Distribution of gaseous and particulate organic composition during dark α-pinene ozonolysis

Distribution of gaseous and particulate organic composition during dark α-pinene ozonolysis M. Camredon, J. F. Hamilton, M. S. Alam, K. P. Wyche, T. Carr, I. R. White, P. S. Monks, A. R. Rickard, and W. J. Bloss School of Geography, Earth & Environmental Science, University of Birmingham, Birmingham, B15 2TT, UK Department of Chemistry, University of York, York, YO10 5DD, UK Department of Chemistry, University of Leicester, Leicester, LE1 7RH, UK National Centre for Atmospheric Science, University of Leeds, Leeds, LS2 9JT, UK Received: 4 December 2009 – Accepted: 9 December 2009 – Published: 23 December 2009 Correspondence to: M. Camredon (m.camredon@bham.ac.uk) Published by Copernicus Publications on behalf of the European Geosciences Union.


Introduction
Secondary organic aerosol (SOA) has received significant interest because of its potential impact on climate, air quality and human health (e.g., Kanakidou et al., 2005). tematic underestimation of SOA production, increasing broadly with air mass ageing (e.g., Volkamer et al., 2006). The current parameterizations used in 3-D models require improvements to represent (i) SOA formation under a wide range of atmospheric chemical and physical conditions and (ii) SOA chemical speciation and its evolution during air mass ageing. 15 Simulation of SOA formation and composition requires theoretical models that describe in detail (i) the formation of semi-volatile organic compounds (SVOC) from gaseous oxidation (initiated by reactions with OH, O 3 , NO 3 or photolysis) of the VOC precursors, (ii) the partitioning of each individual SVOC between the gaseous and the condensed phases, and (iii) the potential reactivity of the condensed SVOC within the 1. the gaseous chemical pathways leading to the formation of SVOC are far from fully identified. Gaseous oxidation produces a multitude of individual SVOC, possibly formed after several oxidation steps of the precursor (e.g., Aumont et al., 2005). The majority of SVOC remain uncharacterised in the gaseous as well as in the condensed phases (e.g., Hallquist et al., 2009). Furthermore, large uncertain- 25 ties still exist over the degradation mechanisms of multi-functionalised organics (e.g., Kroll and Seinfeld, 2008). Therefore assumptions and simplifications are implemented in the chemical schemes which represent the formation of gaseous SVOC; 27840 facility consists of two outdoor atmospheric simulation chambers located on the roof of the Centro de Estudios Ambientales del Mediterraneo (CEAM) building. Technical details concerning the simulation chambers are given in Becker et al. (1996). Briefly, each hemispheric photo-reactor is ∼200 m 3 in volume and is constructed from fluorineethene-propene film (FEP). The reactors are surrounded by retractable steel covers 10 allowing both dark and photolysis experiments to be carried out. Fans located inside the chambers ensure rapid homogenized mixing. Experiments are carried out at ambient temperature and close to the atmospheric pressure. Losses from sampling into the analytical instruments, and any leakage, are compensated for by periodical introductions of purified air into the chamber (both chambers are marginally over pressurised 15 in order to prohibit ambient air ingress). The analytical instrumentation is located on a platform directly under each reactor.

Analytical instrumentation
The analytical instrumentation coupled to the EUPHORE chamber during the experiments are listed in Table 1. The facility is equipped with a range of instruments measur-20 ing the environmental conditions inside the chamber (temperature, pressure, humidity and photolysis rates). Standard monitors were used to follow the temporal evolution of inorganic species (O 3 , CO, NO x ). The OH and HO 2 radical levels were measured on-line using Laser Induced Fluorescence (LIF) (Bloss et al., 2004). Measurements of the temporal evolution of the gas phase degradation products 25 were made using traditional analytical techniques permanently available at EUPHORE, Interactive Discussion namely a formaldehyde monitor and various chromatographic instruments (Gas Chromatography-Mass Spectrometer, GC-MS, and a High Performance Liquid Chromatography, HPLC). Details concerning the permanent EUPHORE instrumentation are also given in Becker et al. (1999). The chamber instrumentation was supplemented by a Chemical-Ionization-Reaction Time-Of-Flight Mass-Spectrometer (CIR-TOF-MS) 5 (Blake et Wyche et al., 2007Wyche et al., , 2009, for the detection of a wide range of VOC at the ppb-sub-ppb (parts-par-billion by volume) level. The temporal evolution of aerosol physical characteristics (size, number and volume distribution) was provided by Scanning Mobility Particle Sizer (SMPS), combining a Differential Mobility Analyser (DMA) and a Condensation Particle Counter (CPC). Aerosol 10 samples were collected onto filters at the end of each experiment and the composition was analysed off-line. Filters were extracted into high purity water (Hamilton et al., 2008), and the mass distribution profile of the sample was obtained using direct infusion of the water soluble extract (via a syringe pump) into an ElectroSpray Ionisation source coupled to an ion trap Mass Spectrometer (ESI-MS). The chemical structures 15 of SOA components were obtained using reverse phase Liquid Chromatography, to separate individual components, and fragmentation patterns obtained using collision induced dissociation in the ion trap MS (LC-MS n ) (Hamilton et al., 2006(Hamilton et al., , 2008.

Experimental protocol
The experiments were performed at levels of reactants approaching (i.e. relevant 20 to) atmospheric conditions (see Table 2). In order to exclude photochemical effects, the experiments were performed in the dark with the chamber housing closed (j (NO 2 )<2×10 −5 s −1 ). CO was used as the OH scavenger. The amount of CO added inside the chamber was selected in order to have more than 95% of the α-pinene chemical removal due to O 3 . An aliquot of SF 6 was introduced to the chamber as an inert tracer to determine the dilution rate. For the experiment with an OH scavenger, the carbon monoxide (manufactured purity >99%, used as supplied) was injected directly. Ozone, produced from 5 a silent discharge in pure oxygen (Linde, purity of 99.999%) with a typical production rate of 20 ppb min −1 , was supplied to the simulation chamber in a flow of purified air.
Subsequently a known liquid volume of α-pinene (manufactured purity >99%, used as supplied) was evaporated by heating through a purified air stream flowing into the chamber. The aerosol sample for ESI-MS and LC-MS n chemical analysis was col-10 lected on 47 mm quartz-fibre filters for a period of ∼1 h at a flow rate of 74 L min −1 , commencing when the α-pinene level had fallen to ∼20% of its initial peak. Samples were immediately frozen and kept below −15 • C until analysis.

SOA modelling
3.1 Gaseous chemical scheme 15 The complete detailed gas-phase oxidation scheme for α-pinene was extracted from the Master Chemical Mechanism version 3.  Bloss et al., 2005b). The protocol applied to derive the α-pinene oxidation scheme is presented in detail in Jenkin et al. (2000) and Saunders et al. (2003).
During NO x -free dark ozonolysis experiments, the oxidation of α-pinene can be initiated by both O 3 and OH. The first oxidation stages of these degradation processes as represented in the MCMv3.1 (under NO x -free conditions) are briefly described below.

5
The first stages of O 3 -initiated α-pinene oxidation are shown in Fig. 1. The initial step consists of the concerted addition of ozone to the >C=C< double bond of α-pinene, leading to the formation of an energy-rich primary ozonide. This ozonide is expected to undergo rapid bond cleavage and isomerisation, giving two ring opened excited Criegee intermediates (CI) with a carbonyl substituent (denoted CI O 3 1 and CI O 3 2 in Fig. 1). The CI O 3 1 and CI O 3 2 are assumed to decompose to yield OH and peroxy radicals PR O 3 1, PR O 3 2 or PR O 3 3. These peroxy radicals react with HO 2 and other RO 2 leading to the formation of stable species either directly (S O 3 1-S O 3 8) or after further RO 2 /RO conversions and intermolecular isomerisations. These chemical pathways lead to the formation of various secondary organics including the first generation ox- 15 idation products previously observed such as formaldehyde, acetone, pinalic-3-acid, pinic acid and 10-hydroxy-pinonic acid (S O 3 9-S O 3 11, respectively). The CI O 3 2 may also be stabilized by collisions. The thermally stabilized Criegee (PR O 3 4) may undergo bimolecular reactions with H 2 O, CO and/or other oxygenated organics (or NO in the ambient atmosphere) to produce pinonaldehyde (S O 3 14) and pinonic acid (S O 3 15).

20
Despite considerable study, large uncertainties still exist over the possible fates of the Criegee intermediates that lead to the formation of observed secondary products during the O 3 -initiated α-pinene oxidation (e.g., Johnson and Marston, 2008).
The initial oxidation pathways involved in the OH-initiated α-pinene oxidation are shown in Fig. 2. The initial step proceeds through the addition of OH to the >C=C< 25 double bond of α-pinene, forming a secondary or tertiary hydroxyl-substituted alkyl radical. These alkyl radicals react rapidly with O 2 to form the corresponding peroxy radicals (denoted PR OH 1 and PR OH 2 in Fig. 2). A fraction of the tertiary alkyl radical is also assumed to isomerise by ring opening prior to O 2 addition, giving the peroxy Introduction

Conclusions
References Tables  Figures   Back  Close Full Screen / Esc

Printer-friendly Version
Interactive Discussion radical PR OH 3. These peroxy radicals recombine with HO 2 and other RO 2 to form stable species (S OH 1-S OH 6) or their corresponding alkoxy radicals (AR OH 1, AR OH 2 and AR OH 3). The alkoxy radicals AR OH 1 and AR OH 2 are assumed to exclusively produce pinonaldehyde (product S OH 7) after C−C bond scission and H abstraction by O 2 . The AR OH 3 radical is considered to decompose into acetone (S OH 8) and a peroxy radical PR OH 4 which then follow the usual RO 2 reactions to give the S OH 9-S OH 11 stable species. Theoretical studies have demonstrated the existence of a number of alternative pathways in the OH-initiated α-pinene oxidation involving intramolecular isomerisations (e.g., Peeters et al., 2001;Fantechi et al., 2002;Vereecken and Peeters, 2004;Vereecken et al., 2007). These processes were not considered in our model, and it is important to note that the existence of these chemical pathways could lead to a different distribution of organic species formed during the gaseous oxidation and have a substantial impact on SOA formation.

Gas/particle partitioning of organic compounds
The gas/particle partitioning of organic compounds was implemented as described in 15 Camredon et al. (2007). The absorption of each SVOC produced during α-pinene ozonolysis was represented assuming a thermodynamic equilibrium between the gas and particle phases, as described in Pankow (1994a,b). The thermodynamic equilibrium for each SVOC was calculated using Raoult's law:

20
where P i is the equilibrium partial pressure of a species i , x i its mole fraction in the condensed phase, P vap i its vapour pressure as a pure liquid at the temperature of interest and γ i its activity coefficient in the condensed phase. The gas/particle equilibria were applied to the full set of stable (i.e. non-radical) organic species formed during α-pinene oxidation (around 180 species as described in the MCMv3.1). As aerosol particles are expected to be composed of organic molecules with a similar structure, we make the approximation that γ i may be set to unity (e.g., Seinfeld and Pankow, 2003 vapour pressure (probably subcooled) of the organic species was estimated using the Myrdal and Yalkowsky method (Myrdal and Yalkowsky, 1997) coupled with the Joback structure/property relationship for boiling point estimates (e.g., Reid et al., 1986), considering its reliability for SVOC generated during gas-phase oxidation (Camredon and Aumont, 2006). No chemical reactions were implemented in the condensed phase.

Box model
The gas-phase kinetic and the gas/particle partitioning modules were implemented in a zero-dimensional box model (Martin, 2009). The two modules were solved by operator splitting and treated as consecutive operators. The gas-phase kinetic module was integrated with the CVODE solver as downloaded from the SUNDIALS (SUite of 10 Nonlinear and DIfferential/ALgebraic equation Solvers) website (https://computation. llnl.gov/casc/sundials/main.html). The absorptive gas/particle partitioning module was solved using the iterative method described in Pankow (1994b). The partitioning processes between the gas and condensed phases were solved with an operator time step of 1 min, as an operator time step below 5 min was required to prevent the introduction 15 of errors associated with operator splitting. The box model was initialised with the environmental conditions and initial concentrations shown in Table 2 in order to simulate the experimental conditions. The simulation was initialised at a time point corresponding to the precursor injection for the α-pinene+O 3 experiment and the ozone injection for the α-pinene+O 3 +CO experiment 20 (shown at time zero in Fig. 3). The observed injection profiles were implemented as a constant flux that reproduced the observed α-pinene (for the α-pinene+O 3 experiment) or ozone (for the α-pinene+O 3 +CO experiment) concentrations ( Table 2). The simulations were performed with environmental parameters (temperature, relative humidity and dilution rate) averaged over each experiment duration, these parameters 25 varying minimally. The chamber mixture is diluted during the experiment because of the periodical introductions of purified air in the chamber to compensate for losses from analytical sampling and also through leakage. The dilution rate was determined 27848 Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version Interactive Discussion by following SF 6 concentration during each experiment as measured by FT-IR (Fourier Transform InfraRed), and was well described by a single first order decay (Table 2). Aerosol formed during an experiment is affected by both dilution and by depositional losses onto the chamber surfaces. Both processes were combined into a global firstorder loss with no dependence on aerosol size. At the end of each experiment, the 5 majority (>80%) of the precursor (α-pinene) was consumed and the particle size distribution of the aerosol remained unchanged. We assume that at this point, dilution and wall loss of particles were the dominant processes affecting SOA evolution. The aerosol loss rate was calculated from the decay of the total particle volume concentration at this point, prior to starting SOA filter collection.

Temporal profile of SOA mass
The simulated temporal profile of SOA mass is compared with that observed in Fig. 3 for the α-pinene+O 3 and α-pinene+O 3 +CO EUPHORE experiments. The observed SOA mass concentration was determined assuming an aerosol density of 1.2 g cm −3 , as measured previously for SOA from α-pinene ozonolysis (e.g., Zelenyuk et al., 2008). 15 The temporal evolution of the precursors (α-pinene and O 3 ) is also shown. The simulated α-pinene concentration differs by less than 10 ppb from the observations. The ozone decay is however underestimated, especially for the experiment performed without a scavenger, with a maximum difference in O 3 concentrations of 20 ppb. A rapid formation of SOA is observed as soon as the α-pinene is oxidized. SOA mass max-20 ima of 230 and 90 µg m −3 were observed for the α-pinene+O 3 and α-pinene+O 3 +CO experiments after approximately 60 and 90 min, respectively. The simulated temporal profile of SOA mass is relatively well captured by the model both in terms of the shape and the magnitude. The modelled formation of SOA starts as soon as the simulation begins to reach a maximum after ∼60 and ∼80 min for the α-pinene+O 3 and the α-pinene+O 3 +CO experiments, respectively, which is temporally similar to the observed maxima. The maximum SOA mass concentration is overestimated by the Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version Interactive Discussion model by around 40 µg m −3 (17%) and 55 µg m −3 (61%) for the α-pinene+O 3 and α-pinene+O 3 +CO experiments, respectively. For these two α-pinene experiments, the model simulates the observed SOA mass within a factor of 2, an acceptable performance considering the uncertainties in SOA formation, especially condensed phase reactions and vapour pressure estimates. Sensitivity studies on these two parameters 5 are presented in Sect. 4.1 and 4.2. The simulated SOA composition is compared with that observed in Sect. 5.

Condensed phase reactivity
No condensed phase reaction was considered in the model. If we assume that the products of any condensed phase reaction are less volatile than their precursor reac-10 tants, the aerosol mass simulated here can be seen as a lower limit (e.g., Kroll and Seinfeld, 2008). In order to assess the influence of SVOC reactivity in the condensed phase on SOA mass, a sink of condensed SVOC was implemented in the model. Each condensed SVOC, SVOC (a) i , was considered to follow a pseudo first order loss process with a kinetic constant k This non-volatile organic species, nvoc, was considered to remain in the condensed phase and to be inert chemically, representing a permanent sink of SVOC to the condensed phase. This loss process of each condensed SVOC was implemented in the 20 box model at the same time step as the gas/particle partitioning processes of each organic, i.e. with an operator time step of 1 min. Simulations were performed considering a chemical lifetime for condensed SVOC (τ

Printer-friendly Version
Interactive Discussion maximum is increased by 35 (13%), 60 (22%) and 130 µg m −3 (48%) during the α-pinene+O 3 experiment and by 40 (28%), 60 (41%) and 95 µg m −3 (66%) during the α-pinene+O 3 +CO experiment, in comparison with the SOA maximum simulated with no condensed phase reactivity, for τ (a) react =60, 30 and 5 min, respectively. Generally, assuming the condensed phase lifetime of SVOC is of the order of 5 min or longer, 5 the implementation of condensed phase reactions in the model would increase the simulated SOA mass by less than a factor of 2.

Vapour pressure estimates
The results of this study may be compared with that of Jenkin (2004), in which the MCM was used to simulate SOA formation during α-pinene chamber ozonolysis experiments, 10 performed under environmental conditions similar to this work. In the earlier study, negligible SOA concentrations were simulated in comparison with experiment, unless the equilibrium constants for gas/particle partitioning were increased by two orders of magnitude. Our approach differs from that of Jenkin (2004)   The uncertainty in vapour pressure estimates is the most likely reason for the overestimation of SOA mass in these experiments. An increase of all vapour pressures (P vap M&Y,J ) by a factor of 2 in the model would lead to an accurate reproduction of the observed SOA mass in both experiments (see Fig. 3). 15 SOA formation from α-pinene ozonolysis was studied with a particular focus on the distribution of gaseous and particulate organic compounds. The organic speciation is illustrated here with modelled versus measured comparisons performed for the α-pinene+O 3 EUPHORE experiment. Interactive Discussion salient points required to interpret the CIR-TOF-MS m/z spectra are presented here. During the current experiments, sample air from the EUPHORE chamber was delivered continuously to the CIR-TOF-MS through a heated 50 cm long Teflon sampling line. Once within the instrument, the analyte molecules underwent chemical ionization via reaction with the hydronium ion, H 3 O + . Using this method the reagent reacts with the analyte, R, via direct proton transfer to form a positively charged ion RH + :

Gaseous phase VOC and SVOC
The proton addition occurs predominantly at the site of highest electronic density on the analyte, i.e. in order: acid, ketone, aldehyde, alcohol and hydroperoxide functional groups, for moieties of relevance to tropospheric oxidation (Blake et al., 2009). Once 10 formed the RH + ion can undergo fragmentation to produce an ionised daughter fragment (F + ) and a neutral molecule (N) (McLafferty and Turecek, 1993): For molecules containing acid, aldehyde and/or alcohol functionalities, the most common fragmentation mechanism following proton transfer ionisation is water elimination 15 (Smith and Spanel, 2005), e.g. for an alcohol ROH: Following ionisation, analyte ions are accelerated by an electric field into a time-offlight mass spectrometer for analysis. The analyte R is then observed in its proto-  Table 3. The mass resolution (m/∆m) of the CIR-TOF-MS during the current experiments was around 1500. The CIR-TOF-MS spectra presented here were normalized to 10 6 H 3 O + ion counts, background subtracted and averaged over 10 min (Blake et al., 2004).

5
A direct comparison of the simulated gaseous concentrations with the CIR-TOF-MS mass spectrum is difficult owing to the presence of numerous fragment ions in the spectrum. A simplified representation of this fragmentation has been considered to create a "simulated gaseous mass spectrum". The simulated gaseous mass spectrum has been constructed as follows: 1. m/z peaks for each species R: ions considered were RH + giving a m/z [M+1] + peak, and in some cases fragmented forms, such as RH + -H 2 O giving a m/z [(M+1)−18] + peak and/or other F + i fragments giving a peak at their corresponding m/z through the following assumptions. For compounds having a characterized CIR-TOF-MS signature the relative contribution of each m/z peak was used, as 15 given in Table 3. For other compounds it has been assumed that (1) protonated species having a chain length shorter than or equal to 3 carbons do not fragment and (2) protonated species containing an acid, aldehyde or alcohol moiety (following the acid, ketone, aldehyde, alcohol, hydroperoxide priority order for the protonated site on a multifunctional species) dehydrate at a value of 40% (repre-20 sentative of the known fragmentation fractions; see Table 3). No fragmentation was considered for the other compounds; 2. intensity of the m/z peaks for each species R: the intensity of each m/z peak was assumed to be proportional to the simulated gaseous concentration of the species R scaled by their relative contribution; 3. total m/z intensity at a resolution of 1 Da: the total intensity of each integer m/z peak was calculated by summing the intensities of all the ions between m/z−0.5 and m/z+0.5.
It is important to note that for compounds which do not have characterised fragmentation patterns (i.e. the majority of secondary species), this approach neglects fragments 5 other than dehydrated ions of the protonated analyte. Furthermore, this method considers that the simulated signal intensity is proportional to the concentration, independent of the species, whereas in practice each species has a different ionisation efficiency, and hence a different CIR-TOF-MS response. In addition, the fraction of the analyte that dehydrates is fixed at 40%, a value that in reality is likely to differ largely between 10 species (Smith and Spanel, 2005). A quantitative comparison between the simulated and CIR-TOF-MS signal intensity is therefore not possible, but a qualitative comparison with the observations may be made.

Comparison of the gaseous mass spectra
The simulated gaseous mass spectrum is compared with the observed CIR-TOF-MS Interactive Discussion most of the observed peaks are not present in the simulation. These signals (such as at m/z 43, 93, 95 or 99) are likely to be fragment ions that are not considered in our simple fragmentation approach. Species have been assigned to the major peaks detected by the CIR-TOF-MS on the basis of characterized CIR-TOF-MS signatures (see Table 3) and previous identi-5 fication of secondary products formed from α-pinene oxidation (e.g., Yu et al., 2008). The proposed structures for these major compounds are shown in Table 4. The main peaks arise from formaldehyde (m/z 31), formic acid (m/z 47), acetone (m/z 59), acetic acid (m/z 61), α-pinene (m/z 137, 81) and pinonaldehyde (m/z 169, 151, 109, 107). The simulated peaks for many of these major observed m/z signals are consistent with 10 the CIR-TOF-MS measurements. The exceptions concern the m/z of acetic acid which has a high intensity in the CIR-TOF-MS signal but is simulated to be formed at very low concentration and formic acid that is not produced in the simulation. Formic and acetic acids signals present within the measurements are most likely a result of offgassing from the chamber walls, which is often observed with such experiments (e.g., 15 Rickard et al., 2009); such reactions not represented in our chamber mechanism. It should also be noted that other larger molecular weight species can produce fragment ions of m/z 61 using the proton transfer technique. The simulated contribution of the identified compounds to the total concentration of gaseous species at a given molar mass (±0.5 g mol −1 ) is also given in Table 4. These assigned compounds are calcu-20 lated to make a dominant contribution (>70%) to their mass channel, indicating that the assigned species are the main product found at these m/z. However other species detectable by CIR-TOF-MS could also potentially contribute to an observed m/z peak. For example, glycolaldehyde is simulated to contribute 27% of the total concentration at m/z 61 compared with 73% from acetic acid, and 2-hydroxopinan-3-one which is 25 simulated to contribute 15% to the total concentration at m/z 169 compared with 85% from pinonaldehyde.
Apart from pinonaldehyde, the major peaks present in the CIR-TOF-MS spectrum correspond to volatile species. Gaseous SVOC, of low volatility and hence most important for SOA formation, are generally expected to have a higher molecular weight and therefore are to be found at higher m/z. Figure 5b focuses on the gaseous mass spectrum for m/z greater than the protonated α-pinene ion (RH + at m/z 137). Structures for the gaseous SVOC detected by the CIR-TOF-MS have been suggested on the basis of the simulated speciation, and are shown in Table 4 together with their 5 simulated contribution to the total concentration of gaseous species having the same molar mass (within ±0.5 g mol −1 ). Most of these species, such as pinonic acid (RH or norpinonaldehyde (RH + at m/z 155) have been previously identified as secondary products formed during α-pinene oxidation (e.g., Christoffersen et al., 1998;Yu et al., 10 1998Yu et al., 10 , 1999Glasius et al., 2000;Koch et al., 2000;Larsen et al., 2001;Lee et al., 2006;Yu et al., 2008;Ma et al., 2008). At the same mass channel as pinonic acid (RH + at m/z 185), an hydroxypinonaldehyde is simulated to have a large contribution to the m/z peak (63% compared with 26% from pinonic acid). An isomer of this hydroxyl pinonaldehyde has previously been identified by Yu et al. (1998Yu et al. ( , 1999  oxide is not included in the MCMv3.1, but could have a contribution to the observed m/z 153 peak. However, the contribution of pinene oxide to the m/z 153 peak is expected to be low due to its important fragmentation in the CIR-TOF-MS (see Table 3).
The observed peaks at m/z 149 and m/z 165 are absent in the simulated spectrum. Yu et al. (2008)  have not been detected by the CIR-TOF-MS either as a protonated parent ion or as a dehydrated fragment ion. However, species having a similar molar mass were found in the condensed phase (see Sect. 5.2). These species were possibly not detected by CIR-TOF-MS because: (i) they contribute mainly to the composition of the condensed phase, with gaseous concentrations under the CIR-TOF-MS detection limit (simulated 5 concentrations were between 10 and 50 ppt) or (ii) they have very low vapour pressures and hence could condense onto the walls of the CIR-TOF-MS inlet line.

The observed ESI-MS mass spectrum
The mass spectral distribution of condensed organics was measured by ESI-MS. A de-10 tailed description of the instrument is given by Hamilton et al. (2008). As for the CIR-TOF-MS only details required to interpret the ESI-MS m/z spectra are briefly presented. The aerosol filter sampled at the conclusion of the experiment was extracted by sonication using water as a solvent (e.g., Hamilton et al., 2008). The extracted solution was directly infused into the electrospray ionisation source using a syringe pump and 15 analysed using an ion trap mass spectrometer to obtain a mass distribution. Electrospray ionisation is a soft ionisation method expected to result in minimal fragmentation. The mass spectrometer was used both in positive and negative ionization modes. In positive ionisation, ions are produced by protonation or cationisation with sodium ions (Na +

The simulated condensed phase mass spectrum
A "simulated condensed phase mass spectrum" has been produced from the modelled SOA composition in the same way as for the gaseous mass spectrum, for comparison with the observed ESI-MS mass spectrum: 3. total m/z intensity at a resolution of 1 Da: the total intensity of each m/z peak was calculated by summing the intensities of all of the ions between m/z−0.5 and m/z+0.5.
As for the simulated gaseous signal, this approach assumes that the simulated signal 15 intensity is proportional to the concentration, independent of species identity, whereas in practice the ESI-MS response will be dependent on the species' ionisation efficiency. While a quantitative comparison is not possible, a qualitative comparison between the simulated and ESI-MS measured mass distributions may be made.

Comparisons of the condensed mass spectra 20
The simulated condensed phase mass spectra are compared with the ESI-MS measured spectra in Fig. 6  [R+Na] + adducts at m/z between approximately 330 and 430 Da) can be clearly identified. These two regions are also present in the measured negative ionisation mode spectrum. However the signal in the oligomeric region is rather noisy and most of the peaks are probably due to contamination and in-source creation of acidic dimer artefacts (Müller et al., 2009). These are not formed in positive ionisation. The oligomers 5 observed in positive ionisation mode are not artefacts of the instrument (they can be separated using LC). This oligomeric region is likely the result of accretion reactions of two monomers either in the aerosol or the gas-phase. These oligomers are homo or heterodimers (referred to hereafter simply as "dimers"). The simulated condensed phase mass spectra are plotted on the lower y-axis in each case. The observed and simulated peaks in the monomer region are in reasonable agreement for both the positive and negative ionisation spectra. The good correspondence for the positive ionisation indicates that the mass distribution of the products is well simulated, while the agreement for negative mode gives confidence in the assignment of acidic functionality. Therefore, the formation of the majority of the observed condensed phase monomer 15 species can be explained by gas/particle partitioning of the SVOC formed during gasphase oxidation. All of the peaks present in the simulated mass spectra were observed by the ESI-MS. A few species detected by the ESI-MS in the monomer region (such as [R+Na] + adducts at m/z 225 or 241 and a small section between 240 and 280 Da) were not present in the simulation. Gaseous compounds with a similar molar mass 20 were not detected by the CIR-TOF-MS. However, not enough is yet known to establish whether these species are formed during gaseous oxidation of α-pinene and/or during condensed phase reactions. The mass distribution in the dimer region is not simulated as no accretion reaction was implemented within the model. Structures for the condensed monomer species have been assigned to the major 25 peaks detected by the ESI-MS in the positive and negative ionisation modes on the basis of the simulated speciation. The proposed species with their simulated contribution to the total concentration of condensed species at the same molar mass (±0.5 g mol −1 ) are shown in Table 5. Aerosol extracts were also separated using LC and subjected to Collision Induced Dissociation (CID) to obtain fragmentation patterns, which were used to predict chemical structures of the SOA components. Simulated SOA components that can be identified in the collected aerosol, based on fragmentation patterns, are given in Table 5. Organic acids such as pinonic acid ( and previous studies on SOA formation from α-pinene oxidation (e.g., Christoffersen et al., 1998;Hoffmann et al., 1998;Yu et al., 1999;Jang and Kamens, 1999;Glasius et al., 2000;Koch et al., 2000, Iinuma et al., 2004Jaoui et al., 2005). The [R+Na] + adduct of pinonic acid at m/z 207 has the highest observed peak intensity of the positive mode. However a hydroxypinonaldehyde is simulated to contribute 54% to the total concentration of the [R+Na] + adduct at m/z 207 and pinonic acid 35%. An isomer of hydroxypinonaldehyde has previously been identified in the condensed phase by Yu et al. (1999). Norpinic acid was also detected here within the condensed phase by 15 LC-MS n and has also been observed in previous studies (e.g., Kavouras et al., 1998;Glasius et al., 2000;Iinuma et al., 2004). In this study the concentration of norpinic acid is simulated to be very low, in both the condensed and gaseous phases. terized to date owing to difficulties in their accurate detection and quantification.

Reactivity in the condensed phase
As noted, the high molecular weight species detected by ESI-MS are probably homo or heterodimeric species. Among the reactions suggested to explain the formation of  Fig. 7. Combination reactions occurring in the gaseous phase followed by gas-particle partitioning, such as peroxy radical self reactions (e.g., Ziemann, 2002) and stabilized Criegee reactions with other oxygenates (Sadezky et al., 2006;Heaton et al., 2007;Zahardis and Petrucci, 2007), could also represent a signifi- 25 cant pathway for formation of the observed dimers (e.g., Müller et al., 2009)

Comparisons of the condensed dimer mass spectra
The simulated condensed-phase dimer mass spectra for those accretion reactions considered are compared with the observed mass spectrum under positive ionisation mode in Fig. 8. Many of the observed features in the dimer region of the measured spectrum ([R+Na] + adducts at m/z between 300 and 500) are consistent with 5 all of the accretion reactions considered, however peroxyhemiacetal and hemiacetal formation reactions fail to represent dimers with the lowest masses ([R+Na] + adducts at m/z<360) and lead to dimers with higher masses than those observed ([R+Na] + adducts at m/z>440) (Fig. 8a,b). Hemiacetal formation has previously been found to be thermodynamically unfavourable (Barsanti et al., 2004). Accretion reactions involv- 10 ing dehydration, such as aldol condensation and esterification, result in a better agreement between the simulated and observed dimer mass spectra (Fig. 8c,d). Dehydration has already been observed to take place during dimer formation in SOA from terpene ozonolysis (Reinhardt et al., 2007;Walser et al., 2008;Müller et al., 2009). Aldol condensation in isolation does not explain the formation of all the observed masses, 15 with around one third of observed peaks missing (Fig. 8c). Aldol condensation has been found to be favourable thermodynamically (Barsanti et al., 2004) but may be acid catalysed (e.g., Kroll et al., 2008), probably not representative of our experimental conditions. Esterification reactions result in a modelled dimer mass distribution in very good agreement with the observed ESI-MS spectrum (Fig. 8d)  Interactive Discussion having this molar mass. A dimeric species formed by esterification and highlighted in red in Fig. 9 is in agreement with the chemical analysis. Each accretion reaction has been investigated independently here whereas in practice these reactions could very well occur in parallel in the condensed phase. The qualitative agreement apparent in Fig. 8d lead us to infer that esterification reactions between the condensed monomers 5 may be the dominant processes forming dimeric species during α-pinene ozonolysis experiment. We note however that the various dimeric species differ in stability, hence during the analysis, the less thermodynamically stable dimers, such as hemiacetals or peroxyhemiacetals, could preferentially converted back to monomers, and therefore not been detected by ESI-MS. Esters would then be the main species observed by 10 ESI-MS due to their stability.

Conclusions
SOA formation from dark α-pinene ozonolysis has been studied with an emphasis upon the composition of gaseous and condensed phase SVOC, interpreted qualitatively in terms of their measured and simulated mass distributions. In the course of EUPHORE 15 chamber simulations, the gas-phase oxygenated organic product evolution and distribution was followed using an online CIR-TOF-MS. The semi-volatile oxygenated organic distribution in the condensed phase was acquired through offline ESI-MS analysis of SOA filters taken at the peak SOA mass loading. A detailed chamber box model designed to simulate SOA formation has been developed for comparison with 20 the experimental results, coupling an equilibrium gas/particle partitioning module to the α-pinene oxidation scheme extracted from the MCMv3.1. The simulated temporal profile of SOA mass for these specific α-pinene ozonolysis experiments is in reasonable agreement with the observations, both in terms of the shape and magnitude. A sensitivity analysis showed that the large differences in published model performances 25 with respect to the simulation of the temporal evolution of SOA mass formation from α-pinene ozonolysis may be largely a consequence of the use of different methods for SVOC vapour pressure estimation. The uncertainties associated with vapour pressure estimates are crucial for the development of reliable organic gas/particle partitioning. However the evaluation of vapour pressure estimation methods for the purpose of SOA modelling is still limited by the availability of experimental data for organic species of interest, i.e. multifunctional species having vapour pressures lower than 10 −6 atm.

5
Comparisons of the simulated mass spectrum with the CIR-TOF-MS mass spectrum show a similar mass distribution for gaseous organics. The simulated composition for the major peaks detected by the CIR-TOF-MS is in agreement with previous identifications of products formed during α-pinene ozonolysis. Identification of the main peaks of gaseous SVOC present in the CIR-TOF-MS spectrum has been proposed on the basis 10 of the simulated composition. The ESI-MS measurements show the presence in the condensed phase of monomer and oligomer type organics. Comparisons of the simulated condensed phase mass spectrum with the ESI-MS mass spectrum show that most of the peaks in the monomer region are in reasonable agreement, i.e. explained in the simulation by the gas/particle partitioning of SVOC formed during the gas-phase ox- 15 idation of α-pinene. A chemical structure for the principal condensed-phase monomers detected has been proposed on the basis of the simulated composition and known LC-MS fragmentation patterns.