Surface ozone data are retrieved from the World Data Centre for Reactive Gases (WDCRG) and archived at the NOAA Global Monitoring Laboratory (https://www.esrl.noaa.gov/gmd/ozwv/surfoz/data.html, last access: 1 December 2020). The measurements of surface ozone are made by several brands of dual-cell UV absorption monitors, which relate UV absorption to ozone concentration following the Beer–Lambert law. Details can be found in the articles VRS in Skov et al. (2004, 2020) and Alert in Bottenheim et al. (2002) or in review articles by e.g. Gaudel et al. (2018), Oltmans et al. (2010), and Cooper et al. (2014). In general, the technique has a detection limit of about 1 ppbv and an uncertainty (95 % confidence interval) of about 1 ppbv for VMRs below 10 ppbv and about 2 ppbv for more typical surface VMRs of 30–40 ppbv (Galbally et al., 2013; Tarasick et al., 2019a). Ozone-sonde data from the three Canadian stations used here can be found at the World Ozone and Ultraviolet Radiation Data Centre (WOUDC). During the period of interest here, all ozone-sondes used were electrochemical concentration cells (ECCs; Komhyr, 1969), manufactured by Environmental Science (EN-SCI) Corp. All sondes used the conventional neutral-buffered 1 % potassium iodide sensing solution. The data records of the Canadian sites have recently been re-evaluated (Tarasick et al., 2016). Based on the typical ozone sensor response time of 25–40 s (Smit and Kley, 1998) and assuming a typical balloon ascent rate of 4–5 m s-1, the ozone-sondes have a vertical resolution of about 100–200 m. Measurement precision is ±3 %–5 %, and the overall uncertainty in ozone VMRs is less than 10 % in the troposphere (Kerr et al., 1994; Smit et al., 2007; Tarasick et al., 2016, 2019a, 2020).

Ozone-sonde releases are normally once per week, although additional releases are often scheduled during observational campaigns in the Arctic spring. Despite their low frequency of observation compared to surface monitoring, ozone-sondes have been used successfully to study boundary layer processes like ODEs (e.g. Bottenheim et al., 2002; Tarasick and Bottenheim, 2002) and long-range transport (e.g. Oltmans et al., 2010; He et al., 2011; Tarasick et al., 2019b).

In addition to the ozone measurements, several other datasets are employed in this study: tropospheric columns of BrO from the Global Ozone Monitoring Experiment-2 (GOME-2; Callies et al., 2000) instrument on board the Meteorological Operational Satellite-A (MetOp-A) and lower-tropospheric profiles of BrO from ground-based multi-axis differential optical absorption spectroscopy (MAX-DOAS) at Eureka, Canada.

The GOME-2 tropospheric columns of BrO used in this study are described in further detail by Blechschmidt et al. (2016). In summary, tropospheric BrO vertical columns (VCDtrop) were obtained based on the approach of Begoin et al. (2010) for deriving BrO total slant column densities by the DOAS (Platt, 1994) method using a 336–347 nm fitting window (Afe et al., 2004) and on Theys et al. (2011) for stratospheric correction. The latter involves the use of a climatology of stratospheric vertical column densities (VCDs) of BrO estimated by the BASCOE (the Belgian Assimilation System for Chemical ObsErvations) chemical transport model (Errera et al., 2008; Viscardy et al., 2010). The stratospheric VCDs were converted to slant columns by application of a stratospheric-air-mass factor and then subtracted from total slant columns. A tropospheric-air-mass factor was applied for conversion to VCDtrop assuming that all BrO is located and well mixed within the lowermost 400 m of the troposphere over ice or snow with a surface reflectance of 0.9. A sensitivity study for a BEE case showed that the GOME-2 tropospheric BrO column has a moderate sensitivity to the stratospheric BrO column; e.g. a variation in the VCDstrat of 15 %–30 % leads to a change in VCDtrop of about 0.5 to 1 × 1013 molecules cm2, respectively (Zhao et al., 2016). The influence of clouds on GOME-2 BrO retrievals and the implications for studying bromine explosion events using GOME-2 data are discussed in Blechschmidt et al. (2016). GOME-2 tropospheric BrO column maps (0.5×0.5∘ grid) and time series based on subsetted data of VCDtrop (all measurements having their centre within a distance of < 40 km from the ground station) at the Resolute, Eureka, and Alert sites are used here.

MAX-DOAS measurements were performed at the Polar Environment Atmospheric Research Laboratory (PEARL) Ridge laboratory (610 m) in Eureka. Spectra were recorded in the UV using a grating spectrometer (1200 groves/mm grating) with a cooled (200 K) charge-coupled device (CCD) detector at 0.4–0.5 nm resolution. Elevation angles of 30, 15, 10, 8, and 5∘ (6∘ in 2011) were used in the elevation scans, and measurements were only taken with solar elevation above 4∘. Differential slant column densities (dSCDs) of BrO and the oxygen dimer (O4) were retrieved using the settings described in Zhao et al. (2016). Reference spectra for the DOAS analysis were interpolated from zenith measurements taken before and after each elevation scan. The dSCDs were converted to profiles using a two-step optimal-estimation method (Frieß et al., 2011). First, aerosol extinction profiles were retrieved from O4 dSCDs, and then the extinction profiles were used as a forward model parameter in the BrO retrieval. The retrievals were performed on a 0–4 km altitude grid with 0.2 km resolution. Due to the altitude of the instrument (610 m) and the lack of low or negative elevation angles, the retrieved profiles are only sensitive to well-mixed BrO in a deep boundary layer and to lofted BrO events.

The Cambridge parallelised-Tropospheric Offline Model of Chemistry and Transport (p-TOMCAT) has a horizontal resolution of 2.825∘ × 2.825∘ (longitude × latitude) and 31 vertical layers from the surface to about 10 hPa (∼ 31 km) at the top layer. Sea ice coverage and sea surface temperatures are monthly and taken from the Hadley Centre Sea Ice and Sea Surface Temperature dataset (Rayner et al., 2003). The p-TOMCAT non-local vertical diffusion scheme is taken from the National Centre for Atmospheric Research Community Climate Model, Version 2. This scheme determines the planetary boundary layer (PBL) height explicitly and takes account of large-scale eddy transport that can occur throughout the boundary layer even when part of it is statically stable. Implementation and validation of the PBL scheme was carried out by Wang et al. (1999). The model behaviour in terms of vertical mixing of atmospheric tracer and air mass transport has been reported in Russo et al. (2011) and Hoyle et al. (2011).

The ozone photochemistry scheme applied to the model has been detailed in previous studies (Law et al., 1998, 2000; Savage et al., 2004), with updates including an isoprene chemistry scheme, same as the one implemented in the UKCA model by Young et al. (2009) according to the method of Pöschl et al. (2000); a hydrolysis reaction of N2O5 on aerosols and cloud droplets (Yang et al., 2005); a tropospheric-bromine scheme involving both gaseous-phase reactions (Yang et al., 2005) and heterogeneous reactions (Yang et al., 2010); and a Fast-J photolysis scheme developed by Voulgarakis et al. (2009b), which is not used in this study. They found that N2O5 hydrolysis can cause net NOx loss at high latitudes by up to 60 % in the Northern Hemisphere and ∼ 80 % in the Southern Hemisphere (Yang et al., 2005). They found that including halogen-related heterogeneous reactions on aerosols and cloud droplets can significantly increase polar BrO partitioning by a factor of ∼ 3 (Yang et al., 2010). This heterogeneous reaction scheme for halogen reactivation was also implemented to the UKCA model (Yang et al., 2014; Dennison et al., 2019; Ming et al., 2020).

Ozone is dry-deposited in the bottom model layer with dry-deposition velocity inferred from the study of Ganzeveld and Lelieveld (1995) by Giannakopoulos (1998). The original dry-deposition velocity over ocean and snow (= 0.05 cm s-1) is reduced to 0.01 cm s-1 in this study following recent modelling work by Hardacre et al. (2015) and Luhar et al. (2018) as well as Helmig et al. (2007a). Since p-TOMCAT only covers part of the stratosphere with a top layer height of ∼ 31 km, a simplified stratospheric-chemical scheme has to be used, including a pre-prescribed top boundary condition for ozone. Therefore, the p-TOMCAT model is quite different from the UKCA model in the upper troposphere and lower stratosphere. However, it is unlikely that the downwards transport of air mass in the polar region may significant influence near-surface bromine. Recent changes to p-TOMCAT, the tropospheric-halogen-chemistry scheme, include updates to dry- and wet-deposition schemes as reported in Legrand et al. (2016). Tropospheric bromine comes from three emission sources: (i) very-short-lived substances (VSLSs) of bromocarbons following the work of Warwick et al. (2006) with reduced flux for CH2Br2 (Yang et al., 2014), (ii) open-ocean sea spray (Yang et al., 2005, Breider et al., 2009; Parrella et al., 2012), and (iii) sea-ice-sourced SSA in polar regions following the work of Yang et al. (2008, 2010, 2019). Here we define total inorganic bromine BrY= HOBr + HBr + BrO + Br + BrONO2+ BrNO2 + 2⋅ Br2+ BrCl.

A process-based SSA transport dry- and wet-deposition scheme has been implemented in the model by Levine et al. (2014) based on the work of Reader and McFarlane (2003). The open-ocean sea spray emission scheme follows Jaeglé et al. (2011), and the sea-ice-sourced SSA scheme follows the latest work of Yang et al. (2019). Both open-ocean-sourced and sea-ice-sourced SSA (denoted as OO and SI, respectively) are tagged in 21 size bins covering dry NaCl diameter of 0.02–20 µm in order to track their history for online calculation of their surface density for heterogeneous reaction rates.

All parameters applied in this study for the Arctic SSA simulation are directly taken from our recent SSA modelling work by Yang et al. (2019), including a 3.5 times Antarctic snow salinity for the Arctic. The Antarctic Weddell Sea cruise data (Frey et al., 2020) are a probability of surface snow salinity, which is different to the constant salinity value (= 0.3 psu, practical salinity unit) used in Legrand et al. (2016), Zhao et al. (2017), and Rhodes et al. (2017). The trebled snow salinity assumption is taken from Yang et al. (2008) to reflect the likelihood that Arctic snow is more saline than in the Antarctic due to reduced precipitation. This assumption is partly justified by surface snow [Cl-] concentrations observed in the two poles. For instance, an averaged surface snow (top 1–2 cm) [Cl-] concentration of 368 µM is derived from the Weddell Sea, Antarctic (https://ramadda.data.bas.ac.uk/repository/entry/show?entryid=853dd176-bc7a-48d4-a6be-33bcc0f17eeb, last access: 1 December 2020). In the Arctic, Pratt et al. (2013) reported a mean surface [Cl-] concentration of 1121 µM (top 1 cm) over coastal sea ice near Barrow, Alaska, and Krnavek et al. (2012) reported a much higher surface [Cl-] concentration of 21 058 µM over first-year sea ice and 63 217 µM over multi-year sea ice over a slightly deeper depth of 2–3 cm below the surface. They are about 3, 57, and 172 times the Weddell Sea surface salinity. The relative higher salinity in the Arctic is partly related to less precipitation as already mentioned. For instance, the depth of snowpack on sea ice near Barrow, Alaska, is in a range of 10–40 cm (Krnavek et al. (2012), while in the Weddell Sea, the mean snow depth over first-year ice (FYI) is 20.9 and 50.0 cm over multi-year ice (MYI) (Frey et al., 2020).

Other parameters used in this study include a mean snow age of 3 d for the Arctic following the recent work of Huang and Jaeglé (2017). We assume that the evaporation rate of blowing snow particles is controlled by the moisture gradient between the surface of the particle and the ambient air, an evaporation function of dmidt=di, with mi being water mass and di being diameter of snow particle), i.e. the classic mechanism in Yang et al. (2019). For blowing-snow size distribution, we used a shape parameter α=3 and a scale parameter β=37.5 µm, with an SSA production ratio N=20 (i.e. 20 SSA particles formed from one saline wind-blown snow particle during sublimation). This set of parameters corresponds to the SI_Classic_B×20 run in Yang et al. (2019) and is one of the best parameter sets that matched the Weddell Sea SSA in the size range of 0.4–10 µm. Also, this set gave the highest SSA mass loading in polar regions (Yang et al., 2019).

To parameterize bromide release from SSA in the Arctic, two different patterns of bromine depletion factor (DF) are used. Table 1 contains a seasonal DF scheme with a maximum value of 0.53 in May and a minimum of 0.07 in December. This seasonal scheme is derived from the bulk SSA bromide depletion strength from Dumont d'Urville (Legrand et al., 2016), Cape Grim, and Macquarie Island (Ayers et al., 1999) in the Southern Hemisphere with a 6-month shift in the phase in order to apply to the NH. Since a similar year-round in situ dataset from the Arctic is not available, we could not justify this seasonal DF pattern, which demands further systematic measurements in the Arctic. As used in previous modelling studies, a size-dependent (non-seasonal) DF scheme for the NH is used for comparison (Supplement Table S1), which is derived from previous work of Yang et al. (2008, 2010) and Breider et al. (2009). Note that we simply apply these DF schemes to all SSA emitted and do not distinguish between the open-ocean-sourced and the sea-ice-sourced SSA in terms of bromide release. However, this approach may introduce bias as freshly emitted sea spray is alkaline with pH > 8 and needs acidification first by absorbing sulphate or nitrate before bromide can be liberated to the atmosphere through heterogeneous reactions (e.g. Breider et al., 2009). In contrast, snowpack in the Arctic is largely acidified with pH of 4–6 due to local acidity contamination (e.g. de Caritat et al., 2005). The difference in initial conditions between sea spray and sea-ice-sourced SSA may affect bromide release in both timing and strength, which has not been considered by our models. Thus, we may overestimate the open-ocean-sourced SSA effect in polar regions as the alkaline buffering effect is not considered.

UKCA, a version of the UK Earth System Model with Chemistry and Aerosols, has a dynamical core from the Met Office Unified Model (UM; Morgenstern et al., 2009). A nudged model version 7.3 is used in this study with a horizontal resolution of 3.75∘ × 2.5∘ and 60 vertical layers from the surface to ∼ 84 km. The tropospheric-chemistry scheme was built on the scheme in the p-TOMCAT model but contains a comprehensive stratospheric-chemistry scheme for climate studies (Braesicke et al., 2013; Banerjee et al., 2014; Ming et al., 2020). In terms of SSA production, the same schemes for open-ocean sea spray and for sea-ice-sourced SSA as in the p-TOMCAT are used apart from the fact that the SSA in the UKCA runs is no longer being tagged and tracked for online calculation of heterogeneous reaction rates. Therefore, the emitted SSAs are just used for bromide emission. For heterogeneous reactions, the aerosol surface area density is calculated using the archived monthly climatology aerosol dataset taken from the CLASSIC scheme (Johnson et al., 2010). In the p-TOMCAT model, heterogeneous reactions occur also on cloud droplets, but UKCA does not include such reactions on cloud droplets. Therefore, in the free troposphere, the BrO partitioning in UKCA may be lower than that in p-TOMCAT, which may result in more soluble inorganic-bromine species being washed out by precipitation in UKCA, as discussed in Sect. 4.

Note that the Unified Model UM-UKCA is a complex chemistry–climate coupling model covering the whole atmosphere, including both troposphere and stratosphere. In many aspects of dynamics and chemistry, it behaves quite differently from the p-TOMCAT CTM. A detailed comparison of model characteristics in vertical mixing and transport of tropical boundary layer tracers was performed by Russo et al. (2011) and Hoyle et al. (2011). The bottom model layer of UKCA, in which chemical compounds such as ozone undergo dry deposition, is ∼ 20 m thick, while in p-TOMCAT it is ∼ 60 m thick. All released SSA and bromine (in the form of Br2) are put in the bottom model layer before they are further vertically mixed and horizontally transported. These differences in model vertical resolution may affect model output even if other factors are the same. Although the two models are quite different, e.g. in absolute values of chemical compounds, the relative changes in response to changes in bromine loading, for example, are still informative and are our major interest and focus of the discussion.

Table 2 lists major model experiments performed in this study. The two model base runs, pTOMCAT_control and UKCA_control, contain reactive-bromine emissions from both sea-ice-sourced and open-ocean-sourced SSA and VSLS bromocarbons. The pTOMCAT_No_Br run does not include any bromine emission; therefore, it is a model run without bromine chemistry. The pTOMCAT_VSLS and UKCA_VSLS runs only contain bromocarbons as a source of reactive bromine (without bromine from open ocean and sea ice). The pTOMCAT_SI_VSLS and UKCA_SI_VSLS runs only contain sea-ice-sourced SSA and bromocarbons as sources of reactive bromine. Similarly, the pTOMCAT_OO_VSLS and UKCA_OO_VSLS runs only contain sea-spray-sourced SSA and bromocarbons as sources of reactive bromine. By checking the differences between experiments from the same model, we expect to separate individual bromine source contributions to Arctic boundary layer bromine mixing ratio and ozone mixing ratio. Similarly, by checking the differences between the two model responses, we will see model-induced uncertainty, e.g. in both the bromine mixing ratio and ozone mixing ratio. This is because both models employ very similar bromine emissions. Therefore, the differences are mainly due to different model configuration either in their physical aspect – including precipitation, boundary layer dynamics, land use, etc. – or in their chemistry aspect involving key atmospheric species such as ozone or OH and heterogeneous reactions.

Apart from the pTOMCAT_Fixed_DF run, in which a fixed (non-seasonal) DF scheme is used (Table S1), all model experiments apply the same bromine DF scheme shown in Table 1. A multiple-year integration (2006–2008) is performed, with averaged outputs used as a climatology for comparison. Several spring runs in 2010, 2011, and 2013 are made with more frequent outputs for ODE and BEE comparisons: 1-hourly output frequency in pTOMCAT_control and 3-hourly output frequency in UKCA_control are used for further analysis and model data comparisons. These years are selected because either significant ODEs or BEEs are observed at one or more sites.

To investigate model sensitivity to key parameters, such as snow salinity, DF, cut-off size, and SSA spectrum, we performed additional model experiments (Table 3) with a range of uncertainty for each parameter. For example, pTOMCAT_high_salinity applies a snow salinity 10 times that of Weddell Sea, and pTOMCAT_low_salinity applies a salinity 1 times that of the Weddell Sea; pTOMCAT_2×DF applies a doubled DF, and pTOMCAT_0.5×DF applies a halved DF; pTOMCAT_SSA20µm applies a large cut-off threshold with a dry NaCl radius of 20 µm, and pTOMCAT_SSA5µm applies a small cut-off threshold of 5 µm; pTOMCAT_spectrum_1 applies the same parameters as in the control run but a small N=10, and pTOMCAT_spectrum_2 applies a different parameter set with N=1 and different α,β, and dmidt functions (see Table 3); the corresponds to the SI_Base run in Yang et al. (2019). This sensitive experiment is only integrated for 1 year (2007) with results are compared to the pTOMCAT_control run result, as discussed in Sect. 5.

Figure 2 shows observed monthly mean surface ozone VMRs at the six Arctic locations: two inland (Summit and Pallas) and four coastal (Alert, Barrow, Tiksi, and VRS) sites. Spring ozone maxima of > 50 ppbv at Summit and ∼ 45 ppbv at Pallas were observed. However, at coastal sites, springtime ozone is depleted, with low VMRs of 15–20 ppbv, which are comparable to or even lower than their summer ozone minimum in July–August. The summer minimum is thought to be attributable to enhanced ozone photo-dissociation, where NOx levels are low, and increased dry deposition to plants (e.g. Hatakka et al., 2003; Engvall Stjernberg et al., 2011). At higher-latitude sites such as Alert and VRS that are within the polar dome and surrounded by Arctic tundra with sparse vegetation, there is normally still snow coverage even in mid-summer, so the local effect of dry deposition to plants may not be as significant as at Pallas or other sites located further south. However, the long-range transport from lower latitudes of ozone affected by summer plants may result in vegetation having an effect on these sites. For example, the suppressed high-latitude summer ozone in Siberia is related to deposition loss to vegetation during long-range transport into the Arctic (Engvall Stjernberg et al., 2011). Figure 2 also shows that model runs without bromine chemistry (pTOMCAT_No_Br) and with bromocarbons only (pTOMCAT_VSLS) can generate spring ozone maxima at all six sites. When open-ocean-sourced reactive bromine is included (orange line in Fig. 2), the spring ozone peak is reduced significantly. The OO-sourced bromine can cause ozone reductions in all seasons (Fig. 3), with a maximum reduction of > 10 ppbv in April and a minimum reduction (1–2 ppbv) in summer. However, the OO-sourced reactive bromine does not alter the ozone seasonality pattern as the spring ozone peak remains (Fig. 2). On the other hand, sea-ice-sourced reactive bromine (red line in Fig. 2) can significantly perturb the ozone seasonal cycle by removing the spring ozone peak completely. On average, SI-sourced bromine can cause a maximum ozone loss of > 10 ppbv in April at coastal sites (Fig. 3), similar to the OO-sourced reactive-bromine effect. In autumn, SI-sourced bromine only weakly influences the bromine budget and ozone loss. Model runs which contain bromine sources from both OO and SI (black line in Fig. 3) can cause a peak of annual ozone loss of > 25 ppbv (in monthly mean) in April at coastal sites, giving the best match to the observations (Fig. 2). However, inclusion of halogen chemistry leads to severe underestimation of spring ozone at Summit and Pallas.

A similar effect of the SI- and OO-sourced reactive bromine on surface ozone can be seen in the UKCA runs but with net ozone loss only half of that seen in the p-TOMCAT runs (Fig. S1). As discussed below, this difference is consistent with the difference in total inorganic bromine BrY between the two models: a spring surface layer BrY maximum of 10–30 pptv is simulated in pTOMCAT_control (Fig. 3), which is about twice that (5–10 pptv) in UKCA_control (Fig. S1). This is consistent with zonal mean (April) BrY differences between the two models as shown in Fig. S2. Since both models employ a very similar bromine emission flux (e.g. SSA production driven by ECMWF data and with the same bromine depletion factor), the difference in BrY between the two models is likely due to the difference in removal process of inorganic-bromine species, such as HBr, HOBr, Br2, and BrONO2, which are dry- and/or wet-deposited. Previous model simulations have shown that, on the global scale, precipitation wash-out is responsible for ∼ 90 % of the removal of tropospheric BrY (Yang et al., 2005). In polar regions, where the precipitation rate is relatively low, dry deposition to the surface is another efficient pathway for inorganic-bromine removal in surface layer. The different approaches in chemical scheme applied by the models may also affect inorganic-bromine deposition rate through influencing partitioning of inorganic-bromine species. This is because some species (e.g. HBr, HOBr) are very soluble, while others are not (e.g. BrO). A higher BrO partitioning is expected at a higher ozone concentration and vice versa. Therefore, an overestimated ozone is expected to have a negative feedback to bromine removal and net ozone loss via bromine chemistry. In addition, p-TOMCAT considers heterogeneous reactions on cloud droplets, while UKCA does not; this difference may explain why BrO partitioning in p-TOMCAT is higher than that in UKCA, especially in the free troposphere, where BrO partitioning can be as large as 50 % (Fig. S2). In addition, the higher BrO partitioning in p-TOMCAT also attribute less BrY removal by dry and wet depositions.

Comparing the surface layer BrO/BrY ratio between the two models (Fig. S3), we can see that both UKCA_control and pTOMCAT_control give a very similar spring peak, with a ratio around 20 %–30 % at coastal sites, with an exception at VRS, where a ratio of up to > 60 % is simulated in the UKCA_control run. The largest discrepancy appears in summer at some costal sites such as Alert, Tiksi, and VRS, where a second summer peak ratio is simulated in the UKCA_control, which is likely attributed to the obviously overestimated summer ozone concentrations by this model (Fig. S4).

From Figs. 2 and S1 we can see that, on average, surface BrO VMRs at inland sites are smaller than those at coastal sites in both model outputs. For example, in April, mean BrO is ∼ 1 pptv at Summit and ∼ 0.5 pptv at Pallas; at coastal sites, VMRs are between 2 and 7 pptv in the pTOMCAT_control run. In terms of BrY, as shown in Fig. 3, in April, both OO- and SI-sourced bromine contributes roughly the same amount (6–8 pptv) at the two inland sites of Summit and Pallas. At coastal sites (Alert, Barrow, Tiksi, and VRS), the OO-sourced bromine contributes one-sixth to half of the SI-sourced BrY, i.e. 4–5 pptv vs. 8–30 pptv. The large (and small) gradient in BrY of the SI-sourced (and the OO-sourced) bromine between inland and coastal sites indicates that SI-sourced bromine is locally sourced, while OO-sourced bromine is remotely transported and thus has a smaller horizontal gradient in VMR. In addition, VSLS bromocarbons only have a relatively small contribution of ∼ 0.5 pptv BrY in spring–summer (Fig. 3), corresponding to an ozone loss of ∼ 1 ppbv. As shown in Fig. S7, VSLS contribution to tropospheric BrY over the Arctic increases from the near-surface layer ∼ 0.5 pptv (in April and July) to ∼ 2 pptv at ∼ 200 hpa. In spring (April), it only accounts for 2 %–4 % of the total BrY in the surface layer and ∼ 40 % at 200 hpa; in July, it accounts for 15 %–20 % in the surface layer and > 60 % at 300–400 hpa.

Figure 4 shows NH summer (July) surface ozone frequency distribution at four coastal sites from both observation and p-TOMCAT runs. Climatology clearly indicates a single ozone peak distribution with peak VMRs around ∼ 20 ppbv. The p-TOMCAT model successfully reproduces this summer single peak distribution frequency, though it overestimates it by a few ppbv, e.g. at Alert and VRS. Similar to what is reflected in Fig. 3, the bromine effect on ozone loss is small, only 1–2 ppbv. In April, the observed ozone distribution frequency is quite different from the summer pattern as a flat distribution across ozone bins is observed with a large ozone depletion fraction at ozone VMRs < 10 ppbv (Figs. 5 and 6). Although both p-TOMCAT (Fig. 5) and UKCA (Fig. 6) fail to reproduce this flat distribution pattern, the two model runs with SI-sourced bromine implemented can largely reproduce ozone depletion fraction (at ozone < 10 ppbv). It is interesting to note that although OO-sourced bromine alone can cause April monthly mean ozone to drop by 5–10 ppbv at coastal sites (dashed blue line vs. solid blue line in Figs. 5 and 6), it alone cannot generate any ozone depletion at VMRs < 20 ppbv, indicating that this remotely sourced bromine to the Arctic is not responsible for coastal ODEs; rather, it only affects background ozone. On the other hand, SI-sourced bromine can cause ozone depletion with a significant fraction of ozone VMRs < 10 ppbv (dashed orange line in Figs. 5 and 6), supporting the suggestion that locally sourced bromine (from sea ice) is responsible for spring ODEs.

As discussed in Sect. 4.3, the failure of models in reproducing the flat ozone distribution in spring is likely attributable to the coarse resolution of the models used. For instance, the horizontal resolutions of ∼ 2.8∘ ×  2.8∘ in p-TOMCAT and 3.75∘ × 2.5∘ in UKCA mean that any sub-grid-scale events will not be captured and represented by the model, and a finer-resolution model may be needed to have better representation of the observations.