Two chemical transport models were used in this study, DEHM (the Danish Eulerian Hemispheric Model) and GEM-MACH (Global Environmental Multiscale model – Modelling Air quality and Chemistry). Brief descriptions of the two models and their setup for the year 2015 simulations are provided in this section. Key model features and configurations are summarized in Appendix A (Table A1). The year 2015 was selected on the basis that it was one of the years featured in the recent AMAP (Arctic Monitoring and Assessment Program) assessment of short-lived climate forcers (AMAP, 2021) and a reference year for the ECLIPSE (Evaluating the Climate and Air Quality Impacts of Short-Lived Pollutants) v6b emission dataset, which was used by all the models that participated in the AMAP assessment (Whaley et al., 2022) as well as by the two models in this study.

Ozone observations from multiple platforms were used for comparison with model simulations in this study, including surface O₃ observations from 8 Arctic ground sites, 7 buoys, and a research vessel over the Arctic Ocean, as well as O₃ vertical profile observations from ozonesondes and research aircraft. In addition, observations of bromine monoxide (BrO) vertical column density (VCD) obtained from multiple axis differential optical absorption spectroscopy (MAX-DOAS) measurements were also used to compare with model results. Table 1 lists all the sites and observational data used in this study.

Arctic lower-tropospheric O₃ is influenced by transport from lower latitudes, photochemical production from anthropogenic and biogenic ozone precursors of both local and distant origins and atmospheric removal processes (such as dry deposition and (photo-)chemical loss through reactions with biogenic VOCs and surface sourced reactive halogens), as well as stratospheric–tropospheric exchange. All of these sources and processes, which are represented in the models in this study at varying degrees of complexity (see Sect. 2 above), vary seasonally, which gives rise to the seasonal variations of Arctic O₃. Figure 2 shows the model-simulated monthly mean O₃ concentrations over the Arctic for January, April, July, and October (representative of each of the four seasons) at three model levels, the lowest (surface level), near 900 hPa, and near 700 hPa (GEM-MACH simulation shown in Fig. 2a and DEHM simulation in Fig. 2b). The GEM-MACH-model-simulated O₃ over the Arctic shows distinctively different seasonal patterns near the surface and aloft and between the central Arctic Ocean and subarctic regions. Over the central and western Arctic Ocean (Eurasian and North American side) close to the surface, this model computes the lowest O₃ in spring as a result of the O₃ depletion events (ODEs) from the prevalence of bromine explosions during this period, in broad agreement with an earlier report of a full-year of surface ozone measurements over the central Arctic Ocean (Bottenheim et al., 2009). The highest ozone from the GEM-MACH simulation is found in fall (October). In contrast, at higher altitudes, O₃ is highest in springtime. The same is also true for the inland subarctic regions. The springtime ozone maximum is thought to be driven by transport from the stratosphere, since intrusion events are more frequent during this season, and by photochemical production from the NO_x released from thermal decomposition of PAN (Walker et al., 2012). The model-simulated O₃ over subarctic boreal regions also displays a spring maximum. The model-simulated low O₃ over summer in these regions can be attributed to both the loss through O₃ reactions with biogenic VOCs (e.g., isoprene) under low-NO_x conditions and enhanced dry deposition. The DEHM-simulated O₃ over the Arctic does not show a clear springtime minimum at the lowest model level. The model simulation shows a general spring maximum over the Arctic throughout the lower troposphere, except for over the very centre of the Arctic Ocean (> 80° N) where the modelled (April) monthly mean O₃ concentration is slightly lower than surrounding areas at the lowest model level. The DEHM-simulated monthly mean O₃ for July shows clear enhancement at elevated levels (particularly at the near 900 hPa level) over northern Alaska and Chukchi Sea, extending into central Arctic Ocean, which is likely contributed by boreal wildfires (see discussions later in Sect. 4.2). Except near the surface and during spring, the two models are quite consistent with each other in simulating O₃ over the Arctic particularly during winter (January) and fall (October). The two models also behaved similarly in simulating O₃ at higher altitude (e.g., near the 700 mb level). Both models simulated low surface O₃ concentrations over northern Eurasia and northern Europe during winter. The low ozone can be argued to be attributable to reduced photo-chemical production and enhanced titration by NO emissions from local sources within the darker and shallower boundary layer during winter, as well as dry deposition. Both model simulations also show low O₃ over subarctic boreal regions in summer, but the low O₃ simulated in GEM-MACH extends to a deeper layer compared to the DEHM simulation. On the other hand, the DEHM-simulated surface O₃ concentrations over the Arctic Ocean during summer are higher than those in the GEM-MACH simulation, which is also the case at higher altitudes (i.e., near the 900 and 700 mb levels).

Figure 3 shows the spatial distributions of the times when the annual maximum and minimum monthly mean O₃ concentrations occur at the three model levels seen in Fig. 2 (left panels from the GEM-MACH simulation; right panels from the DEHM simulation). At the 700 hPa level, the two models are consistent with each other in showing that the annual O₃ maximum occurs in spring months (April and May) over the Arctic, while the annual O₃ minimum occurs in winter (December and January) and late fall (November), with the exception over the Beaufort Sea and the Canadian Northwest Territories where the GEM-MACH-simulated annual O₃ minimum occurs in late summer months (July and August). Near the surface, the two models differ over the Arctic Ocean stemming from the model's differing ability to simulate the springtime ODEs which are prevalent over the Arctic Ocean sea ice (Bottenheim et al., 2009). The GEM-MACH simulation shows annual minimum monthly O₃ in spring months (April and May), due to modelled strong ODEs (see discussion later in Sects. 3.2 and 4.1), and maximum in fall (October), while DEHM simulates annual maximum monthly O₃ in spring over the Arctic (much like the upper levels) due to considerably fewer ODEs simulated by the model (see Sects. 3.2 and 4.1). It is evident that the springtime O₃ depletion process plays a central role in driving the O₃ seasonal cycle at low altitude levels over the high Arctic in the GEM-MACH simulation. Away from the Arctic Ocean and the Canadian Archipelago overland, the two models are again consistent in producing an annual maximum O₃ in spring and minimum O₃ in late summer and early fall over Alaska, Northwest Territories, and the eastern Russian Arctic.

To evaluate the models' ability to simulate Arctic boundary layer O₃, the modelled surface (or lowest model level) O₃ concentrations are compared with observations from ground-based monitoring sites and surface mobile platforms (O-buoys and Mirai cruise). To do this, the modelled O₃ concentrations are extracted at the ground-based sites and following buoy tracks and ship paths from the nearest model grid cells and hours and compared with hourly observations. Existing model evaluations related to tropospheric ozone assessment (e.g., Monks et al., 2015b; Whaley et al., 2023; Young et al., 2018) have been mostly performed on long-term annual and monthly averages. With the two regional models used in this study run at much higher spatial resolutions, as compared to the global models employed in the previous assessment studies, we can examine model simulations and compare with observations at much finer temporal resolutions (e.g., hourly) here.

Figure 4 shows the O₃ time series comparisons at the eight Arctic monitoring sites described in Sect. 2.2.1. Overall, both DEHM and GEM-MACH simulations captured the observed O₃ seasonal as well as synoptic-scale variations at these Arctic ground sites. The three Arctic coastal sites, Utqiaġvik, Villum, and Tiksi, are strongly influenced by the spring ODEs, which are captured reasonably well by the GEM-MACH simulation. DEHM was less successful in capturing the springtime ODEs at these sites. The modelling of ODEs will be examined in more detail later in Sect. 4.1. The seasonal variation in the observed O₃ at the subarctic inland sites (Tustervatn, Pallas, and Esrange) follows the typical pattern of a maximum in spring and minimum in summer, with greater variability in summer and fall. The model simulations from both DEHM and GEM-MACH follow the observed O₃ variations closely throughout the year. The GEM-MACH simulation shows a larger low bias at the two northern European boreal sites (Pallas and Esrange) particularly during the spring and summer seasons, while the DEHM performed better (particularly at Esrange); this will be discussed further in the statistical model evaluation below.

The two high-elevation sites (Zeppelin and Summit) exhibit somewhat different O₃ seasonal patterns. The Zeppelin site, situated at 474 m above the Arctic Ocean, is situated approximately half of the time above the top of the atmospheric boundary layer (Dekhtyareva et al., 2018). The observed O₃ time series in 2015 displays an overall maximum in April and a minimum in July, in contrast to the Arctic coastal sites. This is consistent with the seasonal patterns based on longer-term (multi-year) observations (e.g., Whaley et al., 2023). However, it is evident from the time series in Fig. 4 that the site is sporadically impacted by springtime ODEs during April and May in 2015. Previous observations of ODEs at this site have been reported by others (e.g., Berg et al., 2003; Eneroth et al., 2007; Lehrer et al., 1997; Solberg et al., 1996). The O₃ observation at Summit has a gap between the end of July and the end of October in 2015. The incomplete observed O₃ time series shows no clear trend over the first 5 months (January–May) of 2015 before increasing over June to reach a maximum in July. This is a departure from the seasonal trend shown in Whaley et al. (2023) based on multi-year data (2003–2018), which showed a maximum in May. Both Zeppelin and Summit surface observations display high O₃ events in July 2015. As will be discussed later in Sect. 4.2, there is an indication that these events may be associated with transport of wildfire plumes in the free troposphere. Again, model simulations from both DEHM and GEM-MACH compare well with the observations at these sites, capturing the observed seasonal and synoptic-scale variations (also evident from the statistical evaluation shown in Table 2), though neither model simulation was able to fully capture the July high O₃ events observed at Summit.

Statistical evaluations of model performance were conducted on the hourly time series. Table 2 shows selected seasonal and annual model performance scores at the 8 Arctic ground sites, including normalized mean bias (NMB), Pearson correlation coefficient (r), and unbiased root-mean-square-error (URMSE), while the corresponding monthly scores are shown in the Supplement (Fig. S2). The seasonal scatter plots (colour-coded for each month separately) of model versus observations at the 8 surface sites are shown in Fig. S3. The evaluation (Table 2) shows that both models underpredict wintertime Arctic surface ozone at all sites, with GEM-MACH having a greater negative bias at Utqiaġvik, Villum, Pallas, and Esrange. At coastal sites, DEHM has significant positive bias during the spring months due to its under-representation of the springtime ODEs, while the GEM-MACH model has considerably better performance scores. It is interesting to note the significant positive bias in both models during the summer months at the coastal sites, except for a small negative bias in GEM-MACH at Villum, which is largely driven by the month of June values; see Fig. S3b. Neither DEHM nor GEM-MACH currently includes iodine chemistry, which can play a prominent role in ozone destruction over polar oceans during (as well as after) the time of springtime bromine explosions (Benavent et al., 2022; Fernandez et al., 2024; Mahajan et al., 2010; Raso et al., 2017; Wittrock et al., 2000).

At the two northern European boreal sites, Pallas and Esrange, the models are generally biased low throughout the year. GEM-MACH has the greatest difficulty in simulating surface ozone accurately at these two sites, particularly during summer, as evident by the relatively poor performance scores shown in Table 2 (and Fig. S3) compared to other sites, while DEHM performed considerably better at these sites. This may be partly attributable to the difference in modelled O₃ dry deposition velocities over the boreal land cover between GEM-MACH and DEHM. Clifton et al. (2023) examined O₃ dry deposition velocity formulations across contemporary regional chemical transport models, including the formulations used in GEM-MACH (based on Wesely, 1989) and DEHM (as in Simpson et al., 2012). They showed that the formulation used in GEM-MACH (“GEM-MACH Wesely”) significantly overestimated O₃ dry deposition velocities over the European boreal forest during summer compared to an estimate based on ozone flux measurements. In contrast, the formulation used in DEHM (“DO3SE”) was shown to produce O₃ dry deposition velocities in much closer agreement with those derived from observations over the summertime European boreal forest.

Overall, the two regional models seem to demonstrate better skill in capturing the observed seasonal variations in the Arctic surface ozone, compared to the large-scale global atmospheric chemistry models reported in previous assessments (e.g., Law et al., 2023; Whaley et al., 2023; Young et al., 2018) where the models showed a large spread in simulated surface O₃ concentrations and inability to reproduce the observed seasonal cycles at some of the Arctic sites. Besides the implementation of the processes involved in springtime ODEs in the Arctic, the better performance from the two independent regional models in this study can be attributed, at least in part, to better resolved atmospheric dynamics and boundary layer processes modelled at finer spatial and temporal scales.

The model simulations are also compared with buoy and ship observations in Fig. 5. As described in Sect. 2.2.2, the O₃ observations were available from the six O-buoys (8, 11, 12, 13, 14, and 15) and the Japanese research vessel Mirai for different time periods in 2015 with their tracks over various parts of the Arctic Ocean (Fig. 1). In Fig. 5, the time series of O₃ from different platforms are collated into single plots for observations and two model results, respectively, to illustrate that the composite O₃ seasonal patterns shown in the observations over the Arctic Ocean are virtually consistent with those observed at the Arctic coastal sites; i.e., the spring period is dominated by ODEs followed by a brief rebound before decreasing to its summer minimum and then recovery in the fall. Like the observations, the modelled O₃ time series along the buoys and ship tracks are also consistent with those modelled at the Arctic coastal site (Utqiaġvik shown, as an example, in Fig. 5). The similarity between O₃ observations over the Arctic Ocean and the coastal sites was also found in other studies (e.g., He et al., 2016; Sommar et al., 2010; Bottenheim et al., 2009) with the exception of springtime. The model–observation comparisons for individual buoys and ships, including time series, scatter plots, and statistical scores (i.e., normalized mean bias, NMB; Pearson correlation coefficient, r; and unbiased root-mean-square error, URMSE), are provided in the Supplement (Fig. S4). The two models generally track the buoys and ship observations well, particularly for the latter half of the year. The GEM-MACH model was able to simulate the observed ODEs (O-buoy 11 and 12) during spring. Outside the spring period, the two models exhibit a similar performance in simulating surface O₃ over the Arctic Ocean compared against observations on the buoys (O-buoy 8, 13, 14, and 15) and the ship (R/V Mirai), as indicated in the statistical evaluation (Fig. S4). Similar to the comparisons at the coastal sites in Fig. 4, the model-simulated surface O₃ is biased low over the winter season along the buoy tracks (e.g., O-buoy 11 and 12 over January and February, O-buoy 8, 13, 14, and 15 over November and December; Fig. S4). It is notable that both models simulated the O₃ observations on the R/V Mirai cruise (September 2015) very well (Fig. S4), which is in contrast to a previously identified challenge in simulating the O₃ observations from the multi-year (2013–2018) Mirai cruises in the Arctic (all during September) where global models significantly underpredicted the surface O₃ concentrations (Kanaya et al., 2019). Kanaya et al. (2019) suggested that the dry deposition of O₃ over the ocean may be overrepresented in their model (a dry deposition velocity of ∼ 0.04 cm s⁻¹ over open ocean was used in their case), which may be responsible for the model under-prediction of O₃. As mentioned earlier, GEM-MACH in this study uses a parameterization representing iodide-mediated O₃ deposition over the open ocean (Sarwar et al., 2015) for the Arctic simulation, which can result in a dry deposition velocity smaller than the original GEM-MACH's fixed value of 0.03 cm s⁻¹ over high-latitude open oceans, while the O₃ dry deposition velocity of ∼ 0.05 cm s⁻¹ over open ocean is used in DEHM (see Appendix 1). This suggests that the model representation of O₃ dry deposition may only be partially responsible for the global model underprediction of O₃ over the Arctic Ocean in the earlier study.

To evaluate the models' abilities to simulate the vertical distribution of O₃ over the Arctic, the modelled vertical O₃ profiles at the Arctic ozonesonde sites (see Sect. 2.2.3) are compared with the ozonesonde observations. For the comparison, both modelled and observed (ozonesondes) profiles were interpolated at 10 m resolution and binned to 100 m intervals. The vertical profiles of model data were extracted over the grid cells nearest to the ozonesonde launching sites and at the hours closest to the launch times. We focus on the lowest 5 km a.s.l. altitude range in this study. Figure 6 includes the seasonal comparisons at the six Arctic sites: Alert, Eureka, Resolute, Ny Ålesund, Scoresbysund, and Sodankylä (observations in black, GEM-MACH in red, and DEHM in purple). For the lowest 5 km, the model simulations and observations are in overall good agreement. The spring (MAM) ozonesonde profiles at Alert, Eureka, and Resolute over the Canadian Archipelago are strongly influenced by the ODEs below 1 to 1.5 km a.s.l. The GEM-MACH model was more successful in capturing the ODEs at these sites, though the modelled ODEs were not as strong as the observations close to the surface. The vertical depths of the ODEs, mostly limited to the lowest 1 km, were simulated well. The DEHM simulation did not capture the observed ozone depletion close to the surface. However, above the boundary layer (∼ 1.5 km), the modelled O₃ profiles from the two models do agree well and are in good agreement with observations. The model-simulated ozone profiles (from both models) are biased low compared to the ozonesonde measurements over the winter months (DJF) at most of the sites, consistent with the model low bias shown at the surface sites. In the case of GEM-MACH, the overall model low bias in winter could, at least in part, be attributable to the chemical lateral boundary condition from the ECMWF-CAMS reanalysis. Both Inness et al. (2019) and Wagner et al. (2021) have found that the CAMS reanalysis (for the period of 2003 to 2018) tends to have a negative bias in surface and tropospheric ozone over the winter season at high latitudes, particularly after 2012/2013, which was linked to a switch in data assimilation procedure. At the Sodankylä site, located in the European boreal region (in close proximity with two of the surface observation sites, Pallas and Esrange), the GEM-MACH-simulated ozone has a significant negative bias throughout the lowest 5 km during summer (JJA). The DEHM simulation also shows a similar negative bias above 1.5 km but recovers in the lowest 1.5 km layer, where the modelled O₃ concentrations are much closer to those observed. The modelled ozone profiles at Ny Ålesund and Scoresbysund also show similar negative biases at altitudes above 2–3 km during JJA months. This may be indicative of insufficient transport in the free troposphere in both models, but the GEM-MACH model's underprediction of ozone close to the surface at the Sodankylä site could be attributed to the model's over-representation of the O₃ dry deposition over the European boreal region, as discussed earlier in Sect. 3.2, and possibly to an over-predicted emissions (and hence concentrations) of biogenic olefins such as isoprene reacting rapidly with O3 (e.g., Gong et al., 2022).

Whaley et al. (2023) compared model-simulated vertical profiles (using monthly mean model output) from 12 different large-scale models to the ozonesonde measurements at the same group of sites as we examined here (see their Figs. 8 and S1). We have plotted the profiles of seasonal relative difference between model simulations and observations (or NMB) in Fig. S5, which can be compared with the results shown in Whaley et al. (2023). Again, the two regional models here show better skills in simulating the observed O₃ vertical profiles over the lowest 5 km of the atmosphere examined (having considerably smaller biases, generally well within ± 25 %, compared to the large spread of relative difference, ± 50 %, in the same altitude range amongst the large-scale global models).

Monthly statistical evaluations for three altitude ranges, 0–1, 1–2, and 2–5 km, are presented in Fig. 7, comparing monthly mean, maximum, minimum, and interquartile range between model and ozonesondes at the six Arctic ozonesonde sites. Note that there were no ozonesonde launches in January and December 2015 at Alert and in June 2015 at Resolute and Eureka. Here, again, the distinctively different ozone seasonal patterns between the lowest altitude range (0–1 km) and the higher altitude range in the free troposphere (2–5 km) are evident at all three ozonesonde sites in the Canadian archipelago (Alert, Eureka, and Resolute). The springtime ozone minimum, occurring in May at Alert and in April at Resolute and Eureka, is prominently seen in the lowest 1 km range, driven by the ODEs. The influence of ODEs can be seen in the 1–2 km altitude range also at these sites. In contrast, ozone in the 2–5 km altitude range exhibits a maximum in late spring (in the month of May) at all sites. The ozonesonde observations in the lowest 1 km altitude range also indicate a maximum in October at the three ozonesonde sites over the Canadian Archipelago, consistent with the GEM-MACH model results shown in Figs. 2 and 3. It is also interesting to notice that the usual summer O₃ minimum observed at the surface sites (see Fig. 4) is evident at lower altitudes (below 2 km) but less evident in higher altitudes (e.g., 2–5 km) from the ozonesonde observations at these Arctic sites. The statistical evaluation shows generally good agreement between the models and the ozonesonde observations for the three selected altitude ranges at most of the sites. Larger discrepancies between the GEM-MACH model and observations are seen in June and July at the Sodankylä site, consistent with the model's underprediction of summertime O₃ at the surface sites in the European boreal region (as discussed above). Again, overall, the two models are seen to have good skills in reproducing the observed O₃ vertical distribution and seasonal cycles over the Arctic (except for the coastal sites where DEHM was unable to reproduce the observed O₃ influenced by ODEs in spring).

As shown from the observations and model results presented in Sect. 3, the springtime ODEs play an important role in driving the Arctic surface O₃ seasonal cycles. The main uncertainty in modelling the springtime ODEs is in quantifying the sources for reactive bromine in the Arctic boundary layer. As described in Sect. 2.1.3, the two models included in this study, DEHM and GEM-MACH, consider different sources of reactive bromine: GEM-MACH adopted a representation of a snowpack bromine source mechanism following Toyota et al. (2011), while DEHM implemented a representation of sea-salt-aerosol-sourced bromine from blowing-snow and open-ocean sea spray following Yang et al. (2010).

In Figs. 8 to 11, we examine model simulations of ODEs at the 4 coastal sites (Utqiaġvik, Villum, Tiksi, and Zeppelin) in more detail; these are the only Arctic coastal sites under the strong influence of ODEs with surface O₃ data available for 2015. Included in Figs. 8 to 11 are the time series of observed and modelled (DEHM and GEM-MACH) surface O₃ for March, April, and May. Along with the O₃ time series are the modelled O₃ deficit (or depletion) due to bromine chemistry (computed from the difference between the modelled surface O₃ concentration with and without the snow-sourced bromine

In the case of GEM-MACH, a sensitivity run was conducted with the snowpack bromine flux turned off, which effectively turned off the bromine chemistry in the simulation. In the case of DEHM, a sensitivity run was conducted by turning off the blowing-snow-sourced bromine, while the bromine sourced from open-ocean sea spray remained active.

While there are relatively abundant surface observations of the Arctic springtime ODEs from the ground-based monitoring sites and mobile platforms (e.g., buoys and research vessels) in the Arctic Ocean (Bottenheim et al., 2009), observations on the vertical structure of ODEs are relatively scarce. Using a differential absorption lidar, Seabrook et al. (2011) observed the vertical structure of springtime ODEs over the Arctic Ocean off the south coast of Banks Island. They found that the observed ODEs were largely confined within the lowest 200–600 m of the atmosphere and were associated with air masses being in contact with sea ice for an extended period of time. Oltmans et al. (2012) analyzed the vertical profiles from the near-daily ozonesonde measurements conducted during 2008 and 2009 spring periods at Barrow (Utqiaġvik) and found that the depletion was confined to approximately the lowest 1000 m, with an average height of the top of the layer at ∼ 500 m.

During the 2015 NETCARE spring field campaign, O₃ measurements were made on board the Alfred Wegener Institute Polar-6 aircraft. Figure 12 shows the ozone vertical profiles taken by the aircraft during the 2015 NETCARE field campaign over the Canadian archipelago (around Ellesmere Island over an ice-covered sea surface) along with the modelled profiles (from GEM-MACH and DEHM) extracted at the flight profiling location and time. Also included are the modelled profiles from the runs with the snow-sourced bromine emissions turned off. The segments of the flight tracks during profiling are shown in the inserted map. A shallow ozone depletion layer, with depth ranging between about 500 m and about 1 km can be seen from the profiles taken over the Arctic Ocean off the west side of Ellesmere Island (2015-04-07, 2015-04-08_2, 2015-04-11_2, and 2015-04-13_2). The profiles taken over the Nares Strait (2015-04-08_1 and 2015-04-09) and over Ellesmere Island (2015-04-10 and 2015-04-11_1) all show a deeper layer, ∼ 2 km, of depleted O₃, likely due to transport and vertical mixing of the near-surface bromine-mediated O₃ depletion. In particular, over the interior of Ellesmere Island, a much deeper layer, up to 4 km, can be impacted by the ODEs due to enhanced mixing (comparing between the model-simulated O₃ profiles with and without bromine corresponding to the flight on 10 April). As shown, the GEM-MACH simulation with snowpack-sourced bromine was able to simulate the vertical structure of the depletion layer reasonably well. There are cases where the model was not able to fully simulate the observed depletion close to the surface (e.g., 2015-04-07, 2015-04-08_2, and 2015-04-10), which may be attributable, at least in part, to model resolution (15 km) and the very shallow mixing height of the Arctic atmosphere (e.g., Gryning et al., 2023). Brockway et al. (2024) describe that BrO (and thus reactive bromine that depleted O₃) over the Alaska North Slope region and over the Beaufort Sea snow-covered sea ice occurred in a shallow, very stable boundary layer up to just a few hundred metres. Occasionally they observed some lofted bromine, but mostly that was below 300 m. The DEHM simulation with the blowing-snow-sourced bromine was not able to reproduce the observed near-surface depletion, although for several flights (e.g., 2015-04-10, 2015-04-11_1, 2015-04-11_2, and 2015-04-13_2), the DEHM simulations do show some modest ozone loss from the blowing-snow-sourced bromine (comparing the two DEHM runs with and without the blowing-snow bromine). It is interesting to notice that the DEHM-simulated vertical O₃ profiles are in close agreement with the GEM-MACH-simulated O₃ vertical profiles without bromine, and all the modelled profiles are in reasonably good agreement with the observed profiles above the atmospheric boundary layer, within the lowest 5 km.

To evaluate the modelled bromine levels, the modelled bromine monoxide vertical column densities (BrO VCDs) are compared to the MAX-DOAS measurements available at the Utqiaġvik site and on O-buoy 10, 11, and 12 during spring 2015. Figure 13 shows the comparison in terms of monthly statistics while the hourly time series comparisons are shown in the Supplement (Fig. S6). The monthly stats for both measured and modelled BrO VCDs were calculated based on the data entries with available measurement. The difference between the two modelled BrO fields is largely due to the bromine sources considered in each model, i.e., snowpack-sourced bromine (based on Toyota et al., 2011) in GEM-MACH and open-ocean- and blowing-snow-sourced bromine (based on Yang et al., 2010) in DEHM. At the Utqiaġvik site, the monthly BrO VCDs simulated with the snowpack-sourced bromine (GEM-MACH) tracked the MAX-DOAS measurement well over the period when the measurement was available (21 February to 10 June 2015). The modelled monthly BrO VCDs with open-ocean- and blowing-snow-sourced bromine (DEHM) were considerably lower than the measurements for the month of March and April. The MAX-DOAS measurements on O-buoys were available for much shorter periods in 2015, 21–10 June for O-buoy 10 and 11 and 21 April–22 May for O-buoy 12. The GEM-MACH-simulated monthly BrO VCDs with the snowpack-sourced bromine were considerably higher than the measured BrO VCDs on Buoy 10 in April, mostly driven by an event at the beginning of the measurement period (Fig. S6). On the other hand, the DEHM-simulated BrO VCDs with open-ocean- and blowing-snow-sourced bromine were significantly lower than the measurements on the buoys. These findings are consistent with the results from Swanson et al. (2022) where simulations using the GEOS-Chem model were conducted for a 10-month period (March–November) in 2015 with different snow-sourced bromine mechanisms (i.e., snowpack and/or blowing snow). The DEHM-simulated BrO VCDs are comparable to those from the Swanson et al. simulation with the blowing-snow bromine mechanism alone (their “BLOW” run). Their study also showed much higher BrO VCDs obtained from the simulations with the snowpack bromine mechanism alone (their “PACK” and “PHOTOPACK” runs, the latter considering an enhanced bromine molar yield from snowpack upon O₃ deposition under sunlit conditions as in Toyota et al., 2011) compared to that with the blowing-snow mechanism alone. The comparison between the GEM-MACH-simulated BrO VCDs from this study with those from Swanson et al. (2022) snowpack-only simulations varies. For example, the GEM-MACH-simulated BrO VCDs compared well with the MAX-DOAS measurement at Utqiaġvik, while both simulations with snowpack-sourced mechanism (“PACK” and “PHOTOPACK”) from Swanson et al. (2022) produced much higher BrO VCDs than the measurement, particularly from the run with enhanced bromine molar yield for sunlit conditions (“PHOTOPACK”). On the other hand, the GEM-MACH-simulated BrO VCDs at the buoy locations are more comparable to those from the two snowpack runs in Swanson et al. (2022). This is partly due to the parameters selected (e.g., the bromine molar yields; see Sect. 2.1.3) for the snowpack bromine source mechanism in the different studies. Also worth mentioning is the dependency of bromine production on O₃ deposition in the snowpack bromine source mechanism of Toyota et al. (2011). GEM-MACH employs a reduced O₃ dry deposition velocity over ice and snow surfaces, 0.01 cm s⁻¹ (following Helmig et al., 2007), while a much higher O₃ dry deposition velocity over the Arctic sea ice, between 0.02 and 0.1 cm s⁻¹, was used in Swanson et al. (2022). The uncertainty in the parameterization of the snowpack bromine source mechanism is examined next.

The current model representations of bromine source mechanisms are highly parameterized, and there are large uncertainties in some of the parameters employed by these parameterizations due to a lack of constraints by available lab or field experiments. Some of the studies adopting the approach of Toyota et al. (2011) for the snowpack bromine source mechanism have chosen parameters in variation to those recommended by Toyota et al. (2011). For example, Swanson et al. (2022) chose to make no distinction between FY and MY sea ice in treating snowpack Br₂ production. Herrmann et al. (2021) considered an enhancement factor β (≥ 1.0), to account for non-flat surfaces such as ice or snow and frost flowers, in computing fluxes from Br₂ surface production. As mentioned in Sect. 2.1.3, in this study, the molar yields for Br₂ production from snowpacks over FY and MY sea ice upon dry deposition of O₃ (Φ1) were set at 0.15 and 0.075, respectively, under sunlit conditions, and at 0.01 and 0.005, respectively, under dark conditions in the GEM-MACH simulation presented so far. These are larger than the original values used in Toyota et al. (2011). They were chosen to partly compensate for the possible under-representation of the Br₂ production from reactive bromine cycling via aerosol heterogeneous chemistry due to under-predicted Arctic haze aerosols in the model (see Gong et al., 2024). To explore the sensitivity to the Br₂ molar yields associated with O₃ dry deposition on snowpacks (Φ1) and the role of reactive bromine cycling through aerosol heterogeneous chemistry, two additional sensitivity runs with GEM-MACH were conducted for the spring period (February to May, where February was a spin-up period). The parameter settings for various GEM-MACH runs are specified in Table 3.

Figure 14 shows the modelled O₃ time series from the various GEM-MACH simulations compared to observations at three coastal sites that were most impacted by ODEs (Utqiaġvik, Villum, and Tiksi) as well at O-buoy 11 (the only buoy with observations during the entire spring O₃ depletion season); plots for additional sites are included in the Supplement (Fig. S7). The Sens-Phi1 run used the molar yields Φ1 close to the values recommended by Toyota et al. (2011), i.e., 0.075 and 0.001 for FYI, under sunlit and dark conditions, respectively. For MYI the molar yields Φ1 were set at 0.01 and 0.001, respectively for sunlit and dark conditions, as opposed to zero in Toyota et al. (2011). As shown in Fig. 14, the model-simulated ODEs are weaker in this case than those simulated from the base run, most significantly during March, the early stage of the O₃ depletion period (e.g., O-buoy 11, Utqiaġvik, and Tiksi in Fig. 11; O-buoy 12, Alert, and Eureka in Fig. S7). In the Sens-aerosol run, the molar yields (Φ1) were kept the same as in Sens-Phi1, but the aerosol heterogeneous reaction rates were enhanced by doubling the total aerosol surface area (considering the model under-prediction of Arctic haze aerosols, as mentioned above) to illustrate the role of reactive bromine cycling through heterogeneous chemistry on aerosol surfaces. The enhanced aerosol heterogeneous chemistry (via the artificially increased aerosol loading) resulted in generally stronger model-simulated ODEs than those from the Sens-Phi1 run shown in Figs. 14 and S7, with somewhat more significant enhancements in the modelled ODEs mostly during mid-April to mid-May (though at Tiksi, the most significant impact from aerosol heterogeneous chemistry is seen during an extended depletion event in the beginning of April). However, the impact of the enhanced aerosol heterogeneous reaction on surface ODEs seems to be rather limited during the initial stages of the depletion season (March).

The comparative roles of snowpack Br₂ emission and the Br₂ production through aerosol heterogeneous chemistry on ODEs are examined here. Figure 15 compares the modelled monthly averaged daily snowpack Br₂ flux and the daily Br₂ production from aerosol heterogeneous chemistry, both in moles per m² (per day), in the lowest 200, 500, and 1000 m of air from the three GEM-MACH runs (Base, Sens-Phi1, and Sens-aerosol) for March 2015. The same plots for April and May are included in the Supplement (Figs. S8 and S9). For the Base run, the March-averaged daily snowpack Br₂ flux is mostly distributed along the coastlines over FY sea ice. Comparing the Br₂ productions from snowpacks and through heterogeneous chemistry on aerosol surfaces, the former (snowpack production) is greater than the latter (aerosol surface chemistry) over the lowest 200 m of the atmospheric column, while for the increased column extent of over the lowest 500 m and 1 km layers, the latter becomes greater. It is particularly noticeable that the atmospheric Br₂ production through the heterogeneous reaction spreads much more widely over the Arctic compared to the snowpack fluxes of Br₂. With the reduced molar yields associated with O₃ dry deposition (Φ1) in Sens-Phi1, the Br₂ production from the snowpacks is reduced significantly; the production through aerosol heterogeneous reaction is also reduced as a result of reduced bromine oxidation products (HBr, HOBr, and BrONO₂) in the air. The snowpack Br₂ flux is further reduced in the Sens-aerosol run, compared to the Sens-Phi1 run, due to reduced O₃ deposition (resulting from enhanced ODEs), while the production of Br₂ in the atmosphere is increased from the enhanced heterogeneous reaction rate (through the doubling of aerosol surface area). By May, the atmospheric Br₂ production through heterogeneous reactions from the Sens-aerosol run exceeds that from the Base run (see Fig. S9). Figure 16 shows the time series of the pan-Arctic (> 66.5° N) integrated daily snowpack Br₂ production and the Br₂ production through aerosol heterogeneous reactions from the three GEM-MACH runs (top two panels in Fig. 16). The reduction in snowpack production of Br₂ from the lower Φ1 values in Sens-Phi1 is largest at the beginning of March, and the difference between the Sens-Phi1 and Base runs in snowpack Br₂ production reduces gradually over time (particularly after April). In contrast, the increase in the atmospheric production of Br₂ due to enhanced heterogeneous reactions in the Sens-aerosol run (as compared to the Sens-Phi1 run) starts small at the beginning of March but gradually increases with time to exceed the atmospheric production in the Base run by mid-April. This contrast is better illustrated from the bottom panel of Fig. 16, showing the difference in snowpack Br₂ production in response to the change in snowpack bromine yield from O₃ dry deposition (Φ1; Base - Sens-Phi1) and the difference in atmospheric Br₂ production (via aerosol heterogeneous reactions) in response to the change in aerosol surface area (Sens-aerosol - Sens-Phi1). The gradual increase in atmospheric production of Br₂ (via aerosol heterogeneous reactions) over March and April may reflect the gradual increase in photolysis and photochemical reactivity over the central Arctic during this time (polar sunrise).

We examine the pan-Arctic O₃ loss from bromine chemistry and its sensitivity to the snowpack and atmospheric production of Br₂ in Fig. 17. The bromine-induced O₃ loss (negative) is derived by subtracting the net O₃ production in the No-bromine run from those in the three runs with bromine, i.e., Base, Sens-Phi1, and Sens-aerosol runs, respectively. Figure 17a shows that the largest O₃ loss (or O₃ depletion) from bromine explosions happens within the lowest 200 m layer, followed by the 200–500 m layer. The O₃ loss associated with bromine above 1 km contributes insignificantly to Arctic ODEs. Figure 17b further illustrates the comparative impact of snowpack production of Br₂ and the atmospheric production of Br₂ from reactive bromine cycling through heterogeneous reactions on aerosol surfaces. The reduced O₃ loss (or increase in O₃) from the lower molar yields associated with O₃ dry deposition on snowpacks in Sens-Phi1 is also most significant within the lowest 200 m of the air; its impact decreases with height. In contrast, the enhanced heterogeneous chemistry reactions (via doubling the aerosol surface area) in Sens-aerosol only have a relatively modest impact on the O₃ loss in the lower atmosphere and are comparable initially at 0–200 and 200–500 m. The impact increases with time, and, by April, the most significant impact on O₃ loss due to enhanced heterogeneous reactions is found in the 200–500 m layer followed by the 500 m–1 km layer. Overall, the bromine-induced O₃ loss seems to be more sensitive to the snowpack production of Br₂ than its atmospheric production via heterogeneous chemistry on aerosols. It is worth pointing out that the Br₂ produced through the heterogeneous reactions on aerosol surfaces is originally from the surface-sourced Br₂ (in GEM-MACH), which then undergoes gas-phase photochemical processing to form compounds like HBr, HOBr, and BrONO₂ which, in turn, can reform Br₂ through heterogeneous reactions on acidic aerosol surfaces (Fan and Jacob, 1992; Michalowski et al., 2000; Saiz-Lopez and von Glasow, 2012). Hence the production of Br₂ through this reactive bromine cycling process and its subsequent impact on ODEs will ultimately depend on the bromine release from the snowpacks (or other sources, e.g., blowing snow and sea spray sea salt) and atmospheric oxidation processes that facilitate the formation of HOBr and BrONO₂. On the other hand, the heterogeneous cycling process allows the atmospheric production of Br₂ to take place at distances far away from the original source locations (snowpacks in this case) through atmospheric transport as seen from Fig. 12 (and Figs. S8 and S9), which is consistent with the findings from the airborne field study of Peterson et al. (2017).

To investigate the impact of northern boreal wildfire emissions on tropospheric ozone in the Arctic, the GEM-MACH base case simulation was repeated with the wildfire emissions turned off within its pan-Arctic limited-area grid. Figure 18 compares the model-simulated July mean ozone concentrations over the Arctic, with and without the wildfire emissions, at three model levels. The impact of wildfires is expected to have a large inter-annual variability due to the differences in characteristics of fire seasons and meteorological conditions each year (e.g., Magnussen and Taylor, 2012). In 2015 the Arctic was mostly impacted by the wildfires in Alaska and northern Canada. Particularly, Alaska had a historically high number of fire events and acreage burnt for that fire season, with most of the fire activity concentrated in the late June to July period (Alaska Interagency Coordination Center, 2016). The model simulations show that the northern boreal wildfire emissions had a modest impact on tropospheric ozone concentration in 2015, most significantly in July. The monthly mean O₃ concentrations over the central Arctic are enhanced by 1–2 ppbv at the surface due to northern boreal wildfires, while the enhancement is higher at elevations (e.g., ∼ 900 and 850 hPa levels) by 3–4 ppbv, representing a 5 %–10 % increase at the surface level and up to 10 %–20 % increase at the elevated levels. However, it is worth noting that the DEHM simulation showed more elevated O₃ levels in the same area over northern Alaska extending into the Chukchi Sea and further into the central Arctic Ocean (See Fig. 2b for July at ∼ 900 and ∼ 700 hPa levels). This is consistent with the area impacted by the wildfires in Alaska.

Also shown in Fig. 18 is the excess (or enhancement) ratio ΔO3/ΔCO, defined as the excess O₃ mixing ratio due to a particular source (wildfire, in this case) to the increased CO from the same source, which is often used to characterize ozone production in smoke plumes (Jaffe and Wigder, 2012). Here ΔO₃ and ΔCO were evaluated from the modelled O₃ and CO concentrations with and without the wildfire emissions. A similar approach was used in Pfister et al. (2006) and Thomas et al. (2013). As expected, ΔO3/ΔCO values are small, ∼ 0.02 ppbv ppbv⁻¹ (surface) and ∼ 0.04 ppbv ppbv⁻¹ (elevated levels), over the fire regions in Alaska and the Canadian Northwest Territories, due to limited excess O₃ from photochemical production and large excess CO from fire emissions in fresh plumes. The ΔO3/ΔCO values are considerably larger over the central Arctic, ∼ 0.1 ppbv ppbv⁻¹ (surface) and ∼ 0.14 ppbv ppbv⁻¹ (elevated levels), due to much lower ΔCO resulting from dilution during long-range transport, as well as continued O₃ production in ageing plumes. The higher O₃ excess ratio at elevated levels compared to the surface (lowest model) level is consistent with the higher O₃ enhancement found at elevated levels in the Arctic due to the northern boreal wildfires. These regional enhancement ratio values may be compared with the wide range of ΔO3/ΔCO values reported from existing studies for high-latitude boreal biomass burning plumes. For example, Jaffe and Wigder (2012) provided a summary of ΔO3/ΔCO estimated from observations by biome and plume age. For boreal and temperate regions, they reported ΔO3/ΔCO values ranging between 0.005 and 0.08 (average of 0.018 ppbv ppbv⁻¹) in fresh plumes (≤ 1–2 d), between 0.11 and 0.18 (average of 0.15 ppbv ppbv⁻¹) in plumes of age 2–5 d, and between 0.035 and 0.59 (average of 0.22 ppbv ppbv⁻¹) in older plumes (age > 5 d). Thomas et al. (2013) found mean ΔO3/ΔCO values of 0.08 and 0.49 in fresh and aged biomass burning plumes (from Canadian boreal forest fires), respectively, based on WRF-CHEM model simulations of the ARCTAS-B field campaign. Arnold et al. (2015) also found similar ΔO3/ΔCO values from the POLMIP model simulations, in the range of 0.039–0.196 ppbv ppbv⁻¹ for fresh fire plumes and 0.14–0.261 ppbv ppbv⁻¹ for aged fire plumes. The July monthly ΔO3/ΔCO values found in this study over the North American boreal fire regions, 0.02–0.04 ppbv ppbv⁻¹, are consistent with the range of values found in previous studies for fresh boreal fire plumes, while the values over the central Arctic, 0.1–0.14 ppbv ppbv⁻¹, are within the range, albeit towards the lower end, of the previously reported values for aged boreal fire plumes. The large variability in estimated wildfire-impacted ΔO3/ΔCO enhancement ratios from various studies can arise from the different approaches used in evaluating the enhancement ratios. By comparing between a scatter technique (based on a linear fit to the O₃-CO concentration scatterplot) and an enhancement technique (based on the evaluation of O₃ and CO excess mixing ratios due to wildfire emissions), Pfister et al. (2006) showed that the ΔO3/ΔCO ratios evaluated using the scatter technique were affected by the selection of biomass-burning-impacted air masses and the degree of mixing in the considered air masses. Much higher enhancement ratios were found in anthropogenic-combustion-impacted air masses than in the boreal-wildfire-impacted air masses, due to the difference in NO_x / CO emissions ratios between these source types. Pfister et al. (2006) also showed that when the variability in the background concentration levels was well characterized (which is not a trivial task for the analysis of observational data while being quite straightforward for the analysis of model results through sensitivity runs like ours), the enhancement technique would be more robust and accurate in evaluating the fire-influenced ΔO3/ΔCO enhancement ratios.

Emissions from biomass burning can also lead to large-scale enhancement in high-latitude NO_y (e.g., Arnold et al., 2015). Figure 19 shows the enhancement ratios (July monthly mean), ΔNOy/ΔCO and ΔPAN / ΔCO, evaluated from the GEM-MACH simulations at three model levels (lowest and levels nearest to pressure levels of 900 and 700 hPa). At the lowest model level, higher ΔNOy/ΔCO values are found over the fire regions, while much lower ΔNOy/ΔCO values are found over the central Arctic due to the efficient removal of NO_y species due to dry deposition. Higher NO_y enhancement ratios over the central Arctic are found at elevated levels, highest (∼ 8 pptv ppbv⁻¹) at the model level close to 700 hPa. Note that higher ΔNOy/ΔCO values are found over the Russian fire region compared to the North American fire region, indicating a more efficient NO_y production in Russian fire plumes. This is likely due to the difference in fire emissions (e.g., NO_x emission factors used by the model) between the two regions. As mentioned in Sect. 2.1.2, the GEM-MACH simulation used different data source for wildfire emissions over North America (CEFFPS) and outside North America (FINN v1.5). PAN, a component of NO_y, is of particular interest as it serves as a reservoir for NO_x and can potentially contribute to O₃ formation in the Arctic from its thermal decomposition (Walker et al., 2012). The modelled PAN enhancement ratios (ΔPAN/ΔCO) due to boreal wildfires are simulated to be ∼ 3–4 pptv ppbv⁻¹ over the North American boreal fire regions at the lowest model level, increasing with height to 6–7 pptv ppbv⁻¹ near 700 hPa, comparable with the ΔPAN / ΔCO ratios reported by Arnold et al. (2015) from the group of models driven by the ECMWF meteorological reanalysis. These values are comparable to those deduced from aircraft measurements in boreal fire plumes during the ARCTAS-B campaign (Alvarado et al., 2010). Over the central Arctic, the PAN enhancement ratio has lower values at low altitudes compared to over the fire regions. In contrast, the ΔPAN / ΔCO values are significantly higher at more elevated levels (e.g. 700 hPa), similarly to the case of NO_y. Also included in Fig. 19 is the evaluated PAN-to-NO_y enhancement ratio (ΔPAN / ΔNO_y) from model simulations. As shown, ΔPAN / ΔNO_y ranges from 40 % close to the surface to greater than 70 % at 700 hPa level in the North American boreal fire region and downwind, indicating a significant portion of NO_y produced from the photochemical processing in the boreal fire plumes being in the form of PAN. Over the Arctic, ΔPAN / ΔNO_y ranges from 20 % near the surface to greater than 50 % at higher levels in the lower troposphere. The smaller fraction of PAN at lower levels could be a result of PAN decomposition leading to releasing NO_x and O₃ formation over the Arctic (referring to the increased O₃ enhancement ratio over the Arctic from the source region; see Fig. 18, rightmost column).

Figure 20 shows the modelled O₃ time series at Zeppelin and Summit sites for 2015 summer period, with and without the wildfire emissions. Also included are the corresponding modelled PM_2.5 time series as along with the aerosol absorption measurements available at these two sites. The time series show the main events of northern boreal wildfire plumes affecting the Arctic during July 2015, which are coincident with the high aerosol events indicated by the aerosol absorption measurements. The enhancements in ground-level PM_2.5 from the fires are much more pronounced than in O₃. The enhancement in PM_2.5 is largely driven by primary particulate matters (e.g., primary organic matters, crustal materials) directly emitted from the fires. O₃ is a secondary pollutant, and its formation depends upon the mix of its precursors in the fire plumes and photochemical processing during their transport. The model results indicate that northern boreal wildfires may raise the summertime background O₃ concentrations in the Arctic. However, our model simulations did not fully reproduce the observed episodic peaks in O₃ concentration time series at Zeppelin and Summit during summer 2015, which could be associated with the transport of biomass burning plumes (Fig. 18). This could be an indication of the model underprediction of O₃ production in boreal fire plumes or that the long-range transport from lower latitudes is not being fully captured by the model's lateral boundary conditions. However, there is also a possibility that the measured O₃ may be biased high at Summit under wildfire-influenced conditions due to an instrument's VOC interference issue (Bernays et al., 2022; Long et al., 2021).

Overall, the model simulations suggest that northern boreal wildfires do exert a modest impact on the Arctic tropospheric ozone by influencing the summertime background concentrations. The enhancement of O₃ concentration over the Arctic appears to be greater in the free troposphere than in the boundary layer. Boreal wildfire plumes can often penetrate above the boundary layer where O₃ produced in fire plumes is less subjected to surface removal (dry deposition). Northern boreal wildfires also lead to the enhancement of NO_y in the Arctic. A significant portion of the NO_y in fire plumes is in the form of PAN, particularly at more elevated levels, which can play a role in O₃ production in the Arctic. It should be noted, however, due to the nature of the limited area model (LAM) configuration used in this study, that the model simulations discussed here (with vs. without wildfire emissions) cannot capture the full impact of Eurasian boreal wildfires as most of the Eurasian boreal fires in 2015 were located outside the GEM-MACH LAM domain.

The GEM-MACH simulations incorporated diagnostics for ozone tendencies from each of the processes, (3-D) advection, vertical diffusion (including deposition at the lowest model layer), and chemistry. This was done to help understand how each of the processes influences the O₃ seasonal patterns in the Arctic. Figure 21 includes plots of the monthly averaged O₃ tendencies from each of the process operators in the GEM-MACH 2015 annual simulation for April and July at two model levels, the lowest and near 850 hPa. In April (spring), the O₃ in the Arctic lower atmosphere near the surface is strongly influenced by chemical loss driven by bromine explosions and ODEs, which is compensated by vertical diffusion (primarily) and advection, driven by the strong O₃ gradients (both vertical and horizontal) created by the chemical loss near the surface. In contrast, O₃ in the central Arctic at the elevated level (∼ 850 hPa) is more strongly influenced by advection in spring with chemistry and vertical diffusion playing smaller roles. In July (summer), the net O₃ chemical tendency over the Arctic varies significantly spatially, from negative over large areas in the high Arctic (perhaps driven by loss through reactions mainly with HO₂, e.g Wang et al., 2003) to positive (net production) at more polluted southerly locations, e.g., over northern Europe and northern Eurasia. There is an indication of net photochemical production of O₃ over the shipping channels along the southwestern coast of Greenland and the Canadian Atlantic coast. There is also considerable net O₃ chemical production over the central and northern coast of Alaska extending over the Beaufort Sea. Figure S10 in the Supplement shows the July monthly net O₃ chemical tendency at various model levels from closest to the surface to near 700 hPa from both the GEM-MACH base annual simulation (with wildfires) and the GEM-MACH simulation without the wildfire emissions in the model LAM domain. The impact of boreal wildfires over central Alaska and northern Canada's Northwest Territories on O₃ production is evident. It is particularly interesting to note the potential interaction between the biomass burning plume and anthropogenic emissions of ozone precursors from Alaskan oil fields (Prudhoe Bay). The net O₃ chemical production extends further into the Arctic with the wildfires than without. The O₃ production in wildfire plumes also reaches higher altitudes than those from anthropogenic sources.

The pan-Arctic O₃ budget for each month of 2015 is presented in Fig. 22. It was computed by vertical integration of the daily tendencies through specific depths of the atmospheric columns (from the surface) and then horizontal integration over the area north of 66.5° N (Arctic Circle), given in gigagrams of O₃ per day, and averaged for each month. The budgets for the lowest 200 m above ground level (a.g.l.), 1 km a.g.l., and 4 km a.s.l. are shown. Within the lowest 200 m of air across the Arctic Circle, the O₃ budget is largely balanced between dry deposition (maximum in summer) and vertical diffusion outside the spring ODE season. During the ODE season, the budget is balanced between the combined loss through dry deposition and atmospheric chemistry and the gain from vertical diffusion. Within the lowest 1 km of air, the O₃ budget is largely balanced between the loss from dry deposition (throughout the year) and chemistry (over spring) and O₃ gains from advection (primarily) and vertical diffusion (much reduced compared to over the lowest 200 m). Over a deeper layer (4 km a.s.l.), the O₃ budget is not always balanced, i.e., with non-zero O₃ net gain/loss. The processes contributing to the Arctic O₃ budget over the lowest 4 km (a.s.l.) are dry deposition and chemistry (both contributing to O₃ loss) and advection (contributing to O₃ gain). Also included in Fig. 19 are the O₃ budgets computed from the GEM-MACH no-bromine (NoBr) run (shown for March to June) and the no-wildfire (NoWF) run (shown for July). It is evident that the O₃ chemical loss in the lowest 200m and up to 1 km is almost entirely due to bromine chemistry, with a minimal contribution from non-bromine chemistry (emerging during May–July). The non-bromine chemical loss of O₃ occurs mainly above 1 km a.g.l. and mainly from May to August in terms of timing. The impact of North American boreal wildfire on the Arctic O₃ budget is reflected in the reduced O₃ loss through chemistry (i.e., offset by the O₃ chemical production in wildfire plumes), most noticeable in the budget over the 4 km layer, indicating that most O₃ production from the North American wildfires is happening at higher altitudes.

While the springtime-bromine-explosion-induced O₃ loss mainly occurs within the lowest 1 km of air in the Arctic, it represents a considerable loss in O₃ tropospheric burden over the Arctic. The reductions in monthly mean partial O₃ columns due to snowpack bromine simulated by GEM-MACH are shown in Fig. S11 for the 3 spring months of 2015, by changes in the tropospheric column (surface to 400 hPa) and the lowest 4 km column (surface to 4 km a.s.l.). The modelled snowpack bromine results in up to 15 % reductions in O₃ tropospheric column loading over the central Arctic (up to 30 % reduction in the lowest 4 km O₃ column). These reductions amount to a 5 %–7 % loss of pan-Arctic (> 66.5° N) tropospheric O₃ burden (8 %–12 % loss of the O₃ burden over the lowest 4 km a.s.l. of air).