We use the Modular Earth Submodel System (MESSy) v2.55.2 in combination with basemodel ECHAM5.2 (ECHAM/MESSy) referred to as EMAC. MESSy is a software framework that combines Earth system components, such as atmosphere, land, ocean, and subsystems of these, in a modular way . This modularity enables the deployment of different basemodels with an identical numerical implementation of geophysical and chemical processes and parameterization.

We base the EMAC model setup on REF-C1SD of the Chemistry Climate Model Initiative (CCMI) 2 . We use specified dynamics (SD) for the historical period 2019/2020, with nudged surface pressure (logarithmic) and nudged temperature, divergence, and vorticity fields from above the boundary layer (model level 8) up to a pressure level of 1 hPa. Sea Surface Temperatures (SSTs) and Sea Ice Cover (SIC) are also prescribed. Nudging data were generated from the European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis v5 (ERA5) .

All simulations were conducted in a T42L90MA resolution with the spectral truncation T42 that translates to a (2.7851×2.8125)° regular Gaussian grid in the polar regions (latitude >±68° in both hemispheres). Our model setup comprises 90 hybrid pressure levels up to 0.01 hPa. These hybrid pressure levels are terrain following, resulting in about 1–3 levels within the Arctic spring-time BL. In the Arctic, these levels translate on average to heights above ground at the upper boundary of each level as follows: level:(1,2,3,…,8)→height:(99,254,538,…,3571)m.

We use the Module Efficiently Calculating the Chemistry of the Atmosphere (MECCA) with the CCMI-base-02 chemistry mechanism in which mercury and iodine chemistry are switched off. This chemical mechanism comprises 261 gas-phase, 12 heterogeneous, and 80 photolysis reactions (Sect. S1 in the Supplement). The explicit calculation of heterogeneous reactions is restricted to the stratosphere and uses a prescribed aerosol surface concentration climatology (1996–2005) computed with the MESSy submodel MADE . In the troposphere and boundary layer, we parameterize the emission of Br2 as described in the following paragraph. We depict the most relevant cycles for ODEs grouped by families in a reaction network (Fig. ). The NO_x cycle can be net-zero regarding ozone destruction and production.

In the following, we briefly describe the most relevant MESSy submodels that handle trace gas emissions and sinks. For a complete list of submodels used in this study, please consult the file switch.nml in the Supplement (EMAC_namelists.zip). The boolean values therein indicate whether the submodels were used.

Trace gas emissions that depend on the state of the Earth system components (atmosphere, land, and ocean) are computed during the model integration in the MESSy submodel ONEMIS . Therein, the subroutine AirSnow handles the emission of bromine from sea ice and snow in the polar regions following the scheme suggested by . The dry deposition flux of O3 on first-year sea ice (FY) triggers Br2 emissions. Assuming that a considerable amount of bromide (Br-) has already been pushed out of multi-year sea ice (MY), recycling is limited by the amount of inorganic bromine (HBr, BrNO3, and HOBr) deposition from gas-phase. A similar assumption is used for snow on land (LS). The distribution of MY sea ice is prescribed. This scheme introduces a critical temperature Tcrit effectively acting as a seasonal limit and a critical solar zenith angle (θcrit) as a trigger that separates dark from sun-lit conditions with two molar yields (ϕ1, ϕ2) parameterizing in-snow photochemistry. The sea salt aerosol mass flux is computed with the LSCE scheme that applies lookup tables depending on wind speed. A detailed comparison of all available sea salt aerosol emission schemes is found in . The corresponding bromine flux is derived by scaling this flux with a Cl to Br molar ratio of 667 and assuming half of this flux is released into the gas-phase .

Ocean-to-atmosphere fluxes of organic substances dissolved in seawater (e.g. dimethylsulfate (DMS), bromoform CHBr3 and dibromomethane CH2Br2) are computed by the AIRSEA submodel. The flux for each species is calculated from concentration gradients in the uppermost ocean layer and the lowermost atmosphere layer. The species-specific Henry's law constant defines the respective solubility. The corresponding concentrations of brominated very short-lived substances (VSLS) in ocean waters are taken from .

The submodels OFFEMIS and TNUDGE handled emissions from preprocessed 4D forcing fields . Emission inventories of greenhouse gases (GHGs), Ozone Depleting Substances (ODSs), and ozone precursors are based on the German Aerospace Centre (Deutsches Zentrum für Luft- und Raumfahrt, DLR) CCMI2 inventory (available until 2019). GHGs included are N2O, CH4, and CO2. ODSs include a wide range of halons, chlorofluorocarbons (CFCs), and hydrochlorofluorocarbons (HCFCs). Also included are potential ozone precursors like Volatile Organic Compounds (VOCs) and Non-Methane-Hydro-Carbons (NMHCs) from biomass and agricultural waste burning as well as in fossil fuel emissions from traffic and non-traffic consisting of VOCs (only acetylene (C2H2)), NMHCs, NO, CO, SO2, NH3 plus aerosol SO4 derived from SO2, organic carbon (OC) and black carbon (BC) emissions. These CCMI2 inventories are not available beyond 2019 for our model, so we substituted the missing data with 2019 emissions.

Emissions of the minor VSLS (CHCl2Br, CHClBr2, CH2ClBr) are based on . The submodel TNUDGE nudges the surface concentration climatologies of the long-lived halons CF3Br and CF2ClBr, and CH3Br towards observed mixing ratios.

Dry deposition is the most important way of removing non-soluble chemical substances from the atmosphere. The corresponding dry deposition velocities (vdd) are computed by the submodel DDEP following the Wesely-scheme : 1vdd=1ra+rb+rc, with aerodynamical resistance (ra) driven by atmospheric turbulence, quasi-laminar resistance (rb) driven by molecular diffusion and turbulence, and surface resistance (rc). For most species, there are no observations of rc. Estimates are calculated based on the Henry's law coefficient (Hi), a reactivity rate coefficient (Rdryi), and rc of O3 and sulfur dioxide (SO2): 2rc,snowi=1Hirc,snowSO2⋅105+Rdryirc,snowO3 with rc,snowO3=2000sm-1 and rc,snowSO2=1sm-1.

Wet deposition of soluble species is calculated via the SCAV submodel . It considers convective and large-scale precipitation separately as well as three modes of scavenging from ice-, liquid- and gas-phase.

For evaluating our model results, we use surface ozone VMR from a network of Arctic monitoring sites, satellite observations of BrO column VCD from the TROPOspheric Monitoring Instrument (TROPOMI) onboard the European Space Agency (ESA) Copernicus Sentinel-5 Precursor satellite, and atmosphere and trace gas observational records from the Multidisciplinary drifting Observatory for the Study of Arctic Climate (MOSAiC) expedition (2019/2020). The ozone monitoring sites and MOSAiC drift track locations' are shown in Fig. .

By construction, the amount of Br2 emitted using the Toyota scheme is sensitive to the assumed age of sea ice. Therefore, we expect more Br2 emissions from regions with a large first-year sea ice concentration (FYSIC). derived a multi-year sea ice concentration (MYSIC) from ERA-interim with the assumption of a static multi-year sea ice distribution computed from the SIC at the seasonal minimum of the previous year. We used the same assumption to derive MYSIC from ERA5 SIC and compared this with an age of sea ice (AoSI) product provided by the Integrated Climate Data Center (ICDC) .

ICDC AoSI uses sea ice drift data from satellite observations to assign an age to the individual ice floats for which drift trajectories are computed. Each grid cell with at least 15 % SIC is treated as a Lagrangian particle and traced every week. The co-existence of ice of varying ages in one grid cell prefers the survival of the older ice because younger and thinner sea ice deforms and melts more easily. This causes an overestimation of the multi-year sea ice cover.

To derive a MYSIC, we summed over all ICDC age classes ≥1 year and average the bi-weekly data over one month. The original resolution of 12.5×12.5km was then remapped onto the target resolution (e.g. T42, T106, Fig. ).

Compared to ERA5-derived MYSIC, ICDC AoSI indicates distinctively less multi-year sea ice in the Central Arctic Ocean and the Canadian archipelago but more east of Greenland, north of Spitsbergen and Alaska (Fig. a). To identify which of the two MYSIC estimate is more realistic, we computed the total area covered by multi-year sea ice. The area derived from ERA5 is consistently ∼ 30 % larger than from ICDC AoSI over the period 1980 to 2020 (Fig. a). The total area of multi-year sea ice based on ICDC AoSI amounts on average to 2.5×106km2 in the 2020s, which is consistent with the range (0.5–2.9) ×106 km² given by for the time period 2009–2019.

We conducted two model experiments that differ only in the applied MYSIC (ERA5, ICDC AoSI) and found that the lower the MYSIC in a grid cell the more Br2 and the less BrCl emissions from ice and snow are simulated (Fig. b). The opposite is true for grid cells with larger MYSIC. This means that strong local sources of Br2 due to small-scale sea ice inhomogeneities may appear smeared out in our applied model resolution.

For our model experiments, we computed BrO VCD from χBrO and used a dynamical tropopause metric based on the model potential vorticity (PV) at 3.5PVU (1PVU=10-6 m² s⁻¹ K kg⁻¹) to separate tropospheric and stratospheric columns. We sampled the modeled total and tropospheric VCD at 13–14 h local time and calculated monthly averages to compare with the satellite retrieval. The mean modeled stratospheric contribution to the total BrO VCD in April was rather uniform in all experiments and amounts to (2.5–3) ×1013molec.cm-2 (Fig. ).

The TROPOMI BrO VCD for April 2019 (Fig. a, b) indicates hotspot regions over the Canadian archipelago and the Kara/Laptev Sea. This pattern is well reproduced in experiments with ICDC AoSI-derived MYSIC (Fig. c, d). The ERA5-derived MYSIC shows an additional hotspot at the Alaskan coast (Fig. a, b) which is absent in the satellite VCD. Therefore, we conclude that MYSIC derived from ICDC AoSI is the better choice.

The threshold temperature Tcrit constrains the occurrences of BEs in time and space during the spring. noted that there is no Tcrit that optimizes modeled ozone at all monitoring sites simultaneously. We propose that relaxing this temperature threshold might improve the agreement without implementing a detailed snow microphysics and snow chemistry scheme in a global CCM. For this purpose, we implemented a sigmoidal temperature dependency of the form 3fcrit(T)=11+eT-Tcrit, with T the surface temperature in a given grid cell. The Br2 and BrCl emission fluxes are then scaled with fcrit(T) and reach 10 % and 90 % of the maximum flux at Tcrit±2.2 °C, respectively. Br2 emissions thus already increase where temperatures are slightly above Tcrit, most notably in coastal regions.

We then looked at model results at Tcrit∈{-15,-10,-2.5}°C. Tcrit=-2.5°C means allowing bromine emission also at temperatures around the freezing point of freshwater. With a higher threshold temperature, the modeled Br2 fluxes consistently show an increase over first-year sea ice (Fig. d, g). Emissions from multi-year sea ice regions, indicated by the BrCl fluxes (Fig. c, h), also display an increase but remain two orders of magnitude lower than emissions from first-year sea ice regardless of the choice of Tcrit. At Tcrit=-2.5°C, the model predicts Br2 emissions in the Gulf of Bothnia, the White Sea, and the Sea of Okhotsk (Fig. g), though this does not result in notably enhanced BrO VCD (Fig. i).

At Tcrit=-2.5°C the BrO VCD in the hot spot regions is closest to the TROPOMI VCD (Fig. b), though more enhanced BrO VCDs are found at the coast east of Greenland and East Siberia than observed. As indicated above, this could be a relict of the relatively low model resolution and associated land-sea mask that does not resolve all topographic features and congruent sub-grid temperature variance. Possibly, the stratospheric BrO climatology and simple AMF used in the TROPOMI retrieval also plays a role for the absolute values. Overall, our modeled BrO VCD in 2019 agrees best with satellite observations (Fig. b) in both amount and spatial pattern of tropospheric BrO VCD at Tcrit=-2.5°C.

To decide on the best model setup, we evaluated the model skill in terms of the coefficient of determination of the linear regression (squared Pearson correlation coefficient, R2) and root-mean-square-error (RMSE) at the Arctic ozone monitoring sites (Sect. ) for spring 2019.

The resulting R2 and RMSE for 2019 at Utqiaġvik, Eureka, Summit, and Zeppelin are listed in Table . The closer R2 is to 1, the better the correlation. The closer the RMSE is to 0, the smaller the difference between the observed and modeled time series. For the corresponding histograms with fitted linear regression curves, see Fig. .

We confirmed that there is not a single combination of parameters that optimizes both R2 and RMSE for all sites. The model experiment with MYSIC from ICDC AoSI and Tcrit=-15°C (sfa010) performs best with respect to R2 (〈R2〉>=0.198). The experiments with Tcrit=-10°C (sfa011, sfa017) are best in terms of RSME (RMSE=0.365). The treatment of freshwater lakes like land snow decreases the model skill slightly compared to the experiments where these are excluded (compare (sfa013, sfa017) with (sfa010, sfa011)). We will need to investigate this further. An experiment (sfa012) with Tcrit=-2.5°C performs well at Zeppelin, but displayed lower modeled χO3 than observed at Eureka and Villum research station Nord in May and June (Fig. ). At higher spatial resolution (15×15km), have shown that, when using an ODE terminator defined by snowmelt and additional bromine emission from multi-year sea ice, observed and modeled O3 VMR at Zeppelin agree well with Tcrit=-2.5°C, but was underestimated at Villum research station Nord. This suggests that BEs near Zeppelin may persist at relatively high temperatures as compared to other ice-covered regions for unidentified reasons.

Based on these quantitative results, we decided on the setup with MYSIC derived from ICDC AoSI, Tcrit=-10°C, and Br2 emissions from sea salt enabled (sfa017). As stated above, tropospheric BrO VCD showed a qualitatively better agreement with Tcrit=-2.5°C.

In Sect. , we found that modeled surface temperature and BL agree reasonably well with observations during the MOSAiC expedition. In the following, we will compare modeled χO3 and χBrO with in situ observations during the MOSAiC expedition for the leg March–May 2020.

We interpolated χO3 and χBrO for the model experiments ref, sfa002, and sfa017 onto the MOSAiC drift track and compared these with observations (Fig. ). We found an ozone depletion of up to 20ppbv with our best setup (sfa017), though the observed ODEs (χO3≤5ppbv) in the Central Arctic are not reproduced in any of our model experiments. During the modeled partial ODEs in April, much more surface BrO is produced than observed, while tropospheric BrO VCD remains lower than observed (compare Fig. b, d). In May, χBrO is better reproduced, while χO3 remains too high.

The photochemical steady state between Br and BrO requires the presence of O3 to maintain BrO production (Fig. ) . As the ozone depletion on pan-Arctic scale remains incomplete in our model experiments, the enhanced χBrO at the surface is probably a direct consequence.

The pattern of observed and modeled BEs show a considerable year-by-year variability. In 2020, TROPOMI tropospheric BrO VCD displays an extended area of increased BrO from the northern tip of the Canadian archipelago to the Laptev Sea (BrOVCD≥4×1013molec.cm-2). In the model experiment, this region decomposes into three hotspots: the northern tip of Greenland and the Canadian archipelago (BrOVCD= (2.5–4) ×1013molec.cm-2), north of the Beaufort Sea, and the Laptev/Kara Sea, north of Novaya Zemliya, (BrOVCD= (0.5–1.5) ×1013molec.cm-2) respectively. A modeled corridor of lower BrO enhancement coincides with the maximum of MYSIC. This implies a need to reconsider the assumptions regarding bromine recycling on multi-year sea ice. These findings support the conclusions of , indicating that regions covered by multi-year sea ice serve as a more significant source of reactive bromine than previously recognized. The modeled BrO VCDs are up to a factor of four lower than observed. As shown in Sect. , these scale with the amount of Br2 emitted.

The monthly maximum of the difference in surface ozone (Fig. a) displays major hotspots (ΔO3max= (30–40) ppbv) in the Canadian archipelago/Baffin Bay/north of Greenland and the Laptev Sea/Eastern Arctic Ocean. These hotspots of ozone depletion colocate partly with regions of high BrO VCD (Fig. d). Ozone depletion on a pan-Arctic scale appears relatively weak on a monthly average (Fig. b) and never exceeds 17ppbv in April compared to an ozone background of (29±8)ppbv. Hence, modeled surface ozone is only depleted by 60±3 % on average. This strongly suggests that ozone depletion is too weak not only along the MOSAiC drift track (Sect. ) but also on larger scales. The hotspots of maximum depletion are located in the Canadian archipelago and the Central Arctic (North of the Aleutian Islands). In March and May, the major hotspot region is again located in the Canadian archipelago, while the secondary hotspots vary both in strength and location (Fig. ).

In the mid-latitudes, an average ozone reduction of 0–5 ppbv compared to the reference simulation is predicted. This average ozone reduction could be due to an advection of ozone depleted as well as bromine enriched air masses from the Arctic. In this regard, we found a pronounced reduction maximum of ozone of up to 25ppbv in the North Atlantic between Iceland and the Faroe Islands. This signal is colocated with prescribed shipping emissions of CO, NO_x, NH3, and SO2 in that area and indicates that chemical ozone depletion could be locally enhanced by advected bromine from the Arctic .