulting in a lifetime increase from 6.8 to 7.2 days. This lifetime effect is due to the new distribution of SOA; it includes more isoprene SOA, which is longer-lived than the previously dominant terpene SOA (see Table 4). Isoprene SOA is longer lived than monoterpene SOA due to the percentage of isoprene oxidized outside of the boundary layer compared with that of the larger terpenes.

3.4.3. Deposition Rates

[35] By altering the Henry's law constant for SOGs, it is possible to investigate different possibilities for SOA deposition. As shown by Henze and Seinfeld [2006], predicted SOA concentrations, especially at higher altitudes, are sensitive to the choice of Henry's law constant for the model. Figure 11 shows the enhancement to SOA concentrations throughout the troposphere by reduction of the Henry's law constant to 103 mol L−1 atm−1, which is a plausible value for what might be expected among SOA species. Some of the additional 25.9 Tg yr−1 of SOA produced in this example is present at the surface but in limited amounts compared with that in the free troposphere, where the long-lived SOGs are now transported. As these vapors reach the upper troposphere, equilibrium is shifted toward the aerosol phase, as indicated by the enhanced aerosol-phase deposition illustrated in Figure 12. The resulting average lifetime of gas-phase secondary organics is increased from 6.8 to 38 days. This effect is akin to adding an aerosol source, and average SOA lifetimes increase from 6.8 days in the basecase to 16 days. The increased concentrations of gas-phase condensable material, coupled with the low volatility of SOA at high altitudes due to low temperatures, create a dramatic increase (344%) in SOA burdens at steady state. Since the bulk of the additional SOA is formed outside the boundary layer, lifetimes are increased beyond what may be expected for primary aerosol species (e.g., black carbon). The higher-volatility species are most sensitive to the gas-phase solubility since they are preferentially partitioned to the gas phase and experience the longest lifetimes in this scenario (up to 41 days). While high-volatility species are not typically considered important for SOA formation, they may become important if slow deposition allows time for sufficient aging of ASOA.

Figure 11.

Annual average absolute difference of total SOA in low-solubility case minus that in basecase. Concentrations are in micrograms per cubic meter at 1 atm and 298 K.

Figure 12.

Annual average volatility-resolved SOA and SOG wet and dry deposition in basecase and low-solubility case. Basecase aerosol deposition is 28.9 Tg yr−1; aerosol deposition in low gas-phase solubility case increases to 54.8 Tg yr−1.

3.4.4. Aging ASOA With Low Water Solubility

[36] To investigate possible synergistic effects between aging and deposition, the “low-solubility aging” case combines the aging mechanism with the changes to deposition included in the low-solubility case. By including aging of anthropogenic SOGs in the low-solubility case, SOA production increased from 54.8 to 75.1 Tg yr−1. Recall that annual SOA production between the basecase and aging case increased from 28.9 to 38.1 Tg yr−1. The overall effect is greater than the sum of the two effects because the additional high-volatility SOGs that accumulate in the low-solubility case act as an SOA precursor when ASOA aging is considered. The ratio of SOA concentrations in the low-solubility versions of the aging case and basecase is shown in Figure 13. The SOA enhancements are on the order of a factor of 2.6 in the midlatitude free troposphere. The additional source of SOA from aging is very similar to that shown in Figure 8 for the ratio of the basecase and aging case. Aging of ASOA may have a much larger effect than reported if SOA vapors are less water soluble than originally assumed. Analogously, if POA is treated as semivolatile and undergoes a similar aging mechanism, as suggested by Shrivastava et al. [2008], the water solubility of gas-phase organic compounds may become even more important.

Figure 13.

Annual average ratio of SOA in low-solubility aging and low-solubility simulations.

3.4.5. Effect of Temperature on SOA Volatility

[37] Volatility behavior of SOG/SOA at the range of temperatures encountered in the troposphere is strongly influenced by the heat of vaporization. Pathak et al. [2007] showed that the effective heat of vaporization of α-pinene SOA is roughly 30 kJ mol−1. Modeling studies have used a range of values, including 30 kJ mol−1 [Murphy and Pandis, 2009], 42 kJ mol−1 [Chung and Seinfeld, 2002; Henze and Seinfeld, 2006] and as high as 150 kJ mol−1 [Strader et al., 1999]. Saathoff et al. [2009], using a two-product modeling framework, suggested that the temperature dependence of α-pinene and limonene SOA was well captured using heats of vaporization of approximately 25 and 60 kJ mol−1 for the two products, respectively. To investigate the model sensitivity to this parameter, we have doubled the value to 60 kJ mol−1. The predicted SOA burden for this case tripled to 1.5 Tg, due partly to an increase in production (net condensation) from 29 to 39 Tg yr−1 and also to a doubling of the average OA lifetime from 6.8 to 14.5 days. This lifetime effect is similar to that noticed for the low-solubility case, where aerosol lifetimes increase due to increased concentrations outside of the boundary layer. The change in partitioning of SOA is shown in Figure 14, which illustrates the relationship between altitude and the fraction of a simulated secondary organic species in the aerosol phase. At the surface, temperatures are warm, but concentrations are high, so SOA partitions to the aerosol phase. As altitude increases, temperature begins to drop, but dilution is the major effect, and SOA evaporates. As temperatures continue to drop, condensation begins again. The temperature effect is stronger for the ΔHvap case, and SOA begins to recondense at a much lower altitude.

Figure 14.

Annual average volatility-resolved ξi (see equation (1)) as a function of altitude in the basecase (solid lines) and the ΔHvap case (dashed lines).

[38] At the surface, the average SOA concentration increases by 15%, but some especially warm areas, namely, the Sahara desert, parts of the Amazonian rain forest, and northeastern India, experience modest decreases in SOA concentrations, also on the order of 15%. This effect is secondary compared with the effects outside the boundary layer.

4. Conclusions

[39] Annual global SOA formation from monoterpenes, sesquiterpenes, isoprene, and anthropogenic sources has been estimated to be 17.2, 3.9, 6.5, and 1.6 Tg yr−1, respectively. Oxidation of anthropogenic SOGs (aging) may contribute an additional 2.7–9.6 Tg yr−1 to SOA formation, adding to the regional background OA and also contributing significantly to free-tropospheric OA. SOA processing through volatility-reducing reactions of SOGs significantly enhances SOA lifetimes (Table 4). Elucidation of the details of the aging mechanism or a more accurate parameterization may help to constrain this possible SOA source. If POA is treated in a similar manner, there will be dramatic changes in model-predicted global free-tropospheric OA concentrations.

[40] Updated yield information has brought model predictions into closer agreement [Chung and Seinfeld, 2002; Tsigaridis and Kanakidou, 2003] with measurements. The model has a modest low bias against IMPROVE data of 41% and 26% for the basecase and aging case, respectively. The parameterization of SOA yields in the volatility basis set framework as applied allows modeling of aerosol partitioning across the atmospherically relevant range of saturation concentrations, 0.1 to 104μg m−3. The revised model has significantly improved performance at predicting low OA concentrations observed in rural and remote regions.

[41] SOA formation, especially at high altitudes, is a strong function of vapor-phase solubility in water and SOA heat of vaporization (ΔHvap). While high-altitude measurements suggest that model-predicted OA concentrations have a low bias aloft [Heald et al., 2005], model emission inventories may not be the main source of this error; rather, this bias can potentially be ascribed to inadequate estimation of OA physical properties or to the poor understanding or model implementation of OA processing.


[42] This research was supported EPA-STAR 83337401. This paper has not been subject to EPA's required peer and policy review, and therefore does not necessarily reflect the views of the EPA. No official endorsement should be inferred.