Ozonolysis of α-phellandrene – Part 1 : Gas-and particle-phase characterisation

The ozonolysis of α-phellandrene, a highly reactive conjugated monoterpene largely emitted by Eucalypt species, is characterised in detail for the first time using a smog chamber at the Guangzhou Institute of Geochemistry, Chinese Academy of Sciences. Gas-phase species were monitored by a proton-transfer-reaction time-of-flight mass spectrometer (PTR-TOF), with yields from a large number of products obtained, including formaldehyde (5– 9 %), acetaldehyde (0.2–8 %), glyoxal (6–23 %), methyl glyoxal (2–9 %), formic acid (22–37 %) and acetic acid (9–22 %). Higher m/z second-generation oxidation products were also observed, with products tentatively identified according to a constructed degradation mechanism. OH yields from α-phellandrene and its first-generation products were found to be 35± 12 and 15± 7 %, respectively, indicative of prominent hydroperoxide channels. An average first-generation rate coefficient was determined as 1.0± 0.7× 10−16 cm3 molecule−1 s−1 at 298 K, showing ozonolysis as a dominant loss process for both αphellandrene and its first-generation products in the atmosphere. Endocyclic conjugation in α-phellandrene was also found to be conducive to the formation of highly condensible products with a large fraction of the carbon mass partitioning into the aerosol phase, which was monitored with a scanning mobility particle sizer (SMPS) and a high-resolution timeof-flight aerosol mass spectrometer (AMS). Nucleation was observed almost instantaneously upon ozonolysis, indicating the rapid formation of extremely low-volatility compounds. Particle nucleation was found to be suppressed by the addition of either NO2 or a Criegee scavenger, with it being proposed that stabilised Criegee intermediates are important for new particle formation in the system. Aerosol yields ranged from 25 to 174 % depending on mass loadings, with both firstand second-generation products identified as large contributors to the aerosol mass. In short, with a high chemical reactivity and aerosol-forming propensity, α-phellandrene is expected to have an immediate impact on the local environment to which it is emitted, with ozonolysis likely to be an important contributor to the significant blue haze and frequent nocturnal nucleation events observed over Eucalypt forests.

Abstract.The ozonolysis of α-phellandrene, a highly reactive conjugated monoterpene largely emitted by Eucalypt species, is characterised in detail for the first time using a smog chamber at the Guangzhou Institute of Geochemistry, Chinese Academy of Sciences.Gas-phase species were monitored by a proton-transfer-reaction time-of-flight mass spectrometer (PTR-TOF), with yields from a large number of products obtained, including formaldehyde (5-9 %), acetaldehyde (0.2-8 %), glyoxal (6-23 %), methyl glyoxal (2-9 %), formic acid (22-37 %) and acetic acid (9-22 %).Higher m/z second-generation oxidation products were also observed, with products tentatively identified according to a constructed degradation mechanism.OH yields from α-phellandrene and its first-generation products were found to be 35 ± 12 and 15 ± 7 %, respectively, indicative of prominent hydroperoxide channels.An average first-generation rate coefficient was determined as 1.0 ± 0.7 × 10 −16 cm 3 molecule −1 s −1 at 298 K, showing ozonolysis as a dominant loss process for both αphellandrene and its first-generation products in the atmosphere.Endocyclic conjugation in α-phellandrene was also found to be conducive to the formation of highly condensible products with a large fraction of the carbon mass partitioning into the aerosol phase, which was monitored with a scanning mobility particle sizer (SMPS) and a high-resolution timeof-flight aerosol mass spectrometer (AMS).Nucleation was observed almost instantaneously upon ozonolysis, indicating the rapid formation of extremely low-volatility compounds.Particle nucleation was found to be suppressed by the addition of either NO 2 or a Criegee scavenger, with it being proposed that stabilised Criegee intermediates are important for new particle formation in the system.Aerosol yields ranged from 25 to 174 % depending on mass loadings, with both first-and second-generation products identified as large contributors to the aerosol mass.In short, with a high chemical reactivity and aerosol-forming propensity, α-phellandrene is expected to have an immediate impact on the local environment to which it is emitted, with ozonolysis likely to be an important contributor to the significant blue haze and frequent nocturnal nucleation events observed over Eucalypt forests.

Introduction
Biogenic sources dominate the global emission budget of volatile organic compounds into the atmosphere, with monoterpenes accounting for a significant fraction of nonmethane hydrocarbons emitted (Guenther et al., 1995;Schurgers et al., 2009;Guenther et al., 2012;Lathière et al., 2006;Sindelarova et al., 2014).Considering source strength, estimated to be 30-127 Tg C year −1 , along with high chemi-Published by Copernicus Publications on behalf of the European Geosciences Union.F. A. Mackenzie-Rae et al.: Ozonolysis of α-phellandrene cal reactivity (Calvert et al., 2000;Atkinson and Arey, 2003), monoterpenes are thought to play an important role in the chemistry of the atmosphere, influencing its oxidative capacity, the tropospheric ozone budget and producing secondary organic aerosol (SOA) with impacts on both health and climate (Hoffmann et al., 1997;Griffin et al., 1999a;Chung and Seinfeld, 2002;Hallquist et al., 2009;Pye et al., 2010).Indeed, the ozonolysis of monoterpenes is thought to be one of the major sources of organic SOA in the atmosphere (Griffin et al., 1999b;Ortega et al., 2012).
Consequently, the gas-phase reaction of ozone with monoterpenes has been the focus of numerous studies, both experimental, with a focus on gas-phase kinetics and particle formation, properties and composition (e.g.Bateman et al., 2009;Berndt et al., 2003;Griffin et al., 1999a;Herrmann et al., 2010;Lee et al., 2006;Ma et al., 2007;Pathak et al., 2007;Saathoff et al., 2009;Shilling et al., 2008Shilling et al., , 2009;;Walser et al., 2008), and theoretical, utilising state-of-the-art computational methods (Zhang and Zhang, 2005;Nguyen et al., 2009).Collectively, research has come a long way to understand the mechanism and product distributions of monoterpene ozonolysis and provided important insights into SOA precursors and production.Accurate chemical mechanisms for the reaction of specific monoterpenes with ozone have since been developed (Camredon et al., 2010;Jenkin, 2004;Leungsakul et al., 2005), whilst more general parameterisations for gas-phase reactions (Jenkin et al., 1997;Saunders et al., 2003) and SOA formation (Odum et al., 1996;Donahue et al., 2006;Stanier et al., 2008) have been implemented into chemical transport models.
Ozonolysis is generally agreed to occur through a concerted cycloaddition of ozone to the olefin bond, forming a 1,2,3-trioxolane intermediate species referred to as a primary ozonide (POZ) (Calvert et al., 2000;Johnson and Marston, 2008).Addition of ozone is highly exothermic with excess energy retained in the POZ structure, resulting in rapid decomposition through homolytic cleavage of the C-C and one of the O-O bonds, which forms, in the case of asymmetrically substituted alkenes, a pair of products containing a carbonyl and a reactive Criegee intermediate (CI).Sufficient vibrational and rotational excitation exists in the CI to permit further unimolecular decomposition which typically occurs through one of two channels; firstly, excited CIs can cyclise to a dioxirane, which then decomposes to a carboxylic acid, ester or lactone, depending on neighbouring substituents, in what is known as the ester or "hot" acid channel, or secondly, when available, excited CIs can isomerise via a 1,5hydrogen shift to form a vinyl hydroperoxide, which subsequently decomposes into a vinoxy radical and a hydroxyl radical in what is known as the hydroperoxide channel.Alternatively, excited CIs can be collisionally stabilised such that bimolecular reactions with trace species (e.g.H 2 O, NO 2 , CO, aldehydes, acids) become important (Johnson and Marston, 2008).The relative prevalence of these competing channels is strongly linked to the structure and conformation of the CI (Vereecken et al., 2015), with the various mechanistic pathways summarised in Fig. 1.
When considered as a whole, research shows significant variability in gas-phase ozonolysis products and SOA yields between different monoterpenes due to their structural differences, highlighting the unique impact different monoterpenes can have on regional atmospheric chemistry.It is therefore important that individual monoterpene variability be accounted for in developing accurate gas-and particlephase models.Nonetheless, current literature has predominantly focused on a small number of the more commonly emitted monoterpenes (e.g.α-pinene, β-pinene, limonene).One monoterpene for which relatively little is known is αphellandrene (structure provided in Fig. 1).One of the most reactive monoterpenes, α-phellandrene has been identified as a major constituent of extracts (Li et al., 1995;Brophy and Southwell, 2002;Pavlova et al., 2015;Maghsoodlou et al., 2015) and in emissions (He et al., 2000;Maleknia et al., 2009) from various Eucalypt species, the world's most widely planted hardwood tree (Myburg et al., 2014).During day-to-day activities and processes, Eucalypts, such as Eucalyptus microtheca, Eucalyptus viminalis and Eucalyptus dives, emit α-phellandrene into the atmosphere, with αphellandrene likely contributing to the intense and frequent particle nucleation events observed over Eucalypt forests -a phenomenon already believed to be caused by monoterpene oxidation (Suni et al., 2008;Lee et al., 2008;Ortega et al., 2009Ortega et al., , 2012)).In the indoor environment, α-phellandrene can be found as an additive to household cleaning products, detergents and air fresheners (e.g.Eucalypt-themed products), with the European EPHECT project reporting αphellandrene at a concentration of 16.7 µg m −3 in a study of a passive air freshener in a 1 m 3 room after 5 h (Stranger, 2013).Maisey et al. (2013) reported similar maximum concentrations of α-phellandrene in Australian dwellings.
The rate constant of α-phellandrene with ozone has been measured in a number of studies with results spanning an order of magnitude (Grimsrud et al., 1975;Atkinson et al., 1990;Shu and Atkinson, 1994), with a rate constant of 3.0 × 10 −15 (± 35 %) cm 3 molecule −1 s −1 favoured (Calvert et al., 2000).High chemical reactivity likely makes ozonolysis a dominant loss process for α-phellandrene in the atmosphere; however, experimental information regarding reaction products is limited to OH radical yields, measured by Herrmann et al. (2010) to be 26-31 and 8-11 % for the ozonolysis of the two double bonds, and acetone yields, which were reported by Reissell et al. (1999) to be minor (< 2 %).Recently, the reaction mechanism was investigated theoretically for the first time by Mackenzie-Rae et al. (2016), who mapped the potential energy surface to first-generation products.A basic overview of the reaction pathways is provided in Fig. 1, with a comprehensive discussion of the reaction mechanism of α-phellandrene with ozone based on findings of Mackenzie-Rae et al. (2016) pertinent to this study provided in the Supplement (Sect.S1).This study aims to experimentally characterise the reaction of α-phellandrene with ozone in detail for the first time by exploring and characterising both the gaseous and particle phases, with the impact of Criegee scavengers and NO 2 on the system addressed.In doing so, the impact of a highly reactive and potentially important monoterpene will be parameterised.

Atmos
2 Materials and method 2.1 Experimental set-up and procedure A total of 11 dark α-phellandrene ozonolysis experiments were conducted using the indoor smog chamber facility at the Guangzhou Institute of Geochemistry, Chinese Academy of Sciences (GIG-CAS).A complete description of the facility and chamber setup is given in Wang et al. (2014).Briefly, the GIG-CAS smog chamber consists of a 30 m 3 fluorinated ethylene propylene (FEP) reactor housed inside a temperature-controlled room.The reactor was flushed with purified dry air for at least 48 h prior to each experiment, until no residual hydrocarbons, O 3 , NO x or particles were detected, with the impact of off-gassing of radicals from the reactor walls during experiments under the dark, dry conditions considered negligible (Wang et al., 2014).Two Tefloncoated fans located inside the reactor ensure rapid homogenisation of introduced species.Liquid reactants were vaporised via injection into a heating system similar to that of gas chromatography, before being carried by nitrogen gas through FEP Teflon lines into the reactor.Ozone was generated using a commercial ozone generator (VMUS-4, Azco Industries Ltd.), with pure oxygen feed gas.Initial mixing ratios of the reactants varied between 10 and 175 ppb for αphellandrene (Aldrich Chemical Company, Inc., USA) and between 56 and 500 ppb for O 3 .α-Phellandrene was injected prior to admission of O 3 into the chamber, with O 3 added through two separate additions in experiments 7 and 10 to facilitate the identification of detected species as either first-or second-generation products.Anhydrous cyclohexane (Sigma-Aldrich, 99.5 %) was added in sufficient quantity in all but two experiments to scavenge > 95 % of OH radicals (Aschmann et al., 1996;Herrmann et al., 2010), with the remaining experiments used to asses the impact of cyclohexane's inclusion.Formic acid (J&K Scientific Ltd., 98 %) was added to experiments 6 and 7 as a stabilised Criegee intermediate (sCI) scavenger to better understand the impact of sCIs on gas-phase species distribution and, importantly, particlephase formation and growth, for which it has been identified as a significant precursor (Bonn et al., 2002;Bateman et al., 2009;Sakamoto et al., 2013;Wang et al., 2016).Prior to O 3 addition in experiment 11, 385 ppb of NO 2 was added through a septum installed in one of the injection ports using a gas-tight syringe, with the inclusion providing an alternative representation of tropospheric nocturnal chemistry in a polluted environment.All experiments had 2.5 µL of acetonitrile injected as a dilution tracer, with the top frame of the reactor periodically lowered to maintain a positive pressure differential inside the reactor.Experimental run times ranged from 205 to 305 min, with a final reactor volume typically between 6 and 8 m 3 .The starting conditions for each experiment are listed in  are similar to the mixing time of the reactor.Consequently only a lower bound of the ozone concentration is known.

Characterisation of gas and particle phases
Volatile organic compounds (VOCs) were measured online with a commercial proton-transfer-reaction time-offlight mass spectrometer (PTR-TOF 2000, Ionicon Analytik GmbH, Austria) (Jordan et al., 2009;Graus et al., 2010), using H 3 O + reagent ions.For data collected in the first seven experiments in Table 1, the PTR-TOF drift tube was operated at 2.2 mbar and 60 • C, with a drift tube field of 600 V cm −1 (E/N = 136 Td).Significant fragmentation was observed under this regime, with a drift tube voltage of 484 V cm −1 (E/N = 112 Td) found to be optimal (Supplement Sect.S3).The refined operating conditions were then used for experiments 8-11.PTR-TOF spectra were collected at a time resolution of 2 s.Data were processed using the PTR-TOF data analyser (Müller et al., 2013), with 30 spectra averaged to improve counts of trace species.A generic H 3 O + rate constant of 2×10 −9 cm 3 s −1 was used for conversion into ppb, except for those species where experimental or theoretical data exist (Cappellin et al., 2012;Tani, 2013).Gas-phase O 3 and NO x were measured online using dedicated monitors (EC9810 and 9841T, Ecotech, Australia), which were calibrated regularly using a Thermo Scientific model 146i multi-gas calibrator unit.In all experiments, excluding experiment 11 where it is added, NO x concentrations were negligible (< 1 ppb).The O 3 analyser experienced significant interference (had a false bias) from α-phellandrene, which was corrected for using PTR-TOF measurements.
Particle number size distributions were measured online with a scanning mobility particle sizer (SMPS; TSI Inc., USA) (Wang and Flagan, 1990), consisting of an electro-static classifier (TSI 3080) fitted with a TSI 3081 differential mobility analyser (DMA) and condensation particle counter (CPC, TSI 3775).Sheath and aerosol flow rates were 3.0 and 0.3 L min −1 , respectively, with voltage inside the DMA varied exponentially from −10 to −9950 V every 240 s to provide a mobility spectrum over particle diameters 14-750 nm.Higher moment size distributions were calculated by assuming spherical particles (Wiedensohler et al., 2012).
A high-resolution time-of-flight aerosol mass spectrometer (AMS; Aerodyne Research Incorporated, USA) was used to measure particle chemical composition in real time (Jayne et al., 2000;DeCarlo et al., 2006).The AMS was operated in the high-sensitivity V mode and high-resolution W mode, switching between modes every 2 min.AMS data were analysed in Igor Pro 6.2 (Wavemetrics) using the ToF-AMS data analysis toolkits Peak Integration by Key Analysis (PIKA) and Sequential Igor Data Retrieval (SQUIRREL).Updates were made to the fragmentation table following a similar method to Chen et al. (2011), with a detailed discussion provided in the Supplement (Sect.S5).Conductive silicon tubes were used as sampling lines for the SMPS and AMS to reduce electrostatic losses of particles, whilst all other instruments had FEP Teflon feed lines.Losses of VOCs and particles in the transfer lines are estimated to be less than 5 % (Liu et al., 2015).
3 Results and discussion 3.1 Gas-phase analysis 3.1.1Peak identification and yields Significant fragmentation was observed in the PTR-TOF upon injection of starting materials into a clean reactor.α-Phellandrene was detected at m/z 137 at 32-34 % depending on drift tube conditions, consistent with fragmentation observed in the PTR-MS studies of Misztal et al. (2012) and Tani (2013) (Supplement Sect.S2).Acetonitrile was found exclusively at m/z 42 and remained constant throughout all experiments, indicating that dilution effects in the reactor are negligible.Despite having a lower proton affinity than water, cyclohexane was detected at m/z 85, although overall sensitivity is greatly reduced.The detection of cyclohexane is likely the result of termolecular reactions in the PTR-TOF (Smith and Španěl, 2005).Meanwhile, in a separate characterisation experiment, formic acid was found at m/z 47, with minor fragments at m/z 48, 49 and 65 (< 2 %).
Observed interferences are expected to impact detection of α-phellandrene's degradation products, biasing signals to lower m/z.Aldehyde, ketone, alcohol, ester and acid-bearing compounds are known to dehydrate following protonation to yield a MH + (−H 2 O) daughter ion (Smith and Španěl, 2005;Blake et al., 2006).Furthermore, multifunctional carbonyl compounds can eject a second water molecule from nascent MH + ions yielding a MH + (−H 2 O) 2 daughter ion, whilst complex acid-bearing molecules have been observed to fragment via the loss of formic acid to produce MH + (−HCOOH) ions and esters through ejection of −OR groups to yield MH + (−ROH) (Španěl et al., 1997;Španěl and Smith, 1998).Uncertainty arising from fragmentation limits quantitative analysis for the majority of species, with standards neither available nor prepared.Nevertheless, Table 2 lists peaks routinely detected by the PTR-TOF across the 11 experiments.Note that m/z includes the addition of H + .
Figure 2 shows time profiles of major species detected by the PTR-TOF during the ozonolysis of α-phellandrene.For clarity, peaks have been corrected for background readings recorded prior to the introduction of ozone.Upon injection of ozone, α-phellandrene is rapidly oxidised forming a number of product ions at low concentrations that continually increase throughout the experiment.Meanwhile, ozone, after rapid initial consumption, slowly decreases throughout the experiment in part due to losses to the reactor walls (Wang et al., 2014).The stability of acetonitrile and cyclohexane signals supports the finding of Wang et al. (2014) that wall losses are relatively minor for volatile organics in the GIG-CAS chamber.
Ignoring conformational isomerism, the ozonolysis of αphellandrene can yield four unique CIs (Fig. 1) (Mackenzie-Rae et al., 2016), with the degradation mechanism of CI3  provided in Fig. 3. Detailed schematics of the remaining CIs are provided in the Supplement (Sect.S1) and lead to products isomeric to those shown in Fig. 3.A focus is on RO 2 -RO 2 radical chemistry which, due to the large rate constant of α-phellandrene with ozone and lack of competing radical termination channels, dominates under the considered reaction conditions.Elucidating the mechanism of α-phellandrene, one expects initially to form a large range of first-generation products; however, none of the product ions detected were observed to decrease over the course of the chamber experiments, suggesting that detected ions in part correspond to secondgeneration species.For example, from the sCIs, one might expect an unsaturated keto-aldehyde or dialdehyde product (Fig. 5), analogous to pinonaldehyde from α-pinene and limonaldehyde from limonene, to be detected at m/z 169.Indeed, this signal was observed, but it continued to increase in concentration after α-phellandrene was consumed, suggest-   28, 39, 40, 41, 42, 43, 44, 54, 55, 56, 57, C 6 H 12 H + Cyclohexane and fragments 58, 67, 68, 69, 70, 82, 83, 84, 85, 86 81, 99, 100, 116, 117 C  ing that the observed m/z 169 is not simply a direct product ion of α-phellandrene.Other major first-generation product ions expected include m/z 185, which corresponds to a range of isomeric species formed through either excited or thermalised CI rearrangement reactions, whereby three oxygen atoms are added for no loss of carbon or hydrogen (e.g.acids, esters, epoxides, secondary ozonides), and m/z 155, which can be formed through radical transfer and subsequent CHO loss in the hydroperoxide channel (Mackenzie-Rae et al., 2016).Both these ions were detected in the PTR-TOF but again had concentrations which increased throughout the experiments, suggesting that they have large contributions from saturated species.This continual increase remained true in experiments which added a large secondary dose of ozone after commencement of the reaction (Fig. S4.1), confirming the discussed ions as saturated.
A similar phenomenon, whereby a distinct lack of firstgeneration products was observed by a PTR-MS, occurred when studying the ozonolysis of α-terpinene (Lee et al., 2006;Ng et al., 2006), a structurally similar endocyclic-conjugated monoterpene.In the studies of Lee et al. (2006) and Ng et al. (2006), first-generation products were observed using identical methods for other monoterpene species including 3-carene, α-pinene, β-pinene, terpinolene and myrcene.It is possible then that for highly reactive monoterpenes such as α-terpinene and α-phellandrene, concentrations of first-generation products do not accumulate sufficiently during experiments for gas-phase detection.However, as discussed Sect.3.1.3and 3.2.1, a simple rate-study analysis shows that residence lifetimes based on gas-phase reactions are sufficient, whilst analysis of saturation concentrations suggests that the majority of predicted first-generation products likely reside in the gas phase.
Recent literature has shown that functionalised organic species experience considerable losses to Teflon chamber walls through gas-wall partitioning (e.g.Matsunaga and Ziemann, 2010;Zhang et al., 2014;Yeh and Ziemann, 2015;Krechmer et al., 2016;La et al., 2016).Observations indicate that organic compounds are not lost to the reactor walls, but rather partition between the gas-phase and Teflon walls in a reversible process that eventually reaches equilibrium, the speed of which is dependent on reactor geometry, turbulence and species diffusivity, and penetration and accommodation in the reactor walls.Based on the work of Krechmer et al. (2016), the timescale for reaching gaswall equilibrium in these experiments is thought to be less than 600 s.Gas-wall partitioning therefore operates quick enough to affect the considered chamber experiments and detection of first-generation products.The relative impact of gaseous wall losses is further explored in Sect.3.2.1;nonetheless, partitioning is strongly dependent on volatility, with losses of highly functionalised first-generation products of α-phellandrene to reactor walls and/or sample lines during transfer into and detection by the PTR-TOF expected (Yeh and Ziemann, 2015;Krechmer et al., 2016;La et al., 2016).
Figure 2 shows that the highest product signal concentrations were observed for low m/z species ( C 3 ).Whether this is an accurate representation of the system or a systematic bias from fragmentation is unknown; however, anecdotally increased counts of low mass species were observed as the energy of the drift tube was raised, suggesting that the latter does have some effect.Major peaks were found at m/z 31, 45, 47, 59, 61 and 73, corresponding to formaldehyde, acetaldehyde, formic acid, glyoxal, acetic acid and methyl gly-oxal, respectively.Although acetone also resides at m/z 59, based on the low gas-chromatographic yields reported in Reissell et al. (1999) the signal is apportioned to glyoxal.As α-phellandrene contains two double bonds, yields in this work were calculated as the slope of the least square regression between the change in concentration of the oxidation product and change in wall-loss-corrected ozone, as shown in Fig. 4, with ozone wall loss rates frequently characterised following the method described in Wang et al. (2014).The average yield from sequential ozonolysis is therefore calculated with results provided in Table 3.In practice, however, calculations are dominated by data points measured after the consumption of α-phellandrene, with the data corresponding to the initial reaction of α-phellandrene comparably limited and often largely excluded to reduce errors associated with having a finite reactor mixing time.As discussed later, this problem is navigated for OH radicals by using a higher PTR-TOF time resolution and measuring yields against αphellandrene consumption; however, mixing ratios of other oxidation products are too low in the initial stages of the experiment to produce reliable yield data in this regime.For experiments that had two additions of ozone (7 and 10), separate yield lines were fitted for data after each addition of ozone with the results then averaged, therefore maintaining www.atmos-chem-phys.net/17/6583/2017/Atmos.Chem.Phys., 17, 6583-6609, 2017 Table 3. Gas-phase molar yields (%) for major α-phellandrene ozonolysis products.
No. Formaldehyde Acetaldehyde Formic acid Glyoxal Acetic acid Methyl glyoxal 1 6.9 ± 2 8.3 ± 2 37 ± 9.0 23 ± 5 13 ± 3 3.7 ± 0.9 2 5.9 the reported yield as an average of the entire ozonolysis system.Formic and acetic acid were both found to be produced with high yields.The fragmentation pattern of acetic acid was determined in a separate calibration experiment, with 88 % residing at m/z 61 and the remaining mass distributed over m/z 43, 62 and 79, corresponding to dehydration to the acylium ion, the 13 C isotope and protonation by a water cluster, respectively.Correcting for fragmentation, yields of formic and acetic acid were found to range from 22 to 37 % and 9 to 22 %, respectively, across the conducted experiments.Yields of formic acid are considerably higher than what has been reported for the ozonolysis of other terpenes, whilst acetic acid yields are consistent with species containing an endocyclic bond (Lee et al., 2006); although a subtle difference in methodology should be noted with Lee et al. (2006) calculating yields with respect to the parent hydrocarbon.The addition of NO 2 was found to reduce yields of both formic and acetic acid to 10 ± 2 and 5 ± 1 %, respectively, with O 3 losses through reaction with NO x accounted for.The addition of NO 2 therefore acts as an inhibitor to acidic group formation, likely by scavenging acyl peroxy radicals to form peroxyacyl nitrates (PANs).Alternatively, NO 2 can impact the chemistry of the system by reacting with stabilised secondary ozonides (SOZs), although no changes in acid product yields were observed in the experiments where sCIs were scavenged, indicating that this channel is negligibly important in forming low-molecular-weight acids.
Heavier second-generation products routinely detected across experiments are listed in Table 2, with yields for a number of these products given in Table 4.The absence of a yield indicates that the peak was not detected by the PTR-TOF, which typically occurred for minor peaks in experiments with lower starting α-phellandrene concentrations.Again, no fragmentation was assumed in determining yields, although some ions do differ by common fragment mass amounts, suggesting that fragmentation may be important.
For example, m/z 185 and 167, m/z 129 and 111 and m/z 115 and 97 all differ by 18 amu, suggesting that the latter masses could be dehydrated fragments.Whilst strong correlation (R 2 > 0.99) between these pairs of peaks is observed, it is not consistent across the entire dataset, suggesting that there exist multiple contributors to the aforementioned signals.Similar instances are also observed for peaks separated by 28 amu (e.g.m/z 143 and 115) and 46 amu (e.g.m/z 185 and 139).
Calculated yields for these larger products were in general < 5 %, with detected products sufficiently volatile such that gas-wall partitioning losses are thought to be minor (see Fig. 9).Again, the presence of OH radicals in experiment 9 had little effect on product yields.Addition of NO 2 to the system in experiment 11 resulted in significantly reduced yields, with overall distribution remaining similar and no new peaks or evidence of nitrate-containing compounds observed.Nonetheless, alkyl nitrates are known to readily lose HNO 3 after protonation in the PTR-TOF, resulting in the formation of bare alkyl ions (D'Anna et al., 2005;Aoki et al., 2007;Duncianu et al., 2017).Proposed structures for some of these larger second-generation products, along with plausible formation mechanisms, are shown in Fig. 3, although it is possible that more than one species contributes to an observed oxidation product mass.A large number of products also remain unidentified, with their m/z unable to be transcribed to plausible, mechanistically derived structures.
Figure 2 shows that product signals at m/z 31, 59, 73 and 87 show a sharp increase upon commencement of the reaction, suggesting that these products are formed directly from α-phellandrene ozonolysis.Nevertheless, a large fraction of the product mass for these ions is generated after αphellandrene consumption, indicating that yields are largely driven by contributions from second-generation species.Slower initial production of the remaining ions suggests that their formation is linked to consumption of first-generation products.Interestingly, the peaks corresponding to the heaviest ions, m/z 167, 169 and 185, have relatively constant temporal profiles which also lack an accelerated increase upon  a second addition of ozone, a feature that is apparent among lighter product ions.Their unique time profiles imply that they are derived from a source secondary to ozonolysis, such as gas-phase accretion reactions, with modelling support for this provided in the Supplement (Sect.S4).Formation of prescribed products after the second ozonolysis is in agreement with the proposed degradation mechanism (Fig. 3), which predicts a number of both small and large species to form upon fragmentation of the carbon backbone.A large fraction of the smaller products come from decomposition of the three-carbon system (C 1 -C 2 -C 7 ; Fig. 1) bridging the conjugated double bonds in α-phellandrene, which segment from the rest of the molecule after the second ozone addition.For example, plausible mechanisms can be traced to methyl glyoxal formation irrespective of the order of addition of ozone to the two double bonds, with the only prerequisite being that the first addition of ozone adds one carbonyl group to the C 1 -C 2 -C 7 system.An example showing this from a proposed first-generation product is provided in Fig. 5. Subsequent decomposition of the C 1 -C 2 -C 7 Criegee biradical fragment can yield products including formaldehyde, formic acid and acetic acid.Meanwhile, functionalisation of the larger seven-carbon system bridging the conjugated bonds in α-phellandrene can give rise to a large number of heavier second-generation products.The m/z 129 is assigned to 2-propan-2-ylbutanedial, which can be formed from a number of pathways (e.g.Fig. 3).The m/z 115 is assigned to 2-propan-2-ylpropanedial, which is formed if a CI from either addition participates in the hydroperoxide channel, resulting in CHO fragmentation.Conversely, if instead of fragmentation, stabilisation occurs after a 1,5-hydrogen shift, then the product detected at m/z 143 shown in Fig. 3 may form.In all instances, detected second-generation products can be formed from a wide variety of predicted firstgeneration products independent of the order of addition of ozone to the two double bonds (Supplement Sect.S1).

Determination of OH yields
The OH radical scavenger, cyclohexane, reacts with OH to form both cyclohexanone and cyclohexanol (Atkinson et al., 1992;Berndt et al., 2003), with cyclohexanone (m/z 99) used as the OH radical tracer in this study.In a character-isation experiment, 98 and 85 % of cyclohexanone (Sigma Aldrich, 99.8 %) was found to reside at m/z 99 when the PTR-TOF drift tube was operated at 112 and 136 Td, respectively, with the remaining mass distributed over dehydrated and cluster peaks at m/z 81, 116 and 117.A minor ozonolysis product is also detected at m/z 99 (Sect.3.1.1);however, the two peaks are resolvable in the PTR-TOF.
A major uncertainty in determining OH yields is the yield of cyclohexanone formed from the reaction of cyclohexane with OH radicals.Atkinson et al. (1992) reported the combined yield of cyclohexanone and cyclohexanol to be 0.55 ± 0.09, with cyclohexanone/cyclohexanol ratios typically ranging from 0.8 to 1.4, depending on the terpene investigated.In contrast, Berndt et al. (2003) reported a cyclohexanone yield of 0.53 ± 0.06.In this study, the OH yield is based on the average of these two findings, with a cyclohexanone yield from the reaction of OH and cyclohexane of 0.41 ± 0.14 used.Scavenging is assumed to be 95 % efficient based on the volume of cyclohexane introduced into the reactor, with the error in this assumption thought to be minimal with respect to the inherent uncertainty in cyclohexanone yields.Background interference from cyclohexane, of which a small portion is oxidised by O + 2 to cyclohexanone in the drift tube, is corrected for (Winterhalter et al., 2009).
OH yields for the initial reaction of ozone with αphellandrene were calculated from the slope of OH produced against α-phellandrene reacted.As both m/z 99 and 137 are major signals in the PTR-TOF, spectra were not averaged during analysis, resulting in a 2 s time resolution.A characteristic OH-production time profile is shown in Fig. 6, which can be separated into three regions.The initial part of the experiment is characterised by a linear section, where α-phellandrene is the primary source of OH radicals.The gradient obtained from linear regression in this regime is equivalent to the OH yield from ozonolysis of the first double bond in α-phellandrene (Fig. 7a).The αphellandrene-dominated regime is short-lived with respect to total OH production time in the reactor, suggesting that first-generation products are also highly reactive and large producers of OH radicals.As the reaction proceeds, fasterreacting first-generation products begin to contribute to the OH budget, whilst α-phellandrene becomes increasingly less Atmos.Chem.Phys., 17, 6583-6609, 2017 www.atmos-chem-phys.net/17/6583/2017/influential.This results in a gradual curve, until essentially all α-phellandrene has been consumed and a vertical path is traced, indicating that first-generation species are now dominating OH radical production.By plotting OH formation against O 3 consumption in the product-dominated regime, the collective first-generation product OH radical yield is obtained (Fig. 7b).Naturally, this method is only applicable to those experiments where the product-dominated regime is attained, which is why no OH radical yields are reported for first-generation products in experiments 8 and 10.OH yields from the reaction of α-phellandrene and its first-generation degradation products with ozone are listed in Table 5.The average OH yield for the reaction of the first double bond in α-phellandrene across the 10 experiments was found to be 35 ± 12 %, whilst the average OH yield from the ozonolysis of the second reacting double bond was 15 ± 7 %.Both these determined values are slightly higher than the values calculated in Herrmann et al. (2010), although they agree 95 % of α-phellandrene reacts before first-generation products start to be consumed.However, as Fig. 6 shows, firstgeneration products contribute to the OH radical budget considerably earlier than this.As a result, Herrmann et al. ( 2010) is likely to have underpredicted OH yields of first-generation products, inadvertently apportioning their contribution to αphellandrene, whose OH yields would subsequently be overpredicted.The method employed in this study is thought to provide a more accurate distinction between OH radical production from α-phellandrene and its first-generation products.From the determined yields, it can be concluded that the hydroperoxide channel does play an important role in the decomposition of α-phellandrene by ozone, supportive of findings from a recent theoretical study (Mackenzie-Rae et al., 2016).

Modelling rate constants and OH yields
The conjugated system in α-phellandrene provides two reactive sites for ozone addition.Based on analogy with rate constants from simpler alkenes, such as cyclohexene (k = 8.1 × 10 −17 cm 3 molecule −1 s −1 ) and 1-methyl-1cyclohexene (k = 1.66×10 −16 cm 3 molecule −1 s −1 ) (Calvert et al., 2000), inductive effects are expected to make the methyl-substituted double bond the more reactive addition site.However, recent theoretical results suggest the contrary (Mackenzie-Rae et al., 2016), with steric effects raising the energy barrier for entry to the more substituted double bond, resulting in addition to the less substituted double bond in αphellandrene being favoured.This finding is consistent with experimental evidence for isoprene, where methacrolein, not methyl vinyl ketone, is the favoured first-generation product (Paulson et al., 1992;Grosjean et al., 1993;Aschmann and Atkinson, 1994;Rickard et al., 1999).Nevertheless, the average energy difference for addition to the two double bonds is minor, with both entry channels expected to be important.Given the high chemical reactivity of both double bonds in α-phellandrene, it is interesting to investigate the reactivity of first-generation products.The following reaction parameterisation was therefore constructed to determine the average rate constant of ozone with all first-generation species: where FG represents all first-generation products, SG represents all second-generation products, wO 3 is ozone lost to the reactor walls and x and y are stoichiometric coefficients representing OH yields from each reaction step.The rate constant for the reaction of ozone with α-phellandrene (k 1 ) was constrained to the literature value (Calvert et al., 2000), whilst a first-order ozone wall loss rate of k 3 = 2-8 ×10 −6 s −1 was used based on a number of calibration ex-periments.Remaining parameters, namely x, y and k 2 , were varied to optimise model performance.The reaction scheme was solved using the online numerical integrator AtChem (https://atchem.leeds.ac.uk/) for all experiments, barring 9 and 11 due to the unconstrained influence of NO 2 and/or OH radicals.
Figure 8 shows the results of the simulation of ozone consumption and OH production for three different experiments, with optimised parameters for each experiment given in Table 5. Considering simplicity, the model performs surprisingly well.Based on all experiments, the average simulated rate constant for the reaction of first-generation products with ozone was k 2 = 1.0 ± 0.7 × 10 −16 cm 3 molecule −1 s −1 .Although the ozonolysis of first-generation products is around 30 times slower than that of α-phellandrene, it is still faster than the ozonolysis of numerous monoterpenes including α-pinene, β-pinene, sabinene, 3-carene and β-phellandrene (Calvert et al., 2000).Using a typical background tropospheric ozone mixing ratio of 30 ppb, atmospheric lifetimes (τ ) of α-phellandrene and its first-generation products can be estimated by The atmospheric lifetime of α-phellandrene is therefore τ 1 ∼ 7.5 min, whilst the average lifetime of first-generation products is calculated to be τ 2 ∼ 3.75 h.Both α-phellandrene and its first-generation products therefore have a relatively short atmospheric lifetime with respect to ozone and are unlikely to be involved in long-range transport phenomena.Instead, complete saturation likely occurs in the chemical environment to which α-phellandrene is emitted, thus impacting the local radical, acid and SOA budgets.Interestingly, increasing the ozone concentration to conditions found in chamber experiments results in first-generation product lifetimes of the order of tens of minutes, which is more than sufficient for gas-phase detection.The inability to detect first-generation products is therefore indicative of an underlying sampling or detection issue.OH production is additionally included in the model to assist in parameterising yields from α-phellandrene and the average of its first-generation products.Experimental assessment of OH yields carries an inherent uncertainty, in that linear regression was used to fit data belonging to a segment of a curve (Fig. 5), with information in the "combination" section notionally discarded.Model parameterisation allows for a more complete description, although mechanistic simplicity renders the results far from quantitative.Instead, its purpose is to both validate experimental findings and allow further constraints to be placed on OH production from the α-phellandrene system.
culated experimentally, whilst yields from first-generation products are consistent with experimental measurements.Overall, net OH radical production is greater from the model than experimental measurements.The source of this discrepancy is likely the limited data used in calculating αphellandrene's OH yields, with a large proportion of αphellandrene consumed in the combination section.Given that the α-phellandrene-dominated regime generally lasted around 2-3 min, which is comparable to the mixing time of the reactor, it is entirely possible that OH production in this regime was not characterised well.The experimental finding for OH radical production from α-phellandrene of 35 ± 12 % is therefore recommended as a lower bound.

SOA formation
First-and second-generation ozonolysis products are highly functionalised polar species with high molecular weights.
It is therefore expected that they should make a significant contribution to the aerosol phase through gas-particle partitioning (Pankow, 1994;Odum et al., 1996).To assess this, the saturation vapour concentration (C * , µg m −3 ) (Donahue et al., 2006(Donahue et al., , 2012) ) of each species was calculated.Vapour pressures were estimated using the Extended Aerosol Inorganic Model (E-AIM) (Clegg et al., 2008), using the structure-based estimator of Nannoolal et al. (2004) for boiling points coupled with Moller et al. (2008) for vapour pressures.This method has been compared extensively with other estimation techniques (Barley and McFiggans, 2010;O'Meara et al., 2014).Activity coefficients were calculated using the UNIversal Functional Activity Coefficient (UNIFAC) method (Fredenslund et al., 1975).The saturation vapour concentrations calculated are shown in twodimensional volatility oxidation space in Fig. 9.
Gas-particle partitioning occurs in competition with gaswall partitioning, a process that is also dependent on species saturation vapour concentrations (Supplement Sect.S6).In parameterising gas-wall partitioning, the Teflon film is often considered to have an equivalent organic aerosol mass concentration (C w ).Values for C w vary significantly in the literature, with Ziemann and co-workers reporting values of C w ∼ 2-40 mg m −3 (Matsunaga and Ziemann, 2010;Yeh and Ziemann, 2015), Zhang et al. (2014) reporting C w values from 0.0004 to 300 mg m −3 and Krechmer et al. (2016) showing values of C w to vary with C * , from C w = 0.016 mg m −3 for C * < 1 up to 30 mg m −3 for C * > 10 4 .The reasons for the large discrepancies between studies are unknown; however, they are likely due to differing deformation and activities of the Teflon walls (Krechmer et al., 2016).Nonetheless, comparing reported values to SOA loadings generated during the chamber experiments reported in this work, it is evident that gas-wall partitioning is at least competitive, if not dominant, compared to gas-particle partitioning.The impact is shown in Fig. 9 by plotting the fraction of an organic species that remains in the gas phase over different saturation vapour concentrations using C w = 5 mg m −3 and an SOA loading of 200 µg m −3 .Under this scenario, gas-wall partitioning dominates, with compounds having C * < 10 2 µg m −3 predominantly residing in the walls with a small fraction in the aerosol phase after equilibrium is established, whereas species with C * > 10 6 µg m −3 remain almost entirely in the gas phase.Compounds with 10 2 < C * < 10 6 µg m −3 will partition to varying extents depending on their volatility and functional group composition between the wall, gas and particle phases (Krechmer et al., 2016).However, no corrections for gas-particle partitioning are made in the present study, given that no product vapour loss rate measurements were made for the GIG-CAS chamber and the large variability in literature values of C w .With-  out correcting for vapour wall losses, SOA yields are likely to be underestimated (Matsunaga and Ziemann, 2010;Zhang et al., 2014;La et al., 2016).
The majority of predicted first-generation and detected second-generation gas-phase products are classified as intermediate volatility compounds (IVOCs) (Donahue et al., 2012).As IVOCs, they are considered to have quite low vapour pressures but nonetheless reside almost exclusively in the gas phase.Of the proposed species, only the firstgeneration acids (e.g.Fig. 3) are classified as semi-volatile organic compounds (SVOCs), a classification given to those species which are expected to have sizeable mass fractions in the aerosol phase.Nevertheless, rapid aerosol formation is observed upon reaction of α-phellandrene and ozone as shown in Fig. 10, with sharp increases in particle number (dN/d log Dp) and volume (dV /d log Dp) concentrations observed.With no aerosol seed, nucleation must be driven by supersaturation of condensible species formed in the initial stages of the reaction.Donahue et al. (2013) argue that nucleation occurs through compounds that have extremely low volatility (ELVOC, C * < 3 × 10 −4 µg m −3 ).For the ozonolysis of other monoterpenes, ELVOC formation has been proposed to occur through gas-phase accretion reactions (Bateman et al., 2009;Heaton et al., 2009;Camredon et al., 2010) and autoxidation processes (Ehn et al., 2014;Jokinen et al., 2015).Meanwhile, to condense onto fresh aerosol but not homogeneously nucleate, vapours need to have saturation concentrations in the 10 −3 -10 −2 µg m −3 range (Donahue et al., 2011;Pierce et al., 2011), placing them in the lowvolatility organic compound bin.Formation of these compounds can be explained through conventional gas-phase chemistry (Donahue et al., 2011).It is therefore evident from Fig. 9 that the simple mechanistic overview provided to explain formation of gas-phase products in Sect.3.1.1and in Mackenzie-Rae et al. ( 2016) is insufficient to account for aerosol observations, with more complex reactions or reaction processes such as autoxidation, oligomerisation and/or heterogeneous oxidation required to develop species of sufficiently low vapour pressure for both particle nucleation and growth (Hallquist et al., 2009).
The maximum number of particles inside the reactor occur within the first few minutes of the reaction commencing (time resolution of the SMPS), with a small average particle diameter (∼ 40 nm).Rapid nucleation is consistent with the findings of Jokinen et al. (2015), who, based on limonene and α-pinene, concluded that endocyclic biogenic VOCs are efficient ELVOC producers upon ozonolysis.Coagulation of the newly formed aerosol decreases the number of particles, whilst further partitioning of low-volatility oxidation products increases the volume, with maximum aerosol concentration attained around 30 min into each experiment.After this point, irreversible wall losses supersedes gains from partitioning, with the volume, and hence mass, of aerosol decreasing inside the reactor.
In the early stages of experiments, the number concentration is a useful proxy for measuring the amount of nucleation occurring in the system (Bonn et al., 2002).As Fig. 11 shows, the addition of a Criegee scavenger systematically reduces initial particle number concentrations, concurrent with a shift of SOA to larger diameters.These changes suggest a reduction in the number of SOA-nucleating agents, implying that the reaction of α-phellandrene sCIs is important in forming ELVOC and IVOC compounds, whilst ruling out the reaction of sCIs with formic acid as a nucleating mechanism.This finding is consistent with experimental literature that is now building around sCIs as a source of new particle formation, whether through intramolecular SOZ formation (Bonn et al., 2002) ucts (Bateman et al., 2009) or oligomer formation through reaction with peroxy radicals (Sadezky et al., 2006(Sadezky et al., , 2008) ) or hydroperoxides (Sakamoto et al., 2013).Processes such as these would all be precluded by the addition of formic acid to the system.Similarly, there is a reduction in the αphellandrene normalised number distribution when NO 2 is added (Fig. 11).Like formic acid, NO 2 can also react with sCIs (Johnson and Marston, 2008) and therefore potentially inhibit particle formation and growth.If this were the case, then results from this ozonolysis study likely represent an upper limit to SOA formation under ambient conditions, although more experiments are necessary to confirm the impact of NO 2 on SOA formation in the α-phellandrene system.Assuming spherical particles, effective aerosol densities were calculated by comparing distributions of vacuum aerodynamic and electric mobility diameters, using the AMS and SMPS, respectively (DeCarlo et al., 2004;Katrib et al., 2005).Results are listed in Table .The average density across all experiments was found to be 1.49 ± 0.2 g cm −3 , indicating that the aerosol exists in a solid or waxy state (Kostenidou et al., 2007).This value is consistent with the SOA density found in the ozonolysis of other monoterpenes under similar conditions, which typically ranges from 1.15 to 1.73 g cm −3 (Bahreini et al., 2005;Kostenidou et al., 2007;Saathoff et al., 2009;Shilling et al., 2009).Because the particles are potentially non-spherical, the quoted effective density represents a lower bound of the true α-phellandrene SOA density, with error from assuming spherical particles expected to be less than 10 % (DeCarlo et al., 2004;Bahreini et al., 2005).It is noted that the densest aerosol was produced in experiment 11, which had NO 2 added, although one experiment is insufficient for reliable conclusions.The aerosol density was found to be insensitive to a range of other experimental parameters, including starting α-phellandrene and ozone concentrations, aerosol loading, aerosol oxidation state and the presence of CI scavengers.These findings are in contrast to studies conducted on α-pinene (Shilling et al., 2009) and β-caryophyllene (Chen et al., 2012), in which particle density was found to decrease as chamber aerosol loadings  increased, in accordance with changes in aerosol oxidation state.
Aerosol densities were used to convert SMPS volume concentrations into mass loadings (µg m −3 ).Wall loss effects were corrected for by assuming a size-independent firstorder loss process (Pathak et al., 2007), by modelling data at the end of each experiment, after gas-aerosol partitioning had reached equilibrium.Calculated wall loss rate constants, which ranged from 0.32 to 0.79 h −1 , were then applied to correct mass loading data for respective experiments.This way, differences between individual chamber simulations are accounted for.Determined wall loss rates are consistent with those found for α-pinene ozonolysis in the chamber (Wang et al., 2014).
The same method was used to correct V-mode AMS data, with results given in the Supplement (Sect.S7).Clustering of points around the 1 : 1 line in Fig. S7.1 indicates general agreement between mass loadings calculated using the AMS and SMPS (Shilling et al., 2008).Nevertheless, densitycorrected SMPS data is preferred in this work, primarily be-  (Pathak et al., 2007).
cause the AMS is known to suffer from transmission losses caused by particles bouncing off the vaporiser and, to a lesser extent, shape-dependent collection losses whilst focusing the particle beam (Matthew et al., 2008;Slowik et al., 2004;Huffman et al., 2005).Whilst it is noted that the SMPS does not measure particles with diameters larger than 750 nm, as shown in Fig. 10, this shortcoming is expected to have minimal impact on reported yields in this work (Wiedensohler et al., 2012).Wall-loss-corrected mass loadings for each experiment are given in Table , along with fractional aerosol yields (Y ).The fractional aerosol yield is defined as the amount of organic particulate matter that is produced ( M o , µg m −3 ) for a given amount of precursor VOC reacted ( HC, µg m −3 ) (Odum et al., 1996) and provides a convenient way of assessing the bulk aerosol-forming potential of an individual VOC.Utilising the gas-particle partitioning framework, aerosol yield can be described as a function of organic aerosol mass concentration (Pankow, 1994;Odum et al., 1996): where α i is the stoichiometric factor and K om,i the temperature-dependent equilibrium-partitioning constant of product i.
A characteristic yield plot is given in Fig. 12. Whilst a large number of products are expected to contribute to the particle phase, SOA yield is best fit using the parameters α 1 = 1.2 ± 0.2 and K om,1 = 0.022 ± 0.02 m 3 µg −1 , with higher-order fits found to be superfluous.The fitted constants offer little physical insight, other than perhaps the average of all α and K om values but nonetheless can be used in regional and global modelling (Chung and Seinfeld, 2002;Tsigaridis and Kanakidou, 2003;Henze and Seinfeld, 2006;Jathar et al., 2016) (Odum et al., 1996).
Figure 12 shows that α-phellandrene produces a large amount of aerosol upon ozonolysis compared to other monoterpenes (Wang et al., 2014;Saathoff et al., 2009;von Hessberg et al., 2009).Formation of the necessary semivolatile organic compounds is likely driven by the presence of two highly reactive endocyclic double bonds, with functionalisation rather than fragmentation dominating for the first addition (Lee et al., 2006).Both experiments where a CI scavenger was added lie below the fitted yield curve, strengthening the argument for sCIs as a source of condensible products.Nevertheless, yields from the two experiments differ by almost a factor of 2 despite having similar starting conditions, with further experiments necessary to better quantify the impact of sCIs on yields.Cyclohexane has been shown to reduce SOA yields in ozonolysis experiments (Bonn et al., 2002;Keywood et al., 2004;Saathoff et al., 2009), although no such effects were observed in this study.
The second addition of ozone, in general, fragments the molecule but in doing so increases relative oxygen content.Thus, the relative contribution of first-and second-generation products to SOA is empirically difficult to predict. Figure 13 shows SOA mass as a function of α-phellandrene reacted, producing time-dependent aerosol growth curves.In all experiments where α-phellandrene was completely consumed, dominant vertical growth profiles are traced.This increase in aerosol mass after complete consumption of parent hydrocarbon is characteristic of compounds with more than one double bond (Ng et al., 2006) and suggests that, when formed, second-generation products make an important contribution to the total aerosol mass.It is therefore likely that a large number of second-generation species fall in the IVOC or SVOC category in Fig. 9.
Whilst concentrations of precursors are somewhat elevated in experiments compared to ambient conditions, results nonetheless show α-phellandrene ozonolysis products to be heavily involved in both particle nucleation and growth processes.In polluted environments (e.g.inner-city forests, consumer products), a high SOA yield results in a large fraction of α-phellandrene partitioning into the particle phase irrespective of gas-phase loadings.Meanwhile, a strong nucleation potential makes α-phellandrene ozonolysis a strong candidate to help explain the intense and frequent nocturnal nucleation events observed in Eucalypt forests (Lee et al., 2008;Suni et al., 2008), which are already believed to be caused by monoterpene oxidation products (Ortega et al., 2009(Ortega et al., , 2012)).Indeed, the reaction conditions used in these experiments better reflect this clean environment, where reactions of RO 2 with HO 2 and other RO 2 radicals dominate along with unimolecular rearrangements.Such conditions favour the formation of low-volatility compounds, with the highest SOA yields for monoterpenes found under low-NO x conditions (Presto et al., 2005;Ng et al., 2007;Capouet et al., 2008;Eddingsaas et al., 2012).Under these conditions, ozonolysis reactions remain important (Perraud et al., 2012;Zhao et al., 2015), which is conducive to autoxidation processes and therefore nascent SOA formation and growth due to enhanced propensity for intramolecular rearrangements (Ehn et al., 2014;Jokinen et al., 2015).SOA yields measured in experiment 11 are consistent, however, with the other ozonolysis experiments in this study (Fig. 12), suggesting that the impact of NO x on SOA yields during the www.atmos-chem-phys.net/17/6583/2017/Atmos.Chem.Phys., 17, 6583-6609, 2017 Experiment no.ozonolysis of α-phellandrene is limited, with sufficient condensible products still able to be produced (Draper et al., 2015).Nonetheless, the reduction in aerosol number concentration in the initial stages of experiment 11 does suggest that formation pathways of ELVOC species (i.e.oligomerisation, autoxidation) are suppressed by the inclusion of NO 2 (Perraud et al., 2012).Detailed modelling studies are required to establish the relative importance of α-phellandrene in different environments, although evidence suggests that it is likely a contributor to nucleation events and aerosol growth in regions where it is emitted.

SOA composition
Resolution in the W mode of the AMS is sufficient to unambiguously identify chemical formulae of detected ions (De-Carlo et al., 2006;Aiken et al., 2007).Ions are formed, however, using high-energy electron impact ionisation (70 eV), resulting in significant fragmentation.The complexity of aerosol produced, along with an unknown number of fragmentation pathways, including the possibility of charge migration and other internal rearrangements, makes it exceedingly difficult to obtain clear structural information about SOA constituents from the AMS.For this reason, filter samples were collected and analysed to identify SOA constituents, with results to be published in a companion paper.Nevertheless, the AMS remains useful for analysing bulk properties of the aerosol to gain further insight into the system.
Bulk elemental composition can be estimated by averaging ion contributions across the entire mass spectrum (Aiken et al., 2007).Raw measured atomic ratios are converted to estimated ratios using the calibration factors of Aiken et al. (2008), namely 0.91 ± 10 % for hydrogen-to-carbon (H / C), 0.75 ± 31 % for oxygen-to-carbon (O / C) and 0.96 ± 22 % for nitrogen-to-carbon ratios, respectively, thus accounting for chemical biases in fragmentation.
Figure 14 shows the typical temporal profile of aerosol composition observed over an experiment.Initial aerosol formation and growth is driven by highly oxygenated species; however, as the organic aerosol (OA) medium grows and less functionalised species begin to partition, the overall oxidation state rapidly decreases, as seen by a drop in O / C and respective rise in H / C ratios.Once gas-particle partitioning slows and aerosol-loss processes dominate, there is a shift in equilibrium, with the more volatile aerosol constituents evaporating back into the gaseous phase.It can therefore be concluded that many of the SOA products generated during the chamber ozonolysis of α-phellandrene in this study are semi-volatile (Donahue et al., 2012).Nitrogen-containing species were found to make little contribution to the aerosol formed in experiment 11, with an average N / C ≈ 0.002 dur-Atmos.Chem.Phys., 17, 6583-6609, 2017 www.atmos-chem-phys.net/17/6583/2017/ing the experiment.Nitrate and PAN functionality is believed to significantly reduce the vapour pressure of constituents (Capouet and Müller, 2006;Pankow and Asher, 2008), with the result implying a small gas-phase concentration.Nevertheless, there exists evidence that organic nitrate contribution to SOA may be kinetically driven, rather than volatility driven (Perraud et al., 2012).
The average oxidation state of carbon (OS c ) in aerosol comprising of carbon, hydrogen and oxygen was parameterised by Kroll et al. (2011) as Although the definition ignores the effects of peroxides, whose oxygen atoms carry an oxidation state of −1, it nonetheless serves as a useful metric for representing the degree of oxidation of organic species in complex aerosol mixtures.Figure 15 shows that OS c decreases from −0.61 to −1.00 as the particle loading increases from 21.5 to 658.1 µg m −3 , suggesting a strong link between mass loading and degree of functionalisation, consistent with the findings of Shilling et al. (2009) for the ozonolysis of α-pinene.
The fastest change in OS c is observed to occur at lower mass loadings.Calculated OS c classifies the aerosol formed throughout the campaign as semi-volatile oxygenated organic aerosol (SV-OOA) (Kroll et al., 2011), consistent with numerous monoterpene and O 3 chamber experiments (Bateman et al., 2009;Aiken et al., 2008;Shilling et al., 2009;Chhabra et al., 2010;Chen et al., 2011).SOA density predictions from elemental ratios using the parameterisation of Kuwata et al. (2012) show some agreement with measured values (Supplement Sect.S8).
A van Krevelen plot of the entire dataset is given in Fig. 16.The impact of CI scavengers, cyclohexane and NO 2 on OA in van Krevelen space is observed to be minor.The important parameter was found to be aerosol mass loadings, with changes resulting in vertical shifts consistent with a change in oxidation state.Ozonolysis reactions are unique, as oxygen can be added and condensible products formed with no loss (and possibly gain) of hydrogen.Because of this, generic functionalisation lines used to characterise reactions in van Krevelen space (Heald et al., 2010;Chhabra et al., 2011) are not applicable.
It is evident from Fig. 16 that the majority of predicted species have a lower O / C ratio compared to what is measured for the bulk of the aerosol.It is therefore unlikely that any of the detected gas-phase species are substantially contributing to the generated aerosol, which instead is dominated by more functionalised products.Whilst it is likely that species comprising the OA are also present in the gas phase, they exist below the detection threshold of, or are lost in detection by, the PTR-TOF.Indeed, the presence of a filter prior to the PTR-TOF inlet may hinder detection of less volatile species, as elevated levels of OA on the filter may coax species into partitioning (Turpin et al., 2000;Kirchstetter et al., 2001).The carbon mass balance for each experiment is shown in Fig. 17.It was calculated by summing the gas-phase yields of all product ions, assuming a carbon of 6 for unidentified products, with SOA yields, whose carbon content was determined from elemental ratios measured in each experiment.The carbon balance ranged from 25 to 131 %.General losses in the system, such as vapour losses to the Teflon walls, affect the ability to close the carbon mass balance for most experiments, with performance worse in those experiments with lower starting α-phellandrene concentrations due to an inability to detect minor gas-phase products.It is evident from Fig. 17 that, despite having lower yields, heavier gas-phase products make a larger contribution to the carbon mass balance than lighter species such as formaldehyde, glyoxal, formic acid and acetic acid, whose nominal yields are higher.Meanwhile, experiment 4 had a carbon mass balance exceeding 100 %, which is thought to be the result of an erroneously high SOA yield (Fig. 12).It is immediately obvious from the carbon mass balances that a large fraction of αphellandrene partitions into the aerosol phase upon ozonolysis, exemplifying the impact α-phellandrene can have on SOA formation and growth.Currently, the species comprising SOA generated from α-phellandrene ozonolysis remains unidentified; however, a complete analysis of filter samples collected during these experiments is underway, in preparation for a follow-on publication.

Conclusions
The reaction of α-phellandrene with ozone was studied in depth for the first time through 11 chamber experiments.In the gas phase, only signals with increasing temporal profiles were detected by the PTR-TOF, indicative of secondgeneration products.Of these, small species (≤ C 3 ) were found to be produced in the highest yields, namely formaldehyde (5-9 %), acetaldehyde (0.2-8 %), glyoxal (6-23 %), www.atmos-chem-phys.net/17/6583/2017/Atmos.Chem.Phys., 17, 6583-6609, 2017 methyl glyoxal (2-9 %), formic acid (22-37 %) and acetic acid (9-22 %), with yields of all products suppressed by the addition of NO 2 .Despite having lower yields, heavier second-generation products were found to make a larger contribution to the carbon mass balance.A small number of second-generation products were tentatively identified based on a constructed gas-phase mechanism, including 2-propan-2-ylpropanedial and 2-propan-2-ylbutanedial. Experimental OH-radical yields of 35 ± 12 % and 15 ± 7 % for α-phellandrene and its first-generation products are in good agreement with those reported in Herrmann et al. (2010) and show the hydroperoxide channel to be an important pathway, with model output from a simple reaction pa-rameterisation suggesting experimental yields to be a lower bound.Meanwhile, modelling provides a rate coefficient of 1.0 ± 0.7 ×10 −16 cm 3 molecule −1 s −1 for the average reaction of first-generation products with ozone at 298 K.This equates to an atmospheric lifetime of around 3.75 h, higher than many other monoterpenes, and suggests that complete saturation of α-phellandrene likely occurs in the environment to which it is emitted.
α-phellandrene was found to form a large amount of aerosol upon reacting with ozone.A homogeneous nucleation burst of fresh aerosol was observed in all experiments within the first few minutes of the reaction, indicating a rapid formation of ELVOC species.Addition of a CI scavenger inhibited nucleation, suggesting that sCIs are important precursors in forming compounds of low volatility in the system.The mechanism behind this remains unknown, although numerous pathways have been proposed in the literature for CIs from other alkenes with more experiments required.Addition of NO 2 was found to reduce initial nucleation, although overall yields remained the same.The average effective SOA density was determined to be 1.49 ± 0.2 g cm −3 with an oxidation state varying from 0.56 to 1.02 depending on mass loadings.SOA growth curves show both first-and secondgeneration species contribute to the particulate phase, driving aerosol growth through to completion of the reaction.SOA yield is best fit by a one-product model with α 1 = 1.2 ± 0.2 and K om,1 = 0.022 ± 0.02 m 3 µg −1 , with the SOA-forming potential from α-phellandrene ozonolysis greater than other monoterpenes previously investigated in the literature.
Atmos.Chem. Phys., 17, 6583-6609, 2017 www.atmos-chem-phys.net/17/6583/2017/High radical, acid and SOA yields, coupled with a high reactivity, result in α-phellandrene having an immediate and significant impact on its local environment.Indeed, it appears likely that ozonolysis of α-phellandrene contributes to the significant blue haze and intense and frequent nocturnal nucleation events observed over Eucalypt forests.Characterisation and parameterisation of both the gaseous and particle phases formed from the ozonolysis of α-phellandrene therefore better our understanding of the impact of biogenic emissions and begin to enable the inclusion of this potentially important monoterpene in future atmospheric models.

Figure 1 .
Figure1.Simplified mechanism showing reaction processes involved during ozone addition to α-phellandrene within conventional frameworks (adapted fromMackenzie-Rae et al., 2016).Carbon labels on α-phellandrene are referred to in the main text.

Figure 2 .
Figure 2. Time profiles of major species detected using the PTR-TOF during the ozonolysis of α-phellandrene in experiment 5.The peak of α-phellandrene observed upon its addition was the result of the reactor fans being switched on immediately prior to the introduction of acetonitrile in this experiment.

Figure 3 .
Figure 3. Partial mechanism for the ozonolysis of α-phellandrene starting from CI3, yielding product masses detected by the PTR-TOF.Similar constructs for the remaining CIs are provided in the Supplement (Sect.S1).

Figure 4 .
Figure 4. Determination of gas-phase product yields in experiment 5.

Figure 5 .
Figure 5. Mechanism of O 3 addition to the proposed m/z 169 structures, yielding pairs of Criegee intermediates and carbonyl-containing products.

Figure 6 .
Figure 6.OH radical production versus α-phellandrene consumption for the first 18 min of experiment 3.

Figure 7 .
Figure 7. OH production from the (a) first and (b) second addition of ozone to α-phellandrene in experiment 3 against α-phellandrene and ozone consumption, respectively.

Figure 8 .
Figure 8. Plot of consumption of ozone and OH production against reaction time for experiments 2 (blue), 3 (yellow) and 5 (green).Experimental data are represented by open circles for O 3 and open diamonds for OH, whilst solid lines are modelled results using parameters listed in Table5.

Figure 9 .
Figure 9. Dots show predicted first-generation and detected second-generation products from the ozonolysis of α-phellandrene in Donahue et al. (2006) space.The grey line shows the fraction of species of different saturation vapour concentrations in the gas phase (F g ) after gas-wall and gas-particle equilibrium is reached, using C w = 5 mg m −3 and an SOA loading of 200 µg m −3 .Formulation of F g is given in the Supplement (Sect.S6).

Figure 11 .
Figure 11.Impact of sCI scavengers on particle nucleation shown by (a) peak particle number distributions scaled for the amount of α-phellandrene reacted in all experiments and (b) particle number distributions evolution over the first hour of experiments 1 and 6.

Figure 13 .
Figure 13.Time-dependent SOA growth curves.The grey line is a fitted one-parameter fit yield curve.

Figure 14 .
Figure 14.H / C and O / C ratios as a function of time for a typical α-phellandrene ozonolysis experiment (experiment 4).

Figure 15 .
Figure 15.Average oxidation state of carbon for increasing SOA loadings generated through α-phellandrene ozonolysis experiments, with the general trend shown.

Figure 16 .Figure 17 .
Figure 16.Van Krevelen plot.Blue dots are for experiments with a CI scavenger (6, 7), red dots are for the experiment without cyclohexane (9), yellow dots are for the experiment with NO 2 added (11) and green dots are for the remaining experiments.Both predicted and detected gas-phase species are shown with open black circles.
a All experiments had acetonitrile (2.5 µL) added as a dilution tracer.b 800 ± 80 ppb added prior to starting experiment.c 385 ± 5 ppb added prior to starting experiment.

Table 2 .
Identified ions detected by the PTR-TOF.Refer to Fig.3for product structures.

Table 5 .
Measured and modelled OH radical yields and modelled rate constants for α-phellandrene ozonolysis experiments.
* Wall-loss corrected . Comparison of SOA yield data for α-phellandrene with other monoterpene ozonolysis experiments.Lines are the best empirical model fits