SEARCH

SEARCH BY CITATION

Keywords:

  • modeling;
  • molecular approach;
  • secondary organic aerosol

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Model Development
  5. 3. Simulation of Organic Aerosols Over Europe
  6. 4. Sensitivity Analysis and Investigations of Organic Aerosol Formation
  7. 5. Conclusion
  8. Acknowledgments
  9. References
  10. Supporting Information

[1] A secondary organic aerosol (SOA) model, the Hydrophilic/Hydrophobic Organic model (H2O), is presented and evaluated over Europe. H2O uses surrogate organic molecules to represent the myriad of SOA species and distinguishes two kinds of surrogate species: hydrophilic species (which condense preferentially into an aqueous phase) and hydrophobic species (which condense only into an organic phase). These surrogate species are formed from the oxidation in the atmosphere of volatile organic compounds. H2O includes several important processes, including the effect of nitrogen oxides (NOX) on SOA formation, the dissociation of organic acids in an aqueous phase, the oligomerization of aldehydes, the non-ideality of the particle phase and the hygroscopicity of organics. Concentrations of organic aerosols were simulated over Europe from July 2002 to July 2003 for comparison with measurements of the European Monitoring Evaluation Program (EMEP). In H2O, primary organic aerosols (POA) are considered as semi-volatile organic compounds (SVOC) present in both the gas phase and the particle phase. Taking into account the gas-phase fraction of SVOC increases significantly organic PM concentrations, particularly in winter, in better agreement with observations. The impacts on organic aerosol formation of ideality, of the choice of the parameterization for isoprene SOA formation, and of the OM/OC ratio of the model were also investigated. Assuming ideality in H2O was found to lead to a small decrease in OM. Compared to a two-product parameterization, the parameterization of Couvidat and Seigneur [2011] for SOA formation from isoprene oxidation leads to a significant increase in isoprene SOA by taking into account their hydrophilic properties and suggests that most models may currently underestimate isoprene SOA.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Model Development
  5. 3. Simulation of Organic Aerosols Over Europe
  6. 4. Sensitivity Analysis and Investigations of Organic Aerosol Formation
  7. 5. Conclusion
  8. Acknowledgments
  9. References
  10. Supporting Information

[2] Fine particulate matter (PM) with an aerodynamic diameter less than 2.5 μm (PM2.5) is regulated because of its impact on human health. Furthermore, PM2.5 degrades atmospheric visibility and influences climate change. Three-dimensional air quality models, which estimate PM2.5 formation, tend to underestimate organic aerosol concentrations due to the complexity of the phenomena involved (gas and particle phase chemistry, oligomerization, hygroscopicity, non-ideality of particulate liquid phases) and to the large number of organic species involved originating from diverse anthropogenic and biogenic sources. Particulate organic matter (OM) represents a large fraction of the particulate mass, typically between 20 and 60% [Kanakidou et al., 2005; Yu et al., 2007; Q. Zhang et al., 2007]. Therefore, efforts have to be made to represent OM as accurately as possible in models. Organic PM is either primary (directly emitted as particles) or secondary (formed in the atmosphere by the absorption of the low-volatility oxidation products of gaseous species into an organic phase or by absorption of species with a high affinity for water into an aqueous phase) [Kanakidou et al., 2005; Carlton et al., 2009]. Both primary organic aerosols (POA) and secondary organic aerosols (SOA) are currently poorly understood [Kanakidou et al., 2005; Carlton et al., 2009; Jimenez et al., 2009].

[3] Five years ago, models used to consider POA as non-volatile whereas experimental studies [Robinson et al., 2007] have shown that some POA are in fact condensed semi-volatile organic compounds (SVOC), which exist in both the gas phase and the particle phase. Consequently, the amount of POA depends on the dilution of the aerosol, temperature (if the temperature decreases, the volatility of SVOC decreases) and SVOC present in the gas phase, which can be oxidized to form less volatile compounds. For example, SVOC oxidation in the gas phase was shown to be an important source of SOA in several modeling studies concerning the MILAGRO campaign [Dzepina et al., 2011; Hodzic et al., 2010; Tsimpidi et al., 2011; Shrivastava et al., 2011]. The representation of POA in emission inventories (which typically suppose that POA are non-volatile) has therefore to be rethought because they are based on PM measurements after some significant dilution of the emissions and do not account for the gaseous fraction of the SVOC present in POA.

[4] SOA can be modeled with different approaches. One approach is the two lumped product method [Odum et al., 1996; Schell et al., 2001], which is based on empirical data obtained in smog chambers. Another approach is the volatility basis set [Donahue et al., 2006], which treats the chemical evolution of an aggregate distribution of semi-volatile material according to their volatilities. Finally, the molecular approach [Pun et al., 2002, 2006; Griffin et al., 2003; Tulet et al., 2006] associates experimental data with several molecular structures, which are surrogates of a large number of SOA species, to extrapolate SOA formation from smog chambers to the atmosphere. In the molecular approach, several processes, which are often not taken into account in the other approaches can be readily estimated (e.g., absorption into an aqueous phase, oligomerization, hygroscopicity, non-ideality) and treated explicitly in the model.

[5] In this work, a model based on the molecular approach is presented: The Hydrophilic/Hydrophobic Organic model (H2O). This model includes primary SVOC, oxidation of several precursors (aromatics, isoprene, monoterpenes, sesquiterpenes) under several conditions (oxidation by OH under high-NOX and low-NOX conditions, oxidation by O3 and by NO3) and several processes (absorption into an organic phase or an aqueous phase, oligomerization, hygroscopicity and non-ideality). The model formulation is described first. Next, this model is evaluated over Europe with measurements available for a one-year period and the results are discussed. Finally, the sensitivity of the model to several key parameters is investigated.

2. Model Development

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Model Development
  5. 3. Simulation of Organic Aerosols Over Europe
  6. 4. Sensitivity Analysis and Investigations of Organic Aerosol Formation
  7. 5. Conclusion
  8. Acknowledgments
  9. References
  10. Supporting Information

2.1. Overview

[6] Previous work based on analysis of experimental data [Saxena et al., 1995] and modeling [Chang et al., 2010] has suggested the importance of the absorption of organic species into aqueous particles. Accordingly, H2O, which is based on the conceptual approach of the AER/EPRI/Caltech (AEC) model [Pun et al., 2002, 2003, 2006; Kim et al., 2011a], distinguishes two types of surrogate SOA species: hydrophilic species (which condense mainly into an aqueous phase) and hydrophobic species (which condense only into an organic phase due to their low affinity with water). Hydrophilic species may condense into an organic phase in the absence of aqueous particles, i.e. at very low relative humidities (see Table 1). Distinction between hydrophobic and hydrophilic compounds is based on their octanol/water coefficient [Pun et al., 2006] or their partitioning between the organic and the aqueous phases [Couvidat and Seigneur, 2011]. Highly oxygenated compounds (like tetrols), organic acids (which can dissociate in an aqueous phase) and aldehydes (which can be hydrated in water and undergo oligomerization [Jang et al., 2005; Pun and Seigneur, 2007]) are generally hydrophilic. Lowly oxygenated compounds with long carbon chains and aromatics are generally hydrophobic. As H2O uses molecular structures, the organic carbon (OC) can be estimated directly from the organic matter for the comparison between the model and OC measurements (see section 3).

Table 1. Properties of the Different Surrogate SOA Species
SurrogateTypeaMolecular StructureMWbHcP0dΔHvapeComments
  • a

    Type A: hydrophilic species, type B: hydrophobic species, type C: hydrophobic non-volatile species, which is not used to compute activity coefficients.

  • b

    Molecular weight [g.mol−1].

  • c

    Henry's law constant [(μg/μg water)/(μg/m3)] at 298 K.

  • d

    Saturation vapor pressure [torr] at 298 K.

  • e

    Enthalpy of vaporization [kJ.mol−1].

  • f

    Deliquescence relative humidity: if RH < DRH, the species is solid (Type B), if RH > DRH, the species is liquid (Type A).

BiMTAmethyl tetrol1360.8051.45 × 10−638.4-
BiPERAmethyl dihydroxy dihydroperoxide1680.1112.61 × 10−638.4-
BiDERAmethyl tetrol1362.804.10 × 10−738.4-
BiMGAAmethyl glyceric acid (MGA)1201.13 × 10−21.4 × 10−543.2pKa = 4.0
BiNGABnitrate derivative of MGA165-1.4 × 10−543.2See equation (1)
BiNIT3Bmethyl hydroxy trinitrate butane272-1.45 × 10−638.4-
 
BiA0DApinonaldehyde1684.82 × 10−52.70 × 10−450See equation (2)
BiA1DAnorpinic acid1702.73 × 10−32.17 × 10−750pKa = 3.2
BiA2DApinic acid1866.52 × 10−31.43 × 10−750pKa1 = 3.4, pKa2 = 5.1, DRH = 0.79f
BiNITANitrooxy-limonene-1-ol215-2.5 × 10−6109-
BiBlPBC15 hydroxy nitrate aldehyde298-6.0 × 10−10175-
BiBmPBC15 oxo aldehyde236-3.0 × 10−7175-
AnBlPBmethyl nitro benzoic acid167-6.8 × 10−850-
AnBmPBmethyl hydroxy benzoic acid152-8.4 × 10−650-
AnClPCNo structure167----

[7] H2O uses the same conceptual structure as AEC as described by Pun et al. [2002]. Activity coefficients are computed with the UNIversal Functional group Activity Coefficient (UNIFAC) thermodynamic model [Fredenslund et al., 1975] but does not take into account the impact of inorganic species on the activity coefficients of hydrophilic organic species. Hygroscopicity is computed with the Zdanovskii-Stokes-Robinson (ZSR) method as used by Meng et al. [1998]. However, H2O was optimized to be computationally efficient by using the parameters of Liu and Zhang [2008] for the method of Newton-Raphson presented in Pun et al. [2002] and by removing useless but time consuming data in the call of UNIFAC. It includes also more organic species than AEC (coming mainly from isoprene and primary SVOC oxidation) and accounts for processes that had not been identified when AEC was developed (oligomerization, impact of NOX on the chemistry of SOA formation). The hydrophilic module was also corrected to take into account infinite dilution as reference for partitioning instead of the mixture dilution (i.e., the mixture of water and the other organic compounds) when Henry's law is used.

[8] To simulate OM formation, H2O is implemented in the Polair3D air quality model [Sartelet et al., 2007] of the Polyphemus air quality platform [Mallet et al., 2007] and is coupled with the Carbon Bond 05 model (CB05) [Sarwar et al., 2008] for the gas-phase chemistry, ISORROPIA [Nenes et al., 1998] for the formation of inorganic aerosol and the SIze REsolved Aerosol Model (SIREAM) [Debry et al., 2007] for simulating the dynamics of the aerosol size distribution. The amount of liquid water and the pH computed with ISORROPIA are used to compute the partitioning of hydrophilic compounds between the gas phase and the aqueous phase as done by Pun et al. [2002]. CB05 was modified to take into account SOA formation [Kim et al., 2011b].

2.2. Secondary Organic Aerosol Module

[9] In H2O, SOA are formed from 4 classes of precursors: aromatic compounds, isoprene, monoterpenes and sesquiterpenes. For these classes of precursors, which include a great number of species, only a few surrogate precursor species are used to represent all the species as customary in gas-phase chemical mechanisms (for example, the compound noted HUM, for humulene, is used to represent all sesquiterpenes). Properties of the surrogate SOA species and reactions added to CB05 are listed in Tables 1 and 2, respectively.

Table 2. Reactions Leading to SOA Formationa
ReactionKinetic Rate Parameter (s−1 or Molecule−1.cm3.s−1)
  • a

    Oxidants may be present as both reactants and products so that a reaction added to CB05 will not affect the original photochemical oxidant concentrations. MeO2 and C2O3 are respectively the methylperoxy radical and the peroxyacetyl radical.

ISOP + OH [RIGHTWARDS ARROW] ISOR + OH2.54 × 10−11 × exp( inline image)
ISOP + NO3 [RIGHTWARDS ARROW] ISON + NO33.03 × 10−12 × exp( inline image)
ISOR + HO2 [RIGHTWARDS ARROW] 0.282 BiPER + 0.030 BiDER + HO22.05 × 10−13 × exp( inline image)
ISOR + C2O3 [RIGHTWARDS ARROW] 0.026 BiMT + 0.219 MACR + C2O38.40 × 10−14 × exp( inline image)
ISOR + MeO2 [RIGHTWARDS ARROW] 0.026 BiMT + 0.219 MACR + MeO23.40 × 10−14 × exp( inline image)
ISOR + NO [RIGHTWARDS ARROW] 0.418 MACR + 0.046 ISON + NO2.43 × 10−12 × exp( inline image)
ISOR + NO3 [RIGHTWARDS ARROW] 0.438 MACR + NO31.20 × 10−12
ISON + OH [RIGHTWARDS ARROW] OH1.30 × 10−11
ISON + NO3 [RIGHTWARDS ARROW] 0.074 BiNIT3 + NO36.61 × 10−13
MACR + NO [RIGHTWARDS ARROW] NO2.54 × 10−12 × exp( inline image)
MACR + HO2 [RIGHTWARDS ARROW] HO21.82 × 10−13 × exp( inline image)
MACR + MeO2 [RIGHTWARDS ARROW] MeO23.40 × 10−14 × exp( inline image)
MACR + NO2 [RIGHTWARDS ARROW] MPAN + NO22.80 × 10−12 × exp( inline image)
MPAN [RIGHTWARDS ARROW] MACR1.60 × 1016 × exp( inline image)
MPAN + OH [RIGHTWARDS ARROW] 0.067 BiMGA + 0.047 BiNGA + OH3.20 × 10−11
MPAN + NO2 [RIGHTWARDS ARROW] 0.067 BiMGA + 0.047 BiNGA + NO33.20 × 10−11
BiPER + hν [RIGHTWARDS ARROW] Degradation products50 times faster than photolysis of H2O2
API + OH [RIGHTWARDS ARROW] 0.30 BiA0D + 0.17 BiA1D + 0.10 BiA2D + OH1.21 × 10−11 × exp( inline image)
API + O3 [RIGHTWARDS ARROW] 0.18 BiA0D + 0.16 BiA1D + 0.05 BiA2D + O35.00 × 10−16 × exp( inline image)
API + NO3 [RIGHTWARDS ARROW] 0.70 BiA0D + 0.10 BiNIT + NO31.19 × 10−12 × exp( inline image)
BPI + OH [RIGHTWARDS ARROW] 0.07 BiA0D + 0.08 BiA1D + 0.06 BiA2D + OH2.38 × 10−11 × exp( inline image)
BPI + O3 [RIGHTWARDS ARROW] 0.09 BiA0D + 0.13 BiA1D + 0.04 BiA2D + O31.50 × 10−17
BPI + NO3 [RIGHTWARDS ARROW] 0.02 BiA0D + 0.63 BiNIT + NO32.51 × 10−12
LIM + OH [RIGHTWARDS ARROW] 0.35 BiA0D + 0.20 BiA1D + 0.0035 BiA2D + OH4.20 × 10−11 × exp( inline image)
LIM + O3 [RIGHTWARDS ARROW] 0.09 BiA0D + 0.10 BiA1D + O32.95 × 10−15 × exp( inline image)
LIM + NO3 [RIGHTWARDS ARROW] 0.69 BiA0D + 0.27 BiNIT + NO31.22 × 10−11
HUM + OH [RIGHTWARDS ARROW] 0.74 BiBmP + 0.26 BiBlP + OH2.93 × 10−10
TOL + OH [RIGHTWARDS ARROW] . + 0.25 TOLP1.80 × 10−12 × exp( inline image)
TOLP + HO2 [RIGHTWARDS ARROW] 0.78 AnClP + HO23.75 × 10−13 × exp( inline image)
TOLP + C2O3 [RIGHTWARDS ARROW] 0.78 AnClP + C2O37.40 × 10−13 × exp( inline image)
TOLP + MeO2 [RIGHTWARDS ARROW] 0.78 AnClP + MeO23.56 × 10−14 × exp( inline image)
TOLP + NO [RIGHTWARDS ARROW] 0.097 AnBlP + 0.748 AnBmP + NO2.70 × 10−12 × exp( inline image)
TOLP + NO3 [RIGHTWARDS ARROW] 0.097 AnBlP + 0.748 AnBmP + NO31.2 × 10−12
XYL + OH [RIGHTWARDS ARROW] . + 0.274 XYLP1.70 × 10−11 × exp( inline image)
XYLP + HO2 [RIGHTWARDS ARROW] 0.71 AnClP + HO23.75 × 10−13 × exp( inline image)
XYLP + C2O3 [RIGHTWARDS ARROW] 0.71 AnClP + C2O37.40 × 10−13 × exp( inline image)
XYLP + MeO2 [RIGHTWARDS ARROW] 0.71 AnClP + MeO23.56 × 10−14 × exp( inline image)
XYLP + NO [RIGHTWARDS ARROW] 0.063 AnBlP + 0.424 AnBmP + NO2.70 × 10−12 × exp( inline image)
XYLP + NO3 [RIGHTWARDS ARROW] 0.063 AnBlP + 0.424 AnBmP + NO31.2 × 10−12

[10] The following nomenclature is used to name the model species. The species names begin either by “Bi” for biogenic compounds or by “An” for anthropogenic compounds. The suffix of the name corresponds to the species type. “A2D” and “A1D” mean that the species are hydrophilic and are respectively a diacid and a monoacid. “A0D” means that the species is hydrophilic and non-dissociative. “NIT” and “NIT3” mean that the compound is an hydrophobic nitrate or trinitrate. “BlP”, “BmP” and “ClP” mean that the compounds are hydrophobic and are more or less volatile (lP and mP for low and medium saturation vapor pressure, respectively). AnBlP, AnBmP and AnClP are formed by oxidation of aromatic species (B and C refer to semi-volatile and non-volatile organic compounds, respectively). BiA0D, BiA1D, BiA2D and BiNIT are formed by oxidation of monoterpenes, BiBlP and BiBmP by oxidation of sesquiterpenes. Finally, BiMGA, BiNGA are acids formed by oxidation of isoprene under high-NOX conditions; BiMT, BiPER and BiDER are formed under low-NOX conditions and BiNIT3 by oxidation of isoprene by the nitrate radical NO3 (see Couvidat and Seigneur [2011] for details).

[11] For aromatic compounds, toluene (TOL) and xylene (XYL) are used as precursors. The precursors react with OH to form oxidation products (TOLP and XYLP, respectively). These radicals may then react preferentially with HO2, MO2 (methylperoxy radical) or C2O3 (peroxyacetyl radical) under low-NOX conditions or with NO or NO3 under high-NOX conditions. Note that all reaction pathways are simulated but some prevail under low-NOX conditions whereas others prevail under high-NOX conditions; they may contribute similarly in the NOX transition regime. AnBlP and AnBmP are formed by oxidation of aromatic compounds under high-NOX conditions and AnClP (which is assumed to be non-volatile based on data from Ng et al. [2007a]) is formed under low-NOX conditions. These compounds are hydrophobic. Their stoichiometric coefficients and saturation vapor pressures are estimated from the experiments of Ng et al. [2007a]. Svendby et al. [2008] found an enthalpy of vaporization between 40 kJ.mol−1 and 60 kJ.mol−1. A mean value of 50 kJ.mol−1 is used here for the enthalpy of vaporization of AnBlP and AnBmP.

[12] For SOA formation from isoprene oxidation, a previously developed model [Couvidat and Seigneur, 2011] is included. It takes into account SOA formation under different conditions (oxidation by OH under low-NOX and high-NOX conditions and oxidation by NO3) and hydrophilic properties of the various compounds. The compound named BiNGA formed under high-NOX conditions can undergo oligomerization with an effective partitioning constant:

  • display math

where Kp,eff is the effective partitioning constant, Kp is the monomer partitioning constant as defined by Pankow [1994], and Koligo represents the ratio of the oligomer mass to the monomer mass (equal to 64.2).

[13] For monoterpenes, the oxidation scheme is based on Pun et al. [2006]. Three precursors are used: API (for α-pinene and sabinene), BPI (for β-pinene and Δ3-carene) and LIM (for limonene and other monoterpenes and terpenoïds). For oxidation by OH and O3, the reaction scheme is taken from Pun et al. [2006] with the same species and the same properties. Only the enthalpies of vaporization of BiA0D and BiA1D are decreased to 50 kJ.mol−1 [Svendby et al., 2008]. For oxidation by NO3, the reaction scheme is based on several studies where organonitrates and aldehydes were formed [Hallquist et al., 1999; Spittler et al., 2006]. BiA0D and BiNIT are used to represent, respectively, the formation of aldehydes and organonitrates. Yields of Hallquist et al. [1999] for aldehydes formed from the oxidation of different monoterpenes are used as yields for BiA0D. Yields of aerosols from Hallquist et al. [1999] and Spittler et al. [2006] are used to determine the stoichiometric coefficients and the saturation vapor pressure of BiNIT. BiA0D undergoes oligomerization via an effective Henry's law constant [Pun and Seigneur, 2007]:

  • display math

where Heff is the effective Henry's law constant of BiA0D, H is the monomer Henry's law constant of BiA0D, and a(H+) is the activity of protons in the aqueous phase. As the parameterization used an effective Henry's law constant, the formation of oligomers could strongly vary with the pH and the amount of water. However, fine particles are generally very acidic [Ludwig and Klemm, 1990; Keene et al., 2004] and oligomer formation from BiA0D will then in fact appear as an irreversible process, even though the parameterization is formulated as a reversible process.

[14] Formation of SOA from sesquiterpene oxidation is determined as in Pun et al. [2006]. HUM represents all sesquiterpenes and two hydrophobic compounds are formed from its oxidation: BiBmP and BiBlP.

2.3. Primary and Aged Organic Aerosol Module

[15] Primary organic compounds are treated as SVOC. Their concentrations in the particle phase depend on the amount of organic matter into which SVOC will condense and temperature (which influences the volatility of compounds). To treat POA as SVOC, the curve of dilution of POA from diesel exhaust of Robinson et al. [2007] was fitted to represent the dilution with three molecules: POAlP for compounds of a low volatility, POAmP for compounds of medium volatility and POAhP for compounds of high volatility. Based on this fitting, POAlP, POAmP and POAhP represent respectively 25%, 32% and 43% of “non-diluted” POA emissions from diesel vehicles with partitioning constants of 1.1 m3.μg−1, 0.0116 m3.μg−1 and 0.00031 m3.μg−1, respectively. Properties of primary SVOC are listed in Table 3. The least volatile compound (POAlP) does not condense entirely on the particle phase at low concentrations of OM. Areas with low concentrations of organic aerosols should then be sensitive to the value of the partitioning constant of POAlP. To study POA dilution, Robinson et al. [2007] used mostly data at concentrations greater than 20 μg.m−3. There is therefore large uncertainties on the partitioning of SVOC at concentrations typical of ambient conditions. As a result, POA evaporation could then be overestimated (in that case SOA, formation from SVOC oxidation may be overestimated) or underestimated (in that case, SOA formation from primary SVOC oxidation may be underestimated). More information on the dilution of primary SVOC especially at low aerosol loading relevant to atmospheric conditions is needed to know if primary SVOC can condense under atmospheric conditions (and form POA).

Table 3. Properties of Primary and Aged SVOC
SurrogateMWaKpbΔHvapcOM/OCd
POAlP2801.11061.3
POAmP2800.0116911.3
POAhP2800.00031791.3
SOAlP3921101061.82
SOAmP3921.16911.82
SOAhP3920.031791.82

[16] By default and due to the lack of data, we suppose that SVOC emitted from other sources have the same dilution behavior as diesel exhaust. Shrivastava et al. [2006] show a very similar dilution behavior between diesel exhaust and wood smoke. A mean organic matter (OM)/organic carbon (OC) ratio of 1.3 for primary SVOC (ratio typical of hydrocarbon aerosols) is chosen.

[17] SVOC in the gas phase can be oxidized, form less volatile compounds and a significant amount of SOA [Robinson et al., 2007; Grieshop et al., 2009]. For one oxidation step, we supposed, following Grieshop et al. [2009], that SVOC volatility is reduced by a factor of 100 and that the molar mass of the SVOC is increased by 40% due to oxygen addition. However, following Pye and Seinfeld [2010], SVOC were supposed to undergo only one oxidation step for the following reasons. First, chamber experiments are conducted on short timescales and provide information only on the first oxidation steps. Second, this reduction of volatility could be less valid for later generations of oxidation for which molecular fragmentation becomes more important. Thus, aging of primary aerosol is taken into account with the following three reactions in the gas phase, which were added to CB05:

  • display math
  • display math
  • display math

with k the kinetic rate constant equal to 2 × 10−11 molecule−1.cm3.s−1 [Grieshop et al., 2009]. SOAlP, SOAmP and SOAhP are the aged SVOC. Properties of aged SVOC are listed in Table 3.

[18] Unlike SOA surrogate compounds, no molecular structure is associated with primary and aged SVOC. Their partitioning between the gas phase and the organic particle phase is calculated with a partitioning constant Kp (values in Table 3) whereas for SOA surrogate compounds the partitioning constant depends on the mean molecular weight of the organic phase and the activity coefficient of the molecule. For the computation of activity coefficients of SOA surrogate compounds, a molecular structure is needed for POAlP, POAmP, POAhP, SOAlP, SOAmP and SOAhP. They are represented by a single POA species with an “average” structure representative of atmospheric PM. This average structure is constituted of 40% of C23H47COOH, 5% of C8H17CH = CHC7H14COOH, 15% of 4-(2-propio)-syringone, 12% of C29H60 and 28% of 2-carboxybenzoic acid based on EPRI [1999]. This is a simplifying assumption as each source of primary organic aerosol emits different molecules with different structures. Some attempts have been made to use molecular surrogates to represent various POA emissions [e.g., Pun, 2008; Lee-Taylor et al., 2011]. However, there is still considerable uncertainty due to a lack of complete molecular data for POA and, consequently, we selected this simple POA representation. Moreover, it is possible that two organic aerosols emitted from two different sources have little affinity with each other, do not mix and evolve separately. In that case, instead of treating aerosols as an internal mixture (all particles in a given size range having the same composition), it would be necessary to treat aerosols as an external mixture (each aerosol group having its own composition). This part of the model can be refined as more experimental data become available on the chemical composition of major POA sources. The molecular structure used for the mixture of POAlP, POAmP, POAhP, SOAlP, SOAmP and SOAhP is not used to compute activity coefficients of these individual compounds (their partitioning uses a simple partitioning constant assuming constant activity coefficients) but to estimate the impact of these anthropogenic SVOC on the partitioning of other SVOC.

[19] Robinson et al. [2007] also argue for the presence of Intermediate Volatility Organic Compounds (IVOC). IVOC were taken into account in several studies using the VBS framework and lead to improvement in the urban to regional ratio of OM [Robinson et al., 2007; Shrivastava et al., 2008]. However, IVOC were not taken into account here for several reasons. First, the amount of emitted IVOC is highly uncertain (1 to 3 times the amount of emitted SVOC, and there is already, a large uncertainty on emissions of SVOC). Second, there is also a large uncertainty on the oxidation mechanism of IVOC. To have a better representation of IVOC oxidation, Pye and Seinfeld [2010] used naphthalene as a surrogate for IVOC and used the yields from smog chamber experiments [Chan et al., 2009; Kautzman et al., 2010] to develop a mechanism of oxidation of IVOC. They found only minor concentrations of SOA from IVOC (only 5% of total OM and 14% of SOA from SVOC). However, Pye and Seinfeld [2010] discussed the fact that naphthalene may not be an appropriate candidate for lumping IVOC. We consider that more work should be done on IVOC emissions and the associated oxidation mechanism before their incorporation in the model.

3. Simulation of Organic Aerosols Over Europe

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Model Development
  5. 3. Simulation of Organic Aerosols Over Europe
  6. 4. Sensitivity Analysis and Investigations of Organic Aerosol Formation
  7. 5. Conclusion
  8. Acknowledgments
  9. References
  10. Supporting Information

3.1. Description of the Simulation Configurations

[20] OM was simulated over Europe from July 2002 to July 2003. During this period, OC was measured one day per week at 12 monitoring stations from the EMEP network. Locations and names of the stations are shown in Figure 1. OC concentrations simulated by the model are estimated using the OM/OC ratios of the surrogate compounds and are evaluated by comparison to measurements.

image

Figure 1. Locations and names of OC monitoring stations.

Download figure to PowerPoint

[21] Input data to Polair3D/Polyphemus were prescribed as follows. Meteorology was obtained from the European Centre for Medium-Range Weather Forecasts (ECMWF) model. Boundary conditions for gaseous species were obtained from the Model for OZone And Related chemical Tracers (Mozart v2.0) [Horowitz et al., 2003] and boundary conditions for particles from ECHAM5-HAMMOZ [Pozzoli et al., 2011]. Anthropogenic emissions of gases and particles were taken from the EMEP inventory [Vestreng, 2003], except for POA and EC, which were taken from Junker and Liousse [2008]. Biogenic emissions were estimated with the Model of Emissions of Gases and Aerosols from Nature (MEGAN) [Guenther et al., 2006].

[22] Because the inventory of Junker and Liousse [2008] only provides emissions of EI-POA (emission inventory POA measured solely as PM and traditionally assumed to be non-volatile) and does not take into account that a part of SVOC is present in the gas phase and is missing from the inventory. Here, the model does not use EI-POA emissions and all primary compounds are supposed to be SVOC. SVOC emissions are estimated by applying a SVOC/EI-POA ratio to transform EI-POA emissions into emissions of SVOC. Junker and Liousse [2008] use an emission factor of 0.5 g/kg-fuel for diesel POA whereas Robinson et al. [2007] give an emission factor of non-diluted POA of 2.5 g/kg-fuel indicating that at least 80% of diesel SVOC are in the gas phase. Therefore, for the reference simulation, a SVOC/EI-POA ratio of 5 for all sources of SVOC is chosen. It is, however, possible that some of the difference may be due to different types of engines and experimental conditions. Junker and Liousse [2008] used a OC/EC ratio of 0.5 and determined the OC emission factor from the EC emission factor. This ratio is derived from U.S. emission sampling experiments [Cooke et al., 1999] where the dilution of the vehicle exhaust is at least by a factor of 100 [Schauer et al., 1999]. Chirico et al. [2011] studied the evolution of the OM/BC ratio (which is approximately equal to 1.3 × OC/EC) at low organic loading in a tunnel. They found that a OC/EC ratio of 0.5 can only be obtained at low organic loading (from 10 to 20 μg.m−3). The emission factor of OC used in Junker and Liousse should then correspond to low organic loading and probable missing emissions should be accounted for. Although different sources could have different SVOC/EI-POA ratios with different dilution behaviors, available experimental data [Shrivastava et al., 2006] suggest that a large part of SVOC can be in the gas phase even at low dilution ratios. At this point, an average ratio of 5 seems to be a reasonable estimate. The uncertainty associated with this parameter is tested below.

[23] For biomass burning emissions, the inventory of Liousse et al. [2010] is used. This inventory does not provide information on injection heights, which is a potentially significant source of uncertainty for air quality modeling. A parameterization was developed to estimate injection heights depending on PM emissions. First, the burned area is estimated by using the emission factors of Wiedinmyer et al. [2006] for different land covers. Once the burned area has been estimated, emissions are injected between two heights (Htop and Hbottom), which depend on the burned area, based on the method of Western Regional Air Partnership [2008]. Injections heights are listed in Table 4.

Table 4. Injection Heights as a Function of the Fire Size Class
 Class
12345
Size (acres)0–1010–100100–10001000–5000>5000
Hbottom (m)0324123821682430
Htop (m)25.6864360052026480

[24] Results from four different simulations (noted SimRef, SimFire, Sim7, and SimEMEP) are presented and evaluated. The first simulation (SimRef) is the reference simulation; it does not include fire emissions, uses the inventory of Junker and Liousse [2008] for anthropogenic carbonaceous emissions and a SVOC/EI-POA ratio of 5. The other simulations differ from the reference simulation via one of these components: fire emissions for SimFire, a SVOC/EI-POA ratio of 7 for Sim7 (which is the ratio for a very low concentration of OM of 1 μg.m−3 in the dilution sampler and, therefore can be considered as the highest possible ratio) and POA and EC anthropogenic emissions from EMEP for SimEMEP (see Table 5 for details of simulation configurations).

Table 5. Configurations of the Different Simulations
NameSimRefSimFireSim7SimEMEP
SVOC/EI-POA ratio5575
Fire emissionsNoYesNoNo
OC & EC inventoryJunker and Liousse [2008]Junker and Liousse [2008]Junker and Liousse [2008]EMEP

3.2. Simulation Results

[25] Performance statistics (RMSE: Root Mean Square Error, MNE: Mean Normalized Error, MNB: Mean Normalized Bias, NME: Normalized Mean Error, NMB: Normalized Mean Bias, MFE: Mean Fractional Error and MFE: Mean Fractional Bias) for the four model simulations are shown in Tables 6 and 7. Comparisons between observed and simulated OC at the twelve EMEP monitoring stations over the entire year are shown in Figure 2 for two of these simulations, SimRef and SimFire.

Table 6. OC Statistics for the Different Simulations at Individual EMEP Stationsa
StationsSimRefSimFire
Mean Measured OCMean Modeled OCRMSECorrelationMean Measured OCMean Modeled OCRMSECorrelation
  • a

    Concentrations and the root mean square error (RMSE) are in μg.m−3.

AT025.573.753.180.575.574.093.130.53
BE054.121.682.860.764.121.782.740.79
CZ034.543.352.350.694.543.802.490.65
DE024.302.242.970.844.302.513.020.70
FI172.081.141.680.682.081.301.700.57
GB461.530.661.130.761.530.711.060.78
IE311.200.461.150.891.200.571.030.76
IT047.792.838.320.137.793.078.170.15
IT085.922.484.580.245.922.714.450.21
NL092.591.321.800.752.591.411.710.76
PT014.100.775.72−0.024.100.885.67−0.02
SE122.121.121.490.832.121.331.760.49
StationsSimEMEPSim7
Mean Measured OCMean Modeled OCRMSECorrelationMean Measured OCMean Modeled OCRMSECorrelation
AT025.574.853.500.575.574.102.480.67
BE054.124.542.300.764.123.421.690.72
CZ034.544.542.740.684.543.282.160.72
DE024.302.892.330.854.302.603.210.62
FI172.081.341.610.632.081.731.860.39
GB461.530.820.960.791.531.050.770.81
IE311.200.560.990.901.200.600.920.92
IT047.793.517.700.317.794.916.770.43
IT085.923.054.110.335.924.073.350.42
NL092.591.721.500.772.591.961.640.61
PT014.100.845.660.034.101.385.200.32
SE122.121.311.360.782.121.471.410.68
Table 7. OC Statistics for the Different Simulations Averaged Over All EMEP Stations
ConfigurationCriteriaOverallWinterSummer
With PT01Without PT01With PT01Without PT01With PT01Without PT01
  • a

    Unit is μg.m−3.

SimRefMean measured OCa3.783.754.334.233.193.23
 Mean modeled OCa1.781.871.841.971.711.77
 Correlation0.520.580.480.550.670.67
 RMSEa3.683.434.604.212.332.33
 MNE53%52%58%55%49%49%
 MNB−48%−46%−50%−47%−45%−45%
 NME57%54%62%58%49%48%
 NMB−53%−50%−58%−53%−46%−45%
 MFE77%73%84%78%69%68%
 MFB−73%−69%−80%−73%−66%−68%
 
SimFireMean measured OCa3.783.754.334.233.193.23
 Mean modeled OCa1.972.072.292.151.851.79
 Correlation0.500.550.430.50.670.67
 RMSEa3.643.384.564.162.282.28
 MNE52%51%57%55%47%46%
 MNB−41%−39%−40%−36%−42%−42%
 NME55%52%60%56%48%47%
 NMB−48%−45%−51%−46%−44%−43%
 MFE71%68%78%72%65%64%
 MFB−65%−61%−69%−62%−61%−60%
 
Sim7Mean measured OCa3.783.754.334.233.193.23
 Mean modeled OCa2.262.392.552.741.942.02
 Correlation0.530.590.480.550.680.69
 RMSEa3.463.184.343.902.162.14
 MNE48%46%51%48%44%43%
 MNB−36%−33%−33%−28%−38%−37%
 NME51%48%55%50%45%44%
 NMB−40%−36%−41%−35%−39%−37%
 MFE53%59%67%60%60%58%
 MFB−55%−50%−56%−47%−55%−53%
 
SimEMEPMean measured OCa3.783.754.334.233.193.23
 Mean modeled OCa2.522.632.732.862.302.37
 Correlation0.630.690.600.680.720.72
 RMSEa3.102.823.903.471.901.90
 MNE42%41%46%44%37%37%
 MNB−22%−19%−18%−14%−26%−25%
 NME44%41%48%44%38%37%
 NMB−33%−30%−37%−32%−28%−27%
 MFE50%47%53%49%46%45%
 MFB−37%−33%−36%−30%−38%−36%
image

Figure 2. Comparisons between observed and simulated OC at EMEP monitoring stations for SimRef and SimFire (as defined in Table 5). Vertical lines separate the winter and summer periods used to calculate statistics in Table 7.

Download figure to PowerPoint

[26] In the four simulations, OC is underestimated both in summer (from May to October) and in winter (from November to April). Correlations are higher in summer (0.67 for SimRef) than in winter (0.48 for SimRef). In winter, correlations are higher when the station PT01 is not taken into account (0.55 against 0.48 for SimRef). This station has a negative annual correlation for SimRef due to the low concentrations given by the model in winter whereas very high concentrations (between 5 and 35 μg.m−3) are observed at the station (as shown in Figure 2). This observed high OC concentration at PT01 is unexplained by the model and by other models [Simpson et al., 2007; Bessagnet et al., 2008]. It may be due to missing emissions in the inventory or the presence of local pollution sources in winter near the station.

[27] Adding forest fires does not improve overall model results significantly as there is no significant change in the global RMSE (3.68 μg.m−3 for SimRef against 3.64 μg.m−3 for SimFire). The RMSE even increases and the correlation decreases at several stations (CZ03, DE02, FI17, SE12). At station SE12, SimFire simulates a large unobserved peak (as shown in Figure 2). This peak may be due to a low fire injection height given by the algorithm for a forest fire of medium intensity (Class 2 or 3), which may actually have been emitted above the planetary boundary layer. On the contrary, some big fires may not contribute significantly to the ground-level OC concentrations due to an injection height above the planetary boundary layer. Even if the results of the model does not improve significantly on average, fires seem to contribute locally to OC in April and May as shown in Figure 2. Further work on implementation of forest fires emissions with emphasis on the injection heights is needed to minimize this source of uncertainty in air quality modeling.

[28] Using the emissions of Junker and Liousse [2008] with a SVOC/EI-POA ratio of 7 or EMEP inventory emissions leads to improved results with higher annual correlations (0.53 for Sim7 and 0.63 for SimEMEP) and lower RMSE (3.46 μg.m−3 for Sim7 and 3.10 μg.m−3 for SimEMEP) in particular in winter. Improvements in model performance are due to higher carbonaceous emissions with the EMEP inventory or the use of a higher SVOC/EI-POA ratio in Sim7. Moreover, in winter, OC is essentially constituted of anthropogenic primary SVOC (POAlP, POAmP and POAhP) and their oxidation products (SOAlP, SOAmP and SOAhP). These results indicate that OC concentrations in winter are very sensitive to anthropogenic primary SVOC emissions (concentrations of SOA from the oxidation of biogenic precursors and aromatics are low) and that the underestimation of OC in the model may be due to missing anthropogenic primary SVOC emissions and, therefore, missing SOA from SVOC oxidation. Results in summer are also better for both simulations (SimEMEP and Sim7). This is supported by Figure 3 showing the relative error in OC as a function of the relative error in EC (used here as a tracer of anthropogenic primary emissions) during winter. This figure shows that these two errors are correlated (r = 0.45) and that we might expect OC concentrations to improve if anthropogenic primary carbonaceous emissions are improved.

image

Figure 3. Scatter diagram of errors (%) in OC and EC for all EMEP monitoring stations in winter (correlation = 0.45); mod and obs stand for modeled and observed.

Download figure to PowerPoint

[29] OM concentrations are shown over Europe in Figure 4 for August 2002 and February 2003. In August 2002, concentrations are higher with SimEMEP than with SimRef. SimEMEP shows greater concentrations of OM over urban areas and over the Mediterranean and the North Sea due to higher emissions from the marine traffic. For February 2003, SimRef shows very high and localized concentrations over central Europe whereas concentrations are significantly lower over the rest of Europe. SimEMEP shows high concentrations of OM over all Europe. The Mean Fractional Error (MFE) and the Mean Fractional Bias (MFB) averaged over EMEP stations (excluding PT01) are, respectively, 47% and −33% for SimEMEP and 73% and −69% for SimRef.

image

Figure 4. Concentrations of OM over Europe in (a and b) August 2002 and (c and d) February 2003 for SimRef (Figures 4a and 4c) and SimEMEP (Figures 4b and 4d).

Download figure to PowerPoint

[30] With SimEMEP, the model performance criteria proposed by Boylan and Russell [2006] (MFE less than 75% and MFB within ±60%) are met for OM concentrations and almost meet the model goal criteria (MFE less than 50% and MFB within ±30%). Furthermore, the MNE obtained with SimEMEP (42%) is lower than that obtained by Bessagnet et al. [2008] (between 46% and 79% at EMEP stations). Correlations are similar between SimEMEP and Bessagnet et al. [2008] at some stations (BE05, DE02, GB46, IE31 and NL09) but are much higher at the other stations with SimEMEP (e.g., 0.67 against 0.11 at AT02). These correlations are also higher than those obtained by Simpson et al. [2007]. The differences between SimEMEP and those earlier studies lie mostly in the treatment of POA as SVOC. Bessagnet et al. [2008] found indeed low concentrations of OC in winter, which are consistent with the concentrations obtained here when non-volatile emissions of EI-POA (i.e, without SVOC emissions) are used, as shown below.

[31] As H2O uses a molecular approach, it is straightforward to estimate the contribution of the different precursors to OM contributions and determine the nature of the aerosol (biogenic or anthropogenic). Figure 5 shows the contributions of precursors and SOA surrogates to OM concentrations over Europe calculated with SimEMEP. In summer, the main contributors are monoterpenes, aged anthropogenic SVOC and, to a smaller extent, isoprene. However, Bessagnet et al. [2008] simulated a very large fraction of SOA originating from isoprene oxidation in summer. One reason is that the periods compared are different (summer 2002 in this study against summer 2003 in Bessagnet et al. [2008]); the period studied by Bessagnet et al. [2008] was hotter and, consequently, with higher biogenic emissions. For example, isoprene emissions are 40% higher in 2003 than in 2002, whereas terpene emissions only increased by 15% from 2002 to 2003 based on MEGAN. In winter, the main contributors are primary and aged anthropogenic SVOC. The higher contribution of POA in winter is due to the decrease of volatility of SVOC (because of colder conditions), as well as the lower emissions of biogenic precursors. Aromatics and sesquiterpenes are found to be only a minor source of SOA. The low concentration of SOA from aromatics oxidation over Europe has been reported in earlier modeling studies [Bessagnet et al., 2008; Henze et al., 2008]. However, compounds formed by oxidation of aromatics remain mostly in the gas phase and it is possible that further oxidation could occur leading to additional SOA formation, which is not taken into account in the model. Overall, sesquiterpenes are a minor precursor of SOA, nevertheless, SOA from sesquiterpene oxidation represent locally up to 0.5 μg.m−3 in August 2002, which is not insignificant. The contribution of sesquiterpenes to OM concentrations over Europe is low because of the low emission rates obtained with the emission model [Guenther et al., 2006] (around 6% of monoterpene emissions). This low emission rate of sesquiterpenes over Europe is consistent with the results of Karl et al. [2009].

image

Figure 5. Contributions of precursors (VOC and SVOC) and surrogate species to OM concentrations averaged over Europe (land and sea) calculated with SimEMEP.

Download figure to PowerPoint

[32] Figure 6 shows the fraction of biogenic organic aerosols calculated with SimEMEP for August 2002 and February 2003. In summer, most of Europe is covered by an organic aerosol, which is mostly biogenic (from 50% to 90%). Only the area around the English Channel is mostly anthropogenic due to the proximity of megacities (London and Paris) and high anthropogenic marine emissions. Organic aerosols from the north of Europe are highly biogenic due to high monoterpene emissions. The large fraction of anthropogenic organic aerosol over the Atlantic ocean is due in part to missing marine emissions of organic aerosols and to the fact that organic aerosols coming from boundary conditions were classified by default as anthropogenic. In winter, the organic aerosol is almost entirely anthropogenic due to low biogenic emissions except in Portugal and in southern Europe where a small fraction of the aerosol is biogenic (around 20%).

image

Figure 6. Fraction of biogenic organic aerosol over Europe calculated with SimEMEP for (left) August 2002 and (right) February 2003.

Download figure to PowerPoint

[33] Aerosols formed from isoprene and monoterpene oxidation are mainly hydrophilic [Pun et al., 2006; Couvidat and Seigneur, 2011] whereas other precursors lead in the model to hydrophobic compounds. Hydrophilic compounds should then be produced during summer when biogenic emissions are high. In August 2002, 95% and 92% of SOA formed from isoprene and monoterpene oxidation are found to be hydrophilic. In this model, 45% of organic aerosols are then constituted of hydrophilic compounds. However, this fraction could be underestimated because this model does not account for aging of aerosols with several oxidation steps. Aging will produce more oxygenated species that could become hydrophilic at some point and SOA found initially to be hydrophobic (SOA formed from oxidation of primary SVOC, aromatics and sesquiterpenes) could then become in part hydrophilic.

[34] As OM is necessary to estimate PM concentrations (and not OC, which is measured), the OM/OC ratio should be evaluated to understand how well the model performs. In that respect, Figure 7 shows the OM/OC ratio calculated with SimRef for August 2002 and February 2003. In summer, the calculated ratio is between 1.55 and 1.65 over most of Europe. The ratio decreases in winter to 1.35–1.55 due to the higher contribution of POA (which are less volatile in winter due to the decrease in temperature) and lower contributions of biogenic SOA. This ratio does not exceed 1.8; this value seems low compared to the ratio of 2.1 estimated by Turpin and Lim [2001] for rural areas. This low ratio is probably due to the low ratios of major SOA compounds, which are between 1.4 and 1.8 for monoterpene SOA compounds and 1.8 for aged anthropogenic SVOC. This ratio can be increased by adding the formation of highly oxidized species in the model with high OM/OC ratios. For this purpose, several processes can contribute significantly: aqueous-phase oxidation, additional oxidation steps or formation of organosulfates. Clearly, such processes should be studied in greater detail and possibly incorporated in future modeling studies. Figure 7 shows the OM/OC ratio calculated with SimEMEP for August 2002 and February 2003. In summer, the OM/OC ratio is slightly lower with SimRef than with SimEMEP. This is due to higher emissions of primary SVOC in SimEMEP that become oxidized with a high OM/OC ratio. On the contrary, the OM/OC ratio in winter is higher with SimRef than with SimEMEP. In winter, the primary SVOC are oxidized more slowly and the contribution of primary compounds with a low OM/OC ratio is higher. In SimEMEP, as the concentration of OM is higher, primary SVOC condense more easily than with SimRef. It results in a higher OM/OC ratio with SimRef.

image

Figure 7. OM/OC ratio over Europe calculated with (left) SimRef and (right) SimEMEP for (top) August 2002 and (bottom) February 2003.

Download figure to PowerPoint

4. Sensitivity Analysis and Investigations of Organic Aerosol Formation

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Model Development
  5. 3. Simulation of Organic Aerosols Over Europe
  6. 4. Sensitivity Analysis and Investigations of Organic Aerosol Formation
  7. 5. Conclusion
  8. Acknowledgments
  9. References
  10. Supporting Information

4.1. Impact of Non-ideality on Organic Aerosol Formation

[35] Organic aerosols are assumed to be non-ideal in H2O and activity coefficients are computed with the UNIFAC model. Ideality (i.e., the state in which activity coefficients are equal to one) is defined according to a reference state. For hydrophobic compounds, the reference state is the pure compound state, whereas for hydrophilic compounds the reference state is infinite dilution in water. Assuming that activity coefficients are equal to one will impact differently the partitioning of hydrophobic compounds and hydrophilic compounds. Organic aerosol concentrations with ideal state assumed were simulated for August 2002 with the SimRef configuration and results are compared with non-ideal organic aerosol concentrations. Figure 8 shows the impact of assuming ideality on hydrophilic compounds, hydrophobic compounds, and OM.

image

Figure 8. Impact of ideality (top) on hydrophilic SOA compounds, (middle) on hydrophobic SOA compounds and (bottom) on OM. (left) Concentrations (μg.m−3) when ideality is assumed and (right) differences of concentrations (μg.m−3) between non-ideal and ideal SOA solutions.

Download figure to PowerPoint

[36] For hydrophilic SOA compounds, assuming ideal conditions decreases the mean hydrophilic SOA concentrations over Europe in August 2002 from 1.32 μg.m−3 to 1.13 μg.m−3 on average. This decrease is observed because in an aqueous aerosol, most organic compounds tend to be stabilized by other organic compounds with which they have more affinity than with water. Assuming ideal conditions does not take this effect into account and decreases their concentrations. Over the continent, a decrease of hydrophilic SOA concentrations up to 1 μg.m−3 is calculated. However, over water, hydrophilic SOA concentrations can in fact slightly increase. This small increase is due to a small increase of particulate concentrations of BiA0D (which is one of the main contributors of hydrophilic SOA by oligomerization). BiA0D has more affinity with water than other organic compounds and condenses more efficiently when there are only low concentrations of these compounds.

[37] For hydrophobic compounds, assuming ideal conditions increases the mean hydrophobic SOA concentrations over Europe in August 2002 from 0.30 μg.m−3 to 0.38 μg.m−3 on average. Assuming ideality leads to greater concentrations over most of Europe except over a small area in the center of Europe. This decrease in the center of Europe is due to the decrease of BiNIT (organonitrate formed by oxidation of monoterpenes with NO3) concentrations over Europe. When activity coefficients are computed, BiNIT has an activity coefficient lower than one, because the assumed POA structure [EPRI, 1999] tends to stabilize it. Therefore, absorption of BiNIT is enhanced and partitions more efficiently. Overall, this decrease in SOA concentrations due to non-ideality is consistent with the results obtained by Pun [2008] using a different model.

[38] For total OM, assuming ideal conditions leads to a decrease of the mean OM concentration over Europe in August 2002 from 2.93 μg.m−3 to 2.76 μg.m−3 on average, because of a greater contribution of hydrophilic SOA than hydrophobic SOA to OM. However, assuming ideality can lead to an increase of the concentrations of some compounds and lead locally to an increase of OM concentrations. This study does not take into account the influence of activity coefficients on primary and aged anthropogenic SVOC. Pun [2008] showed that taking into account the activity of POA as well as SOA species may lead to larger effects (lower OM when non-ideality is taken into account). Experimental data on POA molecular composition are needed to better address this potentially important issue.

4.2. Comparison of Two Parameterizations of SOA Formation From Isoprene Oxidation

[39] The parameterization of Henze and Seinfeld [2006] (HS hereafter), which is based on a two-product Odum parameterization and yields from Kroll et al. [2006], is used currently in most air quality models, including the modeling study of Y. Zhang et al. [2007]. It is compared here for August 2002 with the parameterization of Couvidat and Seigneur [2011] (CS hereafter) used in H2O. The main difference between the two parameterizations is that CS take into account the hydrophilic properties of SOA formed from isoprene oxidation whereas HS implicitly assume that the SOA compounds formed are hydrophobic. Moreover, CS take into account the impact of NOX concentrations on SOA formation.

[40] The CS parameterization leads to higher concentrations of SOA formation from isoprene oxidation and gives a mean isoprene SOA concentration over Europe in August 2002 of 0.22 μg.m−3 against 0.07 μg.m−3 for the HS parameterization. Isoprene SOA concentrations calculated with CS are constituted essentially by compounds formed under low-NOX conditions: BiDER with 0.11 μg.m−3 and BiPER with 0.09 μg.m−3. BiPER is a compound formed under a transitional state and is oxidized to form more volatile compounds whereas BiDER represents SOA at the final state. This transitional mass is not taken into account in HS but it represents a significant mass of isoprene SOA in CS. Concentrations of BiDER are higher than the concentrations of SOA obtained with the two-product Odum parameterization. It shows that the hydrophilic properties of SOA can influence greatly SOA concentrations. Figure 9 shows the impact of using the CS parameterization on isoprene SOA concentrations. Isoprene SOA concentrations are greater with CS except over remote areas, especially water far from sources of isoprene where concentrations of isoprene SOA are low. However, marine emissions of isoprene are not taken into account in the model and CS could give greater concentrations of isoprene SOA over marine areas if marine emissions of isoprene were included [Shaw et al., 2010]. Differences in isoprene SOA are very important near source areas. Differences are even larger when OM concentrations are compared (Figure 10); they can be as high as 1.0 μg.m−3 because of the effect of isoprene SOA on the partitioning of other SOA compounds (via activity coefficients and the increase of absorbing medium).

image

Figure 9. (left) Isoprene SOA concentrations (μg.m−3) calculated for August 2002 with the parameterization of Couvidat and Seigneur [2011] and (right) difference of isoprene SOA concentrations (μg.m−3) between the parameterizations of Couvidat and Seigneur [2011] and Henze and Seinfeld [2006].

Download figure to PowerPoint

image

Figure 10. Difference in OM concentrations (μg.m−3) for August 2002 between the parameterizations of Couvidat and Seigneur [2011] and Henze and Seinfeld [2006] for isoprene SOA.

Download figure to PowerPoint

[41] This comparison confirms the conclusions of Couvidat and Seigneur [2011] that SOA formation from isoprene oxidation could be underestimated by most current models. Moreover, isoprene could be a major contributor to SOA in areas with very high emissions of isoprene such as the Amazon [Robinson et al., 2011].

4.3. Impact of the Volatility of Primary Organic Aerosols

[42] As POA are in fact semi-volatile compounds, the contribution of POA should be more important in winter due to lower temperatures, which lead to a decrease of the volatility of SVOC. Therefore, by assuming that POA are non-volatile and using EI-POA emissions, OM can be expected to be underestimated in winter. To verify this hypothesis, OM was simulated during February 2003 by using the non-corrected non-volatile emissions of EMEP (i.e., ignoring the potential gas-phase fraction associated with POA emissions).

[43] By using the non-volatile POA emissions of EMEP, H2O gives a mean concentration over Europe in February 2003 of 0.62 μg.m−3 of POA and 0.91 μg.m−3 of OM. In that case, POA constitutes 68% of the organic matter. By using the SimEMEP configuration, H2O gives much higher concentrations over Europe for the same period: 1.45 μg.m−3 of POA and 2.97 μg.m−3 of OM. POA constitutes then only 49% of OM. Using SVOC emissions gives higher concentrations of POA in winter but also a higher fraction of SOA due to the oxidation in the gas phase of primary SVOC. Figure 11 shows the impact of using EI-POA emissions on OM concentrations instead of SVOC emissions. Using non-volatile emissions leads to very low concentrations of OM indicating that for Europe using non-volatile POA emissions of EMEP leads to a large underestimation of OM in winter.

image

Figure 11. Impact of the non-volatility assumption on OM for February 2003. (left) OM concentrations in μg.m−3 calculated with non-volatile emissions and (right) difference of OM concentrations in μg.m−3 with and without SVOC emissions. Compare Figure 11 (left) with Figure 4d.

Download figure to PowerPoint

4.4. Investigation of the OM/OC Ratio

[44] We investigate here the possible reasons for the low OM/OC ratio obtained in the model simulations, focusing on summer. In summer, OM is mainly composed of SOA compounds formed by monoterpene oxidation. As these compounds have low OM/OC ratios (1.40 for BiA0D, 1.57 for BiA1D and 1.72 for BiA2D), the OM/OC ratio obtained by the model is low. However, higher OM/OC ratios could be obtained with second-generation products of monoterpene oxidation. The potential formation of 3-methyl-1,2,3-butanetricarboxylic acid (MBTCA) and organosulfates from pinonaldehyde uptake and their effects on the OM/OC ratio are investigated.

[45] MBTCA has been shown to be formed by oxidation of pinonic acid [Szmigielski et al., 2007; Zhang et al., 2010]. This product has a low volatility and a high OM/OC ratio of 2.125. Accordingly, the formation of a new surrogate compound BiA3D is added to the model using the molecular structure of the MBTCA. The following reaction is added to the model:

  • display math

with k = 9.0 × 10−12 molecule−1.cm3.s−1 [Kamens and Jaoui, 2001]. BiA3D is supposed to be hydrophilic and non-volatile (i.e., very high Henry's law constant). This reaction is supposed to only produce less volatile compounds, whereas it could produce shorter and more volatile molecules by fragmentation (i.e., the stoichiometric coefficient of BiA3D could be lower than one). Furthermore, pinonaldehyde can be oxidized in the gas phase to form pinonic acid. Accordingly, the following reaction is added to the model:

  • display math

with k = 9.0 × 10−11 molecule−1.cm3.s−1 [Glasius et al., 1997]. Adding these two reactions increases OM concentrations over Europe in August 2002 from 2.93 μg.m−3 to 3.22 μg.m−3 and OC from 1.84 μg.m−3 to 1.97 μg.m−3. However, formation of BiA3D impacts the OM/OC ratio only slightly because monoterpene SOA are still dominated by BiA0D. The ratio remains between 1.6 and 1.7 over most of Europe as shown in Figure 12a.

image

Figure 12. OM/OC ratio over Europe for August 2002 calculated (a) by assuming BiA3D formation, (b) by assuming organosulfate formation, (c) with increased OM/OC ratio for aged anthropogenic to 2.1 and (d) with all the previous hypothesis.

Download figure to PowerPoint

[46] Lukács et al. [2009] made a first attempt to quantify organosulfate concentrations and found at a monitoring station in Hungary in May–June 2006 between 0.02 and 0.09 μgS.m−3 of water soluble organosulfates, which correspond to concentrations between 0.16 and 0.75 μg.m−3 if the structure of the organosulfate formed from pinonaldehyde is used. According to Lukács et al. [2009], these values are lower bounds as a fraction of the organosulfates may not be water soluble. Organosulfates could then represent a large fraction of aerosols and could strongly impact the OM/OC ratio.

[47] BiA0D (using the structure of pinonaldehyde) was assumed to undergo oligomerization and the effective partitioning is computed according to the parameterization of Pun and Seigneur [2007]. However, this parameterization was derived from the effective partitioning in an acidic particle (using H2SO4, NH4HSO4, (NH4)2SO4) observed by Jang et al. [2005] and could in fact include organosulfate formation. Significant formation of organosulfates from uptake of pinonaldehyde on acidic sulfate aerosols has indeed been observed by Liggio and Li [2006]. To study the potential effect of organosulfate formation from pinonaldehyde, BiA0D was assumed to condense as an organosulfate having a molecular weight of 248 g.mol−1. Concentrations of OC for BiA0D were then calculated by using an OM/OC ratio of 2.1 instead of 1.4. The new OM/OC ratio is shown for August 2002 in Figure 12b. The ratio increases significantly and reaches a value between 1.74 and 1.92 over most of Europe against a ratio between 1.55 and 1.65 in SimEMEP. The effective OM/OC ratio of condensed BiA0D may in reality be somewhere between 1.4 and 2.1.

[48] To study the impact of the OM/OC ratio of anthropogenic aged SVOC, we increased the OM/OC ratio of aged anthropogenic SVOC (PSOAlP, PSOAmP, PSOAhP) from 1.8 to 2.1 as in Pye and Seinfeld [2010]. The new OM/OC ratio over Europe is shown in Figure 12c. Most of Europe has then an OM/OC ratio between 1.56 and 1.74 as all primary SVOC are not oxidized.

[49] Finally, the OM/OC ratio was computed by assuming both BiA3D and organosulfate formation, and an increase of the OM/OC ratio of aged anthropogenic SVOC (PSOAlP, PSOAmP, PSOAhP) from 1.8 to 2.1. The computed OM/OC ratio is shown in Figure 12d.The model calculates then an OM/OC ratio between 1.86 and 1.98 over most of Europe.

[50] One other aspect, which was not studied here, is that the reaction of monoterpenes with OH used the SOA yield for high-NOX conditions instead of that of low-NOX conditions, which would be more relevant for SOA simulation at the continental scale. However, only few experiments of oxidation of monoterpenes were performed under low-NOX conditions [Ng et al., 2007b] and data are insufficient to derive a two-product parameterization. However, results from Ng et al. [2007b] indicate a higher yield of SOA formation under low-NOX conditions, which could lead to significantly greater atmospheric SOA levels [Kim et al., 2009]. Valorso et al. [2011] developed a theoretical model and calculated under low-NOX conditions a high contribution of hydroperoxide compounds with a high OM/OC ratio (around 2.0). The incorporation of a representation of monoterpene SOA for low-NOX conditions could increase the overall calculated OM/OC ratios and change significantly the calculated OM/OC ratio.

5. Conclusion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Model Development
  5. 3. Simulation of Organic Aerosols Over Europe
  6. 4. Sensitivity Analysis and Investigations of Organic Aerosol Formation
  7. 5. Conclusion
  8. Acknowledgments
  9. References
  10. Supporting Information

[51] A model for SOA formation, H2O, that uses the molecular approach, has been developed and compared to OC measurements at 12 stations of the EMEP network over Europe for the period of July 2002 to July 2003. This comparison indicates an underestimation of the model for OC especially during winter, which seems to be due to missing sources in anthropogenic inventory emissions. Moreover, the model gives an OM/OC ratio between 1.55 and 1.65, which seems low for non-urban sites. However, by taking into account the gas-phase fraction of SVOC emissions, instead of assuming that POA emissions are non-volatile, the model is able to reproduce higher OC concentrations during winter with good annual correlations (e.g., 0.6 with anthropogenic emissions from EMEP).

[52] The impact of the ideality of organic PM, of the choice of the parameterization for isoprene SOA formation, and of using non-volatile POA emissions has been studied. Non-ideality has a complex effect on SOA formation because SOA may either increase (for hydrophilic SOA mostly) or decrease (for hydrophobic SOA mostly) depending on the compound. Overall, it was estimated that assuming ideality in H2O leads to a small decrease in OM. The parameterization of Couvidat and Seigneur [2011] for SOA formation from isoprene oxidation leads to a significant increase in isoprene SOA compared to that of Henze and Seinfeld [2006] and suggests that most models may currently underestimate SOA formation from isoprene. Finally, assuming non-volatile POA and using an inventory with missing SVOC emissions may lead to an underestimation of OM in winter because of missing SVOC that could condense due to the decrease of temperature and the associated decrease of volatility.

[53] To improve model results, emission inventories that take into account the gas-phase fraction of semi-volatile POA should be developed. More studies are needed to investigate the formation of SOA from different sources of primary SVOC. Moreover, the molecular composition of primary and aged anthropogenic SVOC should be investigated to include the molecular structures of POA in models and take into account their associated effect of non-ideality on OM. Efforts should be made to characterize and possibly improve the representation of the OM/OC ratio in the model, which currently simulates low ratios in summer probably due to the low ratio of first-generation oxidation products of monoterpenes. Second-generation products for monoterpenes and oxidation of monoterpenes by OH under low-NOX conditions should be investigated.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Model Development
  5. 3. Simulation of Organic Aerosols Over Europe
  6. 4. Sensitivity Analysis and Investigations of Organic Aerosol Formation
  7. 5. Conclusion
  8. Acknowledgments
  9. References
  10. Supporting Information

[54] This work was funded in part by ADEME, the French Agency for the Environment and Energy Management. Thanks are due to Nathalie Poisson, ADEME, Isabelle Coll, LISA, Université Paris-Est, and Anne Monod, Laboratoire Chimie Provence, Université Aix-Marseille, for useful discussions. We thank Luca Pozzoli and Johann Feichter, Max Planck Institute for Meteorology, Hamburg, for providing data from HAMMOZ and Catherine Liousse, Laboratoire d'aérologie, Toulouse, for providing the emission inventory of EC and OC.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Model Development
  5. 3. Simulation of Organic Aerosols Over Europe
  6. 4. Sensitivity Analysis and Investigations of Organic Aerosol Formation
  7. 5. Conclusion
  8. Acknowledgments
  9. References
  10. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Model Development
  5. 3. Simulation of Organic Aerosols Over Europe
  6. 4. Sensitivity Analysis and Investigations of Organic Aerosol Formation
  7. 5. Conclusion
  8. Acknowledgments
  9. References
  10. Supporting Information
FilenameFormatSizeDescription
jgrd17841-sup-0001-t01.txtplain text document1KTab-delimited Table 1.
jgrd17841-sup-0002-t02.txtplain text document3KTab-delimited Table 2.
jgrd17841-sup-0003-t03.txtplain text document0KTab-delimited Table 3.
jgrd17841-sup-0004-t04.txtplain text document0KTab-delimited Table 4.
jgrd17841-sup-0005-t05.txtplain text document0KTab-delimited Table 5.
jgrd17841-sup-0006-t06.txtplain text document1KTab-delimited Table 6.
jgrd17841-sup-0007-t07.txtplain text document2KTab-delimited Table 7.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.