We use the UM-UKCA CCM that is built around the Met Office Unified Model (MetUM) in the HadGEM3-A configuration (Hewitt et al., 2011) at MetUM version 7.3. The model uses a horizontal resolution of 2.5∘ latitude by 3.75∘ longitude, with 60 vertical levels up to 84 km. We use the extended Chemistry of the Stratosphere (CheS+) chemistry scheme, which is an expansion of Morgenstern et al. (2009) where CFC-11, CFC-12, CFC-113, HCFC-22, halon-1211, halon-1301, CH3Br, CH3Cl, CCl4, CH2Br2, and CHBr3 are considered explicitly, resulting in an additional 17 bimolecular and 9 photolytic reactions. The chemical tracers O3, CH4, N2O, CFC-11, CFC-12, CFC-113, and HCFC-22 are all interactive with the radiation scheme. The model and chemistry scheme were used for the recent SPARC Report on the Lifetimes of Stratospheric Ozone-Depleting Substances, Their Replacements, and Related Species (SPARC, 2013; Chipperfield et al., 2014).

As in Morgenstern et al. (2009), heterogeneous reactions on PSCs as well as formation and removal of nitric acid trihydrate (NAT) PSCs follow Chipperfield (1999), with the formation and removal of ice PSCs included in the hydrological cycle. Regarding the NAT PSCs, as described in Chipperfield (1999), the NAT PSC formation follows the equilibrium expression from Hanson and Mauersberger (1988). We note that this is a simple approach; in reality the formation of NAT particles requires supersaturation of HNO3 over NAT to occur (see, e.g., Solomon et al., 2015). The denitrification scheme assumes a relatively slow NAT sedimentation velocity of ∼ 40 m day-1 for pure NAT PSCs and a much faster NAT sedimentation velocity of ∼ 1540 m day-1 in the presence of ice; the latter assumes coating of NAT PSCs onto ice particles (see also Davies et al., 2002). The reactive uptake coefficients for the heterogeneous chemical reactions on NAT/ICE PSCs used in the model are listed in Table S1 of the Supplement. Stratospheric aerosols are prescribed for heterogeneous reactions, as well as in the photolysis and radiation schemes. Note that, unlike on PSCs, the heterogeneous reactions between ClONO2 and HCl as well as between N2O5 and HCl on sulfate aerosols are not included in the scheme. As heterogeneous reactions on liquid aerosols can be important for the springtime ozone (see, e.g., Solomon et al., 2015), the model heterogeneous chemistry scheme does not represent an exhaustive and fully realistic treatment of heterogeneous processes. However, using a similar model version Keeble et al. (2014) obtained a reasonable representation of the springtime Antarctic ozone hole, with the majority of the simulated Antarctic ozone depletion in the model driven by the heterogeneous reactions on the surfaces of NAT and ice PSCs.

The 11-year solar cycle variability is included consistently in both the radiation and photolysis schemes. In the radiation scheme, we use the method employed in HadGEM1 (Stott et al., 2006) and HadGEM2-ES models (Jones et al., 2011). Total solar irradiance (TSI) data are those recommended in the fifth phase of the Coupled Model Intercomparison Project (CMIP5) (Wang et al., 2005; Lean, 2009), processed to force the mean of the 1700–2004 period to be 1365 W m-2 (Jones et al., 2011). A fit to spectral data from Lean (1995) is used to account for a change in partitioning of solar radiation into wavelength bins. In the Fast-JX photolysis scheme (Telford et al., 2013), the change in partitioning of solar irradiance into wavelength bins is accounted for by scaling the bins according to the CMIP5 spectral solar irradiance data (Wang et al., 2005; Lean, 2009) for the years 1981 and 1986, and the long-term evolution of the processed TSI time series. Solar cycle variability for the future period (after 2009) is included as in earlier periods but with a repeating sinusoidal 11-year cycle with an amplitude derived from observed cycle 23 (see Jones et al., 2011; Gray et al., 2013).

The stratospheric climatology and variability, including that in the Arctic region, have been evaluated in similar stratosphere-resolving versions of the MetUM (Osprey et al., 2010; Hardiman et al., 2010) and UM-UKCA (Morgenstern et al., 2009). The model includes parameterised orographic and non-orographic gravity wave drag (Scaife et al., 2002; Webster et al., 2003), and simulates an internally generated quasi-biennial oscillation (QBO) (Scaife et al., 2002). As shown by Morgenstern et al. (2009), the model reproduces the observed anti-correlation between polar ozone and jet strength in March, which is one indicator of the coupling between chemistry and meteorology discussed above.

A comparison between the Arctic and Antarctic mean (65–90∘ N/S) total ozone columns in a similar model version as used here, but with a specified dynamics set-up, i.e. in which the model's temperature and wind fields are nudged towards meteorological reanalysis data, shows a good correlation between simulated and observed polar column ozone over the recent past (R=0.88 and R=0.96 for the Arctic and Antarctic mean, respectively; see Supplement Fig. S1). In the Antarctic, there is some overestimation of the ozone column in the model by up to ∼ 15–30 DU (see Fig. S1). For the cold March 2011, the model shows a positive bias in the zonal mean monthly mean ozone levels in the Arctic lower stratosphere (30–100 hPa) of up to ∼ 0.9 ppb (∼ 41 %; for the 70–90∘ N mean, 50 hPa) compared with MIPAS satellite data (Fischer et al., 2008, not shown). The positive ozone bias is commensurate with a negative bias in the zonal mean monthly mean ClO levels at this altitude. Note that there is some uncertainty in the observed zonal mean monthly mean ClO, in part due to sparse temporal sampling (twice a day) of the MIPAS instrument. In general, all models will show some biases with respect to observations. Importantly, as in many studies, we compare the model for the present-day and future periods in an internally consistent way; therefore, any biases will become less relevant for our study.

The mean age of air in the tropical lower stratosphere, as well as in the northern high-latitudes, is at the young end of observationally derived values, but its gradient between the northern mid-latitudes and the tropics is in fair agreement with observations and other models (SPARC, 2013; Chipperfield et al., 2014). This indicates a significant improvement in its circulation as compared with the older atmospheric model version presented in Morgenstern et al. (2009) and in SPARC Report No. 5 (SPARC, 2010).

We follow the experimental design of the CCMI REFC2 experiment (Eyring et al., 2013). Lower boundary conditions are used for CH4, N2O, H2, CFC-11, CFC-12, CFC-113, HCFC-22, halon-1211, halon-1301, CH3Br, CH3Cl, CCl4, CH2Br2, CHBr3, and CH3CCl3, with values for ODSs specified from WMO (2011); future GHG abundances are specified according to the representative concentration pathway RCP6.0 (Fujino et al., 2006; Hijioka et al., 2008; Taylor et al., 2012). The prescribed stratospheric aerosols are as given by SPARC (SPARC, 2006), following Chemistry Climate Model Validation 2 (CCMVal-2) recommendations (Eyring et al., 2008). Sea-surface temperatures (SSTs) and sea-ice concentrations (SICs) are taken from one ensemble member (r2i1p1) of the HadGEM2-ES RCP6.0 ensemble (MOHC, 2011; Jones et al., 2011). We perform two integrations from 1960 to 2099 and five over a shorter period from November 1980 to December 2080; the latter five were initiated from different initial conditions taken from a supporting perpetual year 1980 integration. As a result of a data issue, a total of five 6-year-long periods is excluded from the analysis of the ensemble (see Supplement for more details). When compared against the ERA-Interim reanalysis (Dee et al., 2011), the integrations show a somewhat weaker and warmer present-day Arctic stratospheric vortex in early/mid-winter, with a slight zonal wind bias of up to ∼ 6 m s-1 in the mid-latitude lower/mid- stratosphere in March (not shown).

We use two diagnostics to estimate the chemical ozone loss in the Arctic region. Firstly, we estimate the ozone loss due to halogen reactions directly from reaction fluxes. We use the diagnostic framework of Lee et al. (2002) in which the rate of odd oxygen (Ox = O3 + O(3P) + O(1D)) destruction is estimated for different catalytic cycles by determining the rates of their rate-limiting steps. The most important halogen-catalysed ozone loss cycles in the polar lower stratosphere are the ClO dimer cycle (Cycle 1; Molina and Molina, 1987), ClO + BrO cycle (Cycle 2; Yung et al., 1980; McElroy et al., 1986) and the more minor ClO + O(3P) cycle (Cycle 3; Stolarski and Cicerone, 1974): ClO+ClO+M→Cl2O2+MCl2O2+hv→Cl+ClOOClOO+M→Cl+O2+M2⋅(Cl+O3→ClO+O2)(NET:2⋅O3→3⋅O2;Cycle1)‾BrO+ClO→Br+ClOOBrO+ClO→BrCl+O2ClOO+M→Cl+O2+MBrCl+hv→Br+ClBr+O3→BrO+O2Cl+O3→ClO+O2(NET:2⋅O3→3⋅O2;Cycle2)‾ClO+O(3P)→Cl+O2Cl+O3→ClO+O2(NET:O3+O→2⋅O2;Cycle3).‾ The rate-limiting step in each cycle is highlighted in bold font. Three further halogen cycles of lesser importance in the winter polar lower stratosphere (the BrO + O(3P) cycle analogous to Cycle 3, and a pair of ClO + HO2 and BrO + HO2 cycles) are also included. In all six cycles, a net loss of two odd oxygen molecules occurs per cycle. Assuming that [Ox] ≈ [O3] and, thus, d[Ox]/dt ≈ d[O3]/dt, we then calculate the rate of ozone loss from each cycle and integrate in time (from 1 November to 30 March) and altitude (from the surface to 25 km), resulting in an estimate of the cumulative ozone loss at each grid point due to these halogen reactions in the polar lower stratosphere. Whilst the rates (i.e. fluxes) of ozone loss are calculated at each grid point and every UM-UKCA time step, we use the zonal mean monthly mean diagnostics for computational ease (except for vortex-average quantities reported in Sect. 3.3, where daily means are used). The cut-off altitude of 25 km was chosen so as to capture most of the region subject to halogen activation on PSC surfaces. Since no further separation is made between the ozone losses due to heterogeneous and homogenous chemistry, the diagnostic is not equivalent to the ozone losses initiated exclusively by the heterogeneous reactions on PSCs, as presented in, for example, Chipperfield and Jones (1999), Chipperfield (2003) and Keeble et al. (2014).

It is important to understand the characteristics of this diagnostic when integrated over the winter. First, we have calculated losses in an Eulerian framework, and so our values are not directly comparable to the Lagrangian estimates that have been calculated during some polar winters (see Rex et al., 2002, and Harris et al., 2002, for detailed comparison). Second, our values are integrated over 24 h and will necessarily be lower than instantaneous rates calculated, for example, from observed, daytime observations of ClO. Third, unless otherwise specified we present data that have been averaged from 65∘ N to the pole. Unlike in the Antarctic, the Arctic vortex is not usually centred on the pole; it is mobile and highly variable, so our diagnostic will include areas where polar halogen chemistry is not active. However, the strength of our diagnostic is that it allows us to compare different model winters against each other. By definition, greater values of the diagnosed loss will be found in those winters where halogen chemistry is more important.

An additional diagnostic for ozone loss used in Sect. 3.3 is a passive ozone tracer, similar to that implemented in Chipperfield and Jones (1999) and Chipperfield (2003). This chemically inert tracer is initialised to the modelled ozone concentrations on 1 November each year. Whilst undergoing no chemical production or loss, it is transported by the circulation until the end of March. The difference between the chemical ozone field and the passive ozone tracer at the end of each winter represents the change in ozone levels due to all chemical processes. This diagnostic was only included in one ensemble member, discussed in Sect. 3.3.

Figure 1a shows time series of March total column ozone averaged over 65–90∘ N (henceforth referred to as Arctic mean) for the seven UM-UKCA ensemble members. An analogous plot for 65–90∘ S in October is shown in Fig. 1b. The corresponding Arctic mean total inorganic chlorine level (Cly) at 20 km is shown in Fig. 1c.

The results in Fig. 1 are consistent with the multimodel means reported in WMO (2011, 2014) and Eyring et al. (2010). There is a reduction in Arctic total column ozone from the 1960s to the late 1990s, with the March ensemble mean dropping by ∼ 15 % to ∼ 400 DU. This is consistent with the modelled increase in Cly over this period. A similar long-term decline is modelled in the springtime Antarctic, where the ozone reduction is much larger in both absolute and percentage terms (125 DU change in October, i.e. ∼ 38 %). This decline in 20th century polar ozone agrees well with observations (red curve; Bodeker total ozone column dataset, v2.8, Bodeker et al., 2005; Müller et al., 2008).

As a result of the Montreal Protocol and its subsequent amendments and adjustments, the increase in stratospheric Cly abundances has ceased and values are projected to decrease over the 21st century, returning to their 1980 levels by about 2060 in both the Antarctic and Arctic (WMO, 2011). In the Antarctic, the average total ozone returns to 1980 levels over approximately the same period and therefore closely follows the evolution of Cly (not shown). In contrast, total column ozone in the Arctic returns to 1980 levels by the late 2030s, which is approximately 15–20 years earlier than the return date for Cly (Fig. 1c). In the Antarctic, the polar vortex is stronger climatologically and the evolution of ozone is largely determined by changes in halogen chemistry, in agreement with Austin and Wilson (2006). In the Arctic, the return date in UM-UKCA is somewhat later than indicated by the multi-model mean from the CCMVal-2 project (Eyring et al., 2010) but is within the range of individual model estimates (WMO, 2011).

The individual curves in Fig. 1a and b highlight that the interannual variability in springtime total column ozone is much larger in the Arctic than in the Antarctic, with modelled values ranging from 346 to 487 DU in the 1981–2000 period. This large variability remains throughout the 21st century in the presence of the gradual reduction in Cly and the long-term increase in mean Arctic ozone. The ensemble mean ozone increases by about 11.5 (±1.3, i.e. ±2 standard errors) DU decade-1 over the 2000–2080 period, with its 11-year running mean increasing from 408 DU in 2000 to 482 DU in 2074. The minimum monthly mean ozone levels at a single location anywhere poleward of 65∘ N increase at a similar rate to the Arctic mean ozone column, although the 11-year running mean value of this quantity in 2000 is ∼ 50 DU lower than the corresponding Arctic mean value. The persistence of the large interannual variability is associated with years with particularly low Arctic ozone throughout the 21st century. In the second half of the century, March column ozone episodically drops not only below the 1980 level but also to values of ∼ 410–415 DU, which are close to the long-term minimum of the 11-year running ensemble mean around the turn of the century. In addition, one ensemble member simulates 383 DU in March of 2063 (see green point in Fig. 1a), which is comparable to the average values routinely found under present-day conditions.

Figure 2 shows probability density functions (PDFs) of Arctic mean March total ozone for five 20-year intervals from 1981 to 2080. As is also evident in Fig. 1a, the means of the PDFs (coloured diamonds in Fig. 2) progressively increase over the 21st century, consistent with the gradual decline in stratospheric Cly (coloured points in Fig. 2) and the super-recovery of ozone (Eyring et al., 2010). Note that the first two decades of the 21st century have, on average, lower ozone compared to 1981–2000 (∼ 14 DU decrease in mode, Kolmogorov–Smirnov (KS) test p value of 0.052), with a suggestion of a change in skewness of the distribution. This results from the rate of reduction in column ozone at the end of the 20th century being larger than the rate of the subsequent recovery, which begins near the turn of the century.

Notably, the distributions of Arctic March column ozone are negatively skewed, which indicates the occurrence of ozone values well below the 20-year mean. Austin and Wilson (2006) reported an increase in the interannual variability in March Arctic column ozone in the future. Clearly, the precise characteristics of the PDFs in Fig. 2, such as their width and skewness, vary between the five periods shown. Although there could in principle be a forced trend in these characteristics with time (e.g. Rex et al., 2004; Austin and Wilson, 2006), such factors will also be influenced by internal variability and decadal variability in SSTs/SICs and solar forcing, which are common to all of the simulations (see Sect. 2). With the exception of an increase in the mean and mode of ozone over the century, we do not find systematic changes in other characteristics of the PDFs, such as their width, with time.

The occurrence of low-ozone episodes in the future, with values episodically dropping by up to ∼ 50–100 DU below the 11-year running mean, is particularly clear in the PDF for the last two decades modelled (2061–2080). In a single transient integration, Langematz et al. (2014) found that springtime Arctic ozone did not generally drop below the 1970–1982 mean beyond around 2060. In the UM-UKCA simulations, years in which the springtime Arctic ozone drops below the 1980 level (∼ 450 DU; see Fig. 1a) continue until at least 2080. This contrasting result may be due to differences, likely dynamical, in the representation of the Arctic winter stratosphere and its variability between the CCMs used in the studies, differences in experimental set-up (e.g. GHGs, SSTs and SICs), or the fact that we have more fully sampled the internal variability of the climate system by using an ensemble of integrations.

Polar ozone is influenced by a complex interplay between chemical and dynamical processes. Over the 21st century, there may be forced trends in factors that affect both sets of processes, including atmospheric abundances of key chemical species, stratospheric temperatures, and the large-scale circulation; all of these may contribute to the evolution of ozone during this century. The following section discusses the temporal evolution of some of the chemical and dynamical processes that affect Arctic ozone in the UM-UKCA ensemble.

The previous sections have examined the future behaviour of Arctic ozone in the UM-UKCA ensemble, as well as its chemical and dynamical drivers. To further illustrate the potential impact of these processes in individual winters, we now present a case study from the ensemble of an anomalously low total ozone event that occurs many decades into the future. Although the mean ozone column in the Arctic region increases steadily throughout the 21st century (Fig. 1a), individual winters continue to occur in which Arctic mean ozone is up to ∼ 50–100 DU lower than the corresponding long-term mean. The green point in Fig. 1a highlights the model year 2063 in one of the ensemble members. This particular year has an Arctic mean March ozone column of 383 DU, which is lower than the long-term minimum of the ensemble mean found in the late 1990s. We compare this to a model year with near-average total ozone from the same period (2060, marked by orange point in Fig. 1a), whose March Arctic mean ozone column of 489 DU is approximately 100 DU higher than in the model year 2063. These two case study model years are analysed below. Note that the years 2063/2060 are used here as a naming convention, rather than in reference to some specified years in the future.

Figure 8 shows time series of the daily Arctic mean total column ozone during the two case study winters of the model years 2063 (solid red) and 2060 (solid black). Column ozone increases throughout autumn and winter in both years due to transport of relatively ozone-rich air from the tropics by the BDC (Strahan et al., 2013). Note that the Arctic mean encompasses not only the stratospheric polar vortex but also some regions outside of it. For this reason, Fig. 8a also shows the vortex averaged column ozone for the two model years for comparison (dashed lines), defined here by the geographic region where daily mean Ertel's potential vorticity on the 850 K potential temperature surface is greater than or equal to 6×10-4 m2 s-1 K kg-1. These show that similar but somewhat smaller differences between the two years are found when the vortex-average column ozone is considered, compared with the Arctic mean.

The differences in column ozone between the two model years are smaller during late autumn and early winter than in subsequent months; this is in broad agreement with comparisons of similar years in observations (Strahan et al., 2013). Larger differences between the two years occur from late December onwards. In 2063, column ozone increases more slowly between mid-winter and spring, particularly in the Arctic mean. On 1 February, the Arctic mean column is 354 DU in 2063, which is 67 DU lower than in 2060. This is associated with an anomalously strong and cold polar vortex (see Figs. 8b and 9b), and relatively weak stratospheric wave driving in early winter (not shown). The strong polar vortex is associated with reduced downward transport of air inside the vortex. Consistently, the DJF mean 30 hPa w∗‾ is a mere -0.94 mm s-1, which is ∼ 1.06 mm s-1 less negative than the 11-year running mean for that period (Fig. 6).

The differences in column ozone between the two model years are even larger in late winter and spring. In 2063, the polar vortex remains strong until early April (Fig. 9b), and total ozone reaches only ∼ 390–395 DU by the end of March. The vortex average column ozone levels off and oscillates around ∼ 360 DU until the beginning of the vortex break-up in mid-April (Fig. 9b). In contrast, in 2060 the polar vortex weakens substantially in mid-February and does not fully recover before the transition to summertime easterlies occurs (Fig. 9a). Consequently, the Arctic mean column ozone rises to ∼ 490–500 DU by the end of March.

Figure 10a and b show time–altitude cross sections of the differences in Arctic mean ozone mixing ratio and ozone concentrations, respectively, between the model years 2063 and 2060. Small differences in ozone of up to ∼ 0.5–1 ppm and ∼ 1×1012 molecules cm-3 are already present in late autumn at ∼ 25 km, which contribute to the differences in column amounts evident in Fig. 8a. However, larger differences appear in the mid- and upper stratosphere from the end of December onwards. This is particularly evident in the mixing ratios, with up to ∼ 2.5–3 ppm less ozone between 30 and 40 km in 2063 compared to 2060. However, owing to the exponential decay in pressure with altitude the deficit is less pronounced in terms of absolute ozone amounts (up to ∼ 0.5×1012 molecules cm-3). This anomaly is predominantly dynamically driven and is transported downward over the course of the winter, with differences in March maximising at 25–30 km. Ozone abundances are also reduced in the lower stratosphere in 2063 throughout winter and spring. Even though the differences in ozone mixing ratios at these levels are smaller compared with those in the mid- and upper stratosphere, the higher ambient pressure results in significant changes in absolute ozone amounts. These therefore make a substantial contribution to the difference in total ozone columns in Fig. 8a. In general, the deficits at all altitudes increase from mid-winter to early spring, with ∼ 1–1.75×1012 molecules cm-3 less ozone in the 10–30 km region in 2063 compared to 2060 (Fig. 10b). The difference in partial ozone column up to 25 km contributes a relative decrease of ∼ 70 DU by the end of March. These are the altitudes where halogen-catalysed ozone destruction plays its most significant role, as evidenced by elevated ClO concentrations in 2063 (Fig. 10c).

Figure 8b shows time series of the Arctic mean temperature at 22.7 km, in the region of NAT PSC formation, in 2060 (black) and 2063 (red). In 2060, the temperatures during early and mid-winter fluctuate around 210 K but do not drop below ∼ 203 K in the Arctic mean. In contrast, the Arctic mean temperature in 2063 oscillates near or below 195 K during most of January and February. The minimum local daily mean temperatures in 2063 reach 181 K (cf. 191 K in 2060). The low temperatures allow the formation of PSCs (see red star in Fig. 5, Sect. 3.2.3, for the potential VPSC), which in turn permits heterogeneous chemical reactions that lead to the activation of chlorine from its reservoir species, HCl and ClONO2, into reactive forms (ClO and Cl2O2; see Fig. S3, Supplement). In early winter in 2063, ClO increases predominantly near the sunlit vortex edges, which could contribute to some ozone loss in the mid-latitudes (Pyle et al., 1994; Millard et al., 2002). From mid-January onwards, Arctic mean ClO levels in the lower stratosphere are ∼ 125–150 ppt at 21.5 km (Fig. S3) and ∼ 100–125 ppt at 19 km (not shown). This is up to ∼ 75–125 ppt higher than in 2060 (Fig. 10c). While the Arctic mean values are relatively modest, local daily mean ClO levels in March reach up to ∼ 300 ppt at 19 km in 2063 (cf. ∼ 80 ppt in 2060). The very low temperatures in the model year 2063 also lead to substantial denitrification such that the formation of PSCs, and halogen activation, at these altitudes from mid-winter is strongly reduced (not shown).

The dashed blue line in Fig. 8a shows vortex-average column ozone amounts in 2063 for a passive stratospheric ozone tracer, which when compared with the full (chemistry and transport) vortex-average ozone field enables a quantification of the relative roles of transport and (all) chemical processes in determining column ozone (see Tegtmeier et al., 2008, for an alternative approach). This passive tracer is initialised to the ozone field on 1 November and undergoes no chemical production or loss but is advected by the circulation. The full ozone field and the passive tracer for 2063 track each other quite well, particularly in early and mid-winter. By the end of March, both are relatively low and different from the ozone calculated in 2060 and it seems, therefore, that a major reason for the low column ozone in 2063 is dynamical in origin. There is also a difference between the full and passive ozone columns in 2063, which increases with time, with the cumulative effect of all chemical processes in the preceding five months averaged over the polar vortex contributing a 24 DU decrease to the total ozone column by the end of March in that year. This shows that, while dynamical processes are the dominant factor contributing to the low column ozone in 2063, chemical processes cannot be neglected. We caution, however, against over-interpretation of this value as it is the result of a complex balance between different chemical loss and production cycles throughout the depth of the atmosphere as well as their interaction with transport throughout the winter.

Using our other diagnostic, i.e. the cumulative (1 November to 30 March) halogen-induced ozone loss in the lower atmosphere (up to 25 km), we find that the 6 main halogen cycles (see Sect. 2.3) result in 39 DU ozone loss in 2063 compared to 18 DU in 2060 (with very similar values of 44 and 23 DU, respectively, if we take vortex average; see also Table S2 and Figs. S4–S5, Supplement, for other definitions of the polar vortex edge and stereographic plots of key quantities simulated on 1 March in the two model years). By this measure, the difference in estimated halogen losses between the two model years (∼ 20 DU) is equivalent to about 20 % of the full difference (∼ 100 DU; see above) in the end of March Arctic mean total column ozone between the two years. Whilst the two diagnostics (i.e. chemical ozone loss derived from the passive ozone tracer and the halogen-induced ozone loss in the lower atmosphere) are not equivalent, both indicate that dynamical processes are likely to be a dominant driver of future interannual variability in Arctic ozone. However, as illustrated by the later case, loss of ozone through halogen chemistry remains a smaller but nonetheless important contributor to the modelled Arctic ozone variability even in the second half of the 21st century.