The atmospheric impacts of monoterpene ozonolysis on global stabilised Criegee intermediate budgets and SO 2 oxidation: experiment, theory and modelling

. The gas-phase reaction of alkenes with ozone is known to produce stabilised Criegee intermediates (SCIs). These biradical/zwitterionic species have the potential to act as atmospheric oxidants for trace pollutants such as SO 2 , enhancing the formation of sulfate aerosol with impacts on air quality and health, radiative transfer and climate. How-ever, the importance of this chemistry is uncertain as a consequence of limited understanding of the abundance and atmospheric fate of SCIs. In this work we apply experimental, theoretical and numerical modelling methods to

Abstract. The gas-phase reaction of alkenes with ozone is known to produce stabilised Criegee intermediates (SCIs). These biradical/zwitterionic species have the potential to act as atmospheric oxidants for trace pollutants such as SO 2 , enhancing the formation of sulfate aerosol with impacts on air quality and health, radiative transfer and climate. However, the importance of this chemistry is uncertain as a consequence of limited understanding of the abundance and atmospheric fate of SCIs. In this work we apply experimental, theoretical and numerical modelling methods to quantify the atmospheric impacts, abundance and fate of the structurally diverse SCIs derived from the ozonolysis of monoterpenes, the second most abundant group of unsaturated hydrocarbons in the atmosphere. We have investigated the removal of SO 2 by SCIs formed from the ozonolysis of three atmospherically important monoterpenes (α-pinene, β-pinene and limonene) in the presence of varying amounts of water vapour in largescale simulation chamber experiments that are representative of boundary layer conditions. The SO 2 removal displays a clear dependence on water vapour concentration, but this dependence is not linear across the range of [H 2 O] [H 2 O] this dependence becomes much weaker. This is interpreted as being caused by the production of a variety of structurally (and hence chemically) different SCIs in each of the systems studied, which dis-played different rates of reaction with water and of unimolecular rearrangement or decomposition. The determined rate constants, k(SCI+H 2 O), for those SCIs that react primarily with H 2 O range from 4 to 310 × 10 −15 cm 3 s −1 . For those SCIs that predominantly react unimolecularly, determined rates range from 130 to 240 s −1 . These values are in line with previous results for the (analogous) stereo-specific SCI system of syn-/anti-CH 3 CHOO. The experimental results are interpreted through theoretical studies of the SCI unimolecular reactions and bimolecular reactions with H 2 O, characterised for α-pinene and β-pinene at the M06-2X/aug-cc-pVTZ level of theory. The theoretically derived rates agree with the experimental results within the uncertainties. A global modelling study, applying the experimental results within the GEOS-Chem chemical transport model, suggests that > 97 % of the total monoterpene-derived global SCI burden is comprised of SCIs with a structure that determines that they react slowly with water and that their atmospheric fate is dominated by unimolecular reactions. Seasonally averaged boundary layer concentrations of monoterpene-derived SCIs reach up to 1.4 × 10 4 cm −3 in regions of elevated monoterpene emissions in the tropics. Reactions of monoterpenederived SCIs with SO 2 account for < 1 % globally but may account for up to 60 % of the gas-phase SO 2 removal over areas of tropical forests, with significant localised impacts on

Introduction
Chemical oxidation processes in the atmosphere exert a major influence on atmospheric composition, leading to the removal of primary emitted species and the formation of secondary products. In many cases either the emitted species or their oxidation products negatively impact air quality and climate (e.g. ozone, which is also a potent greenhouse gas). These reactions can also transform gas-phase species to the condensed phase, forming secondary aerosol that again can be harmful to health and can both directly and indirectly influence radiative transfer and hence climate (e.g. SO 2 oxidation leading to the formation of sulfate aerosol).
Tropospheric gas-phase oxidants include the OH radical, ozone, the NO 3 radical and halogen atoms. Stabilised Criegee intermediates (SCIs), or carbonyl oxides, have been identified as another potentially important oxidant in the troposphere (e.g. Cox and Penkett, 1971;Mauldin III et al., 2012). SCIs are thought to be formed in the atmosphere predominantly from the reaction of ozone with unsaturated hydrocarbons, though other processes may be important under certain conditions, e.g. alkyl iodide photolysis (Gravestock et al., 2010), dissociation of the DMSO peroxy radical (Asatryan and Bozzelli, 2008). Laboratory experiments and theoretical calculations have shown SCIs to oxidise SO 2 (e.g. Cox and Penkett, 1971;Welz et al., 2012;Taatjes et al., 2013), organic (Welz et al., 2014) and inorganic (Foreman et al., 2016) acids (Vereecken, 2017) and a number of other important trace gases found in the atmosphere, as well as forming adducts with NO 2 Vereecken and Nguyen, 2017;Caravan et al., 2017). Measurements in a boreal forest (Mauldin III et al., 2012) and at a coastal site (Berresheim et al., 2014) have both identified a missing process (in addition to a reaction with OH) that oxidises SO 2 to H 2 SO 4 , potentially arising from SCI reactions.
Here, we present results from a series of experimental studies into SCI formation and reactions, carried out under atmospheric boundary layer conditions in the European Photochemical Reactor facility (EUPHORE), Valencia, Spain. We examine the ozonolysis of three monoterpenes with very different structures (and hence reactivities with OH and ozone): α-pinene (with an endocyclic double bond), βpinene (with an exocyclic double bond) and limonene (with both an endo-and exocyclic double bond). We observe the removal of SO 2 in the presence of each alkene-ozone system as a function of water vapour concentration. This allows us to derive relative SCI kinetics for reaction with H 2 O, SO 2 and unimolecular decomposition. Further, we calculate absolute unimolecular rates and bimolecular reaction rates with H 2 O for all α-pineneand β-pinene-derived SCIs at the M06-Scheme 1. Simplified generic mechanism for the reaction of Criegee intermediates (CIs) formed from alkene ozonolysis.
2X/aug-cc-pVTZ level of theory. A global modelling study, using the GEOS-Chem global chemical transport model, is performed to assess global and regional impacts of the chemical kinetics of monoterpene SCIs determined in this study.

Stabilised Criegee intermediate kinetics
Ozonolysis of an unsaturated hydrocarbon produces a primary ozonide that rapidly decomposes to yield pairs of Criegee intermediates (CIs) and carbonyls (Johnson and Marston, 2008). The population of CIs are formed with a broad internal energy distribution, giving both chemically activated and stabilised forms. Chemically activated CIs may undergo collisional stabilisation to an SCI, unimolecular decomposition or isomerisation. SCIs can have sufficiently long lifetimes in which to undergo bimolecular reactions (Scheme 1).
The predominant atmospheric fate for the simplest SCI, CH 2 OO, is reaction with water vapour, which likely occurs with the dimer ((H 2 O) 2 ) (e.g. Berndt et al., 2014;Newland et al., 2015a;Chao et al., 2015;Lewis et al., 2015;Lin et al., 2016a). For larger SCI, both experimental (Taatjes et al., 2013;Sheps et al., 2014;Newland et al., 2015a; and theoretical (Kuwata et al., 2010;Anglada et al., 2011;Anglada and Sole, 2016; studies have shown that their kinetics, in particular reaction with water, are highly structure dependent. The significant double-bond character exhibited in the zwitterionic configurations of mono-substituted SCIs leads to two distinct chemical forms: syn-SCIs (i.e. those where an alkyl substituent group is on the same side as the terminal oxygen of the carbonyl oxide moiety) and anti-SCIs (i.e. with the terminal oxygen of the carbonyl oxide moiety on the same side as a hydrogen group). The two conformers of CH 3 CHOO, which are both mono-substituted, display these properties. This difference in conformer reactivities has been predicted theoretically (Ryzhkov and Ariya, 2004;Kuwata et al., 2010;Anglada et al., 2011;Lin et al., 2016a) and was subsequently confirmed experimentally (Taatjes et al., 2013;Sheps et al., 2014) for the two CH 3 CHOO conformers. The significantly faster reaction of anti-CH 3 CHOO with water is driven by the higher potential energy of this isomer, while more stable SCI, with a methyl group in syn-position, such as syn-CH 3 CHOO or (CH 3 ) 2 COO, react orders of magnitude more slowly with water.
To date, the effects of the water dimer, (H 2 O) 2 on SCI removal have only been determined experimentally for CH 2 OO Chao et al., 2015;Lewis et al., 2015;Newland et al., 2015a;Sheps et al., 2017;Liu et al., 2017) and anti-CH 3 CHOO (Lin et al., 2016b). Theoretical calculations  have predicted the ratio of the SCI + (H 2 O) 2 : SCI + H 2 O rate constants, k 5 /k 3 , of larger and more substituted SCI to be of a similar order of magnitude as for CH 2 OO (i.e. 1.5-2.5 × 10 3 ).
SCIs can also undergo unimolecular isomerisation or decomposition in competition with bimolecular reactions. This is likely to be a significant atmospheric sink for syn-SCIs because of their slow reaction with water vapour (e.g. . Unimolecular reactions of syn-CI/-SCIs are dominated by a 1,4-H shift, forming a vinyl hydroperoxide (VHP) intermediate (Niki et al., 1987;Rickard et al., 1999;Martinez and Herron, 1987;Johnson and Marston, 2008;Kidwell et al., 2016). Decomposition of the VHP formed in this process is an important non-photolytic source of OH, HO 2 and RO 2 in the atmosphere (Niki et al., 1987;Alam et al., 2013;Kidwell et al., 2016), which can also lead to secondary organic aerosol formation . Unimolecular reactions of the anti-CI/SCIs are thought to be dominated by a 1,3ring closure, the acid ester channel, in which the CI/SCI decomposes, through rearrangement to a dioxirane intermediate, producing a range of daughter products and contributing to the observed overall HO x radical yield (Kroll et al., 2002;Johnson and Marston, 2008;Alam et al., 2013).

Monoterpene ozonolysis
Monoterpenes are volatile organic compounds (VOCs) with the general formula C 10 H 16 and are emitted by a wide range of vegetation, but particularly from boreal forests. Total global monoterpene emissions are estimated to be 95 (±3) Tg yr −1 (Sindelarova et al., 2014) -roughly 13 % of total non-methane biogenic VOC emissions. Monoterpene emissions are dominated by α-pinene, which accounts for roughly 34 % of the total global emissions, while βpinene and limonene account for 17 and 9 % respectively (Sindelarova et al., 2014). Monoterpenes (mainly α-pinene and limonene) are also present in indoor environments, in significant amounts where cleaning products and air fresheners are in routine use (of the order of 100 s of ppbv) (e.g. Singer et al., 2006a, b;Sarwar and Corsi, 2007), and where their ozonolysis products can affect indoor chemistry and health (e.g. Rossignol et al., 2013;Shallcross et al., 2014). Monoterpenes are highly reactive due to the presence of (often multiple) double bonds. The oxidation of monoterpenes yields a wide range of multi-functional gas-phase and aerosol products. This process can be initiated by OH and NO 3 radicals or by O 3 , with ozonolysis having been shown to be particularly efficient at generating low-volatility products that can form SOA, even in the absence of sulfuric acid (e.g. Ehn et al., 2014;Kirkby et al., 2016). These highly oxygenated secondary products have received considerable attention in recent years because of their role in affecting the climate through absorption and scattering of solar radiation (the direct aerosol effect). They can also increase cloud condensation nuclei concentrations, which can change cloud properties and lifetimes (the indirect aerosol effect). They have also been shown to have a wide range of deleterious effects on human health (e.g. Pöschl and Shiraiwa, 2015).
The ozonolysis reaction for monoterpenes is expected to follow a similar initial process to that of smaller alkenes, with cyclo-addition at a double bond giving a primary ozonide (POZ), followed by rapid decomposition of the POZ to yield a CI and a carbonyl (Scheme 1). Stabilisation of the large POZs formed in monoterpene ozonolysis is expected to be negligible (Nguyen et al., 2009a). However, a major difference in ozonolysis at endocyclic bonds is that, on decomposition of the POZ, the carbonyl oxide and carbonyl moieties are tethered as part of the same molecule, providing the potential for further interaction of the two. These can react together to form secondary ozonides (SOZ), which may be stable for several hours (Beck et al., 2011). However, while this has been shown to be potentially the major fate in the atmosphere for SCIs derived from sesquiterpenes (C 15 H 24 ) (e.g. Nguyen et al., 2009b;Beck et al., 2011;Yao et al., 2014), formation of SOZ is predicted to be small for monoterpene-derived SCIs because of the high ring strain caused by the tight cyclisation (e.g. Nguyen et al., 2009b). Chuong et al. (2004) predicted formation of a SOZ to become the dominant atmospheric fate for SCIs formed in the ozonolysis of endocyclic alkenes with a carbon number between 8 and 15, while Vereecken and Francisco (2012) suggested that internal SOZ formation is likely to be limited to product rings containing six or more carbons due to ring strain.
No studies have yet directly determined the reaction rates of the large SCIs produced from monoterpene ozonolysis with SO 2 (or any other trace gases). This is owing to the complexities of synthesising and measuring large SCI. However, Ahrens et al. (2014) concluded that the reaction of the C9-SCI formed in β-pinene ozonolysis with SO 2 is as fast as that determined by Welz et al. (2012) and Taatjes et al. (2013) for CH 2 OO and CH 3 CHOO respectively (ca. 4 × 10 −11 cm 3 s −1 ) by fitting to the decay of SO 2 in the presence of the ozonolysis reaction. Mauldin III et al. (2012) calculated significantly slower reaction rates for an additional oxidant (assumed to be SCI) derived from α-pinene and limonene ozonolysis, with k(SCI + SO 2 ) determined to be 6 × 10 −13 and 8 × 10 −13 cm 3 s −1 for α-pinene and limonene-derived SCIs respectively. However, it seems likely that the rates calculated by Mauldin III et al. (2012) may be substantially underestimated due to the assumption of a very long SCI lifetime (0.2 s) in experiments that were performed at 50 % RH. The calculated rates scale linearly with SCI lifetime, and based on reaction rates of smaller SCIs with H 2 O (reported since the Mauldin III et al. work, e.g. Taatjes et al., 2013), it seems likely that the lifetime of the SCI in their experiments would have been more like 0.1-2 × 10 −2 s, increasing the calculated rate constants by more than an order of magnitude and bringing them into much closer agreement with the rates reported by Ahrens et al. (2014).
Unimolecular reactions of the monoterpene SCIs are expected to proceed rapidly through the VHP route if hydrogen is available for a 1,4-H shift. Those SCIs that cannot undergo this rearrangement may undergo unimolecular reactions via the formation of the dioxirane intermediate, but this is expected to be a much slower process (Nguyen et al., 2009a). In contrast to smaller SCI, it has been observed experimentally and predicted theoretically that the VHP route will mainly lead to rearrangement into an acid (also yielding an OH radical) rather than decomposition of the molecule (e.g. . As for the smaller alkenes, monoterpene ozonolysis has been shown to be a source of HO x (e.g. Paulson et al., 1997;Alam et al., 2013), predominantly via the VHP rearrangement. The MCMv3.3.1 (Jenkin et al., 2015) applies OH yields of 0.80, 0.35 and 0.87 for αpinene, β-pinene and limonene respectively.

α-pinene-derived SCI
Decomposition of the α-pinene POZ yields four different C 10 Criegee intermediates (Scheme 2: CI-1a, 1b, 2a, 2b), with the carbonyl oxide moiety at one end and a carbonyl group at the other. Here, CI-1 is a mono-substituted CI for which both syn-(1a) and anti-conformers (1b) exist while the other, CI-2, is disubstituted and two syn-conformers (2a and 2b) exist for it.  infer a relative yield of 50 % for the two basic CIs formed, based on the observation that norpinonic acid yields from the ozonolysis of α-pinene and an enone, which upon ozonolysis yields CI-1, are almost indistinguishable.
The total SCI yield from α-pinene was determined to be 0.15 (±0.07) by Sipilä et al. (2014) in indirect experiments measuring the production of H 2 SO 4 from SO 2 oxidation in the α-pinene ozonolysis system. Drozd and Donahue (2011) also determined a total SCI yield of about 0.15 at 740 Torr from measuring the loss of hydrofluoroacetone in ozonolysis experiments in a high-pressure flow system. The MCMv3.3.1 (Jenkin et al., 1997(Jenkin et al., , 2015Saunders et al., 2003) applies a value of 0.20 based on stabilisation of only the monosubstituted CI-1.

β-pinene-derived SCI
β-pinene ozonolysis yields two distinct conformers of the nopinone C9-CI (Scheme 3: CI-3 and CI-4), which differ in orientation of the carbonyl oxide group, and CH 2 OO. CI-3 and CI-4 are formed in roughly equal proportions with very little interconversion between the two (Nguyen et al., 2009a). The difference in the chemical behaviour of CI-3 and CI-4, which were often not distinguished in earlier studies, arises from the inability of the carbon attached to the 4-membered ring to undergo the 1,4-H shift that allows unimolecular decomposition via the VHP channel. This was noted in Rickard et al. (1999) as being a reason for the considerably lower OH yield (obtained via the VHP route) from β-pinene ozonolysis compared to that of α-pinene. This difference leads to contrasting unimolecular decomposition rates for the two CIs, with Nguyen et al. (2009a) predicting a loss rate of ca. 50 s −1 for CI-3 (via a VHP) and ca. 1 s −1 for CI-4 (via ring closure to a dioxirane). This result is qualitatively consistent with the experimental work of Ahrens et al. (2014), who deter-mine a ratio of 85 : 15 for the abundance of SCI-4 : SCI-3 about 10 s after the initiation of the ozonolysis reaction, as a consequence of the much faster decomposition rate of SCI-3. The potential for bimolecular reactions to compete with decomposition of SCI-3 is very different from the potential for bimolecular reactions to compete with decomposition of SCI-4. Nguyen et al. (2009a) theoretically calculate a total SCI yield from β-pinene ozonolysis of 42 %, consisting of 16.2 % SCI-3, 20.6 % SCI-4 and 5.1 % CH 2 OO. Ahrens et al. (2014) assume an equal yield of CI-3 and CI-4 (45 %) with a 10 % yield of CH 2 OO; 40 % of the total C9-CI are calculated to be stabilised at 1 atm. If all of the CH 2 OO is assumed to be stabilised when it is formed (e.g. Nguyen et al., 2009a), this gives a total SCI yield of 46 %. Earlier experimental studies have tended to determine lower total SCI yields, with Hasson et al. (2001) reporting a total SCI yield of 0.27 from measured product yields (almost entirely nopinone) and Hatakeyama et al. (1984) reporting a total SCI yield of 0.25. Winterhalter et al. (2000) determined a yield of 0.16 (±0.04) for excited CH 2 OO from β-pinene ozonolysis, obtained via the nopinone yield and 0.35 for the stabilised C9-CI, giving a total SCI yield of 0.51 of all the CH 2 OO is assumed to be stabilised. Also, experimental studies have tended to report higher CH 2 OO yields (determined from measured nopinone yields) than theoretical studies. Nguyen et al. (2009a) note that this could be because nopinone can also be formed in bimolecular reactions of SCI-4, hence experimental studies may overestimate CH 2 OO production. The MCMv3.3.1 incorporates a total SCI yield of 0.25 from β-pinene ozonolysis, with a yield of stabilised C9-CI of 0.102 and a CH 2 OO yield of 0.148.

Limonene-derived SCI
Limonene has two double bonds with which ozone can react. Theory suggests that reaction at the endocyclic bond is more likely: Baptista et al. (2011) calculate the reaction at the endocyclic bond to be 84-94 % (dependent on the level of theory applied). Zhang et al. (2006) suggest the reaction at the endocyclic double bond to be roughly 25 times faster than at the exocyclic bond, i.e. leading to a branching ratio of ca. 96 % reaction at the endobond and the current IUPAC recommendation (IUPAC, 2013) suggests about 95 % of the primary ozone reaction to be at the endobond. Leungsakul et al. (2005) reported a best fit to measurements from chamber experiments by assuming an 85 % reaction at the endocyclic bond and 15 % at the exocyclic bond.
2 Experimental study

Experimental approach
The EUPHORE facility is a 200 m 3 simulation chamber used primarily for studying reaction mechanisms under atmospheric boundary layer conditions. Further details of the chamber set-up and instrumentation are available elsewhere (Becker, 1996;Alam et al., 2011), and a detailed account of the experimental procedure, summarised below, is given in Newland et al. (2015a).
Experiments comprised time-resolved measurements of the removal of SO 2 in the presence of the monoterpene-Scheme 3. Mechanism of formation of the three Criegee intermediates (CIs) from β-pinene ozonolysis.
Scheme 4. Mechanism of formation of the four Criegee intermediates (CIs) from limonene ozonolysis.
ozone system as a function of humidity. SO 2 and O 3 abundance were measured using conventional fluorescence (reported precision ±1.0 ppbv) and UV absorption monitors (reported precision ±4.5 ppbv), respectively; alkene abun-dance was determined via FTIR spectroscopy. Experiments were performed in the dark (i.e. with the chamber housing closed; j (NO 2 ) ≤ 10 −6 s −1 ), at atmospheric pressure (ca. 1000 mbar) and temperatures between 287 and 302 K. The chamber is fitted with large horizontal and vertical fans to ensure rapid mixing (ca. 2 min). Chamber dilution was monitored via the first-order decay of an aliquot of SF 6 , added prior to each experiment. Cyclohexane (ca. 75 ppmv) was added at the beginning of each experiment to act as an OH scavenger, such that SO 2 reaction with OH was calculated to be ≤ 1 % of the total chemical SO 2 removal in all experiments.
The experimental procedure, starting with the chamber filled with clean scrubbed air, comprised addition of SF 6 and cyclohexane, followed by water vapour, O 3 (ca. 500 ppbv) and SO 2 (ca. 50 ppbv). A gap of 5 min was left prior to addition of the monoterpene to allow complete mixing. The reaction was then initiated by the addition of the monoterpene (ca. 400 ppbv for α-pinene and β-pinene, ca. 200 ppbv for limonene), and reagent concentrations followed for roughly 30-60 min; ca. 30-90 % of the monoterpene was consumed after this time, dependent on the reaction rate with ozone. Four α-pinene + O 3 , five β-pinene + O 3 and five limonene + O 3 experiments were performed in total, as a function of [H 2 O]. Each individual run was performed at a constant humidity, with humidity varied to cover the range of [H 2 O] = 0.1-19 × 10 16 molecules cm −3 , corresponding to an RH range of 0.1-28 % (at 298 K). Measured increases in [SO 2 ] agreed with measured volumetric additions across the SO 2 and humidity ranges used in the experiments (Newland et al., 2015a). The experimental raw data are avail-able at https://doi.org/10.15124/4e9cd832-9cce-41c8-8335-c88cf32fe244 .

Analysis
A range of different SCIs are produced from the ozonolysis of each of the three monoterpenes (see Schemes 2-4), each with their own distinct chemical behaviour (i.e. yields, reaction rates); it is therefore not feasible (from these experiments) to obtain data for each SCI independently. Consequently, for analytical purposes we necessarily treat the SCI population in a simplified (lumped) manner -see Sect. 2.2.2.
SCIs are assumed to be formed in the ozonolysis reaction with a yield ϕ (Reaction R1). They can then react with SO 2 , H 2 O, acids formed in the ozonolysis reaction or with other species present or they can undergo unimolecular decomposition under the experimental conditions applied (Reactions R2-R5). A fraction of the SCIs produced reacts with SO 2 . This fraction (f ) is the loss rate of the SCIs to SO 2 (k 2 [SO 2 ]) compared to the sum of the total loss processes for the SCIs (Eq. 1): Here, L accounts for the sum of any other chemical loss processes for SCIs in the chamber. With the exception of reaction with acids these loss processes are expected to be negligible, as discussed later. After correction for dilution and neglecting other (non-alkene) chemical sinks for O 3 , the following equation is derived. Corrections include reaction with HO 2 (also produced directly during alkene ozonolysis; Alam et al., 2013;Malkin et al., 2010), which was indicated through model calculations to account for < 0.5 % of ozone loss under all the experimental conditions.
From Eq.
(2), regression of the loss of ozone (dO 3 ) against the loss of SO 2 (dSO 2 ) for an experiment at a given RH determines the product f ϕ at a given point in time. This quantity will vary through the experiment as SO 2 is consumed and other potential SCIs co-reactants are produced, as predicted by Eq. (1). A smoothed fit was applied to the experimental data for the cumulative consumption of SO 2 and O 3 , SO 2 and O 3 , (as shown in Fig. 2) to determine dSO 2 /dO 3 (and hence f ϕ) at the start of each experiment, for use in Eq. (2). The start of each experiment (i.e. when [SO 2 ] ∼ 50 ppbv) was used, as this corresponds to the greatest rate of production of the SCI and hence the largest experimental signals (i.e. greatest O 3 and SO 2 rate of change; greatest precision), and is the point at which the SCI + SO 2 reaction has the greatest magnitude compared with any other potential loss processes for either reactant species (see discussion below).
Other potential fates for SCIs include reaction with ozone (Kjaergaard et al., 2013;Vereecken et al., 2014Vereecken et al., , 2015 (Su et al., 2014;Vereecken et al., 2014), carbonyl products , acids (Welz et al., 2014) or with the parent alkene Decker et al., 2017). Sensitivity analyses using the most recent theoretical predictions (Vereecken et al., 2015) indicate that the reaction with ozone is not significant under any of our experimental method, accounting for less than 1.5 % of SCI loss for anti-SCIs (based on anti-CH 3 CHOO) at the lowest RH (worst case) experiment. Generally, SCI loss to ozone is calculated to be < 1% for all SCI. Summed losses from reaction with SCI (self-reaction), carbonyls and alkenes are likewise calculated to account for < 1 % of the total SCI loss under the experimental conditions applied.
CH 2 OO and CH 3 CHOO have been shown to react rapidly (k = 1-5 × 10 −10 cm 3 s −1 ) with formic and acetic acid (Welz et al., 2014). In ozonolysis experiments, Sipilä Cumulative consumption of SO 2 as a function of cumulative consumption of O 3 and SO 2 versus O 3 for the ozonolysis of α-pinene, β-pinene and limonene in the presence of SO 2 at a range of water vapour concentrations, from 1 × 10 15 to 1.9 × 10 17 cm −3 . Symbols are experimental data, corrected for chamber dilution. Lines are smoothed fits to the experimental data. et al. (2014) determined the relative reaction rate of acetic and formic acids with (CH 3 ) 2 COO (i.e.k 5 /k 2 ) to be roughly three. Organic acid mixing ratios in this work, as measured by FTIR, reached up to a few hundred ppbv, suggesting these will likely be a significant SCI sink in our experiments. We have therefore explicitly included reaction with organic acids in our analysis, incorporating the uncertainty arising from the (unknown) acid reaction rate constant, as described in Sect. 2.2.1.
The water dimer reactions of non-CH 2 OO SCIs are not considered in our analysis. The effect of the water dimer reaction with C 10 and C 9 SCIs (rather than the monomer) is expected to be minor at the maximum [H 2 O] (2 × 10 17 cm −3 ) used in these experiments (< 30 % RH). Further, with analogy to the syn-/anti-CH 3 CHOO system, syn-SCI loss to the dimer (and monomer) will not become competitive at the highest [H 2 O] used here; for anti-SCI, the water monomer will already be removing the majority of the SCIs at the [H 2 O] at which the dimer would become a significant loss process. Hence the dimer reaction is deemed unimportant. For CH 2 OO, the reaction rates with water and the water dimer have been quantified in recent EUPHORE experimental studies, and the values from Newland et al. (2015a) are used in our analysis.
As noted above, a range of different SCIs are produced from the ozonolysis of the three monoterpenes (see Schemes 2-4), each with their own distinct chemical behaviour, which, treated individually, introduce too many unknowns (i.e. yields, reaction rates) for explicit analysis. Consequently for analytical purposes we treat the SCI population in a simplified (lumped) manner: Firstly, we use the simplest model possible, assuming that a single SCI is formed in each ozonolysis reaction (Eq. 3). f In a second model, for each monoterpene, the SCIs produced are assumed to belong to one of two populations, denoted SCI-A and SCI-B. These two populations are split according to the observation that the decomposition rates and reaction rates with water for the smaller SCIs (CH 3 CHOO) have been predicted theoretically (Ryzhkov and Ariya, 2004;Kuwata et al., 2010;Anglada et al., 2011) and shown experimentally (Taatjes et al., 2013;Sheps et al., 2014;Newland et al., 2015a) to exhibit a strong dependence on the structure of the molecule. The syn-CH 3 CHOO conformer, which has the terminal oxygen of the carbonyl oxide moiety in the syn-position to the methyl group, has been shown to react very slowly with water and to readily decompose, via the hydroperoxide mechanism, whereas the anti-CH 3 CHOO conformer, with the terminal oxygen of the carbonyl oxide moiety in the anti-position to the methyl group, has been shown to react fast with water and is not able to decompose via the hydroperoxide mechanism. Vereecken and Francisco (2012) have shown that all SCIs studied theoretically with an alkyl group in the syn-position have reaction rates with H 2 O of k < 4 × 10 −17 molecule cm 3 s −1 (and for SCIs larger than acetone oxide, k < 8 × 10 −18 molecule cm 3 s −1 ).
We thus define two populations, assuming SCI-A (i.e. SCIs that exhibit chemical properties of the anti-type SCI) to react fast with water and not to undergo unimolecular reactions and SCI-B (i.e. SCIs that exhibit chemical properties of the syn-type SCI) to not react with water but to undergo unimolecular reactions. This simplification allows us to fit to the measurements using Eqs. (4) and (5), as shown below. The total SCI yields are determined by our experiments at high SO 2 , and the relative yields of SCI-A and SCI-B are determined from fitting to Eq. (5). These relative yields are then compared to those predicted from the literature. In where γ is the fraction of the total SCI yield (i.e. γ A + γ B = 1). f A and f B are the fractional losses of SCI-A and SCI-B to a reaction with SO 2 . Adapting Eq. (1) to include the two SCI species gives Eq. (4), where k 5 [acid] accounts for the SCI + acid reaction (see discussion of reaction rate constants below).
Equation (4) can be rearranged to Eq. (5) and fitted accord- Using values for γ A and γ B from the literature and varying the assumed values of the reaction of SCIs with acid (k 5 ) allows us to determine k 3 /k A 2 and k d /k B 2 . The assumptions made here allow analysis of a very complex system. However, a key consequence is that the relative rate constants obtained from the analysis presented here are not representative of the elementary reactions of any single specific SCI isomer formed, but rather represent a quantitative ensemble description of the integrated system, under atmospheric boundary layer conditions, which may be appropriate for atmospheric modelling. Additionally our experimental approach cannot determine absolute rate constants (i.e. values of k 2 , k 3 , k d ) in isolation and is limited to assessing their relative values, measured under atmospheric conditions, which may be placed on an absolute basis through use of an external reference value (here the SCI + SO 2 rate constant).

SCI yield calculation
The value for the total SCI yield of each monoterpene, ϕ SCI-TOT , was determined from an experiment performed under dry conditions (RH < 1%) in the presence of excess-SO 2 (ca. 1000 ppbv), such that SO 2 scavenged the majority of the SCI. From Eq. (2), regressing dSO 2 against dO 3 (corrected for chamber dilution), assuming f to be unity (i.e. all the SCIs produced reacts with SO 2 ), determines the value of ϕ min , a lower limit to the SCI yield. Figure 1 shows the experimental data, from which ϕ min was derived.
In reality f will be less than one at experimentally accessible SO 2 levels, as a fraction of the SCIs may still react with trace H 2 O present or undergo unimolecular reaction. The actual yield, ϕ SCI , was determined by combining the result from the excess-SO 2 experiment with those from the series of experiments performed at lower SO 2 , as a function of [H 2 O], to obtain k 3 /k 2 and k d /k 2 (see Sect. 2.2.1), through an iterative process to determine the single unique value of ϕ SCI which fits both data sets, as described in Newland et al. (2015a), but taking into account the proposed model in this paper of there being two SCIs produced. In , other possible SCIs sinks are assumed to be negligible. In these excess-SO 2 experiments, f A ∼ 1 but f B < 1 since k d still represents a significant sink.
γ A (and hence γ B , since γ B = 1 − γ A ) is derived from fitting Eq. (4) to the data from the experiments performed at lower SO 2 for a given ϕ. Using a range of ϕ gives a range of γ . These different values of γ are used with the respective values of ϕ in fitting to Eq. (4) to determine values of k 3 /k 2 and k d /k 2 .

Experimental uncertainties
The uncertainty in k 3 /k 2 was calculated by combining the mean relative errors from the precision associated with the SO 2 and ozone measurements (given in Sect. 2.1) with the 2σ error and the relative error in ϕ, using the root of the sum of the squares of these four sources of error. The uncertainty in k d /k 2 was calculated in the same way.
The uncertainty in ϕ min was calculated by combining the uncertainty in SO 2 and O 3 , as above. The uncertainty in ϕ was calculated by applying the k 3 /k 2 uncertainties and combining these with the uncertainties in ϕ min using the root of the sum of the squares.

Theoretical calculations
The rovibrational characteristics of all conformers of the CIs formed from α-pinene and β-pinene, the transition states for their unimolecular reaction and their reaction with H 2 O were characterised quantum chemically, first using the M06-2X/cc-pVDZ level of theory, and subsequently refined at the M06-2X/aug-cc-pVTZ level. To obtain the most accurate barrier heights for reaction, it has been shown (Berndt et al., 2015;Chhantyal-Pun et al., 2017;Fang et al., 2016a, b;Long et al., 2016;Nguyen et al., 2015) that post-CCSD(T) calculations are necessary. Performing these calculations for the SCIs discussed in this paper, with up to 14 non-hydrogen atoms, is well outside our computational resources. Instead, we base our predictions on high-level CCSD(T)/aug-cc-pVTZ single-point energy calculations, performed for the reactions of nopinone oxides and the most relevant subset of for α-pinene. From the fit we derive relative rate constants for reaction of the α-pinene-derived SCI, SCI-A and SCI-B with H 2 O (k 3 /k 2 ) and decomposition ((k d + L)/k 2 ) assuming that γ A = 0.40 and γ B = 0.60. pinonaldehyde oxides. These data are reliable for relative rate estimates, but it remains useful to further improve the absolute barrier height predictions, as described by Vereecken et al. (2017) based on a data set with a large number of systematic calculations on smaller CIs, allowing empirical corrections to estimate the post-CCSD(T) barrier heights. Briefly, they compare rate coefficient calculations against available harmonised experimental and very high-level theoretical kinetic rate predictions and adjusts the barrier heights by 0.4 to 2.6 kcal mol −1 (depending on the base methodology and the reaction type) to obtain the best agreement with these benchmark results.
Using the energetic and rovibrational data thus obtained, multi-conformer transition state theory (MC-TST) calculations (Truhlar et al., 1996;Vereecken and Peeters, 2003) were performed to obtain the rate coefficient at 298 K at the high-pressure limit. All rate predictions incorporate tunnelling corrections using an asymmetric Eckart barrier (Eckart, 1930;Johnston and Heicklen, 1962). For the reaction of CI + H 2 O, a pre-reactive complex is postulated at 7 kcal mol −1 below the free reactants, while the CI + (H 2 O) 2 reaction is taken to have a pre-reactive complex of 11 kcal mol −1 stability. This pre-reactive complex affects tunnelling corrections: it is assumed to always be in equilibrium with the free reactants.
In view of the high number of rotamers and the resulting computational cost, only a single limonene-derived CI isomer was studied, where the TS for the CI + H 2 O reaction was analysed at the M06-2X/cc-pVDZ level of theory with only a partial conformational analysis; a limited number of the energetically most stable TS conformers thus discovered were re-optimised at the M06-2X/aug-cc-pVTZ level of theory. These data will only be used for qualitative assessments. However, we apply the structure-activity relationships (SARs) presented by  to obtain an estimate of the rate coefficients and assess the role of the individual SCI isomers in limonene ozonolysis. All quantum chemical calculations were performed using Gaussian 09 (Frisch et al., 2010).

GEOS-Chem model simulation
The global chemical transport model GEOS-Chem (v9-02, www.geos-chem.org, last access: 27 April 2018, Bey et al., 2001) is used to explore the spatial and temporal variability of the atmospheric impacts of the experimentally derived chemistry. The model includes HOx-NOx-VOC-O 3 -BrOx chemistry (Mao et al., 2010;Parrella et al., 2012) and a mass-based aerosol scheme. Biogenic monoterpene emissions are taken from the Model of Emissions of Gases and Aerosols from Nature (MEGAN) v2.1 inventory (Guenther et al., 2006(Guenther et al., , 2012. Transport is driven by assimilated meteorology (GEOS-5) from NASA's Global Modelling and Assimilation Office (GMAO). The model is run at 4 • × 5 • resolution, with the second year (2005) used for analysis and first year discarded as spin-up. The model code used for these runs is available at https://doi.org/10.5281/zenodo.1220385 and the model run directory at https://doi.org/10.5281/zenodo.1220387 (GEOS-Chem team v9-02, 2018).
In this study, the standard simulation was expanded to include emissions of seven monoterpene species (αpinene, β-pinene, limonene, myrcene, ocimene, carene and sabinene) from MEGAN v2.1. The ozonolysis scheme for  Supplement Table S1. Figure 1 shows the lower limit to the SCI yield, ϕ min , for the three monoterpenes, determined from fitting Eq. (5) to the experimental data. This gives values of 0.16 (±0.01) for α-pinene, 0.53 (±0.01) for β-pinene and 0.20 (±0.01) for limonene. These ϕ min values were then corrected as described in Sect. 2.2.2 using the k 3 /k 2 and k d /k 2 values determined from the measurements shown in Figs. 3-5 using Eq. (4). The corrected yields, ϕ SCI , are 0.19 (±0.01) for α-pinene, 0.60 (±0.03) for β-pinene and 0.23 (±0.01) for limonene. Uncertainties are ±2σ and represent the combined systematic (estimated measurement uncertainty) and precision components. Literature yields for SCI production from monoterpene ozonolysis are summarised in Table 1. The value derived for the total SCI yield from α-pinene in this work of 0.19 agrees within the uncertainties, with a value of 0.15 (±0.07) reported by Sipilä et al. (2014) and a value of 0.20 applied in the MCMv3.3.1.

SCI yield
The total SCI yield from β-pinene derived in this work, 0.60, agrees reasonably well with the recent experimental work of Ahrens et al. (2014), who derived a total SCI yield of 0.50 (0.40 for the sum of CI-1 and CI-2 and 0.10 for CH 2 OO, which is assumed to be formed almost completely Figure 6. Variation of k 3 /k 2 (k(SCI-A+H 2 O)) / (k(SCI-A + SO 2 ) and k d (k(SCI-B unimol.)) / (k(SCI-B+SO 2 ) as a function of the ratio k 5 /k 2 (k(SCI + acid) / k(SCI + SO 2 )), derived from least squares fit of Eq. (4) to measurements shown in Figs. 3-5 for αpinene, β-pinene and limonene respectively. stabilised). The MCMv3.3.1 applies a total SCI yield of 0.25, of which 0.10 is a C9-CI and 0.15 is CH 2 OO. Earlier studies also tended to derive lower total SCI yields ranging from 0.25 to 0.27 (Hasson et al., 2001;Hatakeyama et al., 1984).
The total SCI yield from limonene derived in this work 5.2 k 3 (SCI + H 2 O) / k 2 (SCI + SO 2 ) and k d /k 2 (SCI + SO 2 ) analysis Figure 2 shows the loss of SO 2 as ozone is consumed by reaction with the monoterpene for each of the three systems. Box modelling results suggest that > 99 % of this SO 2 removal is caused by reaction with SCIs produced in the alkene-ozone reaction (rather than e.g. reaction with OH, which is scavenged by cyclohexane). When the experiments are repeated at higher relative humidity, the rate of loss of SO 2 decreases.   This is as expected from Eq. (1) and suggests that there is competition between SO 2 and H 2 O for reaction with the SCIs produced, in agreement with observations of smaller SCI, which demonstrate the same competition under atmospherically relevant conditions (Newland et al., 2015a, b). However, as the relative humidity is increased further, the SO 2 loss does not fall to (near) zero as would be expected from Eq. (1). This suggests that at high [H 2 O] the amount of SO 2 loss becomes less sensitive to [H 2 O]. This is most likely due to there being at least two chemically distinct SCI species present. This behaviour was previously observed for CH 3 CHOO by Newland et al. (2015a) and fits with the current understanding that the reactivity of SCIs is structure dependent.
To recap Sect. 2.2.1, the analysis presented here considers two models to fit the observations. The first of these (Eq. 3) assumes the formation of a single SCI species, which, in addition to reacting with SO 2 , can react with water, undergo unimolecular reaction or react with acid. It is clearly evident from Figs. 3-5 that this model does not give a good fit to the observations for any of the monoterpene systems studied. Therefore, the results of this (single SCI) approach are not discussed explicitly hereafter. The second of the models (Eq. E5) assumes the formation of two lumped, chemically distinct populations of SCI, denoted SCI-A and SCI-B. SCI-A is assumed to react fast with H 2 O and to have minimal decomposition. Conversely, SCI-B is assumed to have a negligible reaction with water under the experimental conditions applied but to undergo rearrangement via a VHP. We use a least squares fit of Eq. (5) to the data to determine the values of k 3 /k 2 and k d /k 2 . This approach fits the data well  for all 3 monoterpenes and represents the overall attributes of the SCIs that formed but as noted, does not represent an explicit determination of individual conformerdependent rate constants.

α-pinene
The α-pinene system is sensitive to water vapour at the low H 2 O range, with the SO 2 loss falling dramatically when the RH is increased from 0.1 to 2.5 % (Fig. 2). However, at higher RH the SO 2 loss appears to be rather insensitive to [H 2 O].
CI-1 can be formed in either a syn-(1a) or anticonfiguration (1b), whereas both CI-2 conformers are in a syn-configuration (see Scheme 2). For one of the two conformers of CI-2 (CI-2b), the hydrogen atom available for abstraction by the terminal oxygen of the carbonyl oxide group is attached to the carbon on the 4-membered ring. This has been shown in the β-pinene system to make a large difference with respect to the ability of the hydrogen to be abstracted and to undergo the VHP mechanism (Rickard et al., 1999;Nguyen et al., 2009a). This therefore suggests that CI-2b may exhibit characteristics of both SCI-A and SCI-B.  infer a probable equal yield of the two basic CI structures. This would suggest a relative yield for SCI-A of 0.25-0.50 (depending on the precise nature of CI-2b). Fitting Eq. (4) to the data and allowing lambda to vary determines values of γ A = 0.40 and γ B = 0.60 (Fig. 3). In Fig. 3, Eq. (4) is fitted to the α-pinene measurements, assuming k(SCI + acid) / k(SCI + SO 2 ) = 0. This derives a minimum value for k(SCI-A + H 2 O) / k(SCI-A + SO 2 ), the water-dependent fraction of the SCI and a maximum value for k (decomposition: SCI-B) / k(SCI-B + SO 2 ), the water independent fraction of the SCI. The kinetic parameters derived from the fitting are displayed in Table 2. Figure 6 shows the variation in the derived k 3 /k 2 and k d /k 2 values as the ratio k 5 /k 2 , k(SCI + acid) / k(SCI + SO 2 ), which is varied from zero to one. The derived k 3 /k 2 increases by about 40 % from 1.4 (±0.34) × 10 −3 to 2.0 (±0.49) × 10 −3 . The derived k d /k 2 value decreases again by about 40 % from 8.2 (±1.5) × 10 12 cm −3 to 5.1 (±0.93) × 10 12 cm −3 .
The derived limits to the relative rate constants can be put on an absolute scale using the k(SCI + SO 2 ) values for CH 3 CHOO from Sheps et al. (2014) for the synand anti-conformers. For syn this is 2.9 × 10 −11 cm 3 s −1 and for anti this is 2.2 × 10 −10 cm 3 s −1 . The syn-rate constant is applied to the derived k(decomposition:SCI-B) / k(SCI-B + SO 2 ) value and the anti rate constant to the k(SCI-A + H 2 O) / k(SCI-A + SO 2 ) value. It should be noted that the k 2 values are for quite different SCIs to those formed in this study and to our knowledge no structure-specific k(SCI + SO 2 ) have been reported for monoterpene-derived SCI, though Ahrens et al. (2014) determine an average k 2 ∼ 4 × 10 −11 cm 3 s −1 for SCI derived from β-pinene, i.e. a value within an order of magnitude of those determined for the smaller SCIs CH 2 OO, CH 3 CHOO and (CH 3 ) 2 COO (e.g. Welz et al., 2012;Taatjes et al., 2013;Sheps et al., 2014;Huang et al., 2015). Using the Sheps et al. (2014) values yields k(SCI-A + H 2 O) > 3.1 (±0.75) × 10 −13 cm 3 s −1 and k(decomposition:SCI-B) < 240 (±44) s −1 (using the values derived for k(SCI-A + acid) / k(SCI-A + SO 2 ) = 0). This k 3 value is an order of magnitude larger than the rate constants determined for the smaller anti-CH 3 CHOO in the direct studies of Sheps et al. (2014) (2.4 × 10 −14 cm 3 s −1 ) and Taatjes et al. (2013) (1.0 × 10 −14 cm 3 s −1 ). The decomposition value derived for SCI-B is of the same order of magnitude as that for syn-CH 3 CHOO (348 ± 332 s −1 ) and (CH 3 ) 2 COO (819 ± 190 s −1 ) from Newland et al. (2015a) (using updated direct measurement values of k 2 from Sheps et al. (2014) and Huang et al. (2015) for syn-CH 3 CHOO and (CH 3 ) 2 COO respectively) and within the range from the recent paper by Smith et al. (2016) Sipilä et al. (2014) see similar H 2 SO 4 production across the RH range in their study is consistent with these results.

β-pinene
Two recent studies (Nguyen et al., 2009a;Ahrens et al., 2014) have suggested yields of the two C 9 -CI (CI-3 and CI-4; see Scheme 3) obtained from β-pinene ozonolysis to be roughly equal. In these studies Ahrens et al. (2014) assume a CH 2 OO yield of 0.10 while Nguyen et al. (2009a) determine theoretically the yield of CH 2 OO to be 0.05. Another theoretical study (Zhang and Zhang, 2005) predicted a CH 2 OO yield of 0.08. In experimental studies, Winterhalter et al. (2000) determined the CH 2 OO yield to be 0.16 (±0.04) from measuring the nopinone yield and assuming it to be entirely a primary ozonolysis product (i.e. the co-product of CH 2 OO formation) and  determine a summed contribution of 84 % (±0.03) for the two C 9 -CI (i.e. a 16 % CH 2 OO yield). The theoretical studies are somewhat lower than the experimental but Nguyen et al. (2009a) note that CI-4 is likely to form additional nopinone in bimolecular reactions. The CH 2 OO is assumed to all be formed stabilised (e.g. Nguyen et al., 2009a).
SCI-3 is expected to undergo unimolecular reactions at least an order of magnitude faster than SCI-4 (Nguyen et al., 2009a;Ahrens et al., 2014). The reaction of SCI-3 with water is expected to be slow based on the calculations presented in Table 4, with a pseudo first order reaction rate of 0.3 s −1 at the highest [H 2 O] used here, 2 × 10 17 cm −3 , 298 K, whereas the water reaction with SCI-4 is expected to be considerably faster with a pseudo first order reaction rate of 85 s −1 at [H 2 O] = 2 × 10 17 cm −3 , 298 K. This reaction would thus be expected to be competitive with reaction with SO 2 for SCI-4 under the experimental conditions employed. This is in agreement with the observations of , which show a clear dependence of nopinone formation on RH (presumed to be formed from SCI + H 2 O). Fitting Eq. (4) to the data determines values of γ A = 0.41 and γ B = 0.59 (Fig. 4). Using these values, and assuming k(SCI + acid) / k(SCI + SO 2 ) = 0, yields a k(SCI-A + H 2 O) / k(SCI-A + SO 2 ) value of > 1.0 (±0.27) × 10 −4 and a k(decomposition:SCI-B) / k(SCI-B + SO 2 ) value of < 6.0 (±1.3) × 10 12 cm −3 ( Table 2).

Limonene
For the limonene measurements presented in Fig. 2, (dSO 2 / dO 3 ) / dt appears to be non-linear, with a jump in dSO 2 / dO 3 between 120 and 150 ppbv of ozone consumed. This is most evident in the two lowest RH runs (0.2 and 2.0 %). Limonene is the fastest reacting of the systems presented here, with the alkene reaction having consumed 100 ppbv of ozone within the first 5 min. The limonene sample required about 5 min of heating before the entire sample was volatised and injected into the chamber. This therefore may account for the apparent non-linear nature of dSO 2 / dO 3 in Fig. 2. The SO 2 loss in the limonene-ozone system is less affected by increasing H 2 O than for either α or β-pinene (Fig. 5), with the values of f / [SO 2 ] (y axis) varying by roughly a factor of 2 over the RH range applied compared to more than a factor of 3 for the other two systems. Hence it might be expected that there is little formation of H 2 Odependent SCI or that it has a rather slow reaction rate with water.
Placing these values on an absolute scale (using the values derived fork 5 /k 2 = 0), using the CH 3 CHOO synand anti-k(SCI + SO 2 ) determined by Sheps et al. (2014), yields values of < 7.7 (±0.60) × 10 −15 cm 3 s −1 and > 130 (±3) s −1 for k 3 and k d respectively. These values are similar to those derived for the SCI-A and SCI-B formed from βpinene. The k 3 value is a factor of 3 smaller than that determined by Sheps et al. (2014) for k 3 (anti-CH 3 CHOO + H 2 O), 2.4 × 10 −14 cm 3 s −1 . Sipilä et al. (2014) applied a single-SCI analysis approach to the formation of H 2 SO 4 from SO 2 oxidation by the limonene ozonolysis system and determined that, similarly to α-pinene, k(decomp.) k(SCI + H 2 O)[H 2 O] for [H 2 O] < 2.9 × 10 17 cm −3 ; i.e. the system is rather insensitive to [H 2 O]. Our data are consistent with the limonene system being less sensitive to [H 2 O] than the SCI populations derived from the other two monoterpenes reported here.

Experimental summary
The reaction rates of SCI-A (i.e. SCIs that exhibit chemical properties of the anti-type SCI) derived from the three different monoterpenes with a water range from < 0.8 to > 31 × 10 −14 cm 3 s −1 are broadly in line with the derived rates of Sheps et al. (2014) for anti-CH 3 CHOO of 2.4 × 10 −14 cm 3 s −1 . The decomposition rates of SCI-B (i.e. SCIs that exhibit chemical properties of the syn-type SCI) are of the order of 100-250 s −1 . This is in line with those derived for syn-CH 3 CHOO from cis-and trans-but-2-ene ozonolysis and (CH 3 ) 2 COO by Newland et al. (2015a) of 348 (±332) s −1 and 819 (±190) s −1 respectively (assuming k(syn-CH 3 CHOO + SO 2 ) = 2.9 × 10 −11 cm 3 s −1  and k((CH 3 ) 2 COO + SO 2 ) = 2.9 × 10 −10 cm 3 s −1 ; Huang et al., 2015) and recent results from Smith et Table 3. Unimolecular reactions for the CIs derived from α-pinene, β-pinene and d-limonene, as derived by . Barrier heights (kcal mol −1 ) listed estimated post-CCSD(T) energies. al. (2016) of 269-916 s −1 (strongly dependent on temperature) for (CH 3 ) 2 COO decomposition. In this work we only derive relative rates, but the similarity of the k 3 and k d values derived when the k 2 values for synand anti-CH 3 CHOO from Sheps et al. (2014) are applied is consistent with the recent work of Ahrens et al. (2014), suggesting that large SCI, derived from monoterpenes, demonstrate a similar reactivity towards SO 2 to that of smaller SCI. One uncertainty in the derivation of the kinetics presented herein is the reactions of the SCIs produced with organic acids. These acids were present in the experiments (owing to formation in the monoterpene ozonolysis reactions themselves) at levels which may have been a competitive sink for the SCI.

Theoretical results and comparison to experiments
The theoretically predicted rate coefficients for unimolecular reactions of the monoterpene SCIs are listed in Table 3, while those for the reaction with H 2 O are listed in Table 4. These data can be compared with the experimental data obtained in this work.

α-pinene
The theory-based rate coefficients show one pinonaldehyde oxide, CI-1b, with a rate of reaction with water that is significantly faster than the remaining α-pinene-derived CIs. Comparing this rate to the experimental data shows that CI-1b corresponds to SCI-A, with matching rate coefficients within an order of magnitude, i.e. within the expected uncertainty. We thus deduce that SCI-A is CI-1b. The remaining pinonaldehyde oxides, CI-1a, CI-2a and CI-2b, react predominantly through unimolecular reactions, in which theory-based rate coefficients range from 60 to 600 s −1 . All are within a factor of 4 of the experimentally derived population-averaged rate of 240 ± 44 s −1 , i.e. matching within the uncertainty margins. The unimolecular rate coefficients of this set of CIs are sufficiently close that it is not feasible to separate these in the experimental data, so we can only conclude that SCI-B in the  . Values are based on explicit CCSD(T)/aug-cc-pVTZ//M06-2X/augcc-pVTZ calculations and multi-conformer TST, including empirical corrections to reference experimental data, except for limonenederived CIs where the values are predicted using a structure-activity relationship. The rate coefficients for CH 2 OO, CH 3 CHOO and (CH 3 ) 2 COO are within a factor of 4 of evaluated literature data  α-pinene ozonolysis experiments may consist of a mixture of C-1a, CI-2a and CI-2b.

β-pinene
The theoretical analysis for nopinone oxides shows one isomer, SCI-4, that has a fast rate of reaction with water, but a slow unimolecular isomerisation, while the other isomer, SCI-3, shows a fast unimolecular decomposition. These can thus be unequivocally equated to the experimentally obtained SCI-A and SCI-B, respectively, inasmuch as the yield of CH 2 OO is minor. The predicted rate coefficients are within the expected uncertainty intervals of the theoretical data, a factor of 5 for the unimolecular rates and an order of magnitude for the reaction with H 2 O. The experimental rate measurements are defined relative to the reaction rate with SO 2 ; the value adopted for the k(SCI + SO 2 ) reaction therefore influences the derived rate coefficient values. Ahrens et al. (2014) directly measured the SO 2 rate coefficient of the longest-lived SCI (SCI-4) to be ∼ 4 × 10 −11 cm 3 s −1 , but for SCI-3 we assume a similar rate coefficient as syn-CH 3 CHOO + SO 2 determined by Sheps et al. (2014) of 2.9 × 10 −11 cm 3 s −1 . Nopinone oxides are bicyclic compounds, with a bulky dimethyl-substituted 4-membered ring adjacent to the carbonyl oxide moiety. To examine the potential impact of steric hindrance on the SCI + SO 2 reaction, we characterised all sulfur-substituted secondary ozonides (S-SOZ) formed in this reaction (Kuwata et al., 2015;. We find that the tricyclic S-SOZ shows very little interaction between the sulfurbearing ring and the β-pinene substituents, and little change in the ring strain. The energies of the S-SOZ adducts relative to the SCI + SO 2 reactants thus remain very similar to that of CH 2 OO, CH 3 CHOO or (CH 3 ) 2 COO, confirming the quality of our selection of reference rate coefficients.

Limonene
Of the six non-CH 2 OO CIs formed in limonene ozonolysis, CI-5b was predicted to have a fast reaction rate with H 2 O; its oxide substitution patterns is similar to pinonaldehyde oxide CI-1b. The SAR-predicted rate coefficient of CI-5b + H 2 O is within a factor of 2 of the experimentally derived k 3 value for SCI-A, such that we can equate SCI-A to CI-5b with confidence. The SCI-B set of Criegee intermediates then contains the summed population of the remaining five CIs, all of which react slowly with H 2 O. The SARpredicted unimolecular decay rate coefficients range from 15 to 700 s −1 , all within a factor of 9 of the experimentally obtained k d = 130 s −1 ; it should be noted that for limonenederived CIs, no explicit theoretical calculations are available, and the SAR-predictions carry a somewhat larger uncertainty. We have performed an exhaustive characterisation of the conformers of CI-5b. The most stable conformers show an internal complex formation between the oxide moiety and the carbonyl group, similar to those characterised for the bimolecular reaction of CIs with carbonyl compounds (Jalan et al., 2013;Wei et al., 2015). The theoretical study by Jiang et al. (2013) on limonene ozonolysis appears to have omitted internal rotation and cannot be compared directly. It seems likely that the limonene-derived CIs can thus easily undergo internal SOZ formation, which is thought (Vereecken and Francisco, 2012) to be entropically unfavourable, but has a low barrier to reaction. For α-pinene, a similar internal complex formation and SOZ ring closure is not as favourable due to the geometric limitations enforced by the 4-membered ring.
A large number of transition state conformers for CI-5b + H 2 O were characterised, though no exhaustive search was completed. The energetically most favourable structures show interaction between the carbonyl group and the H 2 O co-reactant as it adds onto the carbonyl oxide moiety. Similar stabilising interactions between the carbonyl moiety and the carbonyl oxide moiety were reported recently in cyclohexene-derived CIs (Berndt et al., 2017). This in-  teraction thus lowers the barrier to a reaction, though it is currently unclear whether it enhances the reaction rate, e.g. compared to the α-pinene-derived CI-1b, as these hydrogenbonded structures are entropically not very favourable. The intra-molecular interactions with heterosubstituents could be investigated in future work.
7 Global modelling study

SCI chemistry
A global atmospheric modelling study was performed using the GEOS-Chem chemical transport model (as described in Sect. 4) to examine the global monoterpene-derived SCI budget and the contribution of these SCIs to gas-phase SO 2 oxidation. The existing chemistry scheme in the model is supplemented with monoterpene SCI chemistry based on the experimental results described in Sect. 5 and in Table 5. It should be noted here that this modelling study focuses on the chemical impacts of monoterpene SCIs formed from ozonolysis reactions only. No chemistry for other SCIs derived from isoprene and/or other (smaller) alkenes are incorporated in the adapted model chemical scheme. The monoterpene emissions in GEOS-Chem are taken from MEGAN v2.1 (Guenther et al., 2012). The scheme emits seven monoterpenes: α-pinene, β-pinene, limonene, myrcene, ocimene, 3-carene and sabinene. The monoterpenes are oxidised within the model by OH, NO 3 and O 3 at rates shown in Table S1. Reaction with O 3 leads to the production of monoterpene-specific SCI. Reactions with OH and NO 3 do not lead to the formation of any products, with the reactions only acting as a sink for the monoterpene and the respective oxidant. The SCI yields from the ozonolysis of α-pinene, β-pinene and limonene are derived from the experimental work presented here. SCIs from each monoterpene are split into SCI-A and SCI-B as defined in previous sections. For the other four monoterpenes emitted, the SCI yields and kinetics are derived based on the similarity of the structure to one of the species studied here or previously in the literature. The main SCIs produced in the ozonolysis of myrcene and ocimene are expected to be acetone oxide ((CH 3 ) 2 COO) or 4-vinyl-5-hexenal oxide (CH 2 CHC(CH 2 )CH 2 CH 2 CHOO), since ozone has been sug- gested to react predominantly at the internal double bond (∼ 97 % for myrcene, ∼ 90 % for ocimene Baker et al., 2004). The SCI yield is taken to be 0.30, similar to that of (CH 3 ) 2 COO from 2,3-dimethyl-but-2-ene ozonolysis (Newland et al., 2015a). However, this may be an underestimation since it has been predicted that stabilisation of small CIs increase with the size of the carbonyl co-product, as this coproduct can take more of the nascent energy of the primary ozonide on decomposition due to a greater number of degrees of freedom available (Nguyen et al., 2009a;Newland et al., 2015b). Sabinene is a bicyclic monoterpene with an external double bond and hence is treated like β-pinene. This assumption is backed up by recent theoretical work (Wang and Wang, 2017), who predict similar behaviour from sabinenederived SCIs to the predicted behaviour from β-pinene SCIs by Nguyen et al. (2009a). They predict a SCI yield between 24 and 64 %. 3-carene is a bicyclic monoterpene with an internal double bond and is treated like α-pinene. Figure 7 shows the annually averaged total SCI burden from monoterpene ozonolysis in the surface layer in the GEOS-Chem simulation. A number of interesting features are apparent from this figure and the associated information is given in Table 6: i. The highest annually averaged monoterpene SCI concentrations are found above tropical forests.
iii. > 97 % of the total monoterpene SCI burden is SCI-B.
Annual global monoterpene emissions are dominated by the tropics (Fig. S1 in the Supplement), accounting for > 90 % during the Northern Hemisphere winter months (November-April) and 70 % even during the peak emissions from the northern boreal region during June and July (Sindelarova et al., 2014). Despite annually averaged surface ozone mixing ratios being roughly a factor of 2 higher in the northern middle to high latitudes, monoterpene SCI production is still dominated by the tropics. Annually averaged surface monoterpene SCI concentrations across the northern boreal regions are < 2 × 10 3 cm −3 ; during the summer months (JJA) this value rises to 2-5 × 10 3 cm −3 .
More than 97 % of the total monoterpene-derived SCIs are SCI-B (Table 6). This is because typical water vapour concentrations in the tropics are > 5.0 × 10 17 cm −3 . This gives SCI-A removal rates (i.e. k 3 [H 2 O]) of 2 × 10 3 -1.5 × 10 5 s −1 , whereas removal rates of SCI-B to unimolecular reactions have been determined here to be 1-3 orders of magnitude slower, of the order of 100-250 s −1 . Since the loss of SCI-B is independent of temperature in the model, the highest SCI-B concentrations would be expected to be located in the regions of highest SCI-B production. Recent experimental studies (Smith et al., 2016) have demonstrated a strong temperature dependence for the unimolecular decomposition rate of (CH 3 ) 2 COO between 283 and 323 K (269-916 s −1 ). Therefore, it may be that in reality there would be some geographical variation in the rate of unimolecular loss.
The monoterpene SCI-A + H 2 O reactions are expected to lead to high yields of both large (e.g.  and small (measured in high yield in the experiments presented here) organic acids. Figure 8 shows the seasonal removal of SO 2 by reaction with monoterpene-derived SCI, as a percentage of total gasphase SO 2 oxidation in the surface layer. Monoterpene SCIs are most important (relative to OH) for SO 2 oxidation over tropical forests, where they account for up to 60 % of the local gas-phase SO 2 removal during DJF and MAM in some regions. The reasons for this are two-fold: firstly, the highest modelled monoterpene SCI concentrations are found in these regions (Fig. 7); additionally, OH concentrations in the model are low over these areas (Fig. S2). Historically there have been discrepancies between modelled and observed OH concentrations over tropical forests, with models appearing to underpredict [OH] by up to a factor of 10 (e.g. Lelieveld et al., 2008). It was proposed that this was due to missing sources of OH recycling during isoprene oxidation. During recent years there have been advances in our understanding of isoprene chemistry. GEOS-Chem v-M. J. Newland et al.: The atmospheric impacts of monoterpene ozonolysis 09, used here, includes an isoprene OH recycling scheme largely based on Paulot et al. (2009a, b), with updates from Peeters et al. (2009), Peeters and Müller (2010) and Crounse et al. (2011Crounse et al. ( , 2012, and evaluated in Mao et al. (2013). However, more recent experimental and theoretical work is not yet included.
Annually, monoterpene SCI oxidation accounts for 1.2 % of the gas-phase SO 2 oxidation in the terrestrial tropics. This accounts for the removal of 2.9 Gg of SO 2 . Across the northern boreal forests, monoterpene SCIs contribute 0.7 % to gasphase SO 2 removal annually, removing 0.8 Gg of SO 2 . Globally, throughout the whole of the atmosphere, monoterpene SCIs account for only 0.5 % of gas-phase SO 2 removal, removing 8.1 Gg of SO 2 annually.
It is noted that MEGAN does not contain oceanic monoterpene emissions, which may increase the global importance of SCIs for gas-phase SO 2 removal. Luo and Yu (2010) determined annual global oceanic α-pinene emissions to be 29.5 TgC using a top-down approach, with only 0.013 (Luo and Yu, 2010) -0.26 (Hackenberg et al., 2017) TgC estimated using a range of bottom-up approaches; clearly there are large uncertainties in oceanic monoterpene emissions. At the upper end of this range they could potentially provide a similar contribution to SCI production and subsequent SO 2 oxidation as monoterpenes emitted from the terrestrial biosphere. SCI production could more generally be further amplified by sources such as marine-derived alkyl iodine photolysis. Blitz et al. (2017) recently calculated a revised SO 2 + OH reaction rate (k 1 (1 bar N 2 )(298 K) = 5.8 × 10 −13 cm 3 s −1 ), based on experimental work and a master equation analysis, which is ∼ 40 % lower than the rate given in the most recent JPL data evaluation (Burkholder et al., 2015) ((k 1 (1 bar N 2 )(298 K) = 9.5 × 10 −13 cm 3 s −1 ), which is used in the GEOS-Chem model simulation. Figure S3 shows the increased influence of monoterpene-derived SCIs on gas-phase SO 2 oxidation if the alternative SO 2 + OH rate is used. This increased the impact of monoterpene SCIs to up to 67 % of gas-phase SO 2 removal in regions of the tropical forests during DJF and MAM, with the contribution of monoterpene SCIs to global gas-phase SO 2 oxidation increasing to 0.7 %.
While certain monoterpenes appear to be more important than others with regard to the production of SCIs which will oxidise SO 2 , these results are sensitive to the kinetics used and the assumptions made for the monoterpenes not studied experimentally here. Hence we do not attempt to draw any conclusions about the relative importance of each monoterpene from the modelling. Clearly the most important monoterpenes will be those with high yields of SCI-B, particularly if those SCI-B have a structure that hinders unimolecular decomposition (such as certain β-pinene-derived SCI).

Conclusions
We report results from an integrated experimental (simulation chamber), theoretical (quantum chemical) and modelling (global chemistry-transport simulation) study of the impacts of monoterpene ozonolysis reactions on stabilised Criegee intermediate (SCI) formation and SO 2 oxidation. The ozonolysis of the monoterpenes α-pinene, β-pinene and limonene have been shown to produce a structurally diverse range of chemically distinct SCIs, with some showing limited sensitivity to/reaction with water vapour under nearatmospheric humidity levels. A multi-component system is required to explain the experimentally observed SO 2 removal kinetics. A two-body model system based on the assumption of a fraction of the SCIs produced being reactive towards water (SCI-A; potentially contributing to the significant formation of a range of organic acids in the atmosphere) and a fraction being relatively unreactive towards water (SCI-B), analogous to the structural dependencies observed for the simpler CH 3 CHOO SCI system, has been shown to describe the observed kinetic data reasonably well for all the monoterpene systems investigated and may form a computationally affordable and conceptually accessible basis for the description of this chemistry within atmospheric models. Moreover such an approach is required to accurately predict SCI concentrations, which will be underestimated if a simple average of the properties of the two different SCI classes is used. The atmospheric fate of SCI-B produced from the monoterpenes studied here will be controlled by their removal by unimolecular decomposition. In this work, we have experimentally determined the monoterpene SCI-B decomposition rate to be between 100 and 250 s −1 . This has significant implications for the role of monoterpene-derived SCIs as oxidants in the atmosphere. The fate of SCI-A will be reaction with water or the water dimer, likely leading to the production of a range of organic acids.
A theory-based analysis of the kinetics of the SCIs formed from α-pinene, β-pinene ozonolysis has also been performed, which complements the experimental work. The identification of the likely SCI-A and SCI-B populations and the derived kinetics agree with experimental observations within the respective uncertainties.
A modelling study using the GEOS-Chem global 3-D chemical transport model supplemented with the chemical kinetics elucidated in this work suggests that the global monoterpene-derived SCI burden will be dominated (> 97 %) by SCI-B. The highest annually averaged SCI concentrations are found in the tropics, with seasonally averaged monoterpene SCI concentrations up to 1.4 × 10 4 cm −3 owing to large monoterpene emissions. Across the boreal forest, average SCI concentrations reach between 3 and 5 × 10 3 cm −3 during the Northern Hemisphere summer. Oxidation of SO 2 by monoterpene SCIs is shown to also be most important in the tropics. While oxidation by SCIs contributes < 1% to gasphase SO 2 oxidation globally, over tropical forests this can rise to up to 60 % at certain times of the year. Monoterpene SCI-driven SO 2 oxidation will increase the production of sulfate aerosol -affecting atmospheric radiation transfer and hence climate -and reduce the atmospheric lifetime and hence the transport of SO 2 . These effects will be substantial in areas where monoterpene emissions are significant, in particular over the Amazon, central Africa and SE Asian rainforests.