Simulating secondary organic aerosol from missing diesel-related intermediate-volatility organic compound emissions during the Clean Air for London (ClearfLo) campaign

. We present high-resolution (5 km × 5 km) atmospheric chemical transport model (ACTM) simulations of the impact of newly estimated trafﬁc-related emissions on secondary organic aerosol (SOA) formation over the UK for 2012. Our simulations include additional diesel-related intermediate-volatility organic compound (IVOC) emissions derived directly from comprehensive ﬁeld measurements at an urban background site in London during the 2012 Clean Air for London (ClearfLo) campaign. Our IVOC emissions are added proportionally to VOC emissions, as opposed to proportionally to primary organic aerosol (POA) as has been done by previous ACTM studies seeking to simulate the effects of these missing emissions. Modelled concentrations are evaluated against hourly and daily measurements of organic aerosol (OA) components derived from aerosol mass spectrometer (AMS) measurements also made during the ClearfLo campaign at three sites in the London area. According to the model simulations, diesel-related IVOCs can explain on average ∼ 30 % of the annual SOA in and around London. Furthermore, the 90th percentile of modelled daily SOA concentrations for the whole year is 3.8 µg m − 3 , constituting a notable addition to total particulate matter. More measurements of these precursors (currently not included in ofﬁcial emissions inventories) is recommended. During the period of concurrent measurements, SOA concentrations at the Detling rural background location east of London were greater than at the central London location. The model shows that this was caused by an intense pollution plume with a Published by

Abstract. We present high-resolution (5 km × 5 km) atmospheric chemical transport model (ACTM) simulations of the impact of newly estimated traffic-related emissions on secondary organic aerosol (SOA) formation over the UK for 2012. Our simulations include additional diesel-related intermediate-volatility organic compound (IVOC) emissions derived directly from comprehensive field measurements at an urban background site in London during the 2012 Clean Air for London (ClearfLo) campaign. Our IVOC emissions are added proportionally to VOC emissions, as opposed to proportionally to primary organic aerosol (POA) as has been done by previous ACTM studies seeking to simulate the effects of these missing emissions. Modelled concentrations are evaluated against hourly and daily measurements of or-ganic aerosol (OA) components derived from aerosol mass spectrometer (AMS) measurements also made during the ClearfLo campaign at three sites in the London area. According to the model simulations, diesel-related IVOCs can explain on average ∼ 30 % of the annual SOA in and around London. Furthermore, the 90th percentile of modelled daily SOA concentrations for the whole year is 3.8 µg m −3 , constituting a notable addition to total particulate matter. More measurements of these precursors (currently not included in official emissions inventories) is recommended. During the period of concurrent measurements, SOA concentrations at the Detling rural background location east of London were greater than at the central London location. The model shows that this was caused by an intense pollution plume with a Published by Copernicus Publications on behalf of the European Geosciences Union. strong gradient of imported SOA passing over the rural location. This demonstrates the value of modelling for supporting the interpretation of measurements taken at different sites or for short durations.

Introduction
Ambient airborne particulate matter (PM) has diverse sources and physicochemical properties. It affects the transport, transformation, and deposition of chemical species, and has significant impacts on radiative forcing and on human health (Pöschl, 2005;USEPA, 2009). The elemental and organic carbon (EC and OC) components constitute a substantial proportion of total particle mass (USEPA, 2009;Putaud et al., 2010;AQEG, 2012). However, the characterization and source apportionment of the organic component remains a major challenge (Fuzzi et al., 2006;Hallquist et al., 2009;Jimenez et al., 2009). Understanding the sources of this organic aerosol (OA) is important in order to devise effective reduction strategies for ambient PM concentrations (Heal et al., 2012).
Organic aerosol is typically a complex mixture of thousands of organic species, the majority of which are present at low concentrations (less than a few ng m −3 ). Current levels of scientific understanding, instrumentation, and modelling capability mean that explicit measurement and modelling of all individual OA species is not feasible at present. Measurement of OA by on-line mass spectrometry, such as with the Aerodyne aerosol mass spectrometer (AMS; Canagaratna et al., 2007), and consideration of individual organic marker ions coupled with multivariate statistical techniques such as positive matrix factorization (PMF; Paatero and Tapper, 1994;Paatero, 1997) have facilitated the subdivision of the OA component into empirical categories. These include hydrocarbon-like organic aerosol (HOA), oxygenated organic aerosol (OOA, which can be further split into lowvolatility and semi-volatile oxygenated organic aerosol: LV-OOA and SV-OOA), solid-fuel organic aerosol (SFOA) or biomass burning organic aerosol (BBOA), cooking organic aerosol (COA), and a number of other categories Ng et al., 2010Ng et al., , 2011Lanz et al., 2010;Young et al., 2015a). The SFOA factor includes biomass aerosols that are the so-called BBOA as well as OA from coal and charcoal combustion (Allan et al., 2010).
Even allowing for the uncertainties in defining and measuring OA components, there is a general tendency for atmospheric chemical transport model (ACTM) simulations to underestimate observed amounts of OA and SOA. For example, the AeroCom (Aerosol Comparisons between Observations and Models) project, which includes ∼ 30 global ACTMs and global circulation models (GCMs), has concluded that the amount of OA present in the atmosphere remains largely underestimated (Tsigaridis et al., 2014). Similarly, in an eval-uation of seven global models, Pan et al. (2015) reported a systematic underestimation of OA over South Asia. Global modelling studies of SOA specifically have demonstrated huge uncertainties (up to 10-fold) in total simulated SOA budgets (Pye and Seinfeld, 2010;Spracklen et al., 2011;Jathar et al., 2011).
Currently, ACTMs cannot explicitly simulate all the kinetic and thermodynamic processes associated with the evolving gas-phase chemistry of semi-volatile organic compounds and their partitioning to the particle phase . A widely used heuristic parametrization for simulating OA is the volatility basis set (Donahue et al., 2011(Donahue et al., , 2012. The volatility (in this case the saturation concentration at 298 K, C * ) of gas-phase organic compounds are sorted into bins: low-volatility organic compounds (LVOCs, C * ≤ 0.1 µg m −3 ; with no lower C * , this category also incorporates extremely low-volatility organic compounds, ELVOCs), semi-volatile organic compounds (SVOCs, C * = 1-10 3 µg m −3 ), intermediate-volatility organic compounds (IVOCs, C * = 10 4 -10 6 µg m −3 ), and volatile organic compounds (VOCs, C * ≥ 10 7 ). Thus, organic compounds are distributed across a continuum from particles to gases. Under typical ambient conditions, all LVOCs, some of the SVOCs, and essentially none of the IVOCs or VOCs are in the condensed phase (Donahue et al., 2006). Current emissions inventories, however, only report estimates for VOCs (C * ≥ 10 7 µg m −3 ) and for the particle fraction of the emissions of species with lower volatilities. The main reason for this is that compounds with intermediate volatility (SVOCs and IVOCs) are difficult to quantify, and this is currently not routinely undertaken. Robinson et al. (2007) and Shrivastava et al. (2008) estimated the mass of emitted IVOCs to be 1.5 times that of POA emissions. In their study, this addition of IVOCs = 1.5 × POA was applied to all sources of POAfrom diesel to biomass burning. They based this estimation on chassis dynamometer tailpipe measurements by Schauer et al. (1999). Since then, several regional and global ACTM applications have adopted this factor of 1.5 (e.g. Murphy  Tsimpidi et al., 2010Tsimpidi et al., , 2016Hodzic et al., 2010;Jathar et al., 2011;Fountoukis et al., 2011;Genberg et al., 2011;Bergström et al., 2012;Zhang et al., 2013;Koo et al., 2014). A number of studies, including many of those cited above reporting model underestimation of OA, have highlighted the need for improved measurement and speciation of SVOCs and IVOCs and for these species to be reported in inventories. Jathar et al. (2014) performed emissions and smog chamber experiments on SOA production from petrol and diesel vehicles. Diesel contains hydrocarbons with a higher carbon number (C 8 -C 20 ) than petrol (mainly C 4 -C 10 ). The typical method used for hydrocarbon analysis is gas chromatography (GC); however, as the carbon number increases, the number of potential structural isomers increases exponentially, meaning GC is unable to distinguish individual species in the intermediate-volatility range (Goldstein and Galbally, 2007). The total carbon of this unresolved complex mixture was estimated and Jathar et al. (2014) concluded that these unspeciated organic gases dominate SOA production compared with SOA from the speciated precursors commonly included in emissions inventories (single-ring aromatics, isoprene, terpenes, and large alkenes). Jathar et al. (2014) also performed box-model simulations of the SOA budget for the US, with the addition of unspeciated emissions based on measurements by Schauer et al. (1999), and concluded that petrol contributes much more to SOA than does diesel. This result is similar to that of Bahreini et al. (2012), who, based on measurements in the Los Angeles Basin, California (CA), concluded that the contribution of diesel emissions to SOA was zero within measurement uncertainty. Conversely, Gentner et al. (2012) reported that diesel was responsible for 65-90 % of vehicular-derived SOA based on measurements of gas-phase organic carbon in the Caldecoff Tunnel, CA, and in Bakersfield, CA, and on estimations of SOA yields. The huge dissimilarity in these conclusions, even in the same state in the US, emphasizes the need for continued research into petrol-and diesel-related SOA formation. Furthermore, the US and Europe have very different diesel vehicle profiles: in the US, a negligible proportion of passenger cars are diesel (3 %), whilst on average across Europe 33 % of passenger cars are diesel and this proportion is increasing (Cames and Helmers, 2013). Globally, the demand for diesel fuel is increasing and by 2020 it is expected to overtake petrol as the principal transport fuel used worldwide (Exxon Mobil, 2014).
In this study, we present new high-resolution simulations of SOA formation in a 3-D ACTM model which includes additional diesel-related IVOC emissions derived directly from comprehensive field measurements of IVOCS and VOCs at an urban background site in central London (Dunmore et al., 2015) during the Clean Air for London (ClearfLo) campaign in 2012 (Bohnenstengel et al., 2014). Modelled concentrations are compared with OA components derived by PMF analysis of AMS measurements during the same cam-paign, including comparisons with the long-term measurements (full year) as well as the two month-long intensive observation periods (IOPs) in winter and summer.

Model description
The EMEP4UK model is a regional application of the EMEP MSC-W (European Monitoring and Evaluation Programme Meteorological Sythesizing Centre-West) model. The EMEP MSC-W model is a 3-D Eulerian model that has been used for both scientific studies and policy making in Europe. A detailed description of the EMEP MSC-W model, including references to evaluation and application studies, is available in Simpson et al. (2012), Schulz et al. (2013), and at http://www.emep.int. The EMEP4UK model is described in Vieno et al. (2010Vieno et al. ( , 2014, and the model used here is based on version v4.5. The EMEP4UK model uses one-way nesting from a 50 km × 50 km greater European domain to a nested 5 km × 5 km area covering the British Isles and parts of the near continent. The model has 21 vertical levels, extending from the ground to 100 hPa. The lowest vertical layer is ∼ 40 m thick, meaning that modelled surface concentrations represent the average for a 5 km × 5 km × 40 m grid cell. The model time step varies from 20 s (chemistry) to 5 and 20 min for advection in the inner and outer domains, respectively. The chemical scheme used in this study is EMEP-EmChem09soa with the MARS equilibrium module for gas-aerosol partitioning of secondary inorganic aerosol (Binkowski and Shankar, 1995;Simpson et al., 2012).
The model was driven by output from the Weather Research and Forecasting (WRF) model (http://www. wrf-model.org, version 3.1.1) including data assimilation of 6-hourly model meteorological reanalysis from the US National Center for Environmental Prediction (NCEP)/National Center for Atmospheric Research (NCAR) Global Forecast System (GFS) at 1 • resolution (NCEP, 2000). The WRF configuration was as follows: Lin Purdue for microphysics; Grell-3 for cumulus parametrization; Goddart Shortwave for radiation physics; and Yonsey University (YSU) for planetary boundary layer (PBL) height (see NCAR, 2008, for further information). This configuration is identical to that presented in Vieno et al. (2010), where it is shown to perform very well in comparison with measurements. No further evaluation is presented here.

SNAP1
Combustion in energy and transformation industries SNAP2 Residential and non-industrial combustion SNAP3 Combustion in manufacturing industry SNAP4 Production processes SNAP5 Extraction and distribution of fossil fuels SNAP6 Solvent and other product use SNAP7 Road transport SNAP8 Other mobile sources and machinery SNAP9 Waste treatment and disposal SNAP10 Agriculture UK and from CEIP (EMEP Centre on Emission Inventories and Projections; CEIP, 2015) for the rest of the model domain. All emissions are apportioned across a standard set of emission source sectors, following the sector structure defined in the Selected Nomenclature for Air Pollutant (SNAP; EEA, 2013; Table 1), consistently applied across the whole domain.
Primary PM emissions reported as PM 2.5 and PM 10 in the NAEI and CEIP were speciated into EC, OA from fossil fuel combustion, OA from domestic combustion, and remaining primary PM by source sectors (using splits developed by Kuenen et al., 2014, as in Fig. 1). Default speciation of NMVOC emissions into 14 surrogate groups was used . International shipping emissions from Entec UK Limited (now Amec Foster Wheeler) were used (Entec, 2010). The annual sectoral total emissions are temporally distributed to hourly resolution using hour-of-day, day-of-week, and monthly emission factors for each source sector as incorporated in the EMEP ACTM . Daily emissions of all the aforementioned trace gases and particles from natural fires were taken from the Fire INventory from NCAR version 1.0 (FINNv1; Wiedinmyer et al., 2011). Monthly NO x emissions from in-flight aircraft, soil, and lightning, as well as biogenic emissions of dimethyl sulfide (DMS), are included as described in Simpson et al. (2012). Biogenic emissions of isoprene and monoterpenes are calculated by the model for every grid cell and time step. Estimated emissions of wind-blown dust and sea salt are also included, but these have no impact on the model simulations of OA .

SOA production in the model
The EMEP MSC-W model uses the 1-D volatility basis set (VBS; Donahue et al., 2006) approach for SOA formation, ageing and phase partitioning. The implementation of the VBS framework within the model, including various options for the treatment of volatility distributions and ageing reactions is described by Bergström et al. (2012).
In the model set-up used here, POA is treated as nonvolatile and inert, as is currently assumed by emissions in-  (Table 1) as specified in the NAEI (for year 2012), with each sector split into POA (HOA or SFOA), EC, and remaining PM following Kuenen et al. (2014). The red bars are additional IVOCs (not included in official emission totals) that can be estimated as 1.5 × the POA mass in that sector. They are included in this plot to give an indication of the relative mass of IVOC additions that have been used in other studies.
ventories. Having POA be non-volatile allows us to better identify and isolate the SOA formed from our additional diesel IVOCs. Furthermore, it has been demonstrated by Shrivastava et al. (2011) that a two-species VBS simulates an evolution of oxygen : carbon ratios (O : C) similar to the ninespecies VBS approach. The two bins of Shrivastava et al. (2011) were of volatility 0.01 and 10 5 , which, because material with the lower volatility is always completely in the particle phase under ambient conditions, is similar to our nonvolatile treatment of POA. Five volatility bins (C * = 0.1, 1, 10, 100, 1000 µg m −3 ) are used for SOA production and ageing. The SOA yields for alkanes, alkenes, aromatics, isoprene, and terpenes under high-and low-NO x conditions were taken from Tsimpidi et al. (2010). Note that Tsimpidi et al. (2010) reported yields for the four VBS bins between 1 and 1000 µg m −3 . In this work, the lowest VBS bin (0.1 µg m −3 ) is used for the ageing reactions, as well as for SOA from the additional diesel IVOCs (explained in the next section). SOA from alkanes, alkenes, and terpenes is assumed to have an initial organic matter to organic carbon ratio (OM / OC ratio) of 1.7, while that from isoprene and aromatics is assumed to be 2.0 and 2.1, respectively Chhabra et al., 2010). For comparison, HOA and SFOA were assumed to have OM / OC ratios of 1.25 and 1.70, respectively (as in Bergström et al., 2012, based on Aiken et al., 2008. Both anthropogenic SOA (ASOA; from alkanes, alkenes, and aromatics) and BSOA (from isoprene and terpenes) undergo atmospheric ageing by the hydroxyl (OH) radical in the model (with rate coefficient of 4.0 × 10 −12 cm 3 molecule −1 s −1 ; Lane et al., 2008), resulting in a shift into the next lower volatility bin and a mass increase of 7.5 %.
A constant background OA of 0.4 µg m −3 is used to represent the contribution of OA sources not explicitly included in the model (e.g. oceanic sources or spores; Bergström  2014). This background OA is assumed to be highly oxygenated and is therefore included under modelled SOA when comparing with observations (with an OM / OC ratio of 2.0 it is also assumed to be non-volatile).

Additional IVOCs from diesel
Current emissions inventories report highly volatile anthropogenic VOCs of C * ≥ 10 7 µg m −3 (Passant, 2002). However, diesel vehicles also produce substantial emissions of species with intermediate volatility in the range 10 5 ≤ C * ≤ 10 6 µg m −3 (IVOCs), as has been shown by Dunmore et al. (2015) from measurements made at an urban background site in central London during the ClearfLo project.
In this study, aliphatic IVOC emissions from diesel vehicles were introduced into the model proportionally to on-road transport VOC emissions, using n-pentadecane (C 15 H 32 ) as a surrogate for the following reasons. First, the amount of alkenes in diesel fuel is low (< 5 %; Gentner et al., 2012), so an alkane is the most appropriate surrogate. Second, all n-alkanes up to n-dodecane were individually speciated and quantified during two month-long intensive observation periods (IOPs) during the ClearfLo project and there were strong correlations between all n-alkanes that have a predominately diesel source (Dunmore et al., 2015). Third, the rate constant for the linear alkane is a reasonable representation of the rate constant for all the (unmeasured) branched and cyclic isomers, as demonstrated by Dunmore et al. (2015) for the C 12 n-alkane, dodecane. The bulk of diesel emissions, however, are likely to have higher carbon numbers than were measured by a comprehensive twodimensional gas chromatography (GC × GC) system (Dunmore et al., 2015). The rate coefficient for the reaction between n-pentadecane and OH has been measured in a number of studies (k = 2.07 × 10 −11 cm 3 molecule −1 s −1 ; Atkinson and Arey, 2003) unlike for the majority of branched isomers in this range. Furthermore, measurements of diesel fuel composition have shown that the average carbon number on a percentage weight basis was 14.94 (Gentner et al., 2012) Figure 2. Annual UK NMVOC emissions by SNAP sector (Table 1) as specified in the NAEI (for the year 2012), with the SNAP7 emissions subdivided into petrol and diesel vehicles, and with the additional diesel-associated IVOC emissions input to the model in this study shown in red.
n-pentadecane was considered to be an appropriate surrogate for diesel emissions in general.
In the NAEI, emissions from petrol vehicles dominate the NMVOC emissions from road traffic, but measurements during the ClearfLo winter IOP showed that NMVOCs assigned to diesel vehicles dominated traffic-related NMVOC concentrations. The amount of pentadecane emitted in the model was therefore set to match the measured diesel-(I)VOC (IVOCs + VOCs) to petrol-VOC ratio (Table 2, Fig. 2). This pentadecane addition was then applied to every country in the model domain using the same factor as for the UK. This first approximation is justified because the fleet share of diesel vehicles in the UK is similar to the European average (∼ 30 %; EEA, 2010), but it can introduce errors for specific countries.
For the oxidation products of C 15 H 32 , SOA mass yields were taken from Presto et al. (2010): 0.044, 0.071, 0.41, and 0.30 for the 0.1, 1, 10, and 100 µg m −3 bins, respectively (Presto et al., 2010, did not report a yield for the 1000 µg m −3 bin). These yields are reported for SOA with unit density (1 g cm −3 ). In this work, SOA density was assumed to be 1.5 g cm −3 (Tsimpidi et al., 2010;Bergström et al., 2012) and the yields were increased accordingly.
For the UK, our approach adds 90 Gg of diesel-IVOC emission for the year 2012 (Fig. 2). The 1.5 × POA approach (Shrivastava et al., 2008, based on measurements by Schauer et al., 1999) would only add 31 Gg (Fig. 1). Part of this discrepancy could be attributable to the different methods and circumstances used to derive the additions (this work: 5 weeks of ambient measurements in a megacity; previous estimate: tailpipe laboratory measurements using different instruments). Another possible reason for the difference is an underestimate in POA emissions in the inventory; more POA would increase the amount of proportionally added IVOCs. However, Dunmore et al. (2015) show that lower-carbon-number (and higher-volatility) NMVOCs measured during the ClearfLo campaign were consistent with emissions estimates. This lends confidence to adding IVOCs proportionally to reported NMVOC emissions, rather than proportionally to POA emissions. Nevertheless, we have also performed a model run using the POA-based IVOC estimate, also including the semi-volatile treatment of POA. The emitted semi-volatile POA (SVOCs) and 1.5 × POA IVOCs are assigned to nine VBS bins -0.03 × POA, 0.06 × POA, 0.09 × POA, 0.14 × POA, 0.18 × POA, 0.30 × POA, 0.40 × POA, 0.50 × POA, 0.80 × POA to the bins 0.01-10 6 , respectively -totalling 2.5×POA (Shrivastava et al., 2008). Both SVOCs and IVOCs then go through atmospheric ageing with OH (k = 4.0 × 10 −11 cm 3 molecule −1 s −1 ; Shrivastava et al., 2008). In this case, the additional IVOCs were calculated from POA from all sources, not just trafficrelated. Note that in the UK, most of the additional IVOCs of the POA-based approach would come from SNAP2 (residential and non-industrial combustion emissions; Fig. 1): 18 Gg, whereas only 5 Gg would be added to SNAP7 (road transport; and 8 Gg to remaining sectors). SVOCs and IVOCs that have undergone at least one ageing shift and are in the particulate phase are included under SOA (in addition to ASOA and BSOA from VOCs as in the Base case). Due to the very different absolute amounts and source categories (the latter of which also leads to differences in the spatial pattern and temporal variation of the additional emissions), detailed comparison of the two different additions is not justified, and only annual total OA component budgets of the different addition methodologies are presented.

Summary of model experiments
Three runs of the EMEP4UK modelling system were performed for 2012: -Base: all anthropogenic emissions as in officially reported inventories; emissions of biogenic VOCs are calculated by the model for each advection time step.
-addDiesel: Base + additional diesel IVOCs added proportionally to NMVOC emissions from traffic (2.3 × SNAP7). The additional IVOCs were treated using n-pentadecane as surrogate species. The semivolatile VBS species formed after oxidation of npentadecane were treated in the same way as the ASOA species from VOC oxidation (the same ageing rate and mass increase due to oxygen addition; see Sect. 2.3).
-1.5volPOA: semi-volatile treatment of POA + additional IVOCs added proportionally to all POA emissions (1.5 × POA; as in Shrivastava et al., 2008, based on measurements by Schauer et al., 1999). Inclusion of anthropogenic and biogenic VOCs as in Base.

Comparison with measurements
Modelled OA 2.5 (OA with diameter < 2.5 µm) is compared with non-refractory submicron (NR-PM 1 ) OA measured by Aerodyne AMS instruments at an urban background site in central London and at two rural sites (Xu et al., 2016;Young et al., 2015a, b;Bohnenstengel et al., 2014; site locations are shown in Fig. 3). The error introduced to the comparison by the different size fractions is believed to be small, as measurements at an urban background site in Birmingham, England have shown that 90 % of organic carbon in PM 2.5 is in the submicron fraction (Harrison and Yin, 2008).
Different types of AMS were deployed in this campaign. At the London North Kensington site a compact time-offlight AMS (cToF-AMS) was deployed for a full calendar year (January 2012-January 2013), and a high-resolution time-of-flight AMS (HR-ToF-AMS) was also deployed for the IOPs at the same site. A HR-ToF-AMS was deployed in Detling during the winter IOP, and in Harwell during the summer IOP. PMF analysis was applied to each of the data sets to apportion measured OA into different components . A detailed description of the derivation and optimization of the factors retrieved from the AMS data at Detling can be found in Xu et al. (2016), at London North Kensington in Young et al. (2015a, b) (all three of these analyses were performed with the PMF2 solver), and at Harwell in Di Marco et al. (2016) (using the ME-2 solver). The OM / OC ratios for each of the PMF data sets presented in this study were calculated with the Improved-Ambient method from Canagaratna et al. (2015). A summary of the instruments, measurement periods, and resolved PMF factors is given in Table 3. As our emissions inventory does not include cooking OA (NAEI, 2013), this factor could not be compared with the model.
When AMS measurements and their PMF apportionments are compared, some disagreement is observed, as shown for the two instruments measuring at the same time at the same location at London North Kensington. This is in part due to 11 Jan-24 Jan (2013) * cToF-AMS HOA, SFOA * * , COA OOA1, OO2 * * * As the cToF-AMS was retuned before the summer IOP and retuned to the previous tuning at the end of the IOP, the subsequent data could not be used in the PMF analysis (see Young et al. (2015a) for details). However, for the purpose of the comparison in this study, data from the HR-ToF-AMS, deployed at the same site during the summer IOP, were used to fill in this period. * * PMF analysis revealed the SFOA and OOA2 factors were convolved due to their similar, strong diurnal cycles. Daily averages have been used to estimate their concentrations (Young et al., 2015a). the differences in the types of AMS used, where more chemical information is retrieved from the HR-ToF-AMS, which can subsequently lead to differences in the derived PMF factors from the individual data sets. It should also be kept in mind that PMF was run on each of the full data sets, covering a full year for the cToF-AMS and only four weeks for each of the HR-ToF-AMS IOPs; thus, it is not necessarily expected that the same PMF factors would be derived from the different data sets. Nevertheless, strong correlations between daily averaged primary OA components from the two instruments deployed at the London North Kensington site during the winter IOP are observed (0.95, 0.92, and 0.88 for HOA, SFOA, and COA, respectively), with less strong correlations for SOA (0.77). Scatter plots of these PMF derived OA component concentrations resolved for the cToF-AMS data and HR-ToF-AMS are shown in Fig. 4. This inherent uncertainty in the measurements constrains the expected correlation with the model.
The following numerical metrics were used for model evaluation: -FAC2 (factor of 2): the proportion of modelled concentrations that are within a factor of 2 of the measured concentrations.
-NMGE: normalized mean gross error, which is defined as where M i is the ith modelled value, O i is the corresponding measured value, O is the mean measured value, and n in the total number of observations.
-COE: coefficient of efficiency, which is defined as A COE of 1 indicates perfect agreement between model and measurements. Although the COE does not have a lower bound, a zero or negative COE implies that the model cannot explain any of the variation in the observations (Legates and McCabe, 2013). Seasons are defined as follows: winter -December-January-February (DJF); spring -March-April-May (MAM); summer -June-July-August (JJA); and autumn -September-October-November (SON).

Results
The comparisons between the model results and measurements are presented in the following order. First, comparisons are presented for primary OA, NO x , O 3 , and for secondary inorganic aerosol (SIA) to give an overview of the overall performance of the modelling system. Second, the hourly concentrations of SOA during the two IOPs are evaluated, demonstrating the agreement between the model and measurements at high temporal resolution. Third, the yearlong daily SOA concentrations are compared and the relative impact of diesel VOCs on SOA production in London is shown. Fourth, modelled and measured OM / OC ratios are shown. Finally, annual total ASOA from our method and the previous 1.5 × POA approach are compared.
3.1 POA, NO x , O 3 , SIA: annual data set Figure 5 shows the year-long comparison between the daily averaged model results and the cToF-AMS measurements at the London North Kensington site. The model underestimates primary OA (HOA and SFOA) concentrations (NMB of −54 and −71 %, respectively) but shows good daily correlations (r values of 0.53 and 0.72, respectively). The underestimation of HOA may be caused by a combination of lack of model resolution (e.g. the minor road close to the measurement site can not be resolved with the 5 km grid), and underestimation of PM emissions. Modelled NO x concentrations are relatively less underestimated in comparison to measurements (NMB of −32 %, Fig. 6a), suggesting that HOA emissions may be more underestimated than the emissions of NO x . Concentrations of secondary inorganic pollutants are simulated well by the model in the gas phase (Fig. 6b, with a NMB of −1 % for ozone) and for inorganic PM constituents (Fig. 6c-d), with NMBs of 6 % for SO 2− 4 , −12 % for NH + 4 , and −23 % for NO − 3 .

Hourly comparison of secondary OA: summer IOP
Evaluation statistics between hourly measured and modelled SOA concentrations in July and August 2012 (summer IOP) show excellent agreement (Fig. 7). The values of r for the Base run were 0.67 and 0.55 at North Kensington and Harwell, respectively. The addDiesel experiment yields a modest improvement in the value of r at North Kensington (to 0.76) and a marked improvement in Harwell (to 0.74). The addDiesel run substantially improves the NMB for SOA at the Harwell and London North Kensington sites from −32 to −5 % and from −35 to 0.1 %, respectively (Fig. 7). This means that ∼ 30 % of SOA at both sites during this period can be explained by the diesel IVOCs added into the model using pentadecane as a surrogate. There is also marked improvement of model-measurement COE values at the two sites (Harwell, 0.26 to 0.42; NK, 0.31 to 0.45). The improvement in NMGE is noticeable (Harwell, 54 to 43 %; NK, 59 to 47 %), but smaller than the improvements in the other metrics. It can be seen from the scatter plots in Fig. 8 that most modelled hourly SOA concentrations fall within a factor of 2 of the measured concentrations (FAC2 for the addDiesel experiment is 78 % at Harwell and 62 % at NK). Measured and modelled mean hour-of-day variations of SOA concentrations are presented in Fig. 9, where it can be seen that measured SOA concentrations do not have a very strong diurnal cycle. Interestingly, both sites exhibit dips in measured SOA concentrations in the morning and early evening. Both measured and modelled SOA concentrations in London North Kensington reach a maximum in the afternoon, but SOA of the addDiesel experiment starts this increase earlier than the measurements, meaning that our ASOA production from pentadecane might be too rapid.
During the summer IOP, there were two sustained episodes of increased SOA concentrations: 23 to 28 July and 9 to 13 August (Fig. 7). Only London North Kensington had measurements during the first episode and the elevated concentrations were well captured by the addDiesel simulation (including the highest peak of greater than 16 µg m −3 : 27 July 13:00 GMT, Fig. 10b). Daily averaged SOA maps (Fig. 11) suggest that this first episode arose from a combination of SOA transported from Europe and SOA produced locally in London. A region of elevated concentration around London exists within a general gradient of SOA from continental Europe to southern England. Even daily averaged concentrations are spatially variable during this episode meaning that inaccuracies in some of the modelled peaks can be attributed to uncertainties in the underlying meteorological model. Most of the modelled SOA during this episode was of anthropogenic origin with the addDiesel run yielding a significant portion of ASOA from pentadecane.
For the second sustained episode of high SOA concentrations, from 9 to 13 August, several features remain substantially underestimated even in the addDiesel run. For Harwell, the model captures two of the highest peaks (10 August 22:00 GMT measured: 6.8 µg m −3 , addDiesel: 8.5 µg m −3 ; 12 August 12:00 GMT measured: 7.9 µg m −3 , addDiesel: 7.0 µg m −3 ), but for London North Kensington, the model simulates a minimum during the highest measured concentration (10 August 05:00 GMT measured: 11.9 µg m −3 , addDiesel: 2.0 µg m −3 ). The high concentrations during the first 2 days of this episode were very localized with horizontal widths of just tens of kilometres ( Fig. 12a and b). There was a build-up of pollution caused by high pressure and low boundary layer height (BLH), which led to production of ASOA in London. The high variability in the modelled con-  (Table 3).  (Table 3). J u l 2 1 J u l 2 2 J u l 2 3 J u l 2 4 J u l 2 5 J u l 2 6 J u l 2 7 J u l 2 8 J u l 2 9 J u l 3 0 J u l 3 1 A u g 0 1 A u g 0 2 A u g 0 3 A u g 0 4 A u g 0 5 A u g 0 6 A u g 0 7 A u g 0 8 A u g 0 9 A u g 1 0 A u g 1 1 A u g 1 2 A u g 1 3 A u g 1 4 A u g 1 5 A u g 1 6 A u g 1 7 A u g 1 8 A u g 1 9 A u g 2 0 Hourly SOA concentration,  centrations (for example, the simulated minimum during the measured maximum at North Kensington) is caused by the shifting of this narrow ASOA plume in space (Fig. 10b). On 12 August, this episode was also subject to SOA contribution from Europe (Fig. 12d). During the period of overlapping measurements at Harwell and North Kensington (3-18 August), both the measurements and the model agree with a modest rural to urban increase. Average measured SOA concentrations were 2.4 and 2.6 µg m −3 for Harwell and North Kensington, respectively, whilst average modelled concentrations were 2.3 and 2.5 µg m −3 (for the addDiesel experiment).

Hourly comparison of secondary OA: winter IOP
Both the Detling and London North Kensington sites exhibit good model-measurement hourly correlation (r = 0.63 and 0.64, addDiesel run; Fig. 13). The addDiesel run decreases the NMB for SOA at these sites from −59 to −30 % for Detling, and from −24 to 8 % for London North Kensington. This means that ∼ 30 % of SOA at these sites during this period can be explained by diesel IVOCs. In Detling, there is also a pronounced improvement in the COE, Atmos. Chem. Phys., 16, 6453-6473, 2016 www.atmos-chem-phys.net/16/6453/2016/ from 0.10 to 0.31. In North Kensington, the COE was already high but is increased from 0.27 to 0.30. It can be seen in Fig. 13 as well as Fig. 14 that lower concentrations of SOA (19)(20)(21)(22)(23)(24)(25)(26)(27) are overestimated by the model. This overestimation is caused by the very simplified method of including missing sources of OA using a constant concentration of 0.4 µg m −3 (which is assumed to be highly oxygenated and is therefore included under modelled SOA). As a constant, this background OA does not currently go through atmospheric emission-removal processes in the model. However, the period in question exhibited snowfall, removing much of the aerosol (as can be seen from the very low concentrations measured in both Detling and London North Kensington). Explicit inclusion of additional missing biogenic sources of OA to the model is already part of ongoing development of the model and will be presented in future studies. During the ClearfLo Winter IOP, measured SOA concentrations were higher in Detling than in North Kensington (Fig. 13). This is correctly captured by the simulations and is caused by a steep positive gradient of concentrations from southern England across to the near European continent (Fig. 15). The measured Detling / North Kensington SOA ratio (ratio of average concentrations for this period) was 1.8, while the modelled ratio was 1.1, so the model correctly simulates the direction of the spatial gradient, but underestimates its magnitude. For North Kensington, the model also captures that SOA concentrations are lower on 5 February than on 4 February. In Detling, however, measured concentrations were higher on 5 February, which the model does not reproduce. During the night of 4-5 February, the wind was very strong (> 10 m s −1 ) and there was a small shift between the measured wind direction and the wind direction input to EMEP4UK from WRF. As a consequence, the simulated pollution plume was shifted too much to the east (Fig. 15b), causing the model-measurement discrepancy on this particular occasion.
Even though the additional diesel IVOCs noticeably increased the modelled SOA concentrations during the winter IOP, there is still a marked underestimation of elevated measured SOA concentrations during 15-19 January and 30 January-4 February. During these periods, the observed temperature was colder than the average temperature of the winter IOP (Crilley et al., 2015) and peaks in measured SOA also coincide with elevated concentrations of SFOA (Figs. 5b and 13b). As our modelled SFOA is underestimated by a factor of 4 (NMB of −72 %), it is likely that (i) SOA precursor VOC emissions from domestic heating are also underestimated, and (ii) adding missing IVOCs from this emission sector would contribute to the modelled SOA during these periods. It has been recently shown by Denier van der Gon et al. (2015) that the emission factors used by different European countries for wood combustion PM emissions, even for the same appliance type, can differ by a factor of 5. They constructed a revised inventory, in which each country's emission was updated using an unified emission factor. This resulted in increases of PM (and estimated accompanying IVOC) emission estimates for most countries. Furthermore, London is a smoke control area and therefore no residential emissions of SFOA are assumed by the national emissions inventory. Recent studies have, however, suggested that there are indeed local sources of SFOA in London (Crilley et al., 2015;Young et al., 2015a).

Daily and seasonal secondary OA: annual data set
Time series of daily averaged modelled and measured SOA concentrations for the whole year are shown in Fig. 16. Table 4 gives daily modelled vs. measured SOA evaluation statistics during different seasons at the North Kensington site. Values for autumn are presented with and without the two extreme points (size of the data set n = 91 and n = 89).
For the daily model-measurement comparison, spring has the highest correlation (r = 0.85, both Base and addDiesel; Table 4). This can also be seen from the time series (Fig. 16: March-May) where both model simulations follow most of the measured peaks. The Base run r value for spring was already high, but, nevertheless, the addDiesel run shows a marked improvement for all other model evaluation statistics. FAC2 is increased by 10 %, COE is increased to 0.39, NMB is reduced by 35 %, and NMGE is reduced by 7 %. The NMGE of 38 % remaining in the addDiesel model run is probably governed by uncertainties in meteorology, as well as by uncertainties in the temporal and spatial variability of emissions. During summer, the model captures the majority of the periods of increased SOA mass well (e.g. 28 June, 22-29 July, 15 and 20 August; Fig. 16: June-August), but there is some model underestimation when SOA concentrations were lower (< 2 µg m −3 ). As for spring, the addDiesel experiment improves all model evaluation statistics. More detailed   hourly analysis of the SOA concentrations during the summer IOP (end of July to August) was presented in Sect. 3.2.
The model performance is less good in autumn than during the other seasons. There are some days where the Base case scenario overestimates measured SOA (23-25 October, 21 and 24 November) with the addDiesel run increasing this further. During these days, particle nitrate (NO − 3 ) and ammonium (NH + 4 ) are also substantially overestimated by the model (Fig. 6). This suggests that the overestimations are likely caused by errors occurring during this period in the meteorological forecasts, e.g. missed rain events, rather than by uncertainties in the formation of secondary organic aerosol specifically. The model evaluation statistics for autumn are strongly influenced by the two modelled values on 23 and 24 October (Table 4) average SOA concentration modelled with the addDiesel run by 33 % (2.0 and 1.5 µg m −3 with and without these two points, respectively). Their combined influence on the annual average modelled concentration is 8 %, which is substantially more than any other points of the annual data set. For the winter months, modelled concentrations in January are much lower than measurements, whereas in February the timing of several peaks is well reproduced and even overestimated by the addDiesel experiment. Detailed hourly analysis of the SOA concentrations during the winter IOP has been presented in Sect. 3.3. In December, measured SOA concentrations were much lower than in January, and even though the model captures the highest peak, there is some overestimation in the lowest range (< 0.5 µg m −3 ). Figure 17 shows annually and seasonally averaged measured and modelled SOA. The difference between the Base and addDiesel experiments illustrate the impact of missing IVOC emissions from diesel traffic on SOA formation. As was discussed before, and can be seen from Table 4, IVOC precursors from diesel vehicles reduce the NMB by ∼ 30 %, which as an annual average is 0.6 µg m −3 of additional SOA. Moreover, the 90th percentile of daily averaged SOA concentrations of the addDiesel experiment is 3.8 µg m −3 (which is similar to the measured 90th percentile of 3.2 µg m −3 ), whereas the 90th percentile of the Base case simulation is 2.2 µg m −3 . This means that (i), on 36 days of the year, SOA is a notable component of PM (the annual average PM 2.5 concentration limit value of the European Union Directive 2008/50/EC is 25 µg m −3 ) and that (ii), during those days, the relative contribution to SOA from diesel IVOCs could be greater than 40 % (calculated as the difference between SOA modelled with addDiesel and Base, relative to addDiesel: (addDiesel-Base)/addDiesel). We note that Fig. 17a shows in the addDiesel simulation that the modelled BSOA + Background OA still makes up 53 % of the SOA, as an annual average. This value is based on the assignment of the constant background OA in the model to natural SOA, which is what it is intended to represent. This may have some anthropogenic origin, and more research on the missing (or boundary condition) sources that this background constant represents is needed for accurate attribution of the biogenic vs. anthropogenic relative contributions.

OM / OC ratios
Measured OM / OC ratios for SOA were generally higher than those modelled (1.99-2.34 vs. 1.88-1.97, Table 5). Nevertheless, the measured OM / OC ratio at London North Kensington during the summer IOP was the lowest of the measured range, 1.99, which is a close match to modelled SOA OM / OC ratio for that period, 1.97. Model performance for spring and summer was shown to be very good, but   it is possible that the missing SOA precursors in the colder months (from domestic heating) could yield SOA with higher initial OM / OC ratios, thereby increasing the annual average value. Furthermore, wintertime simulations of SOA in Paris by Fountoukis et al. (2016) also showed large underestimations, and they speculated that this could be pointing towards an SOA formation process during periods of low photochemical activity that is currently not simulated in atmospheric chemical transport models.
3.6 Comparison to the previous (IVOCs = 1.5 × POA) approach Figure 18 shows the annual average HOA, SFOA, BSOA and background OA (BGND OA), and ASOA concentrations at London North Kensington modelled with different assumptions for additional IVOC emissions. As was explained in Sect. 2.4, for the UK, the addDiesel experiment adds 90 Gg of diesel-related IVOCs proportionally to road transport emissions (SNAP7), whereas the IVOCs = 1.5 × POA approach only adds 5 Gg to SNAP7 and another 26 Gg to other sectors (mainly to SNAP2: residential and non-industrial combustion). Therefore, our approach creates a considerably larger amount of SOA from IVOCs (and only from dieselrelated IVOCs) than the previous method. The 1.5volPOA experiment was undertaken using the semi-volatile treatment of POA emissions. This means that the modelled ASOA from this experiment also includes aged semi-volatile POA, possibly giving it potential to create more ASOA than the Base or addDiesel experiments (the organic material added to the model in the 1.5volPOA experiment is 1.0 × POA (as SVOCs) + 1.5 × POA (IVOCs) = 2.5 × POA as introduced by Robinson et al. (2007) and Shrivastava et al. (2008)  can be seen from Fig. 18a and b that treating POA as semivolatile leads to much lower concentrations than the nonvolatile treatment (which already underestimates measured concentrations of HOA and SFOA by −54 and −71 %, respectively; Fig. 5). This is not surprising given that the semivolatile treatment of POA assigns only 3 % + 6 % + 9 % of the POA to the three lowest volatility bins with saturation concentrations of 0.01, 0.1, and 1 µg m −3 , respectively (as given in Sect. 2.4). In a study in Mexico City, Shrivastava et al. (2011) revised this treatment by assuming much higher total semi-volatile and intermediate-volatility POA emissions: 7.5 × the inventory emissions of (particulate) POA. This was justified by the fact that their emission factors of POA were derived from measurements at urban background sites, but, following Robinson et al. (2007), two-thirds of POA would have evaporated by then. Recently, Shrivastava et al. (2015) also used this factor of 7.5 in global simulations. Emission factors used in European inventories are, however, taken from tailpipe measurements with concentrations sufficiently high that most of the semi-volatiles should still be reported in the particulate phase. Therefore the further underestimation of HOA and SFOA concentrations with the volatile treatment could be due to a number of issues: (i) a systematic underestimation of emissions, but for a different reason than in Shrivastava et al. (2011); (ii) the volatility of POA is overestimated by Robinson et al. (2007), or (iii) the evaporation of semi-volatile POA emission is too rapid in the model (instantaneous in our set-up). Figure 18c shows that the lower HOA and SFOA concentrations lead to a very small negative change for the absorptive partitioning of BSOA. Finally, it can be seen from the annual average concentrations of ASOA in Fig. 18d that including aged SVOCs and IVOCs in the simulation doubles the modelled ASOA concentration compared to the Base case scenario (ASOA from officially reported anthropogenic VOCs) but that the ASOA in our 1.5volPOA experiment is still much lower than simulated with our addDiesel experiment. Base -all emissions as in officially reported emissions inventories, and POA is treated as non-volatile; 1.5volPOA -semi-volatile treatment of POA + IVOC emissions added as 1.5 × POA; addDiesel -Base + IVOC emissions from diesel traffic added proportionally to VOC emissions from the on-road traffic source sector (SNAP7). The last two additions are as described in the main text.

Discussion
We show that ∼ 30 % of SOA in London could be produced from completely new estimates of diesel-related IVOC emissions that are not currently included in the emissions inventories. This is one of a very few studies where IVOC emissions are added proportionally to NMVOC emissions (as opposed to addition proportionally to POA emissions). Moreover, previous studies have added IVOCs proportionally to POA from all sources, whereas this study focuses specifically on the impact of diesel IVOCs from on-road traffic emissions (IVOCs = 2.3 × SNAP7 VOCs). There is reason to believe that higher-volatility VOCs are better represented in current emissions inventories than the emissions of PM. Also, the official inventories do not provide the individual contribu-tion of POA to total PM. Therefore, the addition of IVOCs proportionally to NMVOCs may be better constrained than the POA-based approach used in studies so far. The additional emissions are also tied directly to the relevant emission source category. There are several possible uncertainties in our estimate of additional IVOCs, and subsequent SOA production and ageing. As a first approximation, we added IVOCs to each European country based on our measurements in London. This was justified as the diesel usage in the UK is similar to the European average. Furthermore, different European countries might be using different emissions factors for their estimates of NMVOCs from petrol and diesel or have a different average fleet age than the UK. It should be noted that two of the most populous countries in Europe -France and Germany -both have a higher diesel penetration than the UK ,and therefore for western central Europe our addition is rather conservative.
It was seen from the hourly profiles at the London North Kensington site during the summer IOP (Fig. 9b) that both the model and the measurements exhibit a small diurnal cycle (peaking in the afternoon). Even though somewhat counterintuitive (as most of the SOA chemistry is photochemically driven through reaction with the OH radical), an absence of a strong diurnal cycle of SOA has been seen in many European studies (Zhang et al., 2013;Fountoukis et al., 2014;Young et al., 2015a). A relatively small daytime increase in SOA could be explained by the expansion of the boundary layer height (Xu et al., 2015), as well as by contributions from long-range transport. PMF measurements of SOA in Mexico City, on the other hand, revealed a very strong diurnal cycle, peaking around midday (Shrivastava et al., 2011). The fact that during the summer IOP our addDiesel experiment exhibits a slightly stronger diurnal cycle than the measurements (with daytime values slightly overestimated and nighttime underestimated) indicates that the SOA yields could be too high. We assumed an SOA density of 1.5 g cm −3 and increased the yields linearly, as has been done in all other ACTM studies. Actually, increasing the assumed density of SOA from the unit value (1 g cm −3 ) changes the total C OA (condensed-phase OA) on the Odum mass yield plots (Odum et al., 1996) used to derive the yields from the chamber experiment. Therefore, increasing the yields linearly is not exactly correct (N. M. Donahue, personal communication, 2015) and further studies and refinement into the calculation of SOA yields and density would be beneficial.
We use an ageing rate of 4.0 × 10 −12 cm 3 molecule −1 s −1 for both ASOA and BSOA (Lane et al., 2008). This is slower than has been used in some other studies (for example, Tsimpidi et al., 2010, use 4.0 × 10 −11 cm 3 molecule −1 s −1 , which is 10 times faster, and Fountoukis et al., 2011, use 1.0 × 10 −11 cm 3 molecule −1 s −1 , which is 2.5 times faster). A combination of lower initial SOA yields but slightly higher ageing rates could possibly flatten the diurnal cycle of our modelled SOA, matching the measurements better. There-fore, an improvement for the detailed, hourly evolution could be achieved by a sensitivity study of these yields and ageing rates. This does not, however, change the main scope and results of this paper which illustrate the relative impact of the diesel IVOCs on SOA formation.
In the current set-up of the EMEP model, only two PM size fractions are simulated -PM 2.5 and PM 2.5−10 -because only two fractions are included in the emissions inventories for PM used in this study. Even though on an annual basis 90 % of OC 2.5 is in the submicron (OC 1 ) range (Sect. 2.6), the comparison between a modelled OC 2.5 and a measured OC 1 could be introducing larger errors during specific days or hours. Therefore, as AMS measurements become more prevalent, emissions inventories should be reported for all three size classes, PM 1 , PM 1−2.5 , and PM 2.5−10 . This would allow the model to partition SOA into the corresponding fractions, making the direct comparison of modelled SOA 1 to measured SOA 1 possible.
We showed that treating POA as semi-volatile and letting it evaporate led to a great underestimation of HOA and SFOA concentrations compared to measurements at the London North Kensington urban background site. As has been highlighted by a number studies before us (listed in the Introduction), we would also emphasize that a major source of uncertainty in OA modelling is the volatility of primary emissions, an issue that is currently not addressed by official emissions inventories. In our experiment of semi-volatile POA (denoted 1.5volPOA), IVOCs were included from all source sectors. This experiment simulated substantially less ASOA than our addition of IVOCs associated with just the traffic source sector. This means that a combination of the POAbased IVOCs and our addition of diesel IVOCs proportional to NMVOCs would not create a substantial overestimation of SOA concentrations compared to measurements. Nevertheless, further modelling studies (including different assumptions regarding ageing rates, fragmentation, and yields) as well as more measurements of IVOC emissions from different sources are clearly necessary.
In the evaluation of modelled and measured SOA, it was shown that some of the uncertainties in the modelled concentrations are caused by errors in modelled wind vectors. Nevertheless, the underlying meteorological model works well (as demonstrated by comparisons of different pollutants for the whole calendar year), and overall the errors caused by meteorology are believed to be relatively smaller than those introduced by emissions (amount, volatility, composition), or SOA yields and ageing rates.

Conclusions
This study presents annual time series of new high-resolution simulations of SOA formation over the UK (using the EMEP4UK Eulerian atmospheric chemical transport model) that include diesel-related IVOC emissions not currently in-Atmos. Chem. Phys., 16, 6453-6473, 2016 www.atmos-chem-phys.net/16/6453/2016/ cluded in the emissions inventory. The derivation of the magnitude of these additional emissions of SOA precursors and the evaluation of the model simulated SOA were both based on measurements made during the Clean Air for London (ClearfLo) campaign in 2012. The IVOC emissions were added in proportion to the VOC emissions from the specifically relevant on-road traffic source, in contrast to previous studies that have added IVOCs proportionally to primary organic aerosol (POA) emissions from all POA sources. Modelled concentrations of SOA were compared with positive matrix factorization (PMF) analyses of aerosol mass spectrometer (AMS) measurements at a central London urban background location (North Kensington) and at the Detling and Harwell rural background locations outside of London. The model performance in comparison to relatively more well-known components of air pollution, such as NO x , O 3 , and secondary inorganic aerosol, was shown to be very good, providing confidence in the prediction skill of the ACTM system used. Modelled concentrations of SOA were evaluated in four groups: (i) hourly comparison during a summer IOP (intensive observation period), (ii) hourly comparison during a winter IOP, (iii) daily comparison for a full calendar year (including seasonal statistics), and (iv) comparison of OM / OC ratios of all apportioned OA components. To our knowledge, this is the first study where modelled OA components are compared with a year-long data set of PMFapportioned AMS measurements.
During the period of concurrent measurements at all locations, SOA concentrations at the Detling rural background location were greater than at the central London location. The model showed that this was caused by an intense pollution plume with a strong gradient of SOA from mainland Europe passing over the rural location and demonstrates how short periods of measurements can give a different picture compared with longer-term measurements, as well as the value of atmospheric chemistry transport modelling for supporting the interpretation of measurements taken at different sites or for short durations.
The model simulations show that these estimates of dieselrelated IVOCs could explain on average ∼ 30 % of the annual SOA in and around London. The 90th percentile of modelled daily SOA concentrations at the urban background site for the whole year was 3.8 µg m −3 , and the influence of missing diesel-related IVOC precursors was even greater on high percentile SOA days than its contribution to annual average SOA. The magnitudes of these contributions to SOA provide strong additional support for the need to undertake further refinement of the amount and speciation of these precursor emissions for inclusion in official emissions inventories.

Data availability
Processed measurement data used in this study are available through the ClearfLo project archive at the British Atmospheric Data Centre (http://badc.nerc.ac.uk/browse/badc/ clearflo). The model data (input, code, relevant output) are archived at the University of Edinburgh, and Centre for Ecol-ogy&Hydrology and are available on request.