E ﬀ ects of halogens on European air-quality †

Halogens (Cl, Br) have a profound in ﬂ uence on stratospheric ozone (O 3 ). They (Cl, Br and I) have recently also been shown to impact the troposphere, notably by reducing the mixing ratios of O 3 and OH. Their potential for impacting regional air-quality is less well understood. We explore the impact of halogens on regional pollutants (focussing on O 3 ) with the European grid of the GEOS-Chem model (0.25 (cid:1) (cid:3) 0.3125 (cid:1) ). It has recently been updated to include a representation of halogen chemistry. We focus on the summer of 2015 during the ICOZA campaign at the Weybourne Atmospheric Observatory on the North Sea coast of the UK. Comparisons between these observations together with those from the UK air-quality network show that the model has some skill in representing the mixing ratios/concentration of pollutants during this period. Although the model has some success in simulating the Weybourne ClNO 2 observations, it signi ﬁ cantly underestimates ClNO 2 observations reported at inland locations. It also underestimates mixing ratios of IO, OIO, I 2 and BrO, but this may re ﬂ ect the coastal nature of these observations. Model simulations, with and without halogens, highlight the processes by which halogens can impact O 3 . Throughout the domain O 3 mixing ratios are reduced by halogens. In northern Europe this is due to a change in the background O 3 advected into the region, whereas in southern Europe this is due to local chemistry driven by

O 3 ) with the European grid of the GEOS-Chem model (0.25 Â 0.3125 ). It has recently been updated to include a representation of halogen chemistry. We focus on the summer of 2015 during the ICOZA campaign at the Weybourne Atmospheric Observatory on the North Sea coast of the UK. Comparisons between these observations together with those from the UK air-quality network show that the model has some skill in representing the mixing ratios/concentration of pollutants during this period. Although the model has some success in simulating the Weybourne ClNO 2 observations, it significantly underestimates ClNO 2 observations reported at inland locations. It also underestimates mixing ratios of IO, OIO, I 2 and BrO, but this may reflect the coastal nature of these observations. Model simulations, with and without halogens, highlight the processes by which halogens can impact O 3 . Throughout the domain O 3 mixing ratios are reduced by halogens. In northern Europe this is due to a change in the background O 3 advected into the region, whereas in southern Europe this is due to local chemistry driven by Mediterranean emissions. The proportion of hourly O 3 above 50 nmol mol À1 in Europe is reduced from 46% to 18% by halogens. ClNO 2 from N 2 O 5 uptake onto sea-salt leads to increases in O 3 mixing ratio, but these are smaller than the decreases caused by the bromine and iodine. 12% of ethane and 16% of acetone within the boundary layer is oxidised by Cl. Aerosol response to

Introduction
Over the last decade, there has been increasing evidence, from both an observational and modelling perspective, that halogens (Cl, Br and I) play a role in determining the composition of the troposphere. 1 Different studies have emphasised either the regional impact of these species, 2-5 or their global impact. [6][7][8][9][10][11][12][13] They have also tended to focus on the chemistry of chlorine, 3,14 iodine 10,15 or bromine, 6,8,11 with few studies investigating the coupled chemistry of all three. 7,12 The tropospheric chemistry of halogens is complex (see the recent review by Simpson et al. 1 and references within) with signicant uncertainties remaining, particularly in some aspects of the gas-phase chemistry of iodine and in the heterogenous processing of all halogens. Interactions between the halogens and HO x , NO x , and volatile organic compounds (VOC) species leads to halogens having a pervasive inuence throughout the tropospheric chemistry system. 11,12 The chemistry of Br and I is thought to lead to reductions in O 3 and OH mixing ratios globally 8,10-12 whereas the chemistry of Cl is thought to lead to both increases in O 3 due to more rapid oxidation of VOCs 2,16 and decreases due to halogen nitrate hydrolysis reducing O 3 production (via decreasing NO x ). 11 However, the calculated magnitude of these impacts will be critically dependent on the emissions and chemistry of halogens used.
Both biogenic and anthropogenic sources of gas-phase halogen precursors exist, from a mix of oceanic, terrestrial, and anthropogenic sources. 1 The oceanic source of halocarbons can be spatially variable reecting different ecosystems and driving processes. For example, areas of tidal sea-weed can have signicant emissions of iodine precursor gases which vary with the tide state. [17][18][19][20][21][22][23] For iodine, chemistry involving atmospheric ozone and ocean iodide within the surface micro-layer of the ocean leads to the emission of inorganic species (HOI, I 2 ). 24,25 Other sources of halogens into the troposphere can also occur, such as direct emissions (e.g. HCl/Cl 2 (ref. 26 and 27)) or transport from the stratosphere.
The largest emission of bromine and chlorine into the atmosphere comes from sea-salt aerosol. However this aerosol phase chloride and bromide must be liberated by heterogenous chemistry to become a gas-phase source. Different mechanisms allow for activation to the gas phase: acid displacement (e.g. HNO 3 ); uptake of N 2 O 5 to sea-salt to liberate ClNO 2 ; 28 and uptake of other halogen species (HOBr, HOI, BrNO 3 , HOBr, etc.) to liberate di-halogen species (ICl, IBr, Br 2 , BrCl, Cl 2 ). 1,29,30 Measuring the concentration of reactive halogen species in the atmosphere is difficult due to their low mixing ratio and reactivity. Although there remains some debate, recent observations have demonstrated the pervasive existence of bromine and iodine species throughout the troposphere over oceanic regions by a range of techniques. The highest mixing ratios of these species have been found close to tidal sources 17-23 but measurable mixing ratios have been found above the remote ocean 31 and in the upper troposphere. 32 Observations of reactive chlorine species are particularly sparse. However, a relatively large dataset of ClNO 2 observations have now been made 28,33-38 which show a build up at night and then a rapid decrease (due to photolysis) at sunrise. The observations in polluted coastal regions are explicable through the uptake of N 2 O 5 onto sea-salt. 28 However, high mixing ratios of ClNO 2 in continental regions have proved harder to explain due to the short lifetime of sea-salt in the atmosphere. Various explanations have been postulated ranging from non-oceanic sources of both natural and anthropogenic chlorine species, 35 to the movement of chlorine from sea-salt to ne mode sulfate aerosol via gas phase chemistry. 28 Previous model studies of Br and I chemistry have focussed predominantly on their global scale impacts. 6,8,9,11,12 Whereas, studies of the impact of Cl have typically focussed on a smaller hemispheric or regional (air-quality) scale. [2][3][4] The combined impact of all halogens on the regional scale is less well explored. Here, we use a new version of the GEOS-Chem model, which includes a representation of halogen chemistry, 12 run in its regional grid conguration [39][40][41][42] for Europe 43 to explore the roles that halogens may play in controlling European air quality with a focus on O 3 . We focus on the summer of 2015 as this allows us access to an observational dataset made on the North Sea coast of the UK. We explore the model delity against this data and that offered from the UK air quality network. We explore the differing role of halogens in determining both O 3 concentrations through changes to regional scale chemistry and the hemisphere background. We then consider impacts of halogens on oxidation and contribution of atomic chlorine. The relative contribution of the halogen families on O 3 are then considered, and the impacts on aerosol concentrations. Finally we suggest future areas of research to allow better representation of the halogen chemistry of the atmosphere on a regional scale.

Observations
The Integrated Chemistry of Ozone in the Atmosphere (ICOZA) campaign 44 at the Weybourne Atmospheric Observatory (52.95 N, 1.12 E, 45 ) was designed to examine the composition of the atmosphere and local chemical processes at a coastal site in the UK during the summer of 2015 (29 th June to 1 st August). Weybourne is a World Meteorological Organisation (WMO) Global Atmospheric Watch (GAW) programme site. In addition to the standard observations (CO, and O 3 ), additional NO x (NO, NO 2 ), total reactive nitrogen (NO y ), nitryl chloride (ClNO 2 ) and molecular chlorine (Cl 2 ) measurements were made during this period.
The NO, NO 2 and NO y observations were made $4 m above ground level. The NO and NO 2 measurements were made using a dual channel Air Quality Design Inc. (Golden, Colorado, USA) chemiluminescent instrument equipped with a UV-LED photolytic NO 2 converter as described by Reed et al. 44,46 NO y was measured using a Thermo Environmental 42i TL NO x analyser equipped with a molybdenum catalytic converter. A second high temperature (375 C) molybdenum converter was placed upstream directly at the gas inlet. Heated molybdenum catalysts have been shown to convert NO y species such as PAN, HNO 3 and particulate nitrate into NO 2 . 47-50 Limits of detection were 1.5 pmol mol À1 and 1.9 pmol mol À1 averaged over 1 minute for NO and NO 2 , and 50 pmol mol À1 averaged over 1 minute for NO y .
Carbon monoxide (CO) observations are part of the National Centre for Atmospheric Sciences (NCAS) long-term measurement programme and O 3 observations are part of the Department for Environment, Food and Rural Affairs (DEFRA) Automatic Urban and Rural Network (AURN). CO was measured by a Reduction Gas Analyser (RGA3, Trace Analytical, Inc., California, USA) to the WMO CO X2004 scale and O 3 was measured using UV absorption (TE49i, Thermo Fisher Scientic Inc.).
The observations of ClNO 2 and Cl 2 were made with the University of Leicester Chemical Ionization Mass Spectrometer (CIMS). The instrument, manufactured by THS Instruments (Georgia, USA), is based on the CIMS technique described by Slusher et al., 51 and is similar in conguration to the instrument used by Liao et al. 52 The Leicester CIMS was calibrated for Cl 2 , using a certied standard by BOC (5 mmol mol À1 in nitrogen), and for ClNO 2 , using the methodology described by Thaler et al. 53 The detection limit was 8.5 pmol mol À1 for Cl 2 and 5.1 pmol mol À1 for ClNO 2 . The instrument and the measurements are discussed in more detail in Sommariva et al. (in preparation).
Wider UK air-quality observation data (O 3 , NO 2 , PM 2.5 ) from the DEFRA's AURN 54 was extracted for the period of observations using the OpenAir R package. 55

Modelling
We used the GEOS-Chem model (version 10-01, http://www.geos-chem.org), which includes O x , HO x , NO x , and VOC chemistry 56 and a mass based aerosol scheme. 57,58 The model also has a representation of bromine and chlorine chemistry, 8,59 which has been updated further to include (Cl, Br, I) chemistry 11,15 as described by Sherwen et al. 12 The chlorine scheme is described by Schmidt et al., 11 with additions described in Sherwen et al. 15 including further reactions of chlorine and bromine with organics, ClNO 2 emission following N 2 O 5 uptake on sea-salt, 60 and heterogenous iodine cycling to produce IX (X ¼ Cl, Br). 29 The model is run without sea-salt de-bromination following Schmidt et al., 11 and does not contain acid displacement of chlorine or anthropogenic chloride sources. The halogen cross-sections and rates have been updated to the latest NASA-JPL  recommendations. 16 The model includes biogenic emissions (MEGAN 61 ), biomass burning (GFED4 (ref. 62)), biofuel emissions, 63 and aerosols emissions (including dust, 57 sea-salt, 58 and black and organic carbon 64 ) as well as NO x from lightning, 65 soils, 66 and aircra. 67 For anthropogenic emissions, the model uses the Co-operative Programme for Monitoring and Evaluation of the Long-range Transmission of Air Pollutants in Europe (EMEP) emissions (http://www.emep.int) for NO x , 68 SO x , 69 CO, and NH 3 for the latest available year (2013). EMEP anthropogenic VOC emissions are also used here, but for 2012. Emissions for formaldehyde and acetone were scaled from the EMEP acetaldehyde emissions, ethane emissions were scaled from the EMEP propane emission, and a scaling factor was applied to the acetaldehyde emission following the approach taken previously in Dunmore et al. 70 and described in Table SI1 in the ESI. † The halogen emissions used are as described in Sherwen et al. 12 Emissions of organic iodine species are taken from the monthly values of Ordóñez et al. 71 at 1 Â 1 . Emissions of inorganic iodine (HOI, I 2 ) use the parameterisation of Carpenter et al., 24 which describes a dependancy on model parameters of surface O 3 mixing ratio, wind speed, and ocean surface iodide concentration. Ocean surface iodide concentrations are parameterised based on sea-surface temperatures following MacDonald et al. 25 Coastal and tidal processes are not considered here, and the 1 Â 1 resolution of the organic emissions cannot be expected to capture very localised halogen sources.
The GEOS-Chem model is run at two resolutions. A global simulation (4 Â 5 ) generates boundary conditions to allow "nesting" of a domain at a $25 km (0.25 Â 0.3125 ) resolution covering a domain (32.75-61.25 N, À15-40 E) over Europe. The global model is run for two years (1 st January 2004 to 1 st January 2006) with the rst year discarded as "spin up". Using the March 1st 2005 concentrations elds for March 1st 2015, the global model is run for three further months of "spin up" and to cover the observational period in order to generate boundary conditions. The regional model is then run from two weeks prior to the observational period (as "spin up"), before running for the campaign period (29 th June to 1 st August 2015) using the boundary conditions generated by the global model. PM 2.5 is calculated from the model based on the mass of sulfate, nitrate, ammonia, hydrophilic and hydrophobic carbon, sea-salt and dust, assuming relative humidity of 50%. Using the assumed value of 50% relative humidity allows for comparison with DEFRA observations which follows the method prescribed by European Committee for Standardisation (EN 14907). The coarse mode sea-salt and the two largest dust size bins are ignored for the calculation. We have not used the model's secondary organic aerosol scheme in these model simulations. A full description of the PM 2.5 calculation is given in ESI Table SI2. † Model runs performed are described in Table 1. Simulations were performed with halogen chemistry switched on ("HAL") and off ("NOHAL") in both the global (to generate the boundary conditions) and regional model. A simulation was also performed using the boundary conditions calculated with the halogens switched off but with the halogen chemistry in the European domain switched on ("HAL-LOCAL"). A nal simulation ("NOClNO 2 ") was performed with halogen chemistry in both the regional and local version of the model but with the uptake of N 2 O 5 uptake on sea-salt aerosol leading to the production of 2HNO 3 rather than HNO 3 + ClNO 2 .  There are fewer studies assessing the performance of the European grid version of the GEOS-Chem model against observations 43 than for the model's other regional variants (e.g. North America, 41,42 China 39,40 ). Future studies are required to evaluate the model against observations more comprehensively. The AirBase dataset 72 is well suited for this task but this data is not currently available for 2015. Instead here we make some provisional assessment of the model against two observations datasets of standard air-quality pollutants. First, against a subset of observations made at Weybourne as part of the Integrated Chemistry of Ozone in the Atmosphere (ICOZA) campaign and secondly against the observations made as part of the UK AURN network. Once we have evaluated the model against these compounds we turn our attention to its simulation of halogen compounds.

General model performance
A comparison between a sub-set of the observations (O 3 , CO, NO x and NO y ) made as part of the ICOZA campaign and the model ("HAL") are shown as a time-series in Fig. 2 and as an average diel cycle in Fig. 3. The model captures much of the observed synoptic timescale variability in these species. Notable exceptions include the failure to simulate the very high O 3 mixing ratios occurring at the start of the campaign and the high CO mixing ratios in the middle of the campaign. The diel average shows a reasonable ability to reproduce the daily signal in these  compounds other than for CO where the model shows a signicantly larger cycle than is observed. The model has an average low bias (("HAL"-Obs.)/Obs.) of 9.2, 0.7, 2.5, and 11%, for O 3 , NO x , NO y and CO respectively. To give a wider geographical comparison, the model ("HAL") was compared against hourly O 3 , PM 2.5 , and NO 2 observations from the UK AURN air quality network. 54 Sites reporting data and classed as "rural", "rural background" or "urban background" by DEFRA are used for the comparison. Sites inuenced by localised emissions (e.g. roadside sites) are excluded as they are unlikely to provide an appropriate comparison for a model run at 0.25 resolution. A pointby-point comparison between the hourly measured and the spatially and temporally equivalent model values for O 3 is given in the ESI Fig. SI3. † The model fails to capture peak O 3 mixing ratios, which could be expected considering the limited reactive organics present in the model and could also contribute towards the slight underestimate in average O 3 mixing ratios between observation and the "HAL" simulation shown in Fig. 3.
The probability distribution of the O 3 observations, and the model simulation for the AURN sites for the "HAL", "NOHAL","HAL-LOCAL" simulations are shown in Fig. 4 (with equivalent log plots shown for PM 2.5 and NO 2 in ESI Fig. SI4 and SI5 †). The model without halogen chemistry in either the boundary conditions or in the region ("NOHAL") shows substantially higher mixing ratios of O 3 (mean of 34.5 nmol mol À1 , 25 th percentile ¼ 28.5 nmol mol À1 and 75 th percentile ¼ 41.1 nmol mol À1 ) than observed (mean ¼ 27.0 nmol mol À1 , 25 th percentile ¼ 19.0   nmol mol À1 and 75 th percentile ¼ 32.8 nmol mol À1 ). The model without the halogen chemistry in the boundary conditions ("HAL-LOCAL") calculates similarly higher O 3 mixing ratios. However, including halogen chemistry in both the boundary conditions and in the domain leads to a substantial decrease in the modelled O 3 mixing ratios (mean reduction of 26.1%) improving the simulation (mean ¼ 25.5 nmol mol À1 , 25 th percentile ¼ 19.5 nmol mol À1 and 75 th percentile ¼ 31.1 nmol mol À1 ). Unlike for O 3 , where large changes are seen on inclusion of halogens, modest changes are seen for NO 2 and PM 2.5 (ESI plots SI7 and SI8 †). For NO 2 the mean hourly modelled mixing ratio for the "HAL" simulation is 6.7 (25 th percentile ¼ 1.4 and 75 th percentile ¼ 9.5) nmol mol À1 whereas the mean in the "NOHAL" simulation is 7.1 nmol mol À1 . Both can be compared to the observational mean of 7.7 (25 th percentile ¼ 2.6 and 75 th percentile ¼ 10.4) nmol mol À1 . For PM 2.5 the modelled "HAL" mixing ratio was 8.2 (25 th percentile ¼ 4.2 and 75 th percentile ¼ 9.7) mg m À3 with a "NOHAL" mean of 8.6 (25 th percentile We now turn our attention to the model's ability to simulate inorganic halogen compounds over Europe.

Model simulations of reactive halogens in Europe
The simulation of halogens in the global version of GEOS-Chem and its comparison with observations has been discussed previously. 11,12 This provided a rst broad-brush assessment of the mixing ratio of halogens (mainly IO and BrO). It concluded that the model appears to have some skill in simulating IO and BrO mixing ratios, but appears to underestimate Cl species.
Mean surface mixing ratios of key reactive halogens (BrO, IO and Cl) over Europe are shown in Fig. 5 with mixing ratios of total inorganic halogens (X y , X ¼ Cl, Br, I) given in the ESI (SI1 †). We model the highest halogen mixing ratios over the Mediterranean where emissions are greatest. These emissions are notably high for iodine species where the elevated O 3 together with high sea-surface temperature (which determines the ocean iodide mixing ratio in our simulations 24,25 ) leads to a large inorganic iodine ux. A notable difference exists for Cl y (Fig. SI1 †) where a peak can be also be seen in the North Sea/English channel where high mixing ratios of sea-salt and NO x lead to high ClNO 2 production.
Observations of bromine and iodine inorganic species have previously been reported for a few boundary layer locations in Europe, for example Ireland, 17,18 France, [19][20][21][22]73 and Spain. 23 We now compare values reported in the literature to the values calculated in our model for the period of the simulation (for 29 th June to 1 st August 2015). There are undoubtedly, large seasonal and inter-annual variability in these observations, but this comparison allows a rough assessment of the order of magnitude performance of the model.
A number of eld campaigns have occurred over or near tidal coastal zones. IO has been observed at coastal Ireland (Mace Head, 53.3 N, À9.9 E) with peak mixing ratios of between 4 and 50 pmol mol À1 . 74,75 The model predicts a maximum mixing ratio of 0.6 pmol mol À1 here, substantially lower than the observations. IO has also been reported for Brittany (France, 48.7 N, À4.0 E) of between 7.7 (AE0.5) 76  Mixing ratios of IO have been measured by a ship cruise in the marine boundary layer of between 0.4 and 1 pmol mol À1 (30% uncertainty). 31 This cruise did not extend into the Mediterranean region (where we predict highest IO mixing ratios see Fig. 5), but it did nish in the Mediterranean at Cartagena (Spain) in July 2011 with the last daytime average value reported of $0.5 pmol mol À1 (35 N, À8.4 E). For the same location we calculate an average daytime mixing ratio of 0.7 pmol mol À1 .
Observations of iodine dioxide (OIO) have also been reported. At Mace Head, peak OIO mixing ratios have been reported (at night) of between 3.0 (AE0.4) 78 and 13 (AE4) pmol mol À1 . 76 The model predicts substantially lower peak values of 0.09  pmol mol À1 . OIO mixing ratios have also been reported in Coastal France (Brittany, 48.7 N, À4.0 E) of around 9 pmol mol À1 , 19 and with the model calculating signicantly lower mixing ratios, peaking at 0.007 pmol mol À1 . Molecular I 2 has also been observed in Europe in coastal locations including Ireland, Spain and France. In Spain, (42.5 N, À8.9 E) mixing ratios were reported of 300 (AE100) pmol mol À1 . At Mace Head, peak nighttime mixing ratios of between 61 (AE20) 18 and 94 (AE20) 76 pmol mol À1 have been reported and even higher values at nearby Mweenish Bay (53.3 N, À9.8 E) 79 have been found. In France (18.7 N, À8.87 E) mixing ratios of around 50 pmol mol À1 were observed. 19,21 For these locations we calculate far lower maximum mixing ratios of 0.06, 0.04, 0.06, and 0.07 pmol mol À1 , respectively.
In summary, the model signicantly under-predicts reported reactive iodine mixing ratios (IO, OIO, I 2 ) at coastal regions. The most active chemistry in the model occurs in the non-coastal Mediterranean (Fig. 5), a region where we are unaware of published inorganic iodine observations.
Similarly to iodine, only a few bromine observations have been reported for Europe. At Mace Head and Brittany, maximum mixing ratios were reported of 6.5 (ref. 80) and 7.5 (ref. 20) pmol mol À1 . For these locations we predict maximum mixing ratios of 0.8 and 0.5 pmol mol À1 , respectively. Leser et al. 81 reported measurements for a ship cruise from Germany to Capetown in October 2000, which included passing through the English Channel and to the west of Spain. Maximum values were reported of 2.4 pmol mol À1 north of the Canary Islands and a similar value where the English Channel meets the Bay of Biscay. However the rest of the campaign did not report values above the detection limit. For the period the model was run, we predict an average daytime mixing ratio below $0.3 pmol mol À1 in regions of this campaign and even lower mixing ratios in areas with shipping emissions. Fig. 6 shows the observed and modelled time-series and median diel cycle of ClNO 2 mixing ratios at Weybourne in Summer 2015. The observations show a large variability throughout the observational period (Fig. 6) and comparison with the median diel cycle shows a high bias in the model of a factor of $2. The observed hourly-averaged mean daily maximum is 91 pmol mol À1 , with a peak observed of 946 pmol mol À1 . The model compares well in the mean maximum (95 pmol mol À1 ). However modelled peak magnitude is around half the maximum observed value (458 pmol mol À1 ). The reactive uptake parameter used in the model for N 2 O 5 on sea-salt aerosol is 0.005 for dry sea-salt (relative humidity less that 62%) and 0.03 for wet sea-salt. 82 However, if these values are reduced by half then we nd a median peak mixing ratio of 37 pmol mol À1 , closer to the observations. Molecular chlorine (Cl 2 ) was also measured at the site during the ICOZA campaign, but was found to be below the limit of detection (8.5 pmol mol À1 ). The model also does not predict mixing ratios above the limit of detection.

View Article Online
The published continental HCl observations show mixing ratios in the range of tens of pmol mol À1 to a few nmol mol À1 in Italy, 84 Netherlands, 85,86 France, 87 Germany, 88 England, [89][90][91] and Switzerland. 92 The modelled mixing ratios peak at 12 pmol mol À1 . The model therefore signicantly underestimates the HCl mixing ratios. Some of this bias is likely due to a lack of chlorine sources from anthropogenic activities, both organic and inorganic and from aerosol processing of chloride. However, it may also reect excessive loss processes for HCl.
In summary the observational constraints on the modelled halogen concentrations are weak. Much of the observational activity has focussed on process level understanding of halogens at coastal hot spots. For these locations the model appears to systematically underestimate IO, OIO, I 2 and BrO mixing ratios. ClNO 2 mixing ratios inland appear to be underestimated. The model identies the region with the most signicant halogen chemistry as the Mediterranean, a region with a very low number of observations. Fig. 7 shows the difference in the mean surface O 3 mixing ratio between simulations with halogens ("HAL") and without ("NOHAL"). Fig. 8 (top) shows this in percentage terms. O 3 reduces in all locations, and in some locations by a signicant fraction (45% or 28.9 nmol mol À1 ). On average the surface O 3 within the domain drops by 13.5 nmol mol À1 (25%), consistent with previous studies. 5,12,15,93 To assess changes to O 3 within the domain's boundary layer further, we consider the budget of the rapidly interchanging odd oxygen species (O x , dened previously 12 ). Table 2 gives an O x budget for the boundary layer over Europe for the period of the observations (June 29 th to August 1 st 2015) for the simulations with ("HAL") and without halogens ("NOHAL"). Inclusion of halogens leads to a slight decrease in the magnitude of the O x sources of 4%. This is predominantly due to a reduction in the mixing ratio of NO x due to the hydrolysis of halogen nitrates (XNO 3 / aq.HOX + HNO 3 , X ¼ Cl, Br) as discussed on a global scale. 11,12 The O x sink term also decreases (7%) reecting lower O 3 concentrations in the domain. The O x chemical lifetime decreases from 8 days without halogens to 6.5 days with, a 20% reduction. This reduction in the surface O 3 burden consists of two components: a reduction in the background O 3 entering the domain, predominantly from the West (the boundary conditions), and a change to the chemistry occurring within the domain. By running a simulation with the boundary conditions from the global simulation without halogen chemistry, but with halogen chemistry occurring inside the domain ("HAL-LOCAL") we can separate these two factors. Fig. 8 (top) shows the percentage decrease in the O 3 mixing ratio on inclusion of halogens (("HAL" À "NOHAL")/"NOHAL"). The middle panel then shows the decrease which is attributable to the local chemistry (("HAL-LOCAL" À "NOHAL")/ "NOHAL"), with the bottom panel showing the difference between the two panels which we attribute to the global role of halogens in determining the boundary conditions.

European ozone (O 3 )
Over the northern and western part of the domain, the inuence of halogens on the global mixing ratios (as manifested in the boundary conditions) dominates ( Fig. 8 (bottom)). Mace Head (53.3 N, À9.9 E) on the west coast of Ireland is oen used as the default background air quality site for North West Europe. O 3 at Mace Head drops by an average of 12 nmol mol À1 (31%) on the inclusion of halogen chemistry in both the boundary conditions and in the regional model ("HAL"), consistent with previous global studies. 11,12,31 However, this reduction is only 0.51 nmol mol À1 (1.3%) in the simulation where the boundary condition doesn't reect global halogen chemistry ("HAL-LOCAL"). This inuence of the reduced O 3 due to the a reduction in the global background, extends over the European Atlantic regions and into the North Sea. However, its magnitude decreases over continental regions especially in the south of the domain. This is due to the local production of O 3 in these regions and the shorter lifetime of O 3 in continental regions reducing the inuence of boundary conditions compared to marine regions.
Over the southern and eastern part of the domain the global background inuence of halogens plays a less signicant role and it is local halogen chemistry  obvious observational constraints for halogen species here and so their regional inuence is un-assessed. The cumulative distribution functions of surface hourly O 3 mixing ratios over Europe for the differing simulations are shown in Fig. 9. The inclusion of halogens reduces the probability of high O 3 occurring in the model but the difference between the north (>47 N) and the south (<47 N) of Europe is evident. For the north of Europe, only small changes are seen between simulations with only local halogens ("HAL-LOCAL") compared to no halogens at all ("NOHAL"). The median O 3 mixing ratios in the north of Europe (>47 N) are 31.1, 40.0, and 40.5 nmol mol À1 for the "HAL", "HAL-LOCAL", and "NOHAL" simulations, respectively. Local chemistry thus plays little role in determining the median concentrations. However the role of local chemistry becomes more pronounced at the upper end of the O 3 distribution, with the 95 th percentile mixing ratios for these simulations being 54.0, 59.4, and 65.6 nmol mol À1 .
For the south of Europe (<47 N) a larger proportion of change between the simulation with halogens ("HAL") and without ("NOHAL") can be explained by local chemistry ("HAL-LOCAL") and this inuence is felt throughout the O 3 distribution. Fig. 9 shows a reduction in the median O 3 mixing ratio from "HAL" to "HAL-LOCAL" to "NOHAL" of 44.9, 51.1, and 61.0 nmol mol À1 , respectively. Similar reductions can be seen in the 95 th percentile mixing ratios with values of 62.4, 70.7, and 88.1 nmol mol À1 .
The upper end of the O 3 distribution is most important from an air quality perspective. The model shows a decrease in average surface maximum mixing ratios of 19.9 nmol mol À1 on inclusion of halogens. This is greater than the decrease seen in average surface mean mixing ratios (13.5 nmol mol À1 ). For UK legislation, 50 nmol mol À1 (100 mg m À3 ) is important for human health reasons as above this value exceedances are considered. 45.7% of modelled surface O 3 values are above this value when halogens are not included ("NOHAL"), 34.1% when halogens are just considered locally ("HAL-LOCAL") and 18.9% when halogens are considered in all domains ("HAL"). The O 3 mixing ratio of 40 nmol mol À1 is considered an important threshold for ecosystems. 94 We see a decrease in the percentage of hourly surface values above 40 nmol mol À1 from 70.5% in 'NOHAL 00 and 65.9% in "HAL-LOCAL", to 43.3% in "HAL". Halogens reduce the percent of modelled values above 70 nmol mol À1 too, with the values dropping from 15.1% in "NOHAL" to 3.2% in "HAL-LOCAL" and 0.9% in "HAL". Within our model, with our current representation of halogen chemistry, and for the period we have investigated, halogens have a signicant impact on the mixing ratio of modelled O 3 . There are signicant reductions in the mixing ratio of O 3 both in the north and south of Europe but for differing reasons (global background versus local chemistry) with inuences both for the median and higher percentiles of the distribution. There is a need for signicant and further evaluation of the model against an increased observation dataset to develop evidence to support these conclusions but this work suggest that halogens may play a signicant role in determining the distribution of European surface O 3 .

European oxidation
The oxidation of VOCs, CO, and CH 4 in the presence of NO x drives the chemistry of the troposphere. This oxidation is dominated by the OH radical. Within our domain we calculate average boundary layer OH concentrations of 3.53 Â 10 6 , 3.08 Â 10 6 , and 2.89 Â 10 6 molecules cm À3 for the simulations without halogens ("NOHAL"), with local halogens ("HAL-LOCAL") and with global halogens ("HAL"), respectively. The halogens tend to reduce OH mixing ratios (Fig. 10) as they decrease O 3 and thus the production of OH via the primary sources (photolysis of ozone and the subsequent reactions of the photo products with water), and decrease the NO x mixing ratio thus leading to smaller conversion of HO 2 to OH via this route (NO + HO 2 ). The conversion of HO 2 to OH via XO is not large enough to compensate for this. This leads to an average reduction in surface OH mixing ratios of 16%. The largest reductions are simulated where O 3 mixing ratios are reduced and where there is active halogen chemistry which leads to lower NO x mixing ratios due to rapid hydrolysis of halogen nitrates on aerosol.
The inclusion of halogen chemistry brings with it a new oxidant, atomic chlorine (Fig. 5). The average European boundary layer atomic chlorine mixing ratio is 2.1 Â 10 3 atoms cm À3 . This compares with an annual averaged global tropospheric value of 1.3 Â 10 3 atoms cm À3 found by recent global modelling. 7 Daytime modelled Cl mixing ratios at the surface range from 1.5 Â 10 2 to 2.3 Â 10 4 atoms cm À3 , with a maximum hourly value of 2.7 Â 10 5 atoms cm À3 . Within the boundary layer atomic chlorine provides 12, 16 and 9.1% of the sink for ethane, acetone and propane, respectively. It contributes 1.7% of the CH 4 loss. As discussed earlier, a lack of observational constraint results in signicant uncertainties in our simulation of Cl species but these simulations suggest that Cl may play a moderately important role in determining the oxidation of some VOCs within the European domain.  The modelled mean-daily maximum mixing ratio of ClNO 2 is shown in Fig. 11. Peak magnitudes are comparable to those reported in recent modelling work for Northern hemispheric summer of up to 400 pmol mol À1 , 4 and annual values over Europe from global models of 100-140 pmol mol À1 . 12,95 The highest regions for ClNO 2 mixing ratios are seen where shipping emissions are greatest (Fig. 11). By running a simulation without ClNO 2 production ("NOClNO 2 ") the impact of ClNO 2 on O 3 can be assessed.
We nd that increases in O 3 surface mixing ratios on inclusion of ClNO 2 during summertime are modest, as reported previously. 2,4 The maximum increase seen in the average surface O 3 mixing ratio is up to 0.41 (1.2%) nmol mol À1 , which  is within the range of summer enhancement reported previously for the northern hemisphere (0.2-1.6 nmol mol À1 ). 4 Larger changes have been reported in winter time 4 and would be expected if processes increasing chloride concentrations inland were included in the model. In our model, the dominant source of reactive chlorine in the European boundary layer is the production of BrCl from heterogenous routes, 11 rather than the production of ClNO 2 . This source is both more diffuse than the ClNO 2 source which requires high NO x mixing ratios and does not decrease NO x mixing ratios, in contrast to halogen nitrate hydrolysis. It seems likely therefore that when all chlorine sources are considered together they lead to a reduction in O 3 mixing ratios consistent with previous global studies. 12 Signicant uncertainties remain in our fundamental understanding of this heterogenous chlorine chemistry 96 and further laboratory and eld studies are needed to clarify the mechanisms by which chlorine is released from sea-salt.  the ESI. † These changes equate to a domain average decrease of 1.7 and 4.3%, for PM 2.5 and SO 4 2À + NH 4 + + NO 3 À , respectively. NO 3 À shows the largest changes in topographically elevated regions, highlighting the large decreases in NO x seen at these altitudes on inclusion of halogens. 11,12 Small changes are seen in the concentration of SO 4 2À reecting the changes in the oxidants discussed in Section 5. However, halogens may be able to directly impact the production of SO 4 2À through the oxidation on aerosol of SO 2 by hypohalous acids (HOX) on aerosol as has been discussed, 97 which may lead to increased SO 4 2À production.

Conclusions and discussion
We have investigated the impact of Cl, Br and I chemistry on the mixing ratio of O 3 and other pollutants over Europe in the summer of 2015 using the GEOS-Chem model in its European conguration. An initial assessment of the model against observations made at the Weybourne Atmospheric Observatory and from the UK air quality network shows some skill in capturing mean mixing ratios and diel cycle of O 3 , NO 2 , NO y , and PM 2.5 concentrations, however a more extensive assessment of the model in this conguration is needed. Comparisons between observations of ClNO 2 made at Weybourne show a model overestimate on average. However, the model signicantly underestimates ClNO 2 observations reported for more inland regions suggesting some missing processes. The mixing ratios of inorganic bromine and iodine species reported from European sites are signicantly higher than those calculated. This likely reects the lack of realistic representation of coastal processes in the model. Halogen chemistry has a signicant impact on the O 3 mixing ratios calculated over Europe. The north of Europe is mainly sensitive to the reduction in the global O 3 background, whereas the south (notably the Mediterranean) is sensitive to the local halogen chemistry. Chlorine from ClNO 2 leads to small regional increases in O 3 but this is overwhelmed by the decreases caused by other halogens. We nd that mean surface O 3 mixing ratios signicantly reduced by an average of 13.5 nmol mol À1 (25%), with the frequency of hourly mean surface O 3 mixing ratios above 50 nmol mol À1 falling from 46% to 18%. The frequency of occurrence of hourly mean surface ozone mixing ratios above 70 nmol mol À1 falls from 15.1% to 0.9%. Halogen chemistry may therefore play an important role in determining the O 3 exposure over Europe. Oxidant mixing ratios are changed by halogens with OH at the surface dropping due to a reduction in primary production. Atomic Cl leads to some additional oxidation of VOCs, notably for ethane, propane and acetone. Halogens appear to have little impact on aerosol mixing ratios.
Given these simulations it would appear that halogen chemistry may play a signicant role in determining the O 3 mixing ratios found during summertime in Europe, and should be included in model analyses. Further studies are necessary to conrm these ndings and to evaluate whether they have any specic relevance to European air quality policy. For example, do regions change from being NO x or VOC limited on inclusion of the halogens? How does the model respond to future emissions scenarios? It would be surprising if Europe was alone in this sensitivity. Previous global model simulations 12 show other regions where halogens may play a role in determining the O 3 concentrations such as the west coast of the United States and Canada, western India, northern Japan, southern West Africa etc. Air quality simulations for these regions may similarly be sensitive to the inclusion and representation of halogen chemistry.
However, there is little observational constraint on these conclusions. The current set of observations of halogens in Europe are sparse and potentially biased by coastal specic processes. Future efforts to provide observations of atmospheric chlorine, bromine and iodine species in a range of environments, together with ocean iodide observations especially in the Mediterranean would provide a useful constraint here. Continued development of the laboratory measurements, especially of the heterogenous phase chemistry, would also help to provide a better basis for these model simulations and our understanding of the role of halogen chemistry in determining air-quality.