The roles of volatile organic compound deposition and oxidation mechanisms in determining secondary organic aerosol production: A global perspective using the UKCA chemistry-climate model (vn8.4)

The representation of volatile organic compound (VOC) deposition and oxidation mechanisms in the 15 context of secondary organic aerosol (SOA) formation are developed in the United Kingdom Chemistry and Aerosol (UKCA) chemistry-climate model. Impacts of these developments on both the global SOA budget and model agreement with observations is quantified. Firstly, global model simulations were performed with varying VOC dry deposition and wet deposition fluxes. Including VOC dry deposition reduces the global annual-total SOA production rate by 2 32 %, with the range reflecting uncertainties in surface resistances. 20 Including VOC wet deposition reduces the global annual-total SOA production rate by 15 % and is relatively insensitive to changes in effective Henry’s Law coefficients. With precursor deposition, simulated SOA concentrations are lower than observed, with a normalised mean bias (NMB) of -51%. Hence, including SOA precursor deposition worsens model agreement with observations even further (NMB = -66 %). Secondly, for the anthropogenic and biomass burning VOC precursors of SOA (VOCANT/BB), model simulations were 25 performed varying: a) the parent hydrocarbon reactivity, b) the number of reaction intermediates, and c) accounting for differences in volatility between oxidation products from various pathways. These changes were compared to a scheme where VOCANT/BB adopts the reactivity of a monoterpene (α-pinene), and is oxidised in a single-step mechanism with a fixed SOA yield. By using the chemical reactivity of either benzene, toluene or naphthalene for VOCANT/BB, the global annual-total VOCANT/BB oxidation rate changes by -3, -31 or -66 %, 30 respectively, compared to when using α-pinene. Increasing the number of reaction intermediates, by introducing a peroxy radical (RO2), slightly slows the rate of SOA formation, but has no impact on the global annual-total SOA production rate. However, RO2 undergoes competitive oxidation reactions, forming products with substantially different volatilities. Accounting for the differences in product volatility between RO2 oxidation pathways increases the global SOA production rate by 153 % compared to using a single SOA yield. Overall, 35 for relatively reactive compounds, such as toluene and naphthalene, the reduction in reactivity for VOCANT/BB oxidation is outweighed by accounting for the difference in volatility of RO2 products, leading to a net increase


4
The first studies to quantify the SOA yields from aromatic compounds (Odum et al., 1997;Odum et al., 1996) are not high enough to account for the concentrations of aromatic SOA observed in field studies (Tsigaridis and Kanakidou, 2003;Hoyle et al., 2007). For instance, early estimates of SOA yields from aromatic compounds, which were conducted in relatively high nitrogen oxide (NO X = NO and NO 2 ) concentrations, range between 5 and 10 % (Odum et al., 1997;Odum et al., 1996). Consequently, the use of low SOA yields for 5 aromatic compounds in global models results in low global annual-total SOA production rates, ranging from just 0.05 to 2.5 Tg (SOA) a -1 , which are negligible in comparison to biogenic sources (Tsigaridis and Kanakidou, 2003;Hoyle et al., 2007). However, more recent chamber studies suggest the SOA yields from aromatic compounds are strongly influenced by NO X concentrations (Hurley et al., 2001;Song et al., 2005;Ng et al., 2007;Chan et al., 2009). For example, in agreement with early estimates (Odum et al., 1996;Odum et al., 1997), 10 Ng et al. (2007) also observed an SOA yield from aromatic VOCs of 5 -10 % under high-NO X conditions. However, under lower NO X concentrations, Ng et al. (2007) measured substantially higher SOA yields of 37, 30 and 36 % for benzene (C 6 H 6 ), toluene (C 7 H 8 ) and xylene (C 8 H 10 ), respectively. Similarly, under low-NO X conditions, Chan et al. (2009) observed an SOA yield of 73 % from naphthalene (C 10 H 8 ). Zhang et al. (2014) further corroborates this negative sensitivity of SOA yields from aromatic compounds to NO X concentrations, 15 and also highlights how chamber studies frequently underestimate SOA yields due to wall losses.
The exact mechanism describing aromatic oxidation is not yet fully understood, despite considerable progress to date (Kautzman et al., 2010;Li et al., 2016;Al-Naiema and Stone, 2017;Li et al., 2017b;Schwantes et al., 2017). As aromatic oxidation is initiated by the hydroxyl radical (OH), the influence of NO X on SOA production is probably due to reaction of NO with subsequent reaction intermediates or products. Oxidation of 20 the parent aromatic hydrocarbon by OH is followed by addition of molecular oxygen (O 2 ) and isomerization, forming a bicyclic peroxy radical, RO 2 (Koch et al., 2007;Birdsall et al., 2010). Under high-NO X conditions, the peroxy radical reacts with the nitric oxide radical (NO) to form fragmented products, whereas, under low-NOx conditions, the peroxy radical reacts with the hydroperoxyl radical (HO 2 ) to form functionalised products (Ng et al., 2007). Hence, due to the difference in volatility of products, the RO 2 +HO 2 yields a greater mass of SOA 25 compared to the RO 2 +NO pathway. Water vapour may also be involved in the gas-phase oxidation of aromatic compounds (Hinks et al., 2018). However, as both positive (White et al., 2014) and negative (Cocker et al., 2001) correlations between aromatic SOA yields and relative humidity have been observed in chamber studies, the role of water vapour in aromatic oxidation is not yet clear. The exact mechanism describing aromatic oxidation may not be fully understood but the observed influence of NOx on SOA yields suggests that 30 simulating SOA production from aromatic compounds necessitates multigenerational oxidation mechanisms, with SOA yields responding to oxidant availability.
The peroxy radical reaction intermediate, together with competitive NO and HO 2 reactions with varying SOA yields, has been applied to several different SOA schemes. Benzene, toluene and xylene have been incorporated into both global (Henze et al., 2008;Heald et al., 2011) and regional scale (Li et al., 2017a) models. 35 Henze et al. (2008) applied the laboratory-derived yields from Ng et al. (2007) to aromatic compounds (16 Tg (VOC) a -1 ), which resulted in a global annual-total SOA production rate of 4 Tg (SOA) a -1 , with 61% of SOA being produced via the RO 2 +HO 2 pathway. Peroxy radical chemistry has also been applied to IVOCs, which are a mixture of species emitted from both anthropogenic and biomass burning. Pye and Seinfeld (2010) applied the laboratory-derived yields from Chan et al. (2009) to IVOCs (18 Tg (VOC) a -1 ), which resulted in global annual-5 total SOA production rate of 5 Tg (SOA) a -1 , with 75% of SOA being produced via the RO 2 +HO 2 pathway.
Despite peroxy radical chemistry being included in some SOA schemes, the influence on the global SOA budget and model agreement with observations has not been quantified.
The objective of this study is to further develop the SOA scheme within a chemistry-climate model, the United Kingdom Chemistry and Aerosol (UKCA) model. Firstly, the model is updated to include the wet and 5 dry deposition of SOA precursors. Secondly, the mechanism describing SOA formation from anthropogenic and biomass burning VOCs is updated to account for the influence of NOx on SOA yields. Several simulations are conducted to test the sensitivity of SOA to both precursor deposition and oxidation mechanisms. The impact of these model developments on SOA is assessed through a comprehensive comparison with available observations. The paper is organised as follows. The global chemistry-climate model used in this study is 10 described in Section 2; this section also includes a description of the model developments applied to the SOA scheme. Observations used to evaluate the model are discussed in Section 3. Next, the influence of precursor deposition on SOA is investigated (Section 4). In Section 5, the sensitivity of modelled SOA to oxidation mechanisms and VOC reactivity is explored. Concluding remarks and further work are discussed in Section 6.

2 Chemistry-climate model description
In this section, the model is briefly described. This begins with a brief description of the default configuration, followed by the model developments made in this study. The chemistry-climate model used in this study is the 20 United Kingdom Chemistry and Aerosol (UKCA) model (Morgenstern et al., 2009;Mann et al., 2010;O'Connor et al. 2014) which is coupled to the Global Atmosphere 4.0 (GA4.0) configuration (Walters et al., 2014) of the Hadley Centre Global Environmental Model (Hewitt et al., 2011) version 3 (HadGEM3). The atmosphere-only configuration with prescribed sea surface temperature and sea ice fields based on [1995][1996][1997][1998][1999][2000][2001][2002][2003][2004] reanalyses data (Reynolds et al., 2007) was used. The model was run at a horizontal resolution of N96 (1. 875° 25 longitude by 1.25° latitude) with 85 terrain-following hybrid-height levels distributed from the surface to 85 km.
Horizontal winds and temperature in the model were nudged towards ERA-Interim reanalyses for the [1999][2000] period (Dee et al., 2011) using a Newtonian relaxation technique with a relaxation time constant of 6 hours (Telford et al., 2008). There was no feedback from the chemistry or aerosols onto the dynamics of the model; this ensured identical meteorology across all simulations so that differences in SOA were solely due to 30 differences in precursor oxidation mechanisms and deposition.

1 Gaseous chemistry (UKCA)
The United Kingdom Chemistry and Aerosol (UKCA) model used in this study combines the "TropIsop" tropospheric chemistry scheme from O' Connor et al. (2014) with the stratospheric chemistry scheme from 35 Morgenstern et al. (2009). There are 75 species with 285 reactions. This includes odd oxygen (O x ), nitrogen (NO y ), hydrogen (HO x = OH + HO 2 ), and carbon monoxide (CO). Explicit hydrocarbons included are methane, ethane, propane, isoprene and monoterpene. Isoprene oxidation follows the Mainz Isoprene Mechanism (Poschl et al., 2000) which is described in detail in O'Connor et al. (2014). In addition to the aforementioned explicit hydrocarbons, two additional non-explicit VOCs are included; VOC ANT and VOC BB are lumped compounds representing anthropogenic and biomass burning VOCs, respectively. Together, isoprene, monoterpene, VOC ANT and VOC BB , for the precursors of SOA. The reactivity and production of SOA from these species are discussed in further detail in Section 2.5. For bimolecular gas-phase reactions, rate constants are calculated 5 following the Arrhenius expression where ! is a constant, is the ratio of the activation energy over the universal gas constant (E A /R), and T is temperature. The rate constant is then used to calculate the rate of reaction: 10 = ( 2) where is the rate coefficient, and [A] and [B] are concentrations of gases A and B, respectively.

Gaseous wet deposition
Within UKCA, wet deposition of gases is calculated as a first-order process as a function of precipitation, 15 following Walton et al. (1988). For a detailed description of the wet deposition within UKCA, see O'Connor et al. (2014). Within each grid box, the scavenging rate, r, is calculated as follows:

Gaseous dry deposition
Dry deposition refers to the transfer of chemical species from the atmosphere to the surface in the absence of precipitation. Dry deposition of gas-phase species within UKCA has also been described in detail before (O'Connor et al., 2014) so is only described briefly here. The dry deposition velocity ( ! ) is calculated using a resistance-based approach (Wesely, 1989). This approach is analogous to an electrical circuit, where the 5 transport of chemical species is dependent on three resistances, ! , ! , and ! : The aerodynamic resistance term, ! , represents the resistance to transport of chemical species through the boundary layer to a thin layer of air just above the surface. This term is calculated from the wind profile, taking into account the atmospheric stability and the surface roughness: 10 calculated following Fuchs (1971) which is described in Mann et al. (2010). Mineral dust is also included in the model simulations, but treated in a separate aerosol module (Woodward, 2001).

Emissions
The emissions used in this study are all monthly-varying decadal-average, centred on the year 2000.
Anthropogenic and biomass burning gas-phase emissions are prescribed following Lamarque et al. (2010). 5 Biogenic emissions of isoprene, monoterpene and methanol (CH3OH) are also prescribed, taken from the Global Emissions Inventory Activity (GEIA), based on Guenther et al. (1995). A diurnal cycle in isoprene emissions is imposed based on solar zenith angle. POA and BC emissions from fossil fuel combustion are prescribed following Lamarque et al. (2010). POA and BC emissions from savannah burning and forest fires are prescribed, taken from the Global Fire Emissions Database (GFEDv2; van der Werf et al. (2010)). For VOC BB , 10 monthly-mean CO emissions from biomass burning were used to define its spatial distribution (Lamarque et al., 2010) and scaled to reproduce the global annual VOC total emissions from biomass burning estimated from the Emissions Database for Atmospheric Research (EDGAR) (49 Tg (VOC BB ) a -1 ). For VOC ANT , monthly-mean anthropogenic emissions of benzene, toluene and xylene, were taken from Lamarque et al. (2010), and scaled to reproduce the global annual anthropogenic VOC total emissions estimated by EDGAR (127 Tg (VOC ANT ) a -1 ). 15 Scaling both VOC BB and VOC ANT to the emissions type's totals (i.e. biomass burning and anthropogenic, respectively) represents an upper limit for the SOA precursor emissions. The emissions of VOC BB and VOC ANT described here have been used in Kelly et al. 2018, with the corresponding impacts on SOA rigorously evaluated against observations. Briefly, the locations of SOA observations are well suited to constrain the anthropogenic source of SOA, but not so well suited for evaluating VOC BB . Inclusion of VOC ANT in SOA production gives 20 rise to a substantial improvement in model agreement with observations (Kelly et al. 2018).

Default Treatment of SOA
In this section, the current treatment of SOA in the UKCA model is first described, followed by descriptions of new treatments of precursor deposition and oxidation mechanisms. Within the model, SOA is treated by a 25 coupling between the UKCA gas-phase chemistry and GLOMAP-mode. Emitted parent hydrocarbon gases undergo a single-step oxidation, forming a secondary organic gas (SOG) which condenses, forming SOA. This is shown in Eq (8): where VOC is the concentration of the emitted parent hydrocarbon, [o] is the oxidant concentration, k VOC+[O] is the temperature-dependent rate coefficient (Eq (1)), α VOC+ [O] is the stoichiometric coefficient, and SOG is the secondary organic gas. SOG is treated as non-volatile, and although there is evidence that OA is both semivolatile (Robinson et al., 2007;Donahue et al., 2012) and non-volatile (Jimenez et al., 2009;Cappa and Jimenez, 35 2010), SOG condenses irreversibly to form SOA in UKCA. The yield is identical for all oxidation reactions (13 %), regardless of VOC or oxidant. Essentially, the volatility distribution is assumed to be identical for all reactions, irrespective of parent VOC and oxidant. In the model, no SOA precursor undergoes dry or wet deposition.
In this study, SOA production is considered from gas-to-particle partitioning of VOCs oxidation products. S/IVOCs emissions are not considered and aqueous phase SOA prodcution is not included. These include monoterpene, isoprene, VOC BB and VOC ANT . Monoterpene and isoprene contain both single and double 5 carbon bonding and therefore react with ozone (O 3 ) and the hydroxyl (OH) and nitrate (NO 3 ) radicals, forming SOG and subsequently SOA (Eq 8). Note, for isoprene, oxidation in the context of SOA production (Eq 8) occurs independently to isoprene oxidation in the Mainz Isoprene Mechanism described in Section 2.1. Reaction kinetics for isoprene and monoterpene (α-pinene) oxidation are taken from Atkinson and Arey (2003), and are shown in Table 1. As discussed in Section 2.4, VOC ANT and VOC BB are surrogate compounds, which do not 10 retain molecular information, and therefore, do not have laboratory derived rate constants. Initially, the assumption is made that VOC ANT and VOC BB are reduced compounds, with only single carbon bonding and react predominantly with OH. VOC ANT and VOC BB are also assumed to have a similar reactivity to monoterpene towards OH oxidation, but do not react with O 3 or NO 3 . These assumptions in the parent hydrocarbon reactivity are discussed further in Section 2.4.2. As stated above, none of the SOA precursors in this scheme are wet or dry 15 deposited. In summary, the current SOA scheme suffers from a lack of mechanistic detail in oxidation mechanisms, and neglects precursor deposition. In the following sub-sections, modifications to the model are described and the impacts of these processes quantified.

Addition of SOA Precursor Deposition
Precursors of SOA include the emitted parent hydrocarbons (monoterpene, isoprene, VOC ANT , VOC BB ) and the secondary organic product (SOG). Several modifications were made to UKCA to investigate the influence of precursor deposition on SOA. Firstly, wet deposition of the gas-phase species, as described in Section 2.1.1, was extended to include all SOA 5 precursors. The effective Henry's Law coefficient, for all SOA precursors, was either set to 10 5 or 10 9 M atm -1 . These values of H eff were taken from estimates by Hodzic et al. (2014). Secondly, the treatment of dry removal of gas-phase species (section 2.1.2) was extended to include all SOA precursors and they were assumed to have identical surface resistances. Table 2 shows the surface resistances for the SOA precursors over the 9 surface types. The aerodynamic and quasi-laminar surface resistances were calculated online, based on relative molecular mass and meteorology. During field studies over 10 forested regions, organic hydroperoxides (ROOH) were observed to undergo significant dry deposition (Hall et al., 1999;Valverde-Canossa et al., 2006;Nguyen et al., 2015). Surface resistances derived from these field studies range from 5 -40 sm -1 (Hall et al., 1999;Nguyen et al., 2015). Hence, these field-derived surface resistances of ROOH ('Low'; Table 2) were used to provide a lower estimate of the surface resistances of SOA precursors. Surface resistances corresponding to the dry deposition of CO ('High'; Table 2) were used to provide an upper limit of the surface resistances of SOA precursors. 15

Addition of a New Oxidation Mechanism for VOC ANT/BB
As discussed in Section 2.4, initially VOC ANT/BB follows a single-step oxidation mechanism, with a single fixed SOA yield, and with a reactivity based on α-pinene (Table 1). However, of the anthropogenic and biomass burning VOCs related to SOA production, aromatic compounds have been identified as important components in field studies (Von Schneidemesser 20 et al., 2010;Ding et al., 2012;Guo et al., 2012;Peng et al., 2013). Furthermore, environmental chamber studies suggest aromatic hydrocarbons undergo multi-generational oxidation reactions, with SOA yields dependent on oxidant concentrations (Ng et al., 2007;Chan et al., 2009;Kautzman et al., 2010;Li et al., 2016;Al-Naiema and Stone, 2017;Li et al., 2017b;Schwantes et al., 2017). Therefore, in order to examine how SOA is affected by variations in oxidation mechanisms, chamber-derived aromatic oxidation pathways are applied to VOC ANT/BB . This section outlines how the chamber-derived 25 aromatic oxidation pathway, postulated by Ng et al. (2007), is applied to the mechanistic description of SOA production from VOC ANT/BB within UKCA. Figure 1 shows a mechanistic description of SOA production from toluene, accounting for the influence of NOx on SOA production, adapted from Ng et al. (2007). Briefly, toluene undergoes oxidation by OH, followed by addition of oxygen and isomerisation, to form a bicyclic peroxy radical, RO 2 . The bicyclic peroxy radical undergoes competitive 30 reactions with hydroperoxyl radical (HO 2 ) and NO. The HO 2 pathway forms functionalised products, whereas products of the NO pathway are fragmented. Although Figure 1 shows a mechanistic description of toluene oxidation, the oxidation of other methylated aromatic compounds will also follow a similar pathway. This mechanism for aromatic oxidation, as shown in Figure 1, was applied to VOC ANT/BB oxidation. The rate determining step in Figure 1 is the initial oxidation by OH and, therefore, the mechanism can be simplified as follows: where VOC represents VOCANT/ BB, kVOC+OH represents the rate constant for aromatic oxidation by OH, RO2 represents the bicyclic peroxy radical, kRO2+HO2 and αRO2+HO2 represent the rate constant and the stoichiometric coefficient for the RO2+HO2 reaction, 10 respectively, and kRO2+NO and αRO2+NO represent the rate constant and the stoichiometric coefficient for the RO2+NO reaction, respectively. Both RO2 reactions form the same non-volatile species, SOG, but the yields associated with the formation rates of this product are variable ( !! ! !!" ! and !! ! !!" ). Hence, this mechanism allows the sensitivity of SOA production to HO2/ NO to be accounted for. However, note that the differences in volatility between RO2 oxidation products are not explicitly accounted for. Within the model, the difference in volatility distribution between the products of the RO2 reactions 15 are controlled by the stoichiometric coefficients ( !! ! !!" ! and !! ! !!" ). Previous modelling studies use a similar method to treat SOA production via the RO2+HO2 pathway. Assuming that the products from oxidation of explicit aromatic compounds are non-volatile, Henze et al. (2008) uses a stoichiometric yield of around 18 %. Using IVOC emissions based on naphthalene, Pye and Seinfeld (2010) uses a stoichiometric coefficient of 73 %. However, both Henze et al. (2008) and Pye and Seinfeld (2010) treat products from the RO2+NO pathway as semi-volatile, with stoichiometric yields ranging from 20 2 to 107 %, and equilibrium partitioning coefficients ranging from 0.0037 to 3.3150 m 3 µg -1 . The reaction kinetics for aromatic oxidation used here are shown in Table 1. At 298 K, the rate coefficients for the reaction of OH with naphthalene, toluene and benzene are 23.2, 5.62, and 1.22 x10 -12 cm 3 molecule -1 s -1 , respectively (Table 1). At 298 K, the rate coefficients for the reactions of the peroxy radical with HO 2 and NO are 14.7 and 8.42 x10 -12 cm 3 molecule -1 s -1 , respectively (Table 1). Note, these rate coefficients are used for the peroxy radical irrespective of the identity of the parent VOC (i.e. naphthalene, 25 toleuene or benzene).

Model simulations
In this study, 10 simulations were performed to explore the influence of hydrocarbon deposition and oxidation mechanisms on SOA, and are described in Table 3. The duration of all simulations is two years, spanning from 1999 to 2000. The first 30 year was discarded as spin-up, and analysis was performed on the second year -2000. Firstly, a control simulation was conducted, where the oxidation of all parent hydrocarbons (isoprene, monoterpene, VOC ANT and VOC BB ) followed Eq (8) and no SOA precursors were lost by wet or dry deposition processes. Next, the influence of VOC deposition on SOA was explored. To begin with, precursors were assumed to have low surface resistances (Low ; Table 2), thus, testing the upper limit for precursor dry deposition (Dry_High ; Table 3). Next, the strength of precursor surface resistance was increased 35 (High; Table), testing the lower limit for deposition rates (Dry_Low ; Table 3). Next, SOA precursors were treated as soluble and were, therefore, included in the wet deposition scheme. As with dry removal, the upper and lower limits of precursor wet deposition were tested by carrying out two simulations, one with a higher solubility (Wet_High), and one with a lower solubility (Wet_Low). An additional simulation was conducted to test whether the effects of precursor dry and wet deposition on SOA are additive (DryH_WetL). Note, for this simulation, dry and wet deposition are included with low surface resistances (Low; Table 2) and low solubility. Alternative combinations of surface resistances and solubility could have been used to quantify the combined influence of precursor dry and wet deposition on SOA. 5 Next, the influence of VOC oxidation mechanisms on SOA was explored by modifying the mechanistic description of SOA production from anthropogenic and biomass burning VOCs. As discussed in Section 1, oxidation mechanisms within SOA schemes vary substantially. Therefore, in this section, where necessary, changes to VOC ANT/BB oxidation were made in a step-wise fashion, in order to isolate the effects of individual changes. Firstly, the combined effects of the use of a reactive aromatic compound (naphthalene) and introducing a reaction intermediate (RO 2 ) were explored in the Multi_nap simulation, 10 where VOC ANT/BB follows Eq (9). In this simulation, stoichiometric reaction yields of 13 % are applied to both RO 2 oxidative pathways, which is identical to the reaction yield applied to simulations following the single-step mechanism (Eq (8)). The effects of changes to parent VOC ANT/BB reactivity, the chemical fate of the new reaction intermediate, and SOA production from this intermediate are discussed separately, in Sections 5.1.1, 5.1.2, and 5.1.3 respectively. Next, the influence of accounting for the difference in volatility distribution of products between the peroxy radical pathways was accounted for in 15 a further model experiment (Multi_nap_yield), which is discussed in Section 5.1.4. This was achieved by increasing the SOA yield from 13 to 66 % for the HO 2 pathway, whilst leaving the reaction yield for the NO pathway unchanged at 13 %.
A stoichiometric yield of 66 % was selected as this allows quantification of the theoretical upper limit of SOA production from this pathway. Note, RO 2 and SOG have relative molecular masses of 100 and 150 g mol -1 , respectively. Because of these differences in relative molecular masses between between reactants and products, the stoichiometiric yield applied to 20 the conversion of RO 2 and SOG is not equivalent to the mass yield. For example, a stoichiometric yield of 66 % corresponds to a mass yield of 100 %. A mass yield of 100 % would be the case if all reacted RO 2 ended up forming non-volatile products (with no addition of oxygen atoms). Next, the influence of parent hydrocarbon reactivity was explored, whilst maintaining identical reaction mechanisms and yields (Section 5.1.5). In this simulation, VOC ANT/BB adopts the reactivity of toluene (Multi_tol_yield) and benzene (Multi_benz_yield) ( Table 3). Note, for the simulations investigating the influence of 25 oxidation mechanisms on SOA, isoprene and monoterpene oxidation is unchanged. The emissions of all SOA precursors (isoprene, monoterpene, and VOC ANT/BB ) are identical in all the simulations.

Observations used to evaluate modelled OA 30
This section describes the observations used to test the effects of variations in hydrocarbon physicochemical processes on model performance. To make direct comparisons, and provide a consistent method for evaluating model performance, a suite of observations were chosen which are identical to those used in previous studies involving the UKCA model (Kelly et al., 2018).
The Aerosol Mass Spectrometer (AMS) allows on-line detection of submicron non-refractory aerosol (Jayne et al., 2000;Canagaratna et al., 2007). This method was used to measure OA concentrations for all observations utilised in this study. Uncertainties associated with this method are estimated to be between 30 and 50 % (Bahreini et al., 2009 (2007) Observed OA concentrations from several aircraft campaigns were also used. Observation data from these aircraft 15 campaigns, which were originally compiled by Heald et al. (2011) can also be accessed on the AMS global network website (https://sites.google.com/site/amsglobaldatabase/). Aircraft observations utilised in this study are also shown in Figure  This mismatch in time may be particularly important for regions influenced by biomass burning as the interannual variability of this emissions source is substantially high (Tsimpidi et al., 2016).

Simulated SOA budget and concentrations
When precursor deposition is neglected from the model, the simulated global annual-total SOA production rate is 75 Tg (SOA) a -1 (Control; Table 3). The inclusion of VOC dry deposition with high surface resistances (High; Table 2) reduces the global annual-total SOA production rate by only 2 Tg (SOA) a -1 (2 %) (Dry_Low ; Table 3). However, the rate of VOC dry deposition is highly sensitive to the value of surface resistance. The inclusion of VOC dry deposition with lower surface 5 resistances (Low; Table 2) reduces the global annual-total SOA production rate by 24 Tg (SOA) a -1 (32 %) (Dry_High ;   Table 3). Therefore, inclusion of precursor dry deposition reduces the global annual-total SOA production rate by 2-24 Tg (SOA) a -1 , or 2-32 %, with this range reflecting uncertainties in surface resistances (Table 3).
Wet removal also has a substantial impact on SOA. For example, under the assumption of an effective Henry's coefficient of 10 5 M atm -1 , wet deposition reduces the global annual-total SOA production rate by 12 Tg (SOA) a -1 (15 %) 10 compared to when no precursors undergo deposition (Wet_Low ; Table 3). However, as discussed in Section 1, H eff has been calculated to range from 10 5 to 10 9 M atm -1 for VOC precursors of SOA from different sources (Hodzic et al., 2014). In this study, when H eff of SOA precursors is increased to 10 9 M atm -1 , wet removal reduces the global annual-total SOA production rate by only 13 Tg (SOA) a -1 (17 %) (Wet_High ; Table 3). Therefore, the influence of precursor wet deposition on SOA is rather insensitive to uncertainties in the range of effective Henry's coefficients. 15 Generally, global (Hodzic et al., 2016) and regional (Bessagnet et al., 2010;Knote et al., 2015) scale modelling studies suggest that dry deposition of precursor dominates over wet deposition. Therefore, for subsequent simulations, where both dry and wet removal were included in the model (DryH_WetL), surface resistances corresponding to Dry_High, which had the largest impact on global SOA production, were used, along with H eff of 10 5 M atm -1 (Wet_Low). The influence of dry and wet deposition of precursors on the global SOA budget are not additive. The combination of dry and wet deposition 20 of VOCs reduces the global annual-total SOA production rate by 28 Tg (SOA) a -1 (37 %) (DryH_WetL ; Table 3). Overall, deposition of SOA precursors has a substantial impact on the global SOA budget, with the global annual-total SOA production rate from all VOC source ranging from 47 to 75 Tg (SOA) a -1 , with the range reflecting uncertainties in precursor deposition (Table 3).
Prior to including deposition of SOA precursors, biogenic VOCs account for 57 % of the global annual-total SOA 25 production rate, with VOC ANT/BB accounting for the remaining 43 %. By including deposition of SOA precursors, the relative importance of biogenic VOCs to global SOA increase; considering deposition of SOA precursors, biogenic VOCs account for 62 % of the global annual-total SOA production rate, with VOC ANT/BB accounting for the remaining 38 %. Hence, biogenic VOCs appear to be less susceptible to deposition than anthropogenic and biomass burning VOCs. Until now, the impacts of precursor deposition on SOA concentrations have only been quantified over Europe (Bessagnet et al., 2010) and North America (Knote et al., 2015), both of which using regional scale models, and treat SOA as The lifetime of SOA precursors with respect to both oxidation and deposition is small. Hence, SOA precursors undergo very little transport before removal. Therefore, dry and wet deposition rates of VOCs are largest over terrestrial 30 environments, where they are released. Across these simulations where the deposition of SOA precursors is altered, the global-average annual-average SOA lifetime varies from 4.3 to 4.7 days (not shown).

Comparison of simulated and observed OA concentrations
In this section, the influence of SOA precursor deposition on model agreement with observations is quantified. First, simulated SOA and OA concentrations are evaluated against surface observations in the northern hemisphere (NH) and southern hemisphere (SH), respectively. Next, vertical profiles of simulated OA concentrations are compared against aircraft observations. 5 Figure 4 shows SOA concentrations for the simulations described in Table 2 Inclusion of precursor deposition further reduces model agreement with observations. As discussed in Section 4.1, including VOC dry deposition reduces the global annual-total SOA production rate by 32 % (24 Tg (SOA) a -1 ), whereas 25 including VOC wet deposition reduces SOA production by 15 % (12 Tg (SOA) a -1 ) (Table 3). Therefore, the model negative  Overall, the inclusion of precursor deposition influences model agreement with observations somewhat. In particular, inclusion of precursor deposition worsens model negative biases with respect to observations in the NH midlatitudes. However, differences between simulated OA concentrations from these simulations is substantially less than the 30 difference between simulated and observed OA. These results highlight that variations in VOC deposition contribute to considerable uncertainty in both the global SOA budget and have some impact on model agreement with observations.

Influence of aromatic oxidation mechanisms on SOA
In this section, the sensitivity of SOA to hydrocarbon oxidation mechanisms is quantified. Here, oxidation mechanisms for anthropogenic and biomass burning VOCs are modified as described in section 2.4.2. To begin with, the influence of anthropogenic and biomass burning VOC oxidation mechanisms on simulated SOA is explored. Next, the impact on model agreement with observations is evaluated. In all simulations, deposition of SOA precursors is included (Table 3), emissions  5 of all SOA precursors are held constant, and the mechanistic description describing the oxidation of biogenic SOA precursors (monoterpene and isoprene) is held fixed, following Eq (8).

Simulated SOA budget and concentrations
Firstly, the single-step oxidation mechanism of VOC ANT/BB with reactivity based on α-pinene and a fixed reaction yield of 13 10 % is described (DryH_WetL). The global annual-total reaction fluxes and SOA production rates from anthropogenic and biomass burning hydrocarbons are shown in Figure 7. As described in Section 2.3, the global annual-total VOC ANT/BB emission rate is 176 (VOC ANT/BB ) a -1 , which is held fixed across all simulations. In this case, the global annual-total VOC ANT/BB oxidation rate by OH is 94 Tg (VOC) a -1 (DryH_WetL; Figure 7). The remaining 82 Tg (VOC) a -1 undergoes deposition (not shown). For this single-step mechanism, oxidation of the emitted parent hydrocarbon directly forms the non-15 volatile product, SOG, which condenses almost immediately. A fixed reaction yield of 13 % is assumed, resulting in a global annual-total SOA production rate of 18.4 Tg (SOA) a -1 (Figure 7). Note, due to differences in relative molecular masses for VOC ANT/BB and SOG, the stoichiometric yield is not equivalent to the mass yield. Expressed as a fraction of emitted parent VOC (176 Tg (VOC) a -1 ), the overall yield of SOA production from anthropogenic and biomass burning VOCs (18.4 Tg (SOA) a -1 ) is around 10 %. 20 The combination of a single step oxidation mechanism and the assumption of a relatively reactive parent hydrocarbon results in rapid production of SOA. Figure 8 shows the spatial distributions of annual-total surface VOC ANT/BB emissions, annual-average surface OH concentrations, annual-total vertically integrated VOC ANT/BB +OH oxidation rates, and the resulting SOA production rates. As expected, the spatial distributions of VOC ANT/BB emissions mainly reflects anthropogenic activity. Over high emissions regions, OH concentrations are also relatively high. Over India, China, Europe 25 and North America, annual-average OH concentrations are in the range of 32 -130 x10 -3 ppt(v) (Figure 8 b). Therefore, for most major VOC ANT/BB emissions source regions, OH availability is high, resulting in rapid oxidation; reaction fluxes of VOC ANT/BB +OH peak very close to emissions sources (c.f. Figure 8 a, c). However, uncertainty in simulated OH concentrations will be translated into uncertainty in SOA production. OH is the principal oxidising agent of the atmosphere. Therefore, in order to successfully model OH, many other species (e.g. methane) also need to be modelled correctly 30 (Lelieveld et al., 2016). Due to its very short lifetime (~seconds) and low concentrations, OH is difficult to measure (Stone et al., 2012). Alternatively, the OH concentration can be constrained indirectly from the CH 4 lifetime. Overall, the OH concentration is a difficulty quantity to capture in a global model. Also, as shown in Eq (8), oxidation of the parent VOC results in immediate production of the condensing species, SOG. Hence, not only do parent VOCs undergo rapid oxidation, but the product of this reaction is in the form of condensable organic vapours. Therefore, this combination of high parent VOC reactivity with few reaction steps results in extremely 5 localised SOA production from anthropogenic and biomass burning emissions. This is in contrast to other global modelling studies, which predict more regionally distributed SOA production (Pye and Seinfeld, 2010;Tsimpidi et al., 2016).
Differences in the geographical extent to which SOA production occurs may be attributed to precursor reactivity and the number of reaction intermediates. For example, here, the parent hydrocarbon is a VOC, with a rate constant of 52.9 x10 -12 cm 3 molecule -1 s -1 at 298 K (Table 1), forming SOA in a single-step reaction mechanism. Hence, local SOA production is 10 simulated (Figure 8 d). Conversely, SOA production is more regionally distributed when treated from S/IVOC multigenerational chemistry, where the parent hydrocarbon and oxidation products all react relatively slowly (Tsimpidi et al., 2016). High observed OA concentrations over remote regions (Boreddy et al., 2015;Boreddy et al., 2016) provide evidence for the slow and sustained mechanistic description of SOA production from S/IVOCs (Tsimpidi et al., 2016). High observed OA concentrations within industrialised emissions source regions  support the fast mechanistic 15 description of SOA production from VOCs simulated here.
To summarise, the combination of fast reactivity and a single step oxidation mechanism favours extremely localised SOA production, with parent VOCs undergoing rapid oxidation and subsequent condensation close to source.

Initial OH oxidation of parent hydrocarbon
Production of SOA from anthropogenic and biomass burning hydrocarbons is modified in the following sub-sections to follow the mechanism of Eq (9) which include the reaction intermediate. Naphthalene, the most reactive aromatic VOC considered in this study, is first selected (section 2.4.2), with identical reaction yields applied to both RO 2 pathways (Multi_nap simulation; Table 3).
The initial reaction of VOC ANT/BB with OH is compared to that of a single oxidation reaction step (DryH_WetL ;   Table 3). At 298 K, the rate constants for α-pinene and naphthalene oxidation by OH are 52.9 and 23.3 x10 -12 cm 3 molecule -1 s -1 , respectively (Table 1 and 3). The global annual-total VOC ANT/BB oxidation rate reduces by 3 Tg (VOC) a -1 (or 3 %), from 5 94 Tg (VOC) a -1 using the reactivity of α-pinene, to 91 Tg (VOC) a -1 using the reactivity of naphthalene (Figure 7). Therefore, the global VOC ANT/BB oxidation rate is relatively insensitive to a ~50 % reduction in reactivity. When applying a 13 % stoichiometric yield to this reaction sequence (Table 3), this reduction in parent VOC oxidation rate contributes to a marginal change in the global annual-total SOA production rate (0.6 Tg (SOA) a -1 ).
The response of regional VOC oxidation rates to a ~50 % reduction in the reactivity vary in both magnitude and 10 sign. Figure 9 shows the difference in annual-total vertically integrated VOC ANT/BB oxidation rates for the mechanism which include the reaction intermediate (Table 3), relative to the mechanism which doesn not include the reaction intermediate with reactivity based on α-pinene (DryH_WetL; Table 3). Reduced chemical reactivity lowers oxidation rates within emission source regions. For example, over India and parts of Africa, annual-total VOC ANT/BB oxidation rates reduce by up to 0.05 Tg (VOC ANT/BB ) a -1 (Figure 9 a); these changes in annual-total VOC ANT/BB oxidation rates within emissions source 15 regions correspond to reductions between 10 and 30 % (not shown). By contrast, downwind of many emissions source regions, the lower reactivity acts to enhance VOC ANT/BB oxidation rates. For example, over the Arabian Sea, over Southeast China, off the coast of Nigeria, and over the southeast USA, annual-total VOC ANT/BB oxidation rates increase by 0.001 -0.05 Tg (VOC ANT/BB ) a -1 in response to a ~50 % reduction in parent VOC reactivity (Figure 9 a). These changes in annual-total VOC ANT/BB oxidation rates downwind of emissions source regions correspond to reductions which exceed 60 % (not shown). 20 As discussed in Section 5.1, adoption of the reactivity of α-pinene for the VOC ANT/BB +OH reaction results in peak VOC oxidation rates at emission source, with VOCs undergoing very little transport (Figure 8 c). Therefore, by reducing the reactivity by ~50 %, fewer VOC ANT/BB are oxidised at source but transport of VOC ANT/BB away from source is promoted.

Chemical fate of the new reaction intermediate, RO 2 25
Oxidation of the parent VOC forms a new reaction intermediate, the peroxy radical RO 2 . In this case, VOC ANT/BB oxidation results in a global annual-total peroxy radical production rate of 91 Tg (RO 2 ) a -1 (Multi_nap simulation; Figure 7). Introduction of this new reaction intermediate has the potential to either reduce and/or delay SOA production, depending on assumptions regarding the strength of deposition and chemical reactivity of this intermediate. For example, SOA production would be reduced if the peroxy radical undergoes significant deposition, which is dependent on deposition parameters such 30 as surface resistances and solubility (section 2.4.1). Additionally, SOA production could be reduced or delayed if the chemical removal of RO 2 is slow. The influence of introducing the peroxy radical as a reaction intermediate is therefore predetermined by assumptions in deposition parameters and reaction kinetics. In all simulations, RO 2 is assumed to have identical solubility and surface resistances to all other SOA precursors, H eff = 10 5 M atm -1 and 'Low' surface resistances (Table 2). At 298 K, the rate constants for RO 2 oxidation by HO 2 and NO, taken from Atkinson and Arey (2003), are 14.8 and 8.5 x10 -12 cm 3 molecule -1 s -1 , respectively (Table 1). Consequently, of the 91 Tg of RO 2 generated annually, oxidation by NO and HO 2 removes 57 and 34 Tg (RO 2 ) a -1 , respectively (Multi_nap_yield; Figure 7). Deposition of RO 2 is inconsequential at 0.1 Tg (RO 2 ) a -1 (not shown). This extremely low deposition rate is because the chemical removal of the 5 peroxy radical is extremely fast. The global annual-average lifetime of RO 2 with respect to oxidation is ~1 day, which is relatively short in comparison to atmospheric transport timescales. Note, a review of laboratory studies suggests the lifetime of RO2 could be of the order of minutes (Orlando et al., 2012). Therefore, due to marginal deposition and fast oxidation, introduction of the peroxy radical reaction intermediate will probably have no effect on either the SOA production rate or the geographical distribution of SOA production, which are both quantified in the following section (5.1.3). 10 Chemical removal of the peroxy radical via the two oxidative pathways is an important factor in governing the strength of SOA production, as discussed later in Sections 5.1.3 to 5.1.5. RO 2 is chiefly removed by NO, as opposed to HO 2 radicals. This is demonstrated in Figure 10, which shows the relative contributions of the HO 2 and NO peroxy radical oxidative pathways to the total chemical removal of RO 2 (top row) and to SOA production (bottom row). On a global and annual mean basis, removal by NO accounts for 62 % of RO 2 chemical loss (Figure 10 a). Other global modelling studies 15 which consider the peroxy radical as a reaction intermediate from aromatic compounds or IVOCs, also predict RO 2 removal to be dominated by NO. Henze et al. (2008) estimate that, for peroxy radicals generated from benzene, xylene and toluene, 61 % react via the NO pathway. Peroxy radicals generated from IVOCs, with parent hydrocarbon reactivity based on naphthalene, 66 % are consumed by NO (Pye and Seinfeld, 2010). These results suggest that the chemical fate of the peroxy radical is robust despite the likelihood of variations in precursor emissions and oxidant concentrations between this and the 20 aforementioned studies.
The substantial preference for RO 2 radicals to react via the NO pathway instead of the HO 2 pathway can be attributed to differences in oxidant availability (i.e. concentrations) and in reaction rates. Note, in the UKCA model, HO 2 is assumed to undergo wet removal. Firstly, consider the difference in oxidant levels. Figure 11 shows the spatial distribution of annual-average surface concentrations of NO and HO 2 , as well as the ratios NO/HO 2 and (k RO2+NO x NO) / (k RO2+HO2 x 25 HO 2 ). NO is extremely spatially heterogeneous ( Figure 11). Within the model, sources of NOx include the prescribed anthropogenic, biomass burning and soil emissions, as well as lightning-NOx which is calculated interactively. At the surface, the highest annual-average surface NO concentrations (1-23 ppb(v)) are simulated over industrialised and urban regions of North America, China and Europe (Figure 11 a). Over remote marine environments, away from anthropogenic and biomass burning sources, concentrations of NO are low (Figure 11a). In contrast, concentrations of HO 2 are much lower 30 and more evenly distributed across the surface (Figure 11). Over the majority of both continental and marine regions, annualaverage surface HO 2 concentrations range between 2 and 23 ppt(v) (Figure 11b). Therefore, over most environments, NO concentrations are far greater, with annual-average surface NO concentrations ranging from 10 (NO/HO 2 = 10 1 ) to 10,000 (NO/HO 2 = 10 4 ) times more than HO 2 (Figure 11 c). Only in the remote marine environments are HO 2 levels higher in absolute magnitude compared to NO, with simulated annual-average surface HO 2 concentrations reaching 10 times that of NO (NO/HO 2 = 10 -1 ; Figure 11 c). At higher levels, NO/HO 2 reduces, suggesting an increasing importance of the HO 2 pathway at higher altitudes. However, due to the fast chemical reactivity, the majority of SOA production occurs at the surface. High altitude emissions of VOCs from biomass burning plumes may be more suscebitble to forming RO 2 which react with HO 2 . However, in this study, all VOCA NT/BB are emitted at the surface. For the majority of the atmosphere, the 5 difference in the magnitudes of the oxidant concentrations favours the RO 2 +NO pathway over the RO 2 +HO 2 pathway.
Differences in reactivity of RO 2 with respect to the oxidants also affects the fate of this radical. At 298 K, the rate constant for RO 2 +NO is 8.42 x 10 -12 cm 3 molecule -1 s -1 , almost half that of RO 2 +HO 2 (k(298 K) = 14.7 x10 -12 cm 3 molecule -1 s -1 ; Table 1). Therefore, the higher rate constant for oxidation by HO 2 in comparison to NO favours the RO 2 +HO 2 pathway.
The ratio, (k RO2+NO x NO) / (k RO2+HO2 x HO 2 ), combines the difference in rate constants together with differences in 10 the ratio of oxidant concentrations, and ranges from 10 0 to 10 4 over most continental regions, but is as low as 10 -2 over remote marine environments, such as the Pacific Ocean and South Atlantic Ocean (Figure 11 d). Hence, the net effect of differences in oxidant concentrations and rate constants is to favour peroxy radical removal via the NO oxidative pathway ( Figure 10 a; Figure 11d). This preference for the NO radical pathway is enhanced even further by considering the likelihood of RO 2 being co-located with NO. RO 2 is a second generation oxidation product of VOC ANT/BB , which is released by 15 anthropogenic and biomass burning sources. NO emissions are predominantly emitted from anthropogenic and biomass burning sources. Therefore, peroxy radicals are very likely to be formed in NO-rich environments, further favouring the probability of entering the RO 2 +NO pathway. Furthermore, adoption of naphthalene reactivity for VOC ANT/BB , which is still relatively high, prevents transport away from high-NO regions. Overall, peroxy radicals preferentially react via the NO pathway due to relatively higher NO concentrations than HO 2 , despite the HO 2 pathway having a higher rate constant. 20

Production of SOA from new reaction intermediate, RO 2
For this mechanism with parent VOC reactivity based on naphthalene (Multi_nap), the initial oxidation and subsequent reaction of the intermediate were discussed in Sections 5.1.1 and 5.1.2, respectively. In this section, the production of SOA 25 from this mechanism is examined. In this oxidation scheme, identical reaction yields of 13 % are applied for both the HO 2 and NO pathways. For the RO 2 +NO reaction, a global annual-total reaction flux of 57 Tg (RO 2 ) a -1 results in an SOA production rate of 11 Tg (SOA) a -1 (Multi_nap; Figure 7). Similarly, for the RO 2 +HO 2 pathway, a global annual-total reaction flux of 34 Tg (RO 2 ) a -1 results in an SOA production rate of 7 Tg (SOA) a -1 (Figure 7). Hence, the relative contribution of the RO 2 oxidative pathways to SOA production is simply a reflection of the relative contribution of each 30 pathway to RO 2 consumption. Therefore, the RO 2 +NO pathway accounts for 62 % of the global annual-total RO 2 oxidation rate (Figure 10 a), and also accounts for 62 % of the annual-total SOA production rate from anthropogenic and biomass burning hydrocarbons (Figure 10 e). The sum of global annual-total SOA production from anthropogenic and biomass burning sources, from both oxidative pathways, is 17.8 Tg (SOA) a -1 (Figure 7). This is just 0.6 Tg (SOA) a -1 (or 3 %) less than the global annual-total SOA production rate when using a single-step oxidation mechanisms with reactivity based on αpinene (DryH_WetL; Figure 7). Note, this 0.6 Tg (SOA) a -1 reduction in SOA production is solely due to the 3 % reduction in the VOC ANT/BB oxidation rate (Section 5.1.1.). This therefore confirms that, due to the marginal deposition rate of RO 2 , the introduction of the reaction intermediate has no effect on global SOA production. 5 The difference in annual-average surface SOA concentrations for the mechanisms with the reaction intermediate relative to the mechanisms without the reaction intermediate with reactivity based on α-pinene are shown in Figure 12. The effects of a ~50 % reduction in parent VOC reactivity in combination with the introduction of the reaction intermediate on regional annual-average surface SOA concentrations vary in both magnitude and sign but, generally, are small. These differences in SOA concentrations (Figure 12 a and b) closely resemble differences in parent VOC oxidation rates in 10 response to the change in chemical reactivity (Figure 9 a). Over regions where reduced reactivity has lowered VOC ANT/BB oxidation rates, such as India and and industrialised parts of Africa (Figure 9 a) To summarise, moving from a mechanism with no reaction intermediate and with the reactivity of α-pinene and 20 with a single SOA yield, to a mechanism with a reaction intermediate , based on naphthalene , and a single SOA yield has very little effect on SOA production and surface concentrations. The slower reactivity of naphthalene reduces the global VOC ANT/BB oxidation by 3%, contributing to a reduction in the global annual-total SOA production rate of 0.6 Tg (SOA) a -1 (3 %). Introduction of the reaction intermediate, but with no change to reaction yields, has no effect on global SOA. 25

Accounting for the difference in volatility between HO 2 and NO oxidation products
In this section, the effects of accounting for the difference in volatility between RO 2 oxidation products is examined. This is done by altering the reaction yields for RO 2 reactions, whilst maintaining the same chemical mechanism (Eq (9)) and precursor emission rate. As discussed in Section 1, for aromatic compounds, the volatility and, therefore, the amount of SOA 30 produced, depends on the concentrations of NOx (Hurley et al., 2001;Song et al., 2005;Ng et al., 2007;Chan et al., 2009).
One explanation for this relationship is that the HO 2 pathway forms functioanlised products, whereas the NO pathway forms fragmented products. Functionalisation leads to reductions in volatility, whereas fragmentation leads to increases in volatility. Hence, the SOA yield under low-NO X coditions is higher than under high-NO X conditions. Hence in a further simulation, the difference in fragmentation/functionalization between products of different peroxy radical oxidation pathways are accounted for, whereby the yield for the RO 2 +HO 2 reaction is increased from 13 to 66 %, whilst the yield for the RO 2 +NO reaction is left at 13 % (Multi_nap_yield; Table 3). As discussed in Section 2.5, the assumption of a 66 % stoichiometric reaction yield was selected as it corresponds to a 100 % mass yield and therefore allowing the theoretical 5 upper limit of SOA production via the HO 2 pathway to be quantified whilst conserving mass. With a higher molar yield of 66 %, global SOA production from the RO 2 +HO 2 reaction increases to 34 Tg (SOA) a -1 as compared to 7 Tg (SOA) a -1 using a 13 % yield for this reaction (Figure 7). As a consequence of this increase to the hydroperoxyl reaction yield, the HO 2 pathway now accounts for 75 % of SOA production from anthropogenic and biomass burning sources (Figure 10 f), despite only 38 % of the RO 2 radicals reacting via this pathway (Figure 10 b). This is in remarkably good agreement with previous 10 studies. Pye and Seinfeld (2010) also estimate that the HO 2 pathway accounts for 75 % of SOA production from I-VOCs. In addition, Henze et al. (2008) estimates that, for SOA production from benzene, toluene and xylene, 72 % is produced via the HO 2 pathway. Accounting for differences in volatility between RO 2 oxidation products increases the global SOA production rate by 27.3 Tg (SOA) a -1 (or 153 %), from 17.8 Tg (SOA) a -1 when a molar yield of 13 % is applied to both pathways 15 (Multi_nap), to 45.1 Tg (SOA) a -1 when a molar yield of 66 % is applied (Multi_nap_yield). Under these conditions, the overall aerosol yield from anthropogenic and biomass burning VOC emissions is 25 %, which lies within the range from other modelling studies, either based on explicit aromatic compounds or IVOCs, which range from 22 -30 % (Henze et al., 2008;Pye and Seinfeld, 2010).
The relative spatial homogeneity of HO 2 radicals over land and ocean, as shown in Figure 11b, suggests that 20 increasing the yield for this pathway could lead to enhanced SOA production globally. However, as discussed in Section 5.1.1, the naphthalene+OH rate constant results in relatively fast oxidation rates. Therefore, RO 2 radicals are still being generated close to the emissions source. For these reasons, increasing the reaction yield for the HO 2 reaction pathway increases SOA concentrations mainly over major anthropogenic emission source regions (Figure 12 c, d). In response to this increased yield, over India, China, Africa and Europe, annual-average surface SOA concentrations have increased by 0.5 -8 25 µg (SOA) m -3 (Figure 12 c), corresponding to increases of 10 -100 % (Figure 12 d). Note, differences in SOA concentrations are positive everywhere, whereas both positive and negative changes were found when comparing differences in SOA concentrations between. In summary, both globally ( Figure 7) and regionally (Figure 12 d), when accounting for the different SOA yields for the RO 2 oxidative pathways, despite a reduction in VOC ANT/BB global SOA production rates, surface SOA concentrations increase everywhere. Therefore, the lower reactivity in VOC ANT/BB is compensated for by lower 30 volatility products from the HO 2 oxidation pathway leading to net increases in modelled SOA.

Production of SOA from less reactive hydrocarbons
As discussed in Section 2.4, VOC ANT/BB is a lumped species, and, hence, represents a mixture of species with a range of physicochemical properties. In this section, the uncertainty related to its chemical reactivity and the effects on SOA production are explored. At 298 K, the rate constant for aromatic compounds with respect to OH oxidation ranges from 1.22 to 23.2 x10 -12 cm 3 molecule -1 s -1 , respectively (Table 1 and 3). Therefore, adoption of the naphthalene reactivity represents an 5 upper limit for the VOC ANT/BB oxidation rate when considering SOA relevant aromatic compounds. In this section, the VOC reactivity is varied across a series of different aromatic compounds: naphthalene, toluene and benzene (Multi_nap_yield, Multi_tol_yield and Mult_benz_yield; Table 3). However, the mechanistic description and stoichiometric yields describing SOA formation from VOC ANT/BB are identical and follow Eq (9).
Firstly, consider how reactivity affects SOA production among the oxidation mechanisms which include the 10 reaction intermediate (Multi_nap_yield, Multi_tol_yield and Multi_benze_yield). Reducing the chemical reactivity of VOC ANT/BB reduces the global oxidation rate, whilst at the same time, favours the likelihood of RO 2 radicals entering the HO 2 pathway (which has a higher SOA yield than the NO pathway). The global annual-total VOC ANT/BB oxidation rates are 91, 65 and 32 Tg (VOC ANT/BB ) a -1 using the reactivity of naphthalene, toluene and benzene, respectively (Figure 7). Hence, as reactivity is reduced, oxidation is lowered at the expense of deposition. In response to this reduced oxidation rate, fewer RO 2 15 radicals are being generated, which therefore, drives reductions in SOA production. The global annual-total SOA production rates are 45.1, 34.0, 17.9 Tg (SOA) a -1 using the reactivity of naphthalene, toluene and benzene, respectively (Figure 7). However, as the reactivity is reduced, the chances of RO 2 radicals entering the high-yield HO 2 pathway is increased, therefore, slightly offsetting the effects of the reduced RO 2 production rate. The fraction of peroxy radicals entering the HO 2 pathway is 38, 41 and 46 % using the reactivity of naphthalene, toluene and benzene, respectively (Figure 10 d, e and h, 20 respectively). As shown in Figure 11 d, the HO 2 pathway dominates only in remote marine environments. Hence, as the reactivity of the parent hydrocarbon is reduced, VOC ANT/BB oxidation rates close to emissions sources reduce, but increase further downwind (Figure 9 c and d). Therefore, lower reactivity enhances the likelihood of peroxy radicals being generated downwind of emissions sources, where the HO 2 pathway is favoured. These findings are consistent with Henze et al. (2008), who predicted increased fluxes through the HO 2 pathway for peroxy radicals derived from less reactive parent aromatic 25 hydrocarbons. Overall, reduced parent hydrocarbon reactivity reduces the sources of peroxy radicals but favours lower volatility RO 2 + HO 2 oxidation products.
Secondly, consider the net effects of using aromatic oxidation to describe SOA production from VOC ANT/BB (Multi_nap_yield, Multi_tol_yield and Multi_benze_yield), versus using the single-step mechanism with reactivity based on α-pinene (DryH_WetL). Compared to α-pinene, the aromatic compounds, naphthalene, toluene and benzene are 50, 75 and 30 95 % less reactive, respectively ( Table 2). As discussed in Section 5.1.3, using the chemical reactivity of naphthalene compared to monoterpene leads to a 3 % reduction in VOC ANT/BB oxidation, which drives a 0.6 Tg (SOA) a -1 (1 %) reduction in global annual-total SOA production (c.f. DryH_WetL and Multi_nap; Figure 7). However, as shown in Section 5.1.4, this reduction in VOC oxidation is entirely offset by accounting for the high-yield pathway of the RO 2 +HO 2 reaction, leading to a 27.3 Tg (SOA) a -1 (153 %) increase in global annual-total SOA production (c.f. DryH_WetL and Multi_nap_yield; Figure   7). Using the chemical reactivity of toluene compared to α-pinene also reduces the VOC ANT/BB oxidation, but this time by 31 % (c.f. DryH_WetL and Mutli_tol_yield; Figure 7). However, similar to the case of naphthalene, this reduction in VOC ANT/BB oxidation is still outweighed by accounting for the high-yield HO 2 pathway, such that global annual-total SOA 5 production increases by 15.6 Tg (SOA) a -1 (or 85 %), from 18.4 Tg (SOA) a -1 in the single step oxidation mechanism based on α-pinene, to 34.0 Tg (SOA) a -1 in the multi-step oxidation mechanisms based on toluene (c.f. DryH_WetL and Mutli_tol_yield; Figure 7). On the other hand, benzene is considerably less reactive than α-pinene, leading to 66 % reduction in the global annual-total VOC ANT/BB oxidation rate (c.f. DryH_WetL and Mutli_benz_yield; Figure 7). In this case, the reduction in VOC ANT/BB oxidation is so large, that it is not compensated for by accounting for the difference in volatility 10 between RO 2 oxidation products. Hence, using the reactivity of benzene, the global annual-total SOA production rate reduces by 0.5 Tg (SOA) a -1 (or 3 %), from 18.4 Tg (SOA) a -1 in the single step oxidation mechanism based on α-pinene, to 17.9 Tg (SOA) a -1 in the multi-step oxidation mechanisms based on benzene (c.f. DryH_WetL and Mutli_benz_yield; Figure   7). These results demonstrate how, from a global perspective, the combined effects of introduction of the peroxy radical intermediate which also accounts for the difference in SOA yields between HO 2 and NO pathways can either lead to an 15 increase (Multi_nap_yield and Multi_tol_yield) or reduction (Multi_benze_yield) in SOA production that, critically, depends on the assumed chemical reactivity of the parent VOC.
The spatial distribution of SOA is also influenced by these changes in VOC ANT/BB oxidation mechanisms. For cases where reactivity is based on either naphthalene (Figure 12 c and d) or toluene ( Figure 12 e and f), accounting for the high yield HO 2 pathway compensates for reduced reactivity, such that annual-average surface SOA concentrations increase 20 globally in comparison to the single step oxidation mechanism with reactivity based on α-pinene (DryH_WetL). The spatial pattern for the oxidation mechanism based on benzene (Multi_benz_yield) and the oxidation mechanism based on α-pinene (DryH_WetL) are also very different ( Figure 12 g and h), despite only a small difference in the global annual-total SOA production rate (Figure 7); under the reactivity of benzene, VOC ANT/BB is slowed, and newly introduced RO 2 radicals are being formed in downwind environments, leading to reduced SOA concentrations in emissions sources regions, but 25 increased SOA concentrations downwind. Over emissions source regions, such as China, India and North America, annualaverage surface SOA concentrations are lower by up to 4 µg (SOA) m -3 (Figure 12 g). Over continental outflow regions, such as the Arabian Sea and the Bay of Bengal, annual-average surface SOA concentrations have increased by 0.1 -0.5 µg (SOA) m -3 (Figure 12 h). Although the global annual-total SOA production rates are identical, the global annual-average SOA burden is 10 % greater when using benzene as the parent VOC, highlighting the strong spatial gradients in SOA 30 lifetime. Across these simulations where the VOC ANT/BB oxidation scheme is varied, the global-average annual-average SOA lifetime varies from 4.4 to 5.0 days (not shown).
The spatial pattern simulated under in the oxidation mechanism with the reaction intermediate and with reactivity based on benzene is in greater agreement with the more regionally distributed SOA concentrations simulated in models based on S/IVOC sources (Pye and Seinfeld, 2010;Tsimpidi et al., 2016).

Comparison of simulated and observed OA concentrations
In this section, the influence of anthropogenic and biomass burning hydrocarbon oxidation mechanisms on model agreement with observations is quantified. Reduced parent hydrocarbon reactivity combined with accounting for the different SOA yield pathways of the peroxy radical affects model agreement with observations. Figure 13 shows simulated versus observed surface SOA concentrations for the NH from the simulations described in Table 3. In the oxidation mechnaimss which 10 include the reaction intermediate, using naphthalene and toluene, the annual-total SOA production rate increased relative to the single step fast oxidation pathway. This increase was due to the difference in volatility between products of the peroxy radical oxidation pathways, despite the reduction in parent hydrocarbon reactivity. Therefore, simulated SOA concentrations are in closer agreement to observations (Multi_nap_yield; NMB = -46 %; Figure  Global annual-total emissions of benzne and toluene are 5.6 and and 6.9 Tg (C) a -1 , respectively (Henze et al., 2008), whereas emissions of naphthalene are 0.22 Tg (C) a -1 (Pye and Seinfeld, 2010). This suggests benzene and tolune could be more realistic surrogate compounds to represent VOC ANT/BB mchemistry, as opposed to naphthalene. This is due to the slow reactivity of benzene resulting in a 20 small VOC ANT/BB oxidation rate, which is higher downwind of emissions compared to the point of emissions (Figure 12 h). has no effect on the NMB. However, accounting for the with reactivity based on naphthalene or toluene, the NMB reduces (Multi_nap_yield and Multi_tol_yield), but the NMB increases when the reactivity is based on benzene (Multi_benz_yield). Contrastingly, model performance in Europe and North 30 Amercia = (Figure 6 h -j) remains similar as VOC ANT/BB oxidation is modified. This warrants further discussion. As explained in previous sections, the global SOA production rate is extremely sensitive to the mechanisms of VOC ANT/BB oxidation. However, model performance over the pollution and biomass burning influenced regions is relatively insensitive to VOC oxidation mechanisms. This is likely to be a reflection of the location of aircraft campaigns and how they are categorised. For example, the aircraft campaigns categorised as influenced by biomass burning are in North America, but peak biomass burning emissions are located over tropical forest regions of South America and Africa. Furthermore, the aircraft campaigns categorised as influenced by pollution are all in Europe. Again, this does not correspond to the location of peak anthropogenic emissions over Asia. Therefore, mechanisms of anthropogenic and biomass burning oxidation have 5 substantial impacts on simulated SOA production rates, but almost no effect on model agreement with aircraft observations in 'pollution and biomass burning influenced' regions, due to a lack of aircraft coverage.

Conclusions
In this study, the description of both deposition and oxidation for SOA precursors was developed in a global chemistry-10 climate model. Several model integrations were conducted and the treatments of deposition and oxidation mechanisms of SOA precursors were varied. Subsequent effects on the global SOA budget were quantified and simulated OA was evaluated against a suite of surface and aircraft campaigns spanning both the southern and northern hemispheres.
Within UKCA, SOA formation is considered from VOCs -monoterpene, isoprene, a lumped anthropogenic VOC (VOC ANT ) and a lumped biomass burning VOC (VOC BB ). Under the assumption that no precursors undergo deposition, the 15 global annual-total SOA production rate is 75 Tg (SOA) a -1 and simulated OA concentrations are generally lower than observed (NMB = -50 %). Extending deposition to include SOA precursors has substantial impacts on both the global SOA budget and model agreement with observations. Including SOA precursor dry deposition reduces the global annual-total SOA production rate by 2 -24 Tg (SOA) a -1 (2 -32 %), with the range reflecting uncertainties in surface resistances.
Including SOA precursor wet deposition reduces the global annual-total SOA production rate by 12 Tg (SOA) a -1 (15 %) and 20 is relatively insensitive to changes in effective Henry's Law coefficient. The effects of dry and wet deposition on the global SOA budget are not additive; the inclusion of both these processes reduces the global annual-total SOA production rate by 28 Tg (SOA) a -1 (37 %). Inclusion of VOC deposition generally increases model negative biases with respect to observations. For SOA, across northern hemisphere mid-latitude sites, inclusion of both dry and wet deposition of VOCs increases the NMB from -50 to -66 %. However, for OA, over Manaus (Brazil), when precursor deposition is neglected from 25 the model, simulated OA concentrations exceed observed OA concentrations.
Production of SOA from aromatic compounds, which are typically emitted from anthropogenic and biomass burning activities, has been partially elucidated by environmental chamber studies. Briefly, parent aromatic hydrocarbons are oxidised by the hydroxyl radical (OH) to form a reaction intermediate, the peroxy radical (RO 2 ). RO 2 undergoes competitive reactions; with HO 2 the products are non-volatile, whereas with NO the products are semi-volatile. Hence, higher HO 2 30 concentrations favour higher yields of SOA.
The influence of VOC oxidation mechanisms on the global SOA budget was also examined. For the anthropogenic and biomass burning sources of SOA (VOC ANT/BB ), a series of simulations were performed with varying a) parent hydrocarbon reactivity, b) number of reaction intermediates, and c) accounting for differences in volatility between oxidation products from various pathways. The global annual-total SOA production rate from anthropogenic and biomass burning sources is 18.4 Tg (SOA) a -1 when the parent hydrocarbon, VOC ANT/BB , undergoes a single-step oxidation, with a fixed reaction yield of 13 %, and a reactivity based on α-pinene. Using the reactivity of naphthalene, toluene or benzene, the global annual-total VOC ANT/BB oxidation rate changes by -3, -31 or -66 %, respectively, when compared to using the reactivity of α-5 pinene. Increasing the number of reaction intermediates, by including RO 2 as a product of VOC ANT/BB oxidation, slightly delays SOA production but has no effect on the global SOA production rate. Hence, when the reactivity of VOC ANT/BB is reduced from α-pinene to naphthalene, in combination with the introduction of the reaction intermediate, the global annualtotal SOA production rates changes by just -0.6 Tg (SOA) a -1 (or -3 %), from 18.4 Tg (SOA) a -1 to is 17.8 Tg (SOA) a -1 .
However, the subsequent competitive chemical reactions of RO 2 control the volatility distribution of products. To account 10 for this, the reaction yield for the RO 2 +HO 2 pathway was increased from 13 to 66 %. The reaction yield for the RO 2 +NO pathway was left unchanged, at 13 %. Accounting for the difference in volatility between RO 2 products increases the global annual-total SOA production rate from anthropogenic and biomass burning by 153 %, from 17.8 Tg (SOA) a -1 in the simulation with yields of 13 % for both RO 2 reactions, to 45.1 Tg (SOA) a -1 when the yield for the RO 2 +HO 2 is increased 66 %. 15 Overall, the effects of using aromatic oxidation to describe SOA formation from anthropogenic and biomass burning compounds versus using a single-step mechanism with reactivity based on α-pinene, can be explained in terms of reductions in parent VOC reactivity and accounting for the high-yield HO 2 pathway, as opposed to the introduction of the reaction intermediate. For both naphthalene and toluene, reduced reactivity in comparison to α-pinene is small, and is entirely offset by accounting for the difference in volatility between RO 2 oxidation products. By contrast, benzene is 20 significantly less reactive than α-pinene, and accounting for the different in volatility between RO 2 oxidation products cannot outweigh this. For example, for naphthalene, changes in oxidation rate (-3 %) are outweighed by accounting for the difference in volatility between RO 2 reactions, such that the global annual-total SOA production rate changes by 27.3 Tg for the case of benzene, the substantial change in oxidation rate (-66 %) is not outweighed by accounting for the difference in 30 volatility between RO 2 reactions, such that the global annual-total SOA production rate changes by -0.5 Tg (SOA) a -1 (or -3 %), from 18.4 Tg (SOA) a -1 in the oxidation mechanism with no reaction intermediate and reactivity on α-pinene, to 17.9 Tg (SOA) a -1 in the oxidation mechanisms including the reaction intermediate and with reactivity of benzene. Therefore, from a global perspective, the net effects of increased reaction steps and accounting for the influence of NOx on reaction yields, can either increase (85 -150 %) of reduce (-3 %) SOA production depending on the assumed chemical reactivity of the parent VOC.
These variations in oxidation mechanisms can either improve or worsen model agreement with observations, depending on the chemical reactivity of the parent VOC. In the absence of the reaction intermediate, and with reactivity based on α-pinene, the model underestimates SOA across northern hemisphere mid-latitudes, with an NMB of -66 %. 5 However, the inclusion of the perxy radcial, combined with accounting for the difference in SOA yields for the peroxy radcial reaction pathways, and reactivity based on either naphthalene, toluene or benzene, the NMB across northern hemisphere mid-latitudes is either -46, -56 or -71 %, respectively. These results highlight how, increases to reaction intermediates and accounting for the influence of NOx, has the ability to both improve and worsen model agreement with observations which, crucially, depends on the assumed chemical reactivity of the parent VOC. Global annual-total emissions 10 of benzene and toluene are 5.6 and and 6.9 Tg (C) a -1 , respectively (Henze et al., 2008), whereas emissions of naphthalene are 0.22 Tg (C) a -1 (Pye and Seinfeld, 2010). This suggests benzene and toluene could be more relasitc surrogate compounds to represent VOC ANT/BB mchemistry, as opposed to naphthalene. Note however, that aromatic compound emissions represent only a minor fraction of the global annual-total VOC ANT/BB emission rate, which is 176 Tg (VOC ANT/BB ) a -1 .
In this study, observed OA/SOA concentrations generally exceed simulated OA/SOA concentrations. This is true at 15 the surface and throughout the boundary layer. This model negative bias is very likely due to missing SOA (a) S/IVOC emissions, and (b) aqueous phase SOA production. As a result of these missing SOA source, care should be given when drawing conclusions on how variations in VOC deposition and oxidation mechanisms impact model agreement with observations. For instance, this study begins with a model negative bias, whereby inclusion of SOA precursor deposition worsens the model negative bias. However, if this study were to include S/IVOC emissions and aqueous phase SOA 20 production, it would be possible to begin these series with a positive model bias. If this was the case, the inclusion of SOA precursor deposition would reduce the model positive bias. This study conclusively demonstrates that variations in VOC deposition and oxidation mechanisms do indeed alter the agreement between model and observed OA/SOA concentrations.
However, as the sign of the model bias (i.e. positive or negative) could be sensitive to which SOA source are included, this study does not conclusively demonstrate if these model updates lead to an improvement or worsening of model agreement 25 with observations. These results highlight that the global SOA budget is highly sensitive to hydrocarbon physicochemical processes.
For example, the global annual-total SOA production rate has varied from 47 to 75 Tg (SOA) a -1 due to variations in VOC deposition. The global annual-total SOA production rate from anthropogenic and biomass burning emissions has varied from 17.9 to 45.1 Tg (SOA) a -1 due to variations in VOC oxidation mechanisms. Additional simulations could reach even wider 30 bounds on the global SOA budget. For instance, neglecting SOA precursor deposition combined with VOC ANT/BB undergoing oxidation with NO/HO 2 -dependent yields would results in even higher global SOA production rates. These results suggest that both oxidation and deposition remain significant contributors to uncertainty in the global SOA budget.
Despite the limitations of this study, such as the lack of chemical complexity and geographical coverage of observations, it is apparent that SOA precursor deposition and oxidation contribute considerably towards uncertainties in both the global SOA budget and model agreement with observations. These results highlight the need for greater insight into the physicochemical processes of gas-phase hydrocarbons related to SOA production, together with a greater density of observations. 5 Code and data availability 10 The model used in this study is the Global   (1)) for both existing and new SOA precursors, taken from Atkinson and Arey (2003). Note, VOC ANT/BB reacts with OH, with reaction kinetics based off either monoterpeene, 5 naphthalane, toluene or benzene.  (Hall et al., 1999;Nguyen et al., 2015). 'High' represents surface 5 resistances of CO.      Table 3. Observations, originally compiled by , for the time period 2000-2010, are classified by site type -urban (blue), urban downwind (green) or remote (red), and continent -Asia (squares), North America (circles) and Europe (triangles). Observed oxygenated-OA is assumed to be comparable to  Table 3. For Welgegund, both observed 10 and simulated monthly mean OA concentrations span an entire year. The standard deviations across this year, based on the monthly-mean data, are indicated in blue. For Manaus however, the measurements of OA only span one month, hence, no standard deviation is shown for this site.  Table 3. The standard deviation of the binned observations at each model layer is shown (peach envelope). For each campaign, the normalised mean bias (%) for each simulation is also included in the top right of each panel. 10 Figure 7 -Global annual-total reaction fluxes and total SOA production rate from anthropogenic and biomass burning hydrocarbons, for the simulations described in Table 3. The global annual-total VOC ANT/BB emission rate, of 176 (VOC ANT/BB ) a -1 , is identical across all simulations. surface OH concentrations (ppq(v)), c) the annual-total vertically integrated VOC ANT/BB oxidation rate by OH (Tg a -1 ), and d) the corresponding annual-total SOA production rate (Tg a -1 ), when SOA precursor deposition and a single oxidation step 5 with a yield of 13 % is applied (DryH_WetL; Table 3). Multi_benz_yield (fourth column; d and h) simulations, which are described in Table 3.   Table  3. Multi_tol_yield and d) Multi_benz_yield simulations, described in Table 3. Observations for the time period 2000-, are classified by site type -urban (blue), urban downwind (green) or remote (red), and continent -Asia (squares), North America (circles) and Europe (triangles). Observed oxygenated-OA is assumed to be comparable to simulated SOA. The 1:1 10