Global tropospheric effects of aromatic chemistry with the SAPRC-11 1 mechanism implemented in GEOS-Chem 2

15 The GEOS-Chem model has been updated with the SAPRC-11 aromatics chemical mechanism, 16 with the purpose of evaluating global and regional effects of the most abundant aromatics 17 (benzene, toluene, xylenes) on the chemical species important for tropospheric oxidation 18 capacity. The model evaluation based on surface and aircraft observations indicates good 19 agreement for aromatics and ozone. A comparison between scenarios in GEOS-Chem with 20 simplified aromatic chemistry (as in the standard setup, with no ozone formation from related 21 peroxy radicals or recycling of NOx) and with the SAPRC-11 scheme reveals relatively slight 22 changes in ozone, hydroxyl radical, and nitrogen oxides on a global mean basis (1–4%), although 23 remarkable regional differences (5–20%) exist near the source regions. NOx decreases over the 24 source regions and increases in the remote troposphere, due mainly to more efficient transport of 25 peroxyacetyl nitrate (PAN), which is increased with the SAPRC aromatic chemistry. Model 26 ozone mixing ratios with the updated aromatic chemistry increase by up to 5 ppb (more than 27 10%), especially in industrially polluted regions. The ozone change is partly due to the direct 28 influence of aromatic oxidation products on ozone production rates, and in part to the altered 29 spatial distribution of NOx that enhances the tropospheric ozone production efficiency. Improved 30 representation of aromatics is important to simulate the tropospheric oxidation. 31

Aromatics are released to the atmosphere by biomass burning as well as fossil fuel evaporation and burning (Cabrera-Perez et al., 2016;Na et al., 2004).The dominant oxidation pathway for aromatics is via reaction with hydroxyl radical (OH, the dominant atmospheric oxidant), followed by reaction with nitrate radical (NO 3 ) (Cabrera-Perez et al., 2016;and references therein).The corresponding aromatic oxidation products could be involved in many atmospheric chemical processes, which can affect OH recycling and the atmospheric oxidation capacity (Bejan et al., 2006;Chen et al., 2011).A realistic model description of aromatic compounds is necessary to improve our understanding of their effects on the chemistry in the atmosphere.However, up to now few regional or global-scale chemical transport models (CTMs) include detailed aromatic chemistry.
Despite the potentially important influence of aromatic compounds on global atmospheric chemistry, their effect on tropospheric ozone formation in polluted urban areas remains largely unknown.The main source and sink processes of tropospheric ozone are photochemical production and loss, respectively (Yan et al., 2016).Observation-based approaches alone cannot provide a full picture of ozone-source attribution for the different NMVOCs.Such ozone-source relationships are needed to improve policymaking strategies to address hemispheric ozone pollution (Chandra et al., 2006).Numerical chemistry-transport models allow us to explore the importance of impacts from aromatics and to attribute observed changes in ozone concentrations to particular sources (Stevenson et al., 2006;Stevenson et al., 2013;Zhang et al., 2014).Current global CTMs reproduce much of the observed regional and seasonal variability in tropospheric ozone concentrations.However, some systematic biases can occur, most commonly an overestimation (Fiore et al., 2009;Reidmiller et al., 2009;Yan et al., 2016Yan et al., , 2018a, b;, b;Ni et al., 2018) due to incomplete representation of physical and chemical processes, and biases in emissions and transport, including the parameterization of small-scale processes and their feedbacks to global-scale chemistry (Yan et al., 2014;Yan et al., 2016).
Another motivation for the modeling comes from recent updates in halogen (bromine-chlorine) chemistry, which when implemented in GEOS-Chem decrease the global burden of ozone significantly (by 14%; 2-10 ppb in the troposphere) (Schmidt et al., 2017).This ozone burden decline is driven by decreased chemical ozone production due to halogen-driven nitrogen oxides (NO x = NO + NO 2 ) loss; and the ozone decline lowers global mean tropospheric OH concentrations by 11%.Thus GEOS-Chem starts to exhibit low ozone biases compared to ozonesonde observations (Schmidt et al., 2017), particularly in the southern hemisphere, implying that some mechanisms (e.g., due to aromatics) are currently missing from the model.
A simplified aromatic oxidation mechanism has previously been employed in GEOS-Chem (e.g., Fischer et al., 2014;Hu et al., 2015), which is still used in the latest version v11-02.In that simplified treatment, oxidation of benzene (B), toluene (T), and xylene (X) by OH (Atkinson et al., 2000) is assumed to produce first-generation oxidation products (xRO 2 , x = B, T, or X).And these products further react with hydrogen peroxide (HO 2 ) or nitric oxide (NO) to produce LxRO 2 y (y = H or N), passive tracers which are excluded from tropospheric chemistry.Thus in the presence of NO x , the overall reaction is aromatic + OH = -NO + (inert tracer).While such a simplified treatment can suffice for budget analyses of the aromatic species themselves, it does not capture ozone production from aromatic oxidation products.
In this work, we update the aromatics chemistry in GEOS-Chem based on the SAPRC-11 mechanism, and use the updated model to analyze the global and regional scale chemical effects of the most abundant aromatics in the gas phase (benzene, toluene, xylenes) in the troposphere.Specifically, we focus on the impact on ozone formation (due to aromatics oxidation), as this is of great interest for urban areas and can be helpful for developing air pollution control strategies.
Further targets are the changes to the NO x spatial distribution and OH recycling.Model results for aromatics and ozone mixing ratios are evaluated by comparison with observations from surface and aircraft campaigns in order to constrain model accuracy.Finally, we discuss the global effects of aromatics on tropospheric chemistry including ozone, NO x and HO x (HO x = OH + HO 2 ).
The rest of the paper is organized as follows.Section 2 describes the GEOS-Chem model setups, including the updates in aromatics chemical mechanism.A description of the observational datasets for aromatics and ozone is given in Sect.3. Section 4 presents the model evaluation for aromatics based on the previously mentioned set of aircraft and surface observations, and evaluates modeled surface ozone with measurements from three networks.An analysis of the tropospheric impacts on ozone, NO x , and OH, examining the difference between models results with simplified (as in the standard model setup) and with SAPRC-11 aromatic chemistry, is presented in Section 5. Section 6 concludes the present study.

Model description and setup
We use the GEOS-Chem CTM (version 9-02, available at http://geos-chem.org/) to interpret the importance of aromatics in tropospheric chemistry and ozone production.A detailed description of the GEOS-Chem model is available at http://acmg.seas.harvard.edu/geos/geos_chem_narrative.html.GEOS-Chem has been used extensively for tropospheric chemistry and transport studies (Zhang and Wang, 2016;Yan et al., 2014;Shen et al., 2015;Lin et al., 2016).Here, the model is run at a horizontal resolution of 2.5º each below 850 hPa), as driven by the GEOS-5 assimilated meteorological fields.A non-local scheme implemented by Lin and McElroy (2010) is used for vertical mixing in the planetary boundary layer.Model convection adopts the Relaxed Arakawa-Schubert scheme (Rienecker et al., 2008).Stratospheric ozone production employs the Linoz scheme (McLinden et al., 2000).

Emissions
Global carbon monoxide (CO) and NO x anthropogenic emissions are taken from the EDGAR  (Price and Rind, 1992), and are further constrained by the lightning flash counts detected from satellite instruments (Murray et al., 2012).Soil NO x emissions are described in Hudman et al. (2012).Biogenic emissions of v2.1 with the Hybrid algorithm (Guenther et al., 2012).

Updated aromatic chemistry
In the GEOS-Chem model setup, the current standard chemical mechanism with simplified aromatic oxidation chemistry is based on Mao et al. (2013), which is still true for the latest version v11-02.As mentioned in the introduction, this simplified mechanism acts as strong sinks of both HO x and NO x , because no HO x are regenerated in this reaction, and NO is consumed without regenerating NO 2 .However, it is reasonably well established that aromatics tend to be radical sources, forming highly reactive products that photolyze to form new radicals, and regenerating radicals in their initial reactions (Carter, 2010a, b;Carter and Heo, 2013).A revised mechanism that takes the general features of aromatics mechanisms into account would be much more reactive, given the reactivity of the aromatic products.
This work uses a more detailed and comprehensive aromatics oxidation mechanism: the SAPRC-11 aromatics chemical mechanism.SAPRC-11 is an updated version of the SAPRC-07 mechanism (Carter and Heo, 2013), which is consistent with recent literature.Moreover, SAPRC-11 is able to reproduce the ozone formation from aromatic oxidation that is observed in environmental chamber experiments (Carter and Heo, 2013).Table S1 lists new model species in addition to those in the standard GEOS-Chem model setup.Table S2 lists the new reactions and rate constants.In this mechanism, the tropospheric consumption process of aromatics is mainly reaction with OH.
As discussed by Carter (2010a, b), aromatic oxidation has two possible OH reaction pathways: OH radical addition and H-atom abstraction (Atkinson, 2000).In SAPRC-11, the reactions following abstraction lead to three different formation products, depending on the participating radical and location of the H-abstraction: an aromatic aldehyde (represented as the BALD species in the model), a ketone (PROD2), and an aldehyde (RCHO).The largest yield of aromatic oxidation corresponds to the reaction after OH addition of aromatic rings.The OH-aromatic adduct is reaction with O 2 either forming HO 2 and a phenolic compound (further consumed by reactions with OH and NO 3 radicals), or to form an OH-aromatic-O 2 adduct.The OH-aromatic-O 2 adduct further undergos two competing unimolecular reactions to ultimately form OH, HO 2 , an α-dicarbonyl (such as glyoxal (GLY), methylglyoxal (MGLY) or biacetyl (BACL)), a monounsaturated dicarbonyl co-product (AFG1, AFG2, the photoreactive products) and a diunsaturated dicarbonyl product (AFG3, the reactions of uncharacterized non-photoreactive ring fragmentation products) (Calvert et al., 2002).
Formed from the phenolic products, the SAPRC-11 mechanism includes species of cresols (CRES), phenol (PHEN), xylenols, alkyl phenols (XYNL), and catechols (CATL).Due to their different SOA and ozone formation potentials (Carter et al, 2012), these phenolic species are represented separately.Relatively high yields of catechol (CATL) have been observed in the reactions of OH radicals with phenolic compounds.Furthermore, their subsequent reactions are believed to be important for SOA and ozone formation (Carter et al, 2012).

Simulation setups
In

Aromatics and ozone observations
We use a set of measurements from surface and aircraft campaigns to evaluate the model simulated aromatics and ozone.

Aromatic aircraft observations
For We also employ vertical profiles obtained in 2005 from the CARIBIC (Civil Aircraft for Regular Investigation of the atmosphere Based on an Instrument Container) project, which conducts atmospheric measurements onboard a commercial aircraft (Lufthansa A340-600) (Brenninkmeijer et al., 2007;Baker et al., 2010).CARIBIC flights fly away from Frankfurt, Germany on the way to North America, South America, India and East Asia.Measurements are available in the upper troposphere (50% on average) and lower stratosphere (50%) (UTLS) at altitudes between 10-12 km.To evaluate our results, model data are sampled along the flight tracks, and the annual means from GEOS-Chem model simulations at the 250 hPa level are used to compare with observations between 200-300 hPa.For comparison, annual means for individual site have been used.

Aromatics surface measurements
The KCMP tall tower measurements (at 44.69°N, 93.07°W) have been widely used for studies of surface fluxes of tropospheric trace species and land-atmosphere interactions (Kim et al., 2013;Hu et al., 2015;Chen et al., 2018).A suite of NMVOCs including aromatics were observed at the KCMP tower during 2009-2012 with a high-sensitivity PTR-MS, sampling from a height of 185 m above ground level.We use the hourly observations of benzene, toluene and C 8 (xylenes + ethylbenzene; here consistent with the model speciation) aromatics for our model evaluation.

Ozone observations
Ozone observations are taken from the database of the World Data Centre for Greenhouse Gases

Evaluation of simulated aromatics and ozone
In this part, the SAPRC model simulation results of aromatics (benzene, toluene, xylenes and C 8 aromatics) and ozone from GEOS-Chem are evaluated with observations.Table 1 summarizes the statistical comparison between measured and simulated concentrations over the monitoring stations described in Sect.3. To do the statistical calculations, GEOS-Chem simulation results have been sampled along the geographical locations of the measurements.Over the US, annual mean observed concentrations at the KCMP tall tower are 91.5 ppt for benzene, 56.7 ppt for toluene, and 90.3 ppt for C 8 aromatics (Table 1).The model biases for benzene (8.4 ppt; 9.2%) and C 8 aromatics (−1.4 ppt; −1.6%) are much lower than that for toluene (64.5 ppt; 114%).Figure 3 further shows the observed and simulated monthly averaged concentrations of benzene, toluene and C 8 aromatics.The SAPRC simulation reproduces their seasonal cycles, with higher concentrations in winter and lower mixing ratios in summer, consistent with Hu et al. (2015).The model-observation correlations are 0.89, 0.78 and 0.65 for monthly benzene, toluene, and C 8 aromatics, respectively.The large overestimation of modeled toluene is mainly due to simulated high mixing ratios during the cold season (Fig. 3, October to March).

Tropospheric aromatics
Table 1 shows that in the UTLS, both CARIBIC observed (16 ppt) and GEOS-Chem modeled (12.3 ppt) benzene mixing ratios are higher than toluene concentrations (3.6 ppt for CARIBIC and 1.5 ppt for GEOS-Chem).For benzene, the model underestimates appear to be smaller in the free troposphere (with an underestimate by 23%) than at the surface (36% for EMEP and 32% for EEA).In contrast to benzene, annual mean concentrations of toluene are underestimated by 58% in the UTLS.The geographical variability of benzene is larger than that for toluene (with standard deviation of 4.2 versus 0.7 ppt in model and 15.8 versus 7.5 ppt in observation), probably because of the shorter lifetime of toluene in combination with the lower concentrations in the UTLS for toluene.The model results show smaller spatial variability than the observations.This underestimation for spatial variability in the free troposphere (over 70%) is higher than that at the surface (not shown).
The black lines in Fig. 4 show the tropospheric aromatics profiles during the CALNEX campaign.The measured values peak at an altitude of 0.6-0.8km, with concentrations decreasing at higher altitudes.Although the concentrations in the lower troposphere for benzene (40-100 ppt below 2 km) are lower than mixing ratios for toluene (70-160 ppt below 2 km) and C 8 aromatics (50-120 ppt below 2 km), the benzene mixing ratios (> 30 ppt) in the free troposphere are much higher than those of toluene and C 8 aromatics (< 10 ppt), mainly due to the longer lifetime of benzene.The SAPRC simulation (red lines in Fig. 4) captures the general vertical variations of CALNEX benzene and toluene, with statistically significant model-observation correlations of 0.74 and 0.65 for benzene and toluene, respectively.The model generally overestimates the measured C 8 aromatics below 0.5 km, albeit with an underestimate above 0.5 km, with lower model-observation correlation of 0.37.This overestimation below 0.5 km is also seen for benzene and toluene.

Surface ozone
Table 1 shows an average ozone mixing ratio of 34.1 ppb in 2005 over the regional background WDCGG sites.The annual mean ozone mixing ratios are lower over Europe (from the EMEP dataset), about 30.6 ppb.The ozone mixing ratios are relatively lowest over the US, with an average value of 24.2 ppb, partly due to inclusion of urban and suburban sites that undergo strong titration especially in the cold season.The average over the US rural sites is 32.5 ppb.The SAPRC simulation tends to overestimate the ozone mixing ratios over the US (with a mean bias of +12.1 ppb), whereas it underestimates the mixing ratios over the sites of Europe and background regions with biases of −2.9 ppb and −5.5 ppb, respectively.Figure 5 shows the spatial distribution of the annual mean model biases with respect to the measurements.Unlike the modeled surface aromatics, the simulated ozone spatial variability can be either slightly lower or higher than the observed variability, depending on the compared database: the standard deviation is 10.2 ppb (simulated) versus 13.1 ppb (observed) for AQS sites, 12.8 versus 14.2 ppb for WDCGG sites, 13.2 versus 10.3 ppb for EMEP sites.The temporal variability (temporal correlations of 0.68-0.92) is better captured by the model than the spatial variability (spatial correlations of 0.43-0.54).

Global effects of aromatic chemistry
This section compares the Base and SAPRC simulations to assess to which extent the updated mechanism for aromatics affect the global simulation of ozone, HO x and individual nitrogen species.Our focus here is on the large-scale impacts.

NO y Species
Figure 6 and Table 2 show the changes from Base to SAPRC in annual average surface NO mixing ratios.A decrease in NO is apparent over NO x source regions, e.g., by approximately 0.15 ppb (~20%) over much of the US, Europe and China (Fig. 6).In contrast, surface NO increases at locations downwind from NO x source regions (up to ~0.1 ppb or 20%), including the oceanic area off the eastern US coast, the marine area adjacent to Japan, and the Mediterranean area.The change is negligible (by −0.2%) for the annual global mean surface NO (Table 2).Seasonally, the decrease in spring, summer and fall is compensated partly by the increase in winter (Table 2).
The zonal average results in Fig. 7 show a clear decline in NO in the planetary boundary layer, in contrast to significant increases in the free troposphere, from Base to SAPRC.The free tropospheric NO increases are largest in the remote northern regions with an annual average enhancement up to 5% (Fig. 7), and are particularly large in winter (up to 10%, not shown).For the whole troposphere, the average NO increases by 0.6% from Base to SAPRC (Table 2).
Figure 6 shows that simulated surface NO 2 mixing ratios in the SAPRC scenario are enhanced over most locations throughout the troposphere, in comparison with the Base simulation.Over the source regions, the changes are mixed, with increases in some highly NO x polluted regions (by up to 10%) and decreases in other polluted regions.On a global mean basis, NO 2 is increased (by 2.1% in the free troposphere and 1.0% at the surface, Table 2), due mainly to the recycling of NO x from PAN associated with the aromatics, and the reactions of oxidation products from aromatics with NO or NO 3 (primarily) to form NO 2 and HO 2 .Combing the changes in NO and NO 2 means that the total NO x mixing ratios decrease in source regions but increase in the remote free troposphere.
The NO 3 mixing ratios decrease at the global scale (−4.1% on average in the troposphere, Fig. 7 and Table 2) in the SAPRC simulation, except for an enhancement in surface NO 3 over the northern polar regions and most polluted areas like the eastern US, Europe and eastern China (Fig. 6).
Table 2 shows that nitric acid (HNO 3 ) increases in the SAPRC simulation, both near the surface (by approximately 1.1%) and in the troposphere (by 0.3%).The enhancement in HNO 3 appears uniformly over most continental regions in the northern hemisphere (not shown), due to the promotion of direct formation of HNO 3 from aromatics in the SAPRC simulation.

OH and HO 2
Compared to the Base simulation, OH increases slightly by 1.1% at the surface in the SAPRC simulation (Fig. 8 and Table 2).The largest increases in OH concentrations are found over source regions dominated by anthropogenic emissions (i.e., the US, Europe, and Asia) and in subtropical continental regions with large biogenic aromatic emissions.In these locations, the peroxy radicals formed by aromatic oxidation react with NO 2 and HO 2 , which can have a significant effect on the ambient ozone and NO x mixing ratios.This in turn influences OH, as the largest photochemical sources of OH are the photolysis of O 3 as well as the reaction of NO with HO 2 .Seasonally, a few surface locations see OH concentration increases of more than 10% during April−August (not shown), including parts of the eastern US, central Europe, eastern Asia and Japan.
The OH enhancement (0.2%) is also seen in the free troposphere in the SAPRC simulation (Fig. 9 and Table 2).OH is increased in the troposphere of the northern hemisphere, in contrast to the decline in the troposphere of tropics and southern hemisphere (Fig. 9).These OH changes correspond to the hemispherically distinct changes in aromatics (benzene, toluene, and xylenes), which show a decrease in the northern hemisphere, an increase in the southern hemisphere, and an increase in global mean (by 1%) (not shown).Despite the overall increase in tropospheric OH, CO is increased by ~1% (Table 2) due to additional formation from aromatics oxidation.
Table 2 shows that from Base to SAPRC, HO 2 shows a significant increase at the global scale: 3.0% at the surface and 1.3% in the troposphere, due to regeneration of HO x from aromatics oxidation products.Correspondingly, the OH/HO 2 ratio decreases slightly.These changes mean that, compared to the simplified aromatic chemistry in the standard model setup, the SAPRC mechanism are associated with higher OH (i.e., more chemically reactive troposphere) and even higher HO 2 .From Base to SAPRC, the global average surface ozone mixing ratio increases by less than 1%

Ozone
(Table 2).This small difference is comparable to the result calculated by Cabrera-Perez et al.
( 2017) with the EMAC model, which is based on a reduced version of the aromatic chemistry from the Master Chemical Mechanism (MCMv3.2). Figure 8 shows that the 1% increase in surface ozone occurs generally over the northern hemisphere.Similar to the changes in OH, the most notable ozone increase occurs in industrially-polluted regions.These regions show significant local ozone photochemical formation in both the Base case and the SAPRC simulation.The updated aromatic chemistry increases ozone by up to 5 ppb in these regions.
Increases of ozone are much smaller (less than 0.2 ppb) over the tropical oceans than in the continental areas.In contrast, ozone declines in regions dominated by emissions from biomass burning over the southern hemisphere.Changes elsewhere in the troposphere are similar in magnitude, as shown in Figure 9.
Two general factors likely contribute to the ozone change from Base to SAPRC.In the SAPRC simulation, the addition of aromatic oxidation products (i.e., peroxy radicals) can contribute directly to ozone formation in NO x -rich source regions and also in the NO x -sensitive remote troposphere (i.e., from PAN to NO x and to ozone).The second factor is a change in the NO x spatial distribution, with an overall enhancement in average NO 2 concentrations.The redistribution is mainly caused by enhanced transport of NO x to the remote troposphere (see Sect. 5.1).The enhanced NO x in the remote troposphere enhances the overall ozone formation because this process is more efficient in the remote regions (e.g., Liu et al., 1987).The increased ozone, NO 2 and NO x transport all lead to the aforementioned changes.This is described in detail in section 5.4.
There are notable decreases (more than 5%, Fig. 9) in simulated ozone and OH in the free troposphere (above 4 km) over the tropics (30°S−30°N).A similar decrease is found in modeled NO x (above 6 km, Fig. 7).These decreases are probably related to the upward transport of aromatics (mainly benzene, whose lifetime is longer than the other two species) by tropical convection processes.The aromatics transported to the upper troposphere may cause net consumption of tropospheric OH and NO x , which can further reduce ozone production.
From Base to SAPRC, the modeled ozone concentrations are closer to the WDCGG and EMEP network measurements, but the agreement worsens at the AQS sites (Table 3).For the WDCGG background sites, the annual and seasonal model biases are ~10% smaller in the SAPRC simulation compared to the Base case.For the EMEP stations, although the model results are not improved in summer and fall, the annual model bias is 25% smaller (−2.8 ppb versus −3.5 ppb) in the SAPRC simulation.There are significant overestimates in the Base simulation at the AQS sites, with an annual mean bias of 11.4 ppb.This model overestimation is consistent with the results of previous works (Yan et al., 2016;Fiore et al., 2009;Reidmiller et al., 2009).The recent study of Schmidt et al. (2017) includes a more comprehensive representation of multiphase halogen (Br-Cl) chemistry in GEOS-Chem, which causes a 14% decrease in the global burden of tropospheric ozone and negative ozone biases over the US.Past studies have suggested that the model biases (positive in most models) are a multifaceted problem, such as the effect of coarse resolution and how small-scale processes are represented (Yan et al., 2016).

Discussion of SAPRC aromatic-ozone chemistry
As discussed in Sect.5.3, the increased O 3 mixing ratios from Base to SAPRC are due to the direct impact of aromatic oxidation products (i.e., peroxy radicals) and to the effect of increased Regions with large aromatics emissions show a significant increase of oxidation products from Base to SAPRC.The modeled ozone in these regions increases with increasing NO 2 and its oxidation products.NO and NO 3 are often lower in these regions in the SAPRC scenario because of their reactions with the aromatic-OH oxidation products to form NO 2 and HO 2 .In remote regions and in the free troposphere, ozone production is also enhanced by both NO 2 and HO 2 increases in the SAPRC simulation, but the increase in ozone formation is mainly attributed to the increase in NO x mixing ratios.
NO x concentrations decrease in source regions and increase in the remote regions because of more efficient transport of PAN and its analogues (represented by PBZN here in SAPRC-11).In the SAPRC-11 aromatics chemical scheme the immediate precursor of PAN (peroxyacetyl radical) has five dominant photochemical precursors.They are acetone (CH 3 COCH 3 , model species: ACET), methacrolein (MACR), biacetyl (BACL), methyl glyoxal (MGLY) and other ketones (e.g., PROD2, AFG1).These compounds explain the increased rate of PAN formation.
For example, the SAPRC simulation has increased the concentration of MGLY by a factor of 2. In addition, production of organic nitrates (Table S2) in the model with SAPRC aromatics chemistry may also explain the increase in ambient NO x in the remote regions, due to the re-release of NO x from organic nitrates (as opposed to removal by deposition).Due to recycling of NO x from PANlike compounds and also transport of NO x , NO x increases by up to 5% at the surface in most remote regions and by ~1% in the troposphere as a whole.This then leads to increased ozone due to the effectiveness of ozone formation in the free troposphere.

Conclusions
A representation of tropospheric reactions for aromatic hydrocarbons in the SAPRC-11 mechanism has been added to GEOS-Chem, to give a more realistic representation of their atmospheric chemistry.The GEOS-Chem simulation with the SAPRC-11 aromatics mechanism has been evaluated against measurements from aircraft and surface campaigns.The comparison with observations shows reasonably good agreement for aromatics (benzene, toluene, and ) at locations downwind from these source regions.The total NO x mixing ratios decrease in source regions but increase in the remote free troposphere.This is mainly due to the addition of aromatics oxidation products in the model that lead to PAN, which facilitates the transport of nitrogen oxides to downwind locations remote from the sources.Finally, the updated aromatic chemistry in GEOS-Chem increases ozone concentrations, especially over industrialized regions (up to 5 ppb, or more than 10%).Ozone changes in the model are partly explained by the direct impact of increased aromatic oxidation products (i.e., peroxy radical), and partly by the effect of the altered spatial distribution of NO x .Overall, our results suggest that a better representation of aromatics chemistry is important to model the tropospheric oxidation capacity.
order to investigate the global chemical effects of the most commonly emitted aromatics in the troposphere, two simulations were performed, one with the ozone related aromatic chemistry updates from SAPRC-11 (the SAPRC case), and the other with simplified aromatic chemistry as in the standard setup (the Base case).Both simulations are conducted from July 2004 to December 2005 based on the available observations (Sect.3).Initial conditions of chemicals are regridded from a simulation at 5° long.× 4° lat.started from 2004.Simulations over July-December 2004 allow for a 6-month spin-up for our focused analysis over the year of 2005.For comparison with aromatics observations over the US in 2010-2011 (Sect.3), we extend the SAPRC simulation from July 2009 to December 2011.
aromatics, we use airborne observations from CALNEX (California; May/June 2010) aircraft study over the US.A proton transfer reaction quadrupole mass spectrometer (PTR-MS) was used to measure mixing ratios of aromatics (and an array of other primary and secondary pollutants) during CALNEX.Measurements are gathered mostly on a one-second time scale (approximately 100 m spatial resolution), which permits sampling of the source regions and tracking subsequent transport and transformation throughout California and surrounding regions.Further details of the CALNEX campaign, including the flight track, timeframe, location and instrument, are shown in Hu et al. (2015) and https://www.esrl.noaa.gov/csd/projects/calnex.
3.2).The fact that the anthropogenic RETRO emissions (for year 2000) do not correspond to the year of measurement (2005) may contribute to the above model-measurement discrepancies.The modeled spatial variability of aromatics (with standard deviations of 32.1-66.8ppt) is 18-73% lower than that of the EMEP and EEA observations (41.9-118.4ppt), probably due to the coarse model resolution.The spatial variability in benzene (46-73% lower) is the most strongly underestimated among the three aromatic species.Unlike benzene, simulated concentrations of toluene show a larger standard deviation (66.8 ppt) than the EEA measurements (59.4 ppt), indicating larger simulated spatial variability.Simulation results are thus poorly spatially correlated with observations (R = 0.41-0.49).However, the temporal variability of aromatics is well captured by GEOS-Chem with the correlations above 0.7 for most stations.

Figure 2
Figure2shows a comparison of model results with observations at six stations for benzene, toluene, and xylenes, respectively, followingCabrera-Perez et al. (2016).Model results reproduce the annual cycle at the majority of sites.Aromatics are better simulated in summer than in winter.This feature has been previously found for the climate-chemistry model EMAC for aromatics(Cabrera-Perez et al., 2016) and simpler NMVOCs(Pozzer et al., 2007).In addition, the measurements show larger standard deviations than the GEOS-Chem simulations, with the ratios between the observed and the simulated standard deviations being 2-11.
NO 2 concentrations.The simulated odd oxygen family (O x = O 3 + O( 1 D) + O( 3 P) + NO 2 + 2×NO 3 + 3×N 2 O 5 + HNO 3 + HNO 4 + PAN, Wu et al., 2007; Yan et al., 2016) formation increases by 1-10%, both over the source regions and in the remote troposphere.Although the percentage changes are similar, the driving factors over the source regions are different from the drivers in the remote troposphere.

Figure 7 .
Figure 7. (Left column) Modeled zonal average latitude-altitude distributions of NO (top) and NO 2 (bottom) simulated in the Base scenario.(Right column) The respective changes from Base to SAPRC.

Figure 8 .
Figure 8. Same as Fig. 6 but for OH (top panels) and O 3 (bottom panels).
(Torseth et al., 2012)ss., https://doi.org/10.5194/gmd-2018-196ManuscriptunderreviewforjournalGeosci.Model Dev. Discussion started: 17 August 2018 c Author(s) 2018.CC BY 4.0 License.To evaluate the ground-level mixing ratios of benzene, toluene, and xylenes as well as their seasonal cycles, surface observations of aromatics are collected from two networks (EMEP, data available at http://wwwEMEP, which aims to investigate the long-range transport of air pollution and the flux though boundaries(Torseth et al., 2012), locates measurement sites at which there are minimal local impacts, thus consequently the observations could represent the feature of large regions.EMEP has a daily resolution with a total of 14 stations located in Europe for benzene, 12 stations for toluene, and 8 stations for xylenes (Table1).Here we use the monthly values calculated from the database to evaluate monthly model results.Note that measurement speciation of xylenes (oxylene, m-xylene and p-xylene) in EMEP network does not exactly correspond with the model speciation of xylenes (m-xylene, p-xylene, o-xylene and ethylbenzene) (Hu et al., 2015).The speciation assumption probably can partly account for the xylene model-measurement discrepancy seen in Sect. 4. EEA provides observations from a large number of sites over urban, suburban and background regions (EEA, 2014).However, here we use only rural background sites to do model comparison, as in Cabrera-Perez et al. (2016), because the model horizontal scale cannot simulate direct traffic or industrial influence.This leads to 22 stations available for benzene and 6 stations for toluene.
Table 1 includes the number of locations and time resolutions.The number of sites in EEA for xylenes is only 2, thus we do not include their comparison results in Table 1 due to the lack of representativeness.
4.1 Surface-level aromaticsFor the aromatics near the surface mixing ratios over Europe, observed mean benzene (194.0 ppt for EEA and 166.4 ppt for EMEP) and toluene (240.3 ppt for EEA and 133.1 ppt for EMEP) mixing ratios are higher than observed mean xylene concentrations (42.3 ppt for EMEP).In general, the model underestimates EEA and EMEP observations of benzene (by 34% on average) and toluene (by 20% on average).For benzene, the model results systematically underestimate the annual means (36%) compared to the EMEP database, consistent with the model underestimate of the EEA dataset (32%).The model underestimate for toluene compared to the EMEP dataset (15%) is smaller than that relative to the EEA measurements (25%).The simulation overestimates the xylene measurements in EMEP by a factor of 1.9, in part because the model results include ethylbenzene but the observations do not (see Sect. Geosci.Model Dev.Discuss., https://doi.org/10.5194/gmd-2018-196Manuscriptunder review for journal Geosci.Model Dev. Discussion started: 17 August 2018 c Author(s) 2018.CC BY 4.0 License.xylenes) and ozone.Model results for aromatics can reproduce the seasonal cycle, with a general underestimate over Europe for benzene and toluene, and an overestimate of xylenes; while over the US a positive model bias for benzene and toluene and a negative bias for C 8 aromatics are found.From the Base to the SAPRC simulation, the model ozone bias is reduced by 10% relative to WDCGG observations and by 25% relative to EMEP observations, although the bias increases by 5% at the AQS sites.The simplified aromatics chemistry in the Base simulation under-predicts NO and NO 3 oxidation, and it does not represent ozone formed from aromatic-OH-NO x oxidation.Although the global average changes in simulated chemical species are relatively small (1%-4% from Base to SAPRC), on a regional scale the differences can be much larger, especially over aromatics and NO x source regions.From Base to SAPRC, NO 2 is enhanced by up to 10% over some highly polluted areas, while reductions are notable in other polluted areas.Although the simulated surface NO decreases by approximately 0.15 ppb (~20%) or more in the northern hemispheric source regions, including most of the US, Europe and China, increases are found (~0.1 ppb, up to 20%