Author Responses to Referees and comments on Modelling organic aerosol over Europe in summer conditions with the VBS-GECKO parameterization: sensitivity to secondary organic compound properties and IVOC emissions

1 LISA, UMR CNRS 7583, IPSL, Université Paris Est Créteil and Université de Paris, 94010 Créteil Cedex, France. 2 INERIS, National Institute for Industrial Environment and Risks, Parc Technologique ALATA, 60550 Verneuil-en-Halatte, France. 3 Agence de l’Environnement et de la Maîtrise de l’Energie, 20 avenue du Grésillé BP 90406, 49004 Angers Cedex 01, France. a Now at: CEREA, Joint Laboratory École des Ponts ParisTech – EDF R & D, Université Paris-Est, 77455 Marne la Vallée, France. b Now at: Laboratoire de Métérologie Dynamique, IPSL, CNRS, UMR8539, 91128 Palaiseau Cedex, France.


Introduction
For the past 20 years, fine particulate matter or PM 2.5 (particles with a diameter smaller than 2.5 µm) has been regulated due to their health impacts and the resulting costs (e.g. Lim et al., 2012;WHO Regional Office for Europe and OECD, 2015).
Furthermore, fine particles degrade visibility (e.g. Han et al., 2012) and influence climate change (e.g. Boucher et al., 2013). 5 Organic aerosol (OA) represents a large fraction of the total fine particle mass (e.g. Jimenez et al., 2009). This OA is either primary (directly emitted into the atmosphere) or secondary (formed by gas/particle partitioning of low volatile and/or highly soluble species produced during the oxidation of gaseous organic compounds) (e.g. Carlton et al., 2009;Kroll and Seinfeld, 2008). The secondary organic aerosol (SOA) dominates the primary organic aerosol (POA) in most environments (e.g. Gelencsér et al., 2007;Jimenez et al., 2009). 10 Chemistry-transport models (CTMs) are used to investigate and identify air quality regulation policies. Parameterizations are developed and used in CTMs to represent SOA formation. Different approaches have been followed to describe SOA formation as the two-product model (e.g. Odum et al., 1996;Schell et al., 2001), the molecular approach (e.g. Pun et al., 2002Pun et al., , 2003, the volatility basis set (VBS) approach (e.g. Donahue et al., 2006Donahue et al., , 2012 or the statistical oxidation model (SOM) (e.g. Cappa and Wilson, 2012;Jathar et al., 2015). Parameterizations are constantly improved and additional 15 processes were included in the parameterizations to improve the simulations of SOA concentrations, such as gas-phase aging of organic species (e.g. Rudich et al., 2007), more comprehensive emissions and multiphase chemistry. Robinson et al. (2007) have indeed shown that POA provided in emission inventories is in part composed of semi-volatile organic compounds (SVOCs) (existing both in particle and gas phases) and that a fraction of emitted organic compounds were missing from these inventories: the intermediate-volatility organic compounds (IVOCs) (forming SOA after several 20 oxidation stages) (e.g. Ots et al., 2016;Robinson et al., 2007;Woody et al., 2015). Numerous experimental and modeling studies have since explored the volatility distribution of SVOCs from POA emissions and of IVOC emissions depending on the emission source (e.g. Akherati et al., 2019;Grieshop et al., 2009;Hatch et al., 2018;Jathar et al., 2017;Louvaris et al., 2017;Lu et al., 2018;May et al., 2013aMay et al., , 2013bMay et al., , 2013cWoody et al., 2016).
The development of the VBS-GECKO parameterization explores another track using the results of an explicit model to represent the organic gas-phase chemistry and gas/particles mass transfer, instead of atmospheric chamber data. The VBS-GECKO parameterization for SOA formation (Lannuque et al., 2018) is a VBS-type parameterization with gaseous aging.
VBS-GECKO was optimized based on box modeling results using explicit oxidation mechanisms generated with the 15 Generator for Explicit Chemistry and Kinetics of Organics in the Atmosphere (GECKO-A) modeling tool (Aumont et al, 2005;Camredon et al, 2007). Lannuque et al. (2018) have shown that the VBS-GECKO parameterization is coherent compared to the explicit GECKO-A chemical mechanism. The reliability of the VBS-GECKO parameterization is by design directly linked to the accuracy of the GECKO-A mechanisms. The accuracy of the GECKO-A mechanisms to represent SOA formation has been evaluated against around 50 chamber experiments (Denjean et al, 2015;La et al, 2016 ;McVay et al, 20 2016;Valorso et al, 2011). Some processes relevant for SOA formation in the atmosphere can however be misrepresented in the GECKO-A mechanisms, or are just not included (such as gaseous autoxidation reactions or the reactivity in the condensed phase). The reliability of the VBS-GECKO parameterization to represent SOA formation observed in the atmosphere has thus to be evaluated.
The objectives of this study are (i) to evaluate the behavior of the VBS-GECKO parameterization in the CTM CHIMERE 25 (Menut et al., 2013, Mailler et al., 2017 by comparison with field measurements and previous simulations obtained with the H 2 O parameterization (Couvidat et al., 2012) already implemented in CHIMERE, (ii) to explore the sensitivity of simulated SOA concentrations to organic compound properties (volatility, solubility, aging rates or NO x regime) and (iii) to test the sensitivity of OA concentrations to the uncertainties on IVOC emission fluxes from traffic,. The setup of the CHIMERE model and the implementation of the VBS-GECKO in the CTM are described in section 2. In section 3, the VBS-GECKO is 30 evaluated over Europe for a two-month summer period, the sensitivity to organic compound properties is explored in section 4, and the sensitivity to the uncertainties in IVOC emission fluxes from traffic is investigated in section 5. Finally, results on simulated OA sources and concentrations are discussed in section 6.

The CHIMERE chemical transport model
The evaluation of the VBS-GECKO parameterization and the exploration of SOA sensitivity were performed using the CHIMERE 2017 β version. This version is based on the CHIMERE 2013 version (Menut et al., 2013) which was modified to improve the representation of particles with the implementation of a new aerosol module. Details of the CHIMERE 2017 β 5 version and its evaluation are given in Couvidat et al. (2018).
Briefly, the CHIMERE 2017 β version uses the MELCHIOR2 gas-phase chemical scheme, involving 44 species reacting according to 120 reactions. MELCHIOR2 is a reduced version of the MELCHIOR1 mechanism, obtained by the Carter's surrogate molecule method (Carter, 1990). In CHIMERE, the aerosol evolution is described by a sectional aerosol module (e.g. Bessagnet et al., 2004Bessagnet et al., , 2009Schmidt et al., 2001). The size distribution of aerosol particles is here represented using 9 10 bins, ranging from 10 nm to 10 μm. Aerosol formation is represented in the model by nucleation for sulfuric acid (Kulmala et al., 1998), coagulation between particles (e.g. Debry et al., 2007;Jacobson et al., 1994) and condensation/evaporation via absorption according to the "bulk equilibrium" approach (e.g. Pandis et al., 1993). For inorganic species, the gas/particle equilibrium concentrations are calculated using the ISORROPIA v2.1 (Fountoukis and Nenes, 2007) thermodynamic module. For organic species, the equilibrium concentrations are calculated using the SOAP (Secondary Organic Aerosol 15 Processor) thermodynamic module (Couvidat and Sartelet, 2015). The gaseous formation of secondary organic species able to partition between the gas and the condensed phases (so leading to SOA formation) are represented in the CHIMERE β version using the H 2 O mechanism. Here, the VBS-GECKO parameterization was also implemented. The H 2 O and VBS-GECKO organic aerosol modules are described hereafter.
The chemical speciation of emitted non-methane volatile organic compounds (NMVOCs) is taken from Passant (2002) as 20 described in Menut et al. (2013). POA from emission inventories are considered as SVOC. A factor 5 is apply to residential POA emissions, as wood burning emissions are underestimated in emission inventories (e.g. Denier Van Der Gon et al., 2015). This factor was shown to give satisfactory results on OA estimations (Couvidat et al. 2012(Couvidat et al. , 2018. Biogenic emissions are computed with the Model of Emissions and Gases and Aerosols from Nature MEGAN 2.1 algorithm (Guenther et al., 2012). Dry deposition for gaseous organic species is described using the Wesely (1989) parameterization and according to 25 their Henry's law constants, as described by Bessagnet et al. (2010).

The organic aerosol modules
The purpose of the comparison between the H 2 O and GECKO-VBS mechanisms for SOA formation is to evaluate the reliability of the VBS-GECKO parameterization. The simulations performed with CHIMERE were therefore setup using the same configuration of the model (meteorological data, emissions, deposition, inorganic and organic gaseous chemical 30 mechanism, inorganic and organic gas/particle partitioning…), but implementing either the H 2 O or the GECKO-VBS parameterizations. The implementation of a given parameterization for SOA formation induces anyway some differences, related to the primary compounds considered and/or the processes taken into account in the parameterization. The differences between the H 2 O and VBS-GECKO are mentioned in the following parameterization presentation sections.

The H 2 O reference mechanism
The H 2 O mechanism, described in details by Couvidat et al. (2018), considers SOA formation from the partitioning of hydrophilic species (condensing on an aqueous phase and an organic particulate phase) and hydrophobic species (condensing 5 only on an organic particulate phase owing to their low affinity with water). Distinction between hydrophobic and hydrophilic compounds is based on their octanol/water coefficient (Pun et al., 2006) or their partitioning between the organic and aqueous phases (Couvidat and Seigneur, 2011). H 2 O considers the formation of hydrophilic and/or hydrophobic species from the gaseous oxidation of isoprene, monoterpenes (α-pinene, β-pinene, limonene and ocimene), sesquiterpenes (humulene) and mono-aromatic precursors (toluene and xylenes). Note that in H 2 O, limonene mechanism is used as a 10 surrogate mechanism for ocimene (ocimene having its own OH, NO 3 and O 3 reaction rates). Each emitted SOA precursor is linked to a species of the H 2 O mechanism. POA provided by emissions inventories are split into three emitted SVOCs having different volatilities (saturation vapor pressures at 298 K of 8.9×10 -11 , 8.4×10 -9 and 3.2×10 -7 atm respectively) with a fraction that follows the volatility distribution of POA emissions given by Robinson et al. (2007). In H 2 O, gaseous oxidation of these three compounds with OH leads to hydrophobic species with a lower volatility. No gaseous oxidation is considered for 15 hydrophylic and hydrophobic species in H 2 O. Activity coefficients for the H 2 O species are computed with the thermodynamic model UNIFAC (UNIversal Functional group Activity Coefficient; Fredenslund et al., 1975). H 2 O has been evaluated over Europe (Couvidat et al., 2012(Couvidat et al., , 2018 and the Paris area (Couvidat et al., 2013;Zhu et al., 2016aZhu et al., , 2016b. The H 2 O reference mechanism is presented in table S1 of supplementary material.

The VBS-GECKO parameterization 20
The VBS-GECKO parameterization is described in details in a previous paper by Lannuque et al. (2018). Briefly, VBS-GECKO is a volatility basis set (VBS) type parameterization that represents SOA formation from the partitioning of organic compounds having a low volatility onto an organic aerosol phase. The VBS-GECKO parameterization takes into account for the oxidation of a precursor k (precu k ) (1) the formation of 7 VB k,i , where i is the number of the volatility bin (1 being the most volatile and 7 the less volatile) (reactions R1, R2 and R3), (2) the gas-phase aging of the VB k,i (except for the lowest 25 volatility bin 7) with OH redistributing the matter between the VB k,i (reactions R4) and by photolysis leading to a loss of carbon matter (reactions R4), (3) the gas/particle partitioning of the precursor k (R6) and of the VB k,i (R7) ; the VBS-GECKO follows this structure for a given precursor k: precu k (g) + OH  a k,RRR,1 VB k,1 + a k,RRR,2 VB k,2 +... + a k,RRR,n VB k,7 k precuk+OH (R1) precu k (g) + O 3  b k,RRR,1 VB k,1 + b k,RRR,2 VB k,2 +... + b k,RRR,n VB k,7 k precuk+O3 (R2) 30 precu k (g) + NO 3  c k,RRR,1 VB k,1 + c k,RRR,2 VB k,2 +... + c k,RRR,n VB k,7 k precuk+NO3 (R3) VB k,i (g) + OH  d k,RRR,i,1 VB k,1 + d k,RRR,i,2 VB k,2 +... + d k,RRR,i,n VB k,7 i≠7 k OH =4.10 -11 cm 3 molec -1 s -1 (R4) VB k,i (g) + h carbon lost i≠7  k J acetone (R5) In VBS-GECKO, the production and gaseous aging of the VB k,i for a precursor k are adjusted by stoichiometric coefficients (a k,RRR,i , b k,RRR,i , c k,RRR,i , d k,RRR,i , for reaction (R1) to (R4) respectively) which depend on NO x regime. The formation of more 5 volatile and less volatile bins can be assimilated to fragmentation and functionalization processes, respectively. The stoichiometric coefficients depend on the NO x according to the reaction rate ratio (RRR) of RO 2 with NO: (1) where k RO2+NO (set to 9.0 × 10 −12 cm 3 molec −1 s −1 according Jenkin et al., 1997 at 298 K) and k RO2+HO2 (set to 2.2 × 10 −11 cm 3 molec −1 s −1 according to Boyd et al., 2003, assuming a large carbon skeleton for RO 2 at 298 K) are the rate constants for the 10 reactions of the peroxy radicals with NO, and HO 2 respectively and [NO] and [HO 2 ] the concentration of the radicals. The entire RRR range is covered by linear interpolation of the coefficients between the two closest tabulated values. The photolysis is considered as a limiting process for SOA formation, leading to a loss of matter. The photolysis rate of the VB k,i are based on the acetone one multiplied by an optimized factor  k , different for each precursor k. Precursors and VB k,i condense on an organic particulate phase according to an equilibrium between the gas and the organic particulate phase that 15 follows the Raoult's law (reaction R6 and R7).
The properties of the 7 VB k,i were considered to be independent of the precursor k and set for each volatility bin i to the mean values simulated with explicit GECKO-A simulations. Table 1 gives the molar weights (Mw), saturation vapor pressures (P sat ) at 298 K, effective Henry's law constants (H eff ) at 298K and vaporizations enthalpies (ΔH vap ) used for each VB k,i VBS-GECKO species. The stoichiometric coefficients and factors  k were optimized on explicit GECKO-A 20 simulations of gas-phase oxidation and SOA formation. The stoichiometric coefficients were optimized for 5 RRR values: 0, 0.1, 0.5, 0.9 and 1 (Lannuque et al., 2018). Precursors considered in the current VBS-GECKO parameterization are monoaromatic compounds (benzene, toluene, and o-m-and p-xylenes) and n-alkanes (decane, tetradecane, octadecane, docosane and hexacosane) reacting with OH, and monoterpenes (α-pinene, β-pinene and limonene) and linear 1-alkenes (decene, tetradecene, octadecene, docosene and hexacosene) reacting with OH, O 3 and NO 3 . Note that (1) the parameterization does 25 not represent SOA formation from the partitioning of hydrophilic species, (2) recently identified chemical processes, such as autoxidation reactions or acid-catalyzed pathways, not included in the GECKO-A mechanisms, are thus not considered in the VBS-GECKO parameterization and (3) the high value of the reaction rate of the VB k,i (k OH = 4.10 -11 cm 3 molec -1 s -1 ) was fixed before optimization and is compensated by lower or higher values of optimized coefficients (see details in Lannuque et al., 2018). Tables of optimized stoichiometric coefficients are available in supplementary material of Lannuque et al. (2018). 30 For SOA production from NMVOC oxidation, the former H 2 O parameterization in CHIMERE was replaced by the VBS-GECKO parameterization for terpenes and mono-aromatic compounds. The VBS-GECKO mechanisms were also implemented in CHIMERE for SOA formation from C10 to C13 alkanes and alkenes, gaseous species usually not considered in 3D models as SOA precursors. Each emitted SOA precursor not present in the VBS-GECKO was linked to a VBS-GECKO species. As in the H 2 O mechanism, the VBS-GECKO parameterization for limonene was used as a surrogate mechanism for ocimene. The VBS-GECKO parameterizations for benzene, toluene, o-, m-and p-xylenes were also used as surrogate mechanisms for other emitted mono-aromatic compounds according to their SOA yield and reactivity with OH. ndodecane and tetradecane VBS-GECKO species were used to lump emitted alkanes with 10 to 13 atoms of carbon, 5 according to their carbon chain length. The VBS-GECKO mechanism for 1-decene was applied for all emitted C10 alkenes.
The lumping scheme between emitted NMVOCs and VBS-GECKO species is given in Table 2. The current VBS-GECKO version does not represent SOA production from the oxidation of isoprene and sesquiterpenes. The H 2 O parameterizations for isoprene and humulene were therefore left unchanged in CHIMERE to account for this SOA production. For SOA production for SVOC oxidation distributed from POA emissions, the H 2 O approach was kept unchanged (i.e. distribution of 10 POA emissions into 3 SVOC species and representation of their SOA production using the H 2 O mechanism). In CHIMERE, RRR is calculated in each box at each chemical time step following equation 1. Activity coefficients for the condensation of the VBS-GECKO species into the aerosol particulate phase are fixed to 1 (i.e. ideality of the organic particulate phase is considered). This implementation of the VBS-GECKO in CHIMERE was selected here as the reference configuration and is Changes were then applied to this reference configuration to perform sensitivity tests of SOA formation on secondary organic compound properties (solubility, reactivity with OH, NO x /HO 2 condition dependency and volatility) or IVOC emission fluxes from traffic. For a better readability, the details of these modifications are presented for each sensitivity test in section 4 (properties) and section 5 (IVOC emissions). 20

Simulation setup and field measurements
The model was run to simulate the concentrations of OA over Europe (from 25° W to 45° E in longitude and from 30° to 70° N in latitude) with a horizontal resolution of 0.25° × 0.25° during the July-August 2013 period, SOA formation being expected to be important during summertime. Meteorology was obtained from Integrated Forecasting System (IFS) model of the European Centre for Medium-Range Weather Forecasts (ECMWF). This meteorology has been evaluated in Bessagnet et 25 al. (2016) for the model intercomparison project EURODELTA-III. ECMWF-IFS in the EURODELTA-III project has been shown to be one of the most reliable models to represent meteorological conditions over Europe. Anthropogenic emissions of gases and particles were taken from the European Monitoring and Evaluation Programme (EMEP) inventory (methodology described in Vestreng, 2003) and boundary conditions were generated from the Model for OZone And Related Tracers (Mozart v4.0 (Emmons et al., 2010)). Wildfire emissions were not considered. 30 The VBS-GECKO mechanism was evaluated by comparing the simulated results to the H 2 O mechanism and particulate phase measurements available in the EBAS database (http://ebas.nilu.no/). EBAS is a database hosting observation data of atmospheric chemical composition and physical properties in support of a number of national and international programs ranging from monitoring activities to research projects. EBAS is developed and operated by the Norwegian Institute for Air Research (NILU). This database is populated for example by the EMEP measurements (Tørseth et al., 2012) or the Aerosols, Clouds and Trace gases Research Infrastructure (ACTRIS, http://www.actris.eu/) ones. 48 rural background stations provide measurements for fine particulate matter and were thus selected here for a statistical evaluation: 36 stations for PM 2.5 , 13 for OC PM2.5 (organic carbon in PM 2.5 , obtained by filter calcinations) and 6 for OM PM1 (organic matter in PM 1 , obtained with 5 ACSM). For the comparisons with OC measurements, the OM:OC ratio of the VBS-GECKO volatility bins were assumed to be equal to 1.8, in agreement with typical observed values given by Canagaratna et al. (2015). The location of the selected stations is shown in Fig. 1.a. Among these stations, 7 stations were used for time series comparisons: -the Cabauw (NL0644R, Netherlands), Melpitz (DE0044R, Germany) and Palaiseau (FR0020R, SIRTA, France) rural background stations, located in areas dominantly impacted by anthropogenic air masses (see Figure S1 in supplementary 10 material presenting the mean of the simulated ratios between toluene and α-pinene emission fluxes for the studied period).
These 7 stations were selected among the 48 background station because the measurements at the station provide (1) a direct 15 information on the organic fraction of fine particles, i.e. OM PM1 and OC PM2.5 measurements, and (2) enough data over the studied period to perform time series comparisons .The location of the 7 selected stations is shown in Fig. 1 ( , where c i mod and c i obs are the simulated and observed concentrations of the studied component at the time i. and N the number of available in-situ measurement values. Boylan and Russell (2006) defined two criteria to evaluate the performances of a model. The model performance criteria (described as the level of accuracy that is considered to be acceptable for modeling 25 applications) is reached when MFE ≤ 75% and |MFB| ≤ 50% whereas the performance goal (described as the level of accuracy that is considered to be close to the best values a model can be expected to achieve) is reached when MFE ≤ 50% and |MFB| ≤ 30%. These criteria are currently used to evaluate the reliability of the models (e.g. Ciarelli et al., 2017;Couvidat et al., 2018;Lecoeur and Seigneur, 2013;Mircea et al., 2019). period. The simulated mean OA concentrations range from ~0 μg.m -3 in remote oceanic areas, to ~12 μg m -3 around the Adriatic Sea and in the northern Italy, and are coherent with the expected orders of magnitude and spatial distributions over Europe (Aksoyoglu et al., 2011;Crippa et al, 2014). Figure 2.b presents the relative difference between mean OA mass 5 concentrations simulated with ref-VBS-GECKO and with H 2 O. The ref-VBS-GECKO produces more OA than H 2 O, with a mean OA mass concentration around 30% higher on average over Europe. The increase is particularly important over northern Europe, with maximum differences reaching around +60%. Table 3   At stations influenced by anthropogenic air masses, OA concentrations are weakly influenced by the organic aerosol module (see Fig. 3.a to c and Fig. 4.a and b). OA concentrations simulated with ref-VBS-GECKO are substantially underestimated, differences exceeding -50% for OM PM1 concentrations at the Palaiseau and Melpitz stations as well as for OC PM2.5 concentrations at the Cabauw station. 5

Sensitivity to the parameterization properties
Sensitivity tests were performed to assess the VBS-GECKO parameterization, evaluate the consistency of the modeling results and examine some hypotheses that may explain the gaps between measurement and simulated values. Sensitivities to hydro-solubility, gaseous aging, NO x regimes and volatility were studied comparing results to the non-modified ref-VBS-GECKO version. 10

Sensitivity tests to hydro-solubility and H eff
SOA formation from the gas/particle partitioning of hydro-soluble organic compounds into an aqueous phase is now well recognized (e.g. Bregonzio-Rozier et al., 2016;Carlton et al., 2009;Knote et al., 2014). The effective Henry's law constant (H eff ) is the key parameter which controls this hydrophilic partitioning. Hydro-soluble organic compounds can also be lost at the surface by dry deposition. In CHIMERE, and according to the deposition scheme of Wesely (1989), the stomatal 15 resistance of organic compounds depends on H eff . To analyze the sensitivity of the simulated OA to hydrophilic partitioning and values of H eff , the following two simulations were run:  Hydro-VBS-GECKO. In this model configuration, VB k,i can condense both on organic and aqueous phases of particles. Aqueous-phase partitioning is computed according to Henry's law, assuming the particle phase behave as an ideal well mixed homogeneous aqueous phase. Deposition of VB k,i was already taken into account in the 20 reference model configuration and was kept unchanged. The relative difference on the simulated mean OA concentrations between Hydro-VBS-GECKO (respectively Hydro-VBS-GECKO-high) and ref-VBS-GECKO is given Fig. 5.a (respectively Fig. 5.b) for the two-month period. Figure 5.a shows that considering aqueous phase partitioning of the VBS-GECKO species leads to variations on the simulated mean OA concentrations below ±0.5%. Table 4 shows no significant modification in the statistical results for this simulation. The values of H eff set to each volatility bin increase when the volatility decreases (see Table 1), meaning that the less volatile 30 species are also more prone to condense into the aqueous phase. Adding a hydrophilic partitioning does therefore not increase substantially the concentrations of organic species in the condensed phases.
The Hydro-VBS-GECKO-high configuration increases the mean simulated OA concentrations by ~10%, with a maximal increase reached over Belgium-Netherlands-Luxembourg area (called Benelux hereafter, around +20%, see Fig. 5.b). The contribution of the deposition and the partitioning processes are shown in Fig. 5.c and 5.d respectively. Changes due to 5 deposition appear negligible (below ±0.2%) compared to the changes due to the aqueous partitioning (~+10%). According to the Wesely (1989) parameterization used for deposition, water solubility contributes to the surface resistance only. Knote et al. (2014) have shown that deposition is not limited by the surface resistance for H eff greater than 10 8 mol L -1 atm -1 . In the  Table 4). According to these tests, SOA production 15 due to the hydrophilic partitioning of the various VB k,i of the VBS-GECKO parameterization is expected to be a minor process.

Sensitivity test to gaseous aging rates and OH radical concentrations
In the VBS-GECKO parameterization, the same rate constant is set for the VB k,i reactions with OH (k OH = 4.0×10 -11 cm 3 molec -1 s -1 ). Timescale for gaseous aging is therefore driven by the OH concentrations simulated by the CTM. Simulated OH 20 concentrations depend on the gas-phase chemical mechanisms used in the CTM, with differences on OH concentrations reaching up to 45% between mechanisms (Sarwar et al., 2013). Two simulations were run with modified k OH to examine the sensitivity of SOA production to the rate of chemical aging:  k OH -VBS-GECKO-low. In this model configuration, the VB k,i +OH rate constants are divided by a factor 2, i.e. k OH low = 2.0×10 -11 cm 3 molec -1 s -1 . 25  k OH -VBS-GECKO-high. In this model configuration, the VB k,i +OH rate constants are multiplied by a factor 2, i.e. k OH high = 8.0×10 -11 cm 3 molec -1 s -1 .
The relative difference on the simulated mean OA concentrations between k OH -VBS-GECKO-low (respectively k OH -VBS-GECKO-high) and ref-VBS-GECKO is given Fig. 6.a (respectively Fig. 6.b). A slight variation of simulated OA concentrations is found (lower than ±10%), with simulated OA concentrations decreasing with the decrease of aging rates 30 and vice versa. This result highlights that the gas-phase aging of volatility bins in the VBS-GECKO parameterization promotes functionalization (formation of less volatile bins) rather than fragmentation (formation of more volatile bins), as already shown with tests conducted in box model (Lannuque et al., 2018). The highest relative differences are located over the Mediterranean Sea and North Africa, i.e. areas showing high OH and low OA concentrations (below 4 µg m -3 , see Fig.   2.a). The k OH -VBS-GECKO-high configuration improves statistics, due to an overall increase of the simulated OA concentrations (and contrariwise for the k OH -VBS-GECKO-low configuration) (see Table 4). However, the sensitivity of SOA to the gas-phase aging of the VBS-GECKO volatility bins remains weak and aging rates is likely not a major source of 5 uncertainty.

Sensitivity test to the NO x regime
Similar to the OH discussion above, simulated HO 2 and NO concentrations in CTMs are linked to the gas-phase chemical mechanism used. The concentrations of these two species determine the value of the RRR ratio and therefore drive the aging of the various VB k,i (Lannuque et al., 2018). A sensitivity test was performed to examine the sensitivity of the simulated OA 10 to the chemical regime. HO 2 or NO concentrations can hardly be modified without changing all the simulation conditions.
 RRR-VBS-GECKO-high. In this model configuration, RRR ratio is calculated with k RO2+HO2 divided by 2, i.e. 15 k RO2+HO2 RRRhigh = 1.1×10 -11 cm 3 molec -1 s -1 . Figure 7 presents the mean RRR ratio during the two-month period for both RRR-VBS-GECKO-low ( Fig. 7.a) and RRR-VBS-GECKO-high model configurations ( Fig. 7.b). The entire range of RRR ratio (from remote NO x conditions to high NO x conditions) is covered over Europe with the both model configuration. As expected, the urban, industrial and intense shipping transport areas such as Paris, the Channel, Benelux, northern Italy or Moscow are systematically in the high NO x 20 regime (RRR close to 1) whereas remote areas over the seas (away from shipping tracks) are systematically in the remote NO x regime (RRR close to 0). Between these two extremes, the RRR ratio depends on the environmental and meteorological conditions at the location and, in this sensitivity study, on the model configuration for the RRR calculation. Current parameterizations for SOA formation only consider two extreme regimes corresponding to a high-NO x and a low-NO x condition. Criteria used to define high and low NO x differ from a study to another one but the parameterizations are usually 25 optimized at NO x values typical of rural conditions for low NO x (corresponding to a RRR ratio of ~0.6) and typical of urban conditions for high NO x (corresponding to a RRR ratio of ~1) (e.g. Hodzic et al., 2014;Lane et al., 2008). The range of RRR between 0.0 and 0.6 is therefore not considered in most of the parameterizations, although substantial changes in SOA formation were found within this range of RRR (Lannuque et al., 2018).
The relative difference on the simulated mean OA concentrations between RRR-VBS-GECKO-low (respectively RRR-VBS-30 GECKO-high) and ref-VBS-GECKO is given Fig. 8.a (respectively Fig. 8.b). Results show variations of simulated mean OA concentrations smaller than ~15%. In agreement with previous studies, an increase (decrease) of RRR ratio leads to a decrease (increase) of the simulated OA concentrations (e.g. Donahue et al., 2005;Lannuque et al., 2018;Ng et al., 2007).
As expected, the variation is weaker over areas having either an RRR ratio close to 0 or 1, the NO x regime remaining unchanged among the model configurations. The highest relative differences on OA are found over continental rural areas, i.e. areas showing the largest variation of RRR among the model configurations. Large relative differences are also found over the Mediterranean Sea, owing in part to the low simulated OA concentrations. Similar to the k OH sensitivity tests, the 5 RRR-VBS-GECKO-low configuration increases the overall OA concentrations and improves statistical indicators, and contrariwise for the RRR-VBS-GECKO-high configuration. Sensitivity on RRR values appears weak enough to likely not be a major source of uncertainty for the VBS-GECKO parameterization.

Sensitivity test to volatility and P sat
In the explicit GECKO-A simulations used for the VBS-GECKO optimization, the saturation vapor pressure, P sat , of 10 secondary organic compounds was estimated using structure activity relationships (SAR) (see Lannuque et al., 2018).
Estimated P sat can typically vary within one order of magnitude according to the SAR used (e.g. Valorso et al., 2011). Two simulations were run to examine the sensitivity of SOA to the uncertainties in P sat :  P sat -VBS-GECKO-low. In this model configuration, the nominal P sat values of VB k,i are divided by 10.
 P sat -VBS-GECKO-high. In this model configuration, the nominal P sat values of VB k,i are multiplied by 10. 15 As OA concentration directly contributes to the partitioning, these two simulations can also be considered as a sensitivity test to the simulated OA concentrations.
The relative difference on the simulated mean OA concentrations between P sat -VBS-GECKO-low (respectively P sat -VBS- Fig. 9.a (respectively Fig. 9.b). Shifting the volatility of the VB k,i by one order of magnitude leads to an overall change in the simulated mean OA concentrations of about -25% (+25%) when P sat is 20 increased (decreased). A weaker sensitivity is observed over urban areas, such as Paris or Moscow. This behavior is mainly linked to the simulated volatility of OA in the ref-VBS-GECKO simulations. Figure 10 shows the mean volatility of OA over Europe for the reference configuration. Simulated OA contributors are mainly low volatile species (with mean P sat 298K between 10 -10 and 10 -14 atm), the highest values being found over urban areas (less aged OA), and the lowest values found over areas close to the boundaries of the domain (linked to a boundary effect in the model). A shift in volatilities over these 25 two types of site has a lower impact on OA concentrations, as OA mean volatilities being either too high (mean P sat 298K ≈ 10 -10 atm, upon urban areas) or too low (mean P sat 298K ≈ 10 -14 atm, upon boundary areas) for a change in P sat to substantially impact the partitioning. The largest effect is typically observed over central Europe where OA contributors show intermediate mean volatilities (mean P sat 298K ≈ 10 -12 atm). Statistically, the P sat -VBS-GECKO-low configuration is the only configuration matching the performance goal for all the 30 simulated OA concentrations (OC PM2.5 and OM PM1 ) (see Table 4). For OC PM2.5 , RMSE is however higher than in reference configuration. Simulated OA concentrations appear to be sensitive to uncertainties in the estimated saturation vapor pressures of the numerous OA contributors considered during the development of the VBS-GECKO parameterization.

Sensitivity to IVOC emission fluxes from traffic and transport sources
IVOCs have been shown to be a substantial source of SOA in the plume of megacities (e.g. Hodzic et al., 2010;Tsimpidi et al., 2010). Even if several recent studies have been performed to identify the IVOC speciation of different individual 5 emission sources (e.g. Akherati et al., 2019;Grieshop et al., 2009;Hatch et al., 2018;Jathar et al., 2017;Louvaris et al., 2017;Lu et al., 2018;May et al., 2013aMay et al., , 2013bMay et al., , 2013cWoody et al., 2016), a comprehensive inventory is still not available to represent IVOC emissions by activity sector (gathering several individual emission sources). A large fraction of these IVOCs is thus still not considered in emission inventories. In this section, only IVOC emissions from traffic and transportation sources are treated. Robinson et al. (2007) assumed that IVOC emissions for small off-road diesel engines 10 were equal to 150% of POA emissions, consistent with the Schauer et al. (1999) emission data for 1995 medium-duty diesel vehicles. Recent studies have measured IVOC emissions from (i) exhausts of light-duty gasoline vehicles and (ii) exhausts of both heavy-duty and medium-duty diesel vehicles (Zhao et al., 2015(Zhao et al., , 2016. Experiments on gasoline exhausts were processed on 42 vehicles and experiments on diesel vehicles on 6 vehicles, the selected vehicles being representative of the transportation fleet in North America. In both cases, Zhao et al. (2015Zhao et al. ( , 2016 have shown that a stronger correlation can be 15 found between IVOC and NMVOC emissions (R 2 equal to 0.92 and 0.98 for gasoline and diesel exhausts, respectively) than between IVOC and POA emissions (R 2 equal to 0.76 and 0.61 for gasoline and diesel exhausts, respectively). Zhao et al. (2015Zhao et al. ( , 2016 have estimated that IVOC emissions represent about 4% of NMVOC emissions in cold-start cycle to about 16% in hot-start cycle for light-duty gasoline vehicles, and about 60 ± 10 % of NMVOC emissions for heavy-duty and medium-duty diesel vehicles. 20 In this study, the VBS-GECKO parameterization was used to examine the sensitivity of SOA to IVOC emissions from road traffic (SNAP 7) and other mobile sources and machineries (SNAP 8). The following five model configurations, based on different IVOC emission fluxes, were designed for that purpose: As in the reference model configuration, POA are considered as SVOC in these sensitivity tests for traffic and transport emissions. Primary SVOCs and IVOCs (S/IVOCs) constitute a complex mixture of linear, branched and cyclic alkanes, alkenes and aromatics (Fraser et al., 1997;Gentner et al., 2012;Lu et al., 2018;Schauer et al., 1999Schauer et al., , 2002. The molecular 5 composition of S/IVOCs emitted in the atmosphere by fossil fuel combustion is however still poorly documented. S/IVOCs at emission were thus considered to be distributed into the 9 volatility bins given by Robinson et al. (2007), with the provided fraction of primary SVOCs in each SVOC volatility bin, and of estimated primary IVOCs in each IVOC volatility bin. The VBS-GECKO parameterizations for C 14 , C 18 , C 22 and C 26 1-alkenes and n-alkanes were used as surrogate mechanisms for S/IVOCs (C 14 and C 18 for IVOCs and C 18 , C 22 and C 26 for SVOCs). The C 14 to C 26 VBS-GECKO's n-10 alkanes and 1-alkenes were distributed according to their volatility into the 9 volatility bins of Robinson et al. (2007).
Correspondences are shown in Figure 11, for the example of the IVOC 150POA model configuration. The distribution of alkanes and alkenes was estimated based on (i) the EMEP guidebook (https://www.eea.europa.eu/publications/emep-eeaguidebook-2016), providing speciation data for emissions for various types of vehicles and (ii) the COPERT4 software (Ntziachristos et al., 2009) providing data for a vehicle fleet. Data are only available for light compounds and are here 15 extrapolated to the heavy ones for the needs of the study. Thus, 75 % of the primary S/IVOCs are here assumed to be alkanes and 25 % alkenes. The primary SVOC total emissions and distributions over volatility bins are unchanged between each simulation. The distribution of IVOCs among volatility bins is also unchanged but the total IVOC emissions are modulated according to the 5 IVOC emission scenarios described before (i.e. IVOC 150POA , IVOC 4VOC , IVOC 16VOC , IVOC 30VOC and IVOC 65VOC ). Table 5 gives the speciation of VBS-GECKO species for the various model configurations and the VBS-20 GECKO mechanism for S/IVOCs is presented in Table S3 of supplementary material. Figure 12 shows the mean OA mass concentrations simulated for the 5 IVOC emission configurations, and the absolute and relative differences with the ref-VBS-GECKO simulation without IVOC emissions. Table 6 Table 6). For OC PM2.5 however, the opposite trend is observed with a degradation of the statistical indicators ( Table 6). The ref-VBS-GECKO configuration leads to a slight overestimation of OC PM2.5 concentrations over some stations (e.g. Iskrba, see Fig. 3) and adding the SOA source from IVOCs strengthens the deviation (up to about +30% of RMSE for the IVOC 65VOC configuration), even if the correlation is not significantly modified. 15 IVOC oxidation appears to be a significant SOA source at some locations (e.g. the Cabauw station), especially in the IVOC 30VOC and IVOC 65VOC configuration. However, the resulting OA increase remains too weak to fill in the gaps between observations and simulated data (maximum increase around +40%). For example, time series presented in Figure 13 show that adding IVOC emission increases systematically the simulated OA concentrations, but not enough to explain the OA peaks recorded at the anthropogenic stations (see Fig. 13

Tracking OA sources 30
Apportionment of OA sources is investigated in this section. The study takes into account OA formation from IVOC oxidation and is based on the IVOC 30VOC model configuration. Figure 14 shows the contribution of the various OA sources to the simulated OA concentrations during the July-August 2013 period and the mean daily profiles at two stations located in areas dominantly impacted by anthropogenic air masses (Cabauw and Palaiseau) and two stations located in areas dominantly impacted by biogenic air masses (Birkenes II and Iskrba). SOA constitutes the main fraction of OA whatever the environment. This secondary fraction typically grows from anthropogenic impacted areas (about 70 % at Palaiseau station) to remote areas (about 95 % at Iskrba station). This trend is in agreement with what is usually observed or simulated for 5 summertime periods (e.g. Aksoyoglu et al. 2011;Belis et al., 2013). For the remote stations Birkenes II and Iskrba, respectively 82 and 67% of the simulated OA concentration comes from a biogenic source. Contrariwise, anthropogenic sources are the major OA contributors at anthropogenic impacted stations (65 and 60% of OA at the Cabauw and Palaiseau stations, respectively. Among OA biogenic sources, terpene oxidation is clearly found as the major contributor of OA during the summer period, 10 contributing from 35% (at anthropogenic impacted stations) to 80% (at remote stations) of the total OA mass. The 60% increase of OA mass concentration observed in north Europe between H 2 O and VBS-GECKO parameterizations (see Fig. 2) is also mainly related to SOA formation from terpene, especially ocimene and limonene. In our simulation, SOA produced by isoprene oxidation does not represent a substantial fraction of OA at the selected measurement stations. The major contribution of isoprene SOA to OA reaches about 5% (see Fig. 14.h) and is observed at the Iskrba station during diurnal 15 conditions.
The anthropogenic fraction of OA is found to be dominated by residential biomass burning sources (BBOA). Indeed, according to the temporal factors used in CHIMERE (based on GENEMIS, Ebel et al., 1997;Friedrich, 2000), 4% of annual emissions of residential BBOA occurs during July-August, leading to a non-negligible amount of residential BBOA during summer. This result remains however subject to caution, owing to the large uncertainties in the temporalization of biomass 20 burning emissions in the model. The primary organic fraction (i.e. condensed primary SVOCs) from traffic emissions is found to be substantial in the OA budget only at night in urban areas. On the other hand, the secondary organic fraction produced by traffic emissions can represent about 50% of diurnal anthropogenic OA at stations near urban areas (i.e. Palaiseau and Cabauw). OA formed by the oxidation of mono-aromatic species is found to be negligible over Europe (less than 0.025 µg m -3 on average over the studied domain). Figure 15 shows the contribution of traffic emission to the simulated 25 OA concentrations for the July-August 2013 period for 3 categories of precursors: SVOCs, IVOCs and mono-aromatic compounds. As mentioned above, the OA concentrations from mono-aromatic compound oxidation are negligible compared to concentrations from traffic S/IVOC oxidation. Globally, in our study over Europe, OA concentrations produced from traffic S/IVOC oxidation are of the same order of magnitude. OA from primary SVOCs is locally more important close to sources (i.e. Northern Italy, Moscow, Paris, Gibraltar, etc). OA from IVOC is globally higher far away from the sources, 30 with a higher dispersion over Europe (Fig. 15). This higher dispersion is expected owing to the larger timescale required to produce low volatility species via multistep oxidation processes in the plumes of high emission area.
The distributions of OA within the volatility bins (given in Figure S2 of supplementary material) show similar features from one station to another. The results suggest that OA over Europe has relatively low volatility during summertime. Indeed, the VBS-GECKO contributors to OA have very low volatility: ~80% of the OA contributors from VBS-GECKO are volatility bins 7 to 5 (VB k,7 , VB k,6 and VB k,5 species), i.e. having saturation vapor pressure at 298 K of 10 -14 , 10 -12 and 10 -11 atm respectively.

Conclusions
The VBS-GECKO parameterization for SOA production was developed based on explicit mechanisms generated with the 5 GECKO-A tool. The VBS-GECKO parameterization was fitted using box modelling results for a selected set of parent compounds including terpenes, mono-aromatic compounds, linear alkanes and alkenes and for various environmental conditions, including different NO x regimes, temperatures, OA loads (Lannuque et al., 2018). In this study, the VBS-GECKO parameterization was evaluated in the CHIMERE β 2017 CTM over Europe during summertime.
The VBS-GECKO parameterization shows good performances to simulate OA concentrations over Europe in the summer. 10 Calculated mean fractional biases and mean fractional errors on PM 2.5 , OC PM2.5 and OM PM1 satisfy the performance criteria of Boylan and Russel (2006). The model configuration including the VBS-GECKO parameterization yields to higher OA concentrations compared to the former reference configuration including the H 2 O parameterization. The deviations between the two configurations are especially marked over northern Europe, with an increase factor of ~60%. Outside this area, the OA increases obtained with the VBS-GECKO configuration are slight. Statistically, the use of the VBS-GECKO improves 15 the overall MFB, MFE and RMSE and does not modify significantly correlation coefficients. Tests performed to examine the sensitivity of simulated OA concentrations to hydro-solubility, volatility, aging rates and NO x regimes have shown that the VBS-GECKO parameterization provides consistent results that are not subject to large deviations induced by parameters provided by the gas phase mechanism included in the CTM (e.g. HO x or NO x concentrations). However, the OA concentrations remain underestimated with the VBS-GECKO model configuration, especially in areas with a significant 20 contribution of anthropogenic sources (e.g. reaching a factor of 2.5 for OC PM2.5 at the NL0644R station in Netherlands).
None of the conducted sensitivity test leads to OA variations large enough to fill the gaps between measurements and simulated concentrations at the anthropogenic stations.
The analysis of simulated OA shows that, during summertime, the main fraction is made of secondary matter which represents ~85% of the total mean OA concentration. A large fraction of the simulated OA comes from biogenic sources 25 (between 30 and 85% of the total OA), especially from terpene oxidation which represents ~95% of these biogenic sources.
For the conditions examined in this study, OA formed by the oxidation of mono-aromatic compounds appears to be negligible with maximum mean concentrations of 0.025 µg m -3 over North Sea and Benelux. Note that ignoring SOA production from these precursors in the model would substantially reduce the number of VB k,i species currently considered in the VBS-GECKO parameterization. The simulated OA was found to be made of species having low and extremely low 30 volatilities in remote areas, but also of SVOCs closer to major anthropogenic sources.
Finally, IVOC oxidation was added to examine the contribution of this additional source to the SOA budget. Five model configurations with distinct IVOC emissions from traffic were tested and compared using the VBS-GECKO parameterization in CHIMERE. As expected, considering the emission of IVOCs by traffic and transport sources was found to globally increase background OA concentrations. Although SOA production from traffic IVOC oxidation can locally be significant (up to ~+3 µg m -3 in northern Italy, assuming IVOC emissions represents 65% of NMVOC emissions), this 5 additional OA source remains too small to explain the gap between simulated and measured values at stations where anthropogenic sources are dominant. This first application of this new VBS-GECKO parameterization has been shown to provide consistent results. This outcome creates motivation to extend the exploration to wintertime conditions, and expand the list of parent compounds considered, in particular to include SOA formation from oxidation of isoprene, sesquiterpenes or organics species emitted by residential biomass burning, a prerequisite to extend the evaluation and analysis to wintertime 10 when this source is dominant. This is the subject of ongoing studies. The VBS-GECKO is a heavy parameterization in term of species number. Calculation time is multiplied by two using the complete VBS-GECKO scheme with IVOCs compared to H 2 O. This study has shown that the number of species can be optimized. For example, because of the low influence on OA concentrations, the representation of the SOA formed by the oxidation of mono-aromatic species can be highly simplified and C 10 precursors even removed. 15

Data availability
Daily averages and mean day profiles for the 17 model configurations presented in this article have been made available on Zenodo: https://zenodo.org/record/ 1654297 / (last access: 29 November 2018).

Competing interests
The authors declare that they have no conflict of interest. 20

Author Contribution
VL implemented the parameterization in the air quality model and conducted the simulations and the sensitivity tests. All the authors contributed to design the research, to interpret the data and to write the article.
Simulations were performed using the TGCC-CCRT super computers.

Figure 4 -Measured (black) and simulated mean diurnal profile (in UTC) with the H 2 O model configuration (blue) and the ref-VBS-GECKO model configuration (red) for OM PM1 concentration at stations influenced dominantly by anthropogenic sources (Palaiseau (a) and Melpitz (b)) and by biogenic sources (Hyytiälä (c) and Birkenes II (d))
.         , g and h). Primary and secondary BBOA includes compounds from biomass burning. Traffic SVOC includes C 14 to C 26 VBS-GECKO alkanes and alkenes and SOA from traffic SVOC+IVOC oxidation includes their oxidation products. SOA from terpenes includes all species produced by α-pinene, β-pinene, limonene, ocimene and humulene oxidation.