The impact of different CO2 and ODS levels on the mean state and variability of the springtime Arctic stratosphere

Rising greenhouse gases (GHG) and decreasing anthropogenic ozone-depleting substances (ODS) are the main drivers of the stratospheric climate evolution in the 21st century. However, the coupling between stratospheric composition, radiation and dynamics is subject to many uncertainties, which is partly because of the simplistic representation of ozone (O3) in many current climate models. Changes in ozone due to heterogeneous chemistry are known to be the largest during springtime in the Arctic, which is also a season with very active stratosphere–troposphere coupling. The focus of this study is to investigate the role of varying ozone levels driven by changing GHG and ODS for the Arctic polar cap stratosphere. We use two state-of-the-art chemistry-climate models with ocean coupling in two configurations (prescribed ozone fields vs. interactive ozone chemistry) for three different scenarios: preindustrial conditions—1 × CO2, year 2000 conditions (peak anthropogenic ODS levels) and extreme future conditions—4 × CO2. Our results show that in the upper and middle stratosphere CO2 thermal cooling is the dominant effect determining the temperature response under 4 × CO2, and outweighs warming effects of ozone by about a factor of ten. In contrast, in the lower stratosphere, the effects of O3 warming and CO2 cooling under 4 × CO2 are largely offsetting each other. ODS driven variations in O3 affect both the temperature mean and variability, and are responsible for the tight springtime coupling between composition and dynamics under year 2000 conditions in comparison to simulations under 1 × CO2 or 4 × CO2.


Introduction
Changes in stratospheric composition have received widespread attention since the detection of the Antarctic ozone hole almost 40 years ago (e.g. Farman et al 1985). Today it is well understood that ozone depletion is propelled by ozone depleting substances (ODS) of anthropogenic origin, and chemistry-climate model (CCM) projections suggest that the ozone layer will recover during this century as atmospheric ODS burdens decline as a consequence of the implementation of the Montreal Protocol (Dhomse et al 2018, WMO 2018. For many decades focus in stratospheric research has been given to the southern hemisphere (SH), particularly Antarctica, while the northern hemisphere (NH) and Arctic polar cap received less attention given that ozone losses are substantially more pronounced in the SH. The main reason for the reduced ozone depletion over the Arctic polar cap is that the NH polar vortex is less cold and stable as a result of stronger planetary wave activity in the NH (Waugh et al 1999, Zhou et al 2000, Garfinkel et al 2020, which also manifests in more frequent sudden stratospheric warming (SSW) events (Baldwin et al 2021).
Generally, SSW events can occur throughout the whole winter or spring and the polar vortex can recover its strength after these events. Whereas, the final stratospheric warming event (FSW) marks the transition from westerlies to easterlies and thus also the departure from an environmental regime propelling heterogeneous stratospheric ozone loss. This event can vary in its timing, and influence the tropospheric circulation and surface weather in spring or early summer (Butler et al 2019, Butler andDomeisen 2021). Furthermore, recent research by Friedel et al (2022b) highlights the important active role of ozone for the determination of the FSW. Despite the generally higher SSW frequency and overall more disturbed polar vortex, substantial ozone losses have occurred in recent years also in the Arctic (Manney et al 2011, Khosrawi et al 2017, Manney et al 2020, Kuttippurath et al 2021, when particularly cold conditions conducive for polar stratospheric cloud (PSC) formation persisted throughout winter and spring. ODS are of importance not only for their role in catalytic cycles propelling ozone depletion (e.g. WMO 2018) but also due to their contribution to radiative forcing and thus warming of the surface and troposphere along with CO 2 and other greenhouse gases (GHG) (Ramanathan 1975, Chiodo and Polvani 2022. The rising concentrations of CO 2 have not only led to unprecedented warming of the troposphere but also cooling in the stratosphere (IPCC AR5 2013) and thereby also stratospheric contraction (Pisoft et al 2021). In the stratosphere ODS contribute to the observed cooling via ozone depletion (Molina and Rowland 1974, Farman et al 1985, Anderson et al 1989, Solomon 1990, Randel and Wu 1999, Rieder et al 2014. Ozone and CO 2 are the main contributors to the stratospheric radiative balance. Ozone strongly absorbs the solar UV radiation and thereby heats the stratosphere. On the other hand, CO 2 (and partly ozone) emit in the infrared and cool across stratospheric levels (London 1980). The indirect coupling between CO 2 and ozone in the stratosphere is controlled by the temperature effects of CO 2 influencing the rates of chemical reactions of chlorine, bromine and nitrogen groups with odd oxygen. It is most pronounced in the upper stratosphere, where it is documented that the CO 2 -driven cooling (Fels et al 1980, Rind et al 1990, Shine et al 2003 results in a decrease of the temperature-dependent loss rates of ozone Hitchman 1988, Pitari et al 1992).
These various atmospheric and climatic effects can be resolved fully only in so-called high-top models, which are earth system models (ESMs) that comprise interactive stratospheric chemistry and a wellresolved stratospheric circulation. However, due to high computational costs many climate studies still rely on low-top models or high-top models with a well-resolved stratospheric circulation but without interactive ozone chemistry. Also the ensemble of the Coupled Model Intercomparison Project phase 6 models (CMIP6), in support of the Sixth IPCC Assessment Report includes only a few ESMs with interactive stratospheric chemistry (IPCC AR6 2021). The coupling between the upper troposphere and lower stratosphere is receiving increased attention in recent years. Adequate characterization of the processes involved requires a reliable characterization of stratospheric composition and temperature (Simmons et al 2004, Martineau et al 2016. Given that the importance of ozone chemistry for stratospheric temperatures, polar vortex state and stratosphere-troposphere coupling has been highlighted in recent work (e.g. Xie et al 2008, Haase and Matthes 2019, Rieder et al 2019, Oehrlein et al 2020, we hereby aim to understand the role of different GHG (1 × CO 2 , 4 × CO 2 , year 2000-hereinafter Y2000) and ODS levels (preindustrial and Y2000) on the mean state and variability of the Arctic stratosphere. Thereby we answer the question if CO 2 , or ozone change due to CO 2 or due to ODS is the dominant forcing across different stratospheric levels?

Data and methods
A clear attribution of the impact of ozone (O 3 ) vs. other GHG on stratospheric temperature, and the contribution of interactive ozone chemistry to stratospheric temperature variability requires high-top ESM simulations with transient/specified forcings such as regularly evaluated within the scope of the CCM Validation Activities 1 and 2 (e.g. Eyring et al 2005, Morgenstern et al 2010 and the CCM Initiative Phase 1 and 2 (e.g. Morgenstern et al 2017, Plummer et al 2021. Within this framework simulations spanning multi-decadal or centennial time scales, have become also available for the whole atmosphere community climate model (WACCM) and the solar climate ozone links (SOCOL) model. WACCM is a high-top model, comprising a coupledchemical module that links the thermosphere, mesosphere, stratosphere, troposphere and near-surface atmosphere. The National Center for Atmospheric Research Community Earth System Model is the numerical foundation for WACCM. Much progress in the representation of cross-layer interactions has been accomplished for WACCM in recent years, and the underlying model is well documented (Neale et al 2012, Marsh et al 2013, Smith et al 2014. The model version we use, WACCM4, includes an interactive atmospheric chemistry scheme (Kinnison et al 2007) and is fully coupled to interactive components for ocean, land, and sea ice. The horizontal resolution is 1.9 • latitude × 2.5 • longitude and the model comprises 66 vertical levels, with the model top at 140 km. Besides WACCM we analyze a set of simulations performed with the Max-Planck-Institute Global Ocean/Sea-Ice Model (SOCOL-MPIOM) model (Muthers et al 2014). The model consists of the ocean-sea-ice model MPIOM coupled to the CCM SOCOL version 3 (Stenke et al 2013), which is based on the middle atmosphere model MA-ECHAM5 (Roeckner et al 2003) and comprises a modified version of the chemistry model MEZON (Model for Evaluation of oZONe trends, see Egorova et al 2003). In the atmosphere, the model has a horizontal resolution of T42, corresponding to approximate grid spacing of 2.8 • × 2.8 • , and a vertical resolution of 39 levels from the surface till 0.01 hPa (∼80 km). The chemistry module of the model includes all main reactions important for stratospheric ozone formation and depletion.
Our study focuses on the contribution of CO 2 and ozone levels on Arctic polar cap (60 • -90 • N) mean temperature in the lower (LS, defined here as 70 hPa), middle (MS, defined here as 30 hPa) and upper middle stratosphere (UMS, defined here as 10 hPa) (SPARC Report N • 8 2017), both mean state and variability. To this end we analyze a set of 6 time slice simulations performed with WACCM and SOCOL explicitly tailored to represent past, present and potential future climate states (and varying ODS levels) (table 1) and to explore effects of interactive stratospheric chemistry. These comprise (a) a pair of Y2000 simulations with interactive chemistry (Y2000chem) vs. prescribed chemistry (Y2000nochem), which are the same simulations as analyzed in Rieder et al (2019), (b) a corresponding pair of simulations for a 1 × CO 2 climate (1 × CO 2 chem, 1 × CO 2 nochem), and (c) a corresponding pair of simulations for a 4 × CO 2 climate (4 × CO 2 chem, 4 × CO 2 nochem). The advantage of time slice experiments over the transient set-ups used for CCMI and CCMVal is that we can accurately study (with robust statistics) the impact of ozone chemistry on mean state and variability. Across these pairs, ozone fields in the nochem simulations are specified as the zonal mean 12-month climatology of the interactive ozone chemistry experiments. We note in passing, that the interactivity of stratospheric chemistry does not refer solely to ozone chemistry, but also to other species such as e.g. water vapor concentrations, which are chemically interactive e.g. via methane oxidation. Such effects are also at least partially taken into account through bulk approximations in the nochem schemes.
ODS are held at 1850 levels for the 1 × CO 2 (preindustrial) and 4 × CO 2 (extreme future) simulations. Besides these simulations, an additional simulation is required to cleanly attribute the influence of CO 2 levels on stratospheric temperature, namely a 4 × CO 2 simulation with prescribed preindustrial ozone levels (stemming from 1 × CO 2 chem, referred hereinafter as 4 × CO 2 nochem_1 × CO 2 -O 3 . This simulation allows us to calculate the direct CO 2 effect on stratospheric temperature by differentiating the 4 × CO 2 nochem_1 × CO 2 -O 3 and Table 1. Configuration of the different WACCM and SOCOL time slice simulations analyzed in this study ( * only available for WACCM). Concentrations of other important species in these integrations are as follows: 1 × CO2 (287 ppm for CO2, 0.7 ppm for CH4, 0.2 ppm for N2O, 0 ppb for CFC11 and 0 ppb for CFC12); 4 × CO2 (1148 ppm for CO2, 0.7 ppm for CH4, 0.2 ppm for N2O, 0 ppb for CFC11 and 0 ppb for CFC12); Y2000 (369 ppm for CO2, 1.7 ppm for CH4, 0.3 ppm for N2O, 0.26 ppb for CFC11, 0.5 ppb for CFC12).

Simulation
Configuration 2000s conditions, interactive O3 chemistry Y2000nochem 2000s conditions, prescribed O3 monthly averaged (Y2000chem O3) 1 × CO 2 nochem simulations. Conversely, the direct effect of ozone changes under 4 × CO 2 on temperature can be calculated as the difference between the 4 × CO 2 chem and 4 × CO 2 nochem_1 × CO 2 -O 3 experiments. The combined effect of changes in CO 2 and ozone under 4 × CO 2 can be estimated by differencing the 4 × CO 2 chem and 1 × CO 2 chem experiments. Last but not least, the influence of interactive chemistry on stratospheric temperature can be calculated by differentiating the individual chem vs. nochem pairs. We note, that only for WACCM the full set of these 7 simulations is available (note, no 4 × CO 2 nochem_1 × CO 2 -O 3 for SOCOL), thus our analysis focuses on the set of WACCM simulations and results for SOCOL are used to evaluate model agreement and to corroborate the results presented. We investigate individual contributions of GHG and ODS to stratospheric temperatures at vertical levels spanning from 5 to 70 hPa for the zonal band 60 • to 90 • N. Furthermore, the selection of the month for the final warming event and thus the polar vortex breaking in NH springtime is important, since it is model-dependent. This can be analyzed using the mean zonal wind (at 61 • N) for the classification of the polar vortex strength. It is well known that the polar vortex for WACCM is too persistent (e.g. Eyring et al 2006), which influences the date of the final warming in spring. The FSW marks the complete collapse of the polar vortex (Hardiman et al 2011). By abruptly warming these layers, PSCs dissolve and processes driving heterogenous ozone loss are halted. A recent study by Friedel et al (2022b) documented the occurrence of FSW events for WACCM and SOCOL, finding them to occur most frequently in April. As a result, April is also the month during which the variability of ozone maximizes, and thus any effects on temperature can be largest, and is thus the focus of our analysis. Across our study statistically significant differences among individual data-sets analyzed are assessed by Kolmogorov-Smirnov tests.

Ozone mean state and variability
We start the discussion of the individual CCM simulations by detailing and contrasting the magnitude and variability of ozone concentrations in the individual simulations with interactive chemistry, which is illustrated in figure 1. Ozone abundances (figures 1(a) and (b)) in the Y2000chem simulations are consistently lower compared to 1 × CO 2 chem and 4 × CO 2 chem, particularly in the lower stratosphere for both the WACCM and SOCOL models. This is due to Y2000chem simulations having peak anthropogenic ODS abundances, which effectively deplete ozone in springtime via activation on PSC surfaces (Crutzen and Arnold 1986, Solomon et al 1986, Solomon 1999. In contrast, the highest Arctic ozone abundances emerge in 4 × CO 2 chem (figures 1(a) and (b)), induced by the absence of anthropogenic ODS and the so-called 'ozone super recovery' effect due to climate change (Haigh and Pyle 1982, Jonsson et al 2004. 1 × CO 2 chem ozone levels mostly lie in between the other two experiments and are determined by the absence of (a) anthropogenic ODS and (b) the current/ future elevated CO 2 levels driving the 'super recovery' processes.
In terms of ozone variability (figures 1(c) and (d)) pronounced differences emerge between the three simulations. For the experiments without anthropogenic ODS, the ozone variability in 4 × CO 2 chem is higher than in 1 × CO 2 chem in both models, except 50 hPa in SOCOL. This aligns with the absolute ozone levels in figures 1(a) and (b) in a way that the more ozone is available, the larger the variability driven by meteorology (i.e. extreme vortex states). Namely, in 4 × CO 2 chem more ozone fills the polar regions during the FSWs and also more ozone is isolated during the strong vortex conditions. Besides chemistry, also atmospheric dynamics and transport are of prime importance mainly because the tropospheric wave forcing between these two climate states is significantly different. For example, a recent paper based on CMIP6 simulations (Abalos et al 2021) with higher CO 2 levels (increasing gradually at a 1% yr −1 rate from preindustrial (1850) conditions) shows the acceleration of the BDC and more ozone transport from the tropics to the polar regions. In return, the elevated ozone levels in middle and polar latitudes in the 4 × CO 2 chem experiments influence the local temperatures and thus feedback on the dynamics and the state and isolation of air masses within the polar vortex. Abalos et al (2021) additionally highlights substantial uncertainty in middle and upper stratospheric BDC and polar vortex trends in models and observations. We also see large differences between our two models in terms of seasonal evolution in the westerly winds under all forcings (figure S1). Overall, WACCM has a very strong vortex that forms in autumn and persists till early spring, while in SOCOL the polar vortex is generally weaker and peaks in strength around midwinter. Both features are also well documented in the literature for WACCM (e.g. Calvo et al 2015) and SOCOL (e.g. Sukhodolov et al 2021). WACCM is less sensitive to CO 2 , with only marginal intensification of the mean vortex state in the 4 × CO 2 chem simulation. SOCOL reacts differently, showing a strengthening of the vortex in 4 × CO 2 chem till midwinter (November) and then a weakening from December to April compared to other experiments, which results in a peak strength early offset by 1-2 months. A change in the midwinter state of the polar vortex can also have important implications for the FSW (Hauchecorne et al 2022). Most remarkably, ozone variability at the 30-70 hPa levels in the Y2000chem simulation is clearly different to those in other experiments, confirming earlier results by Rieder et al (2019). Although this experiment reveals the lowest ozone abundance in the LS and MS because of the heterogeneous depletion by anthropogenic ODS, its variability is much larger in both models, and in SOCOL it is additionally skewed towards the negative side. Besides the abundance of ODS, the LS ozone depletion in the NH is strongly dependent on the local temperatures (WMO 2018), which in turn can be influenced by ozone itself, especially during final warming months (Rieder et al 2019, Friedel et al 2022a. As ozone variability affects temperature, it contributes also to dynamical changes (longer lifetime and stronger isolation of the vortex or its earlier weakening), which feeds back into the amplitude of ozone variability itself.

Temperature mean state and variability
We turn next to the analysis of differences in temperature mean state and variability between individual chem and nochem experiments under 1 × CO 2 , 4 × CO 2 and Y2000 conditions over the Northern polar region in springtime. Generally, the temperature distribution is more similar across the different experiments in the LS (figures 2(a) and (e)) than in the UMS (figures 2(b) and (f)). However, we can identify the influence of ODS driven ozone depletion in the LS, as colder temperatures are simulated under Y2000 conditions than in any of the CO 2 experiments, if interactive chemistry is included in the model integrations (Y2000chem) (figures 2(a) and (e)). This result is consistent with recent work (e.g. Rieder et al 2019), documenting that extreme temperatures are rarely or never reached in the LS, when climatological ozone fields ("nochem" configuration) are prescribed. Also, in the mean state a pronounced difference of −0.49 K in WACCM (−0.36 K in SOCOL) is found between Y2000chem and Y2000nochem experiments. Clear differences in LS temperatures emerge also comparing the 1 × CO 2 and 4 × CO 2 experiments in the mean state (figures 3(a) and (e)).
In contrast to the relatively narrow packed temperature distribution in the LS much larger differences emerge among the differently forced integrations in the UMS (figures 2(b) and (f)). Particularly the effect of enhanced stratospheric cooling with increasing levels of CO 2 is clearly seen by contrasting 1 × CO 2 , Y2000 and 4 × CO 2 experiments. Here the mean state difference for the chem (and nochem) integrations of WACCM yields about 7.2 (8.4) K between 1 × CO 2 and 4 × CO 2 and about −5.5 (−5.4) K between 4 × CO 2 and Y2000. This temperature scaling does not come as a surprise. Today, it is well understood that a pronounced thermal cooling effect results from increased CO 2 levels at these altitudes (Brasseur and Solomon 2005, Goessling and Bathiany 2016) due to the emission and absorption behavior of CO 2 molecules. This also explains why the mean temperatures in Y2000 lie between those of the 4 × CO 2 (higher CO 2 amounts) and 1 × CO 2 (lower CO 2 amounts) experiments (figures 2(b) and (f)). Our results for the Arctic polar cap agree qualitatively also well with previous work on global scale  Recent research illustrated the importance of stratosphere-troposphere-surface coupling processes and their modulation by the polar ozone extremes in the Arctic (e.g. Calvo et al 2015, Ivy et al 2017, Friedel et al 2022a Given that LS temperatures are integrating the chemical and dynamical state (Rieder et al 2019), as a starting point in the vertical coupling, we turn now to the analysis of temperature variability among our set of CCM integrations. This analysis is guided by two leading questions: (a) does temperature variability differ between the differently forced experiments in terms of GHG and ODS and (b) to which extent the chemistrydynamics feedback influences it. While the first question is addressed by contrasting the reference 4 × CO 2 chem, 1 × CO 2 chem, Y2000chem experiments, the latter one is studied by comparing the reference chem experiments to their nochem counterparts. The anomalies of the temperature distributions around the mean state are presented as whiskers in figures 2(c) and (g) for the LS in figures 2(d) and (h) for the UMS for WACCM and for SOCOL. For both models, significant differences in the temperature anomalies emerge in the LS between the Y2000chem and Y2000nochem integrations, with wider distribution for Y2000chem. In contrast to Y2000, 4 × CO 2 experiments do not show such an effect for both models, while under 1 × CO 2 conditions the models diverge with SOCOL showing higher variability at both levels for chem and WACCM showing it only for the UMS. The Y2000 findings are consistent with previous work focusing on contemporary conditions (Rieder et al 2019) highlighting the overall importance of ozone abundances (and anthropogenic ODS controlling it via heterogeneous reactions) for lower stratospheric temperatures. In contrast, 1 × CO 2 and 4 × CO 2 experiments do not contain (anthropogenic) ODS and therefore the ozone variability that feedbacks to the state of the vortex is mostly driven by dynamics and transport (figures 1(c) and (d)). This difference between the ODS and non-ODS experiments is increasing with decreasing altitude. The presence of anthropogenic ODS in the chem configuration also results in the coldest temperatures in the LS among all considered modelling setups. Nevertheless, some chemistry-dynamics feedback emerges also in the 1 × CO 2 integrations in the UMS, while the 4 × CO 2 temperature variability appears to be almost fully insensitive to it, despite 4 × CO 2 having more ozone and larger ozone variability at all levels compared to 1 × CO 2 (figure 1). The 4 × CO 2 conditions are characterized by the coldest temperatures in the UMS with the least variability and with marginal difference between chem and nochem. This might suggest that the ozone contribution to the local radiative budget of the polar springtime stratosphere depends on the magnitude of the CO 2 cooling. Among the individual integrations, the 1 × CO 2 chem simulation shows the warmest mean state, which affects the vortex stability in absence of anthropogenic ODS and elevated CO 2 resulting in the largest temperature variability in the UMS.

Net effect of CO 2 and ozone on temperature
Next, we turn to answer the question regarding the net effects of changes in ozone or CO 2 on LS and UMS temperature. To this end we evaluate the temperature effects of (a) changes in ozone without changes in CO 2 (4 × CO 2 chem-4 × CO 2 nochem_1 × CO 2 -O 3 ), (b) changes in CO 2 without changes in ozone (4 × CO 2 nochem_1 × CO 2 -O 3 -1 × CO 2 nochem) and (c) changes following the effective CO 2 effect which includes changes in ozone (4 × CO 2 chem-1 × CO 2 chem). In figure 3, we show the corresponding temperature effect in the mean state (50%percentile, accompanied by the standard deviation) for different vertical levels throughout the NH polar cap springtime stratosphere. We note that the net effects we illustrate here represent the combined effects of chemical, dynamical and radiative processes and their interaction.
At upper stratospheric layers, the CO 2 thermal cooling (and corresponding adjustments in chemistry and dynamics) is the dominant effect, and outweighs warming effects of ozone by about a factor of 10. This is illustrated by a pronounced cooling of −9.5 K (±0.34 K) opposing a warming of +0.95 K (±0.45 K) at 10 hPa. With decreasing altitude, the cooling effect of CO 2 decreases substantially. This is in broad agreement with previous work (e.g. Manabe andWetherald 1967, London 1980) and confirms our analysis of figure 2. In contrast to the US, in the LS ozone is of major importance and the warming effect by ozone almost compensates the cooling by CO 2 . At 70 hPa the warming attributable to the net ozone effect is 2.7 K (±0.64 K). The cooling directly attributed to CO 2 is −3.1 K (±0.51 K), which reduces to a net effective CO 2 effect on temperature of 0.08 K (±0.43 K) taking warming by ozone into account. This also confirms our findings concerning the LS being more sensitive to changes in ozone variability. Therefore, the factors influencing ozone in this region also have a stronger influence on temperature and thus also vortex variability in the LS. A major factor here is the springtime ozone depletion driven by anthropogenic halogen loading (ODS) as clearly seen in the Y2000chem experiment, represented as increased ozone (figure 1) and temperature (figure 2) variability.

Coupling of ozone, zonal wind and temperature
Finally, we turn to the coupling of zonal mean wind, ozone content and temperature in the NH polar cap stratosphere, which is presented in figure 4. The strong springtime feedback loop between ozone variability, temperatures and winds is only present in the LS under Y2000 conditions and driven by the availability of anthropogenic ODS, which are absent in the 1 × CO 2 and 4 × CO 2 experiments. The vortex stability and very low temperatures govern the heterogeneously-driven ozone variability (see figures S2 and S3), which in turn affects back the LS temperatures and the wind speed. The latter two variables are strongly coupled in the LS compared to the UMS (see figures S4 and S5), which is an important part of the mechanism of vertical coupling between the stratosphere and troposphere (Kuroda andKodera 1999, Kidston et al 2015). In the absence of ODS, elevated CO 2 levels increase the ozone content (figure 1), but this does not strengthen the springtime coupling between ozone and dynamics even in the LS (figures 4(a) and (c)). In the UMS, where CO 2 content is the dominant driver for the mean temperature response (figure 3), and also ozone variability is smaller (figure 1), the before-mentioned coupling is absent as illustrated by the flat regression lines between ozone and wind in figures 4(a) and (c). In contrast we find a strong relationship between ozone and zonal mean winds across all experiments earlier in the winter season (see figure S6) which is the manifestation of the role of the vortex strength in modulating the dynamical supply of ozone into the Arctic stratosphere (mixing barrier). In spring, however, this coupling is only prominent in the LS under Y2000 conditions, where also the ozone variability is largest and the coupling is strongly modulated by heterogeneous ozone depletion (cold temperatures driving chemistry while simultaneously stabilizing the vortex).

Conclusion and discussion
It is well established that CO 2 is the main driver of the thermal budget of the UMS and that anthropogenic emissions are elevating CO 2 levels and thus increase the cooling of these stratospheric layers (Ramaswamy et al 2001, Shine and Bourqui et al 2003, Brasseur and Solomon 2005, Aquila et al 2016, Goessling and Bathiany 2016, Maycock et al 2018. Furthermore, elevated CO 2 levels significantly affect the ozone layer through changes in transport and chemistry . Our results (figures 2(b), (f) and 3) show that the amount of CO 2 is the dominant factor determining the radiativelydriven temperature response for the Arctic polar cap stratospheric springtime UMS. The CO 2 induced cooling in a 4 × CO 2 climate is approximately 10 times larger than the warming by ozone at 10 hPa. At lower stratospheric levels, the net effects of the cooling by CO 2 and warming by ozone nearly offset each other for 4 × CO 2 and 1 × CO 2 experiments without ODS (figure 3). In contrast, ODS levels under Y2000 conditions lead to a strong episodic springtime ozone loss in the Arctic, which feeds back on stratospheric temperatures, as illustrated by the tight coupling between ozone and temperature on interannual time-scales in the LS (figures 2(a), (e) and 4(b), (d). Furthermore, zonal mean temperature and zonal wind speed (which is a measure of the vortex strength) are also tightly coupled in the contemporary LS, while no such coupling is found in the UMS (figures S4 and S5). In an atmosphere with elevated CO 2 loadings (in the absence of ODS), despite the elevated ozone levels, no strengthened coupling between ozone and dynamics emerges.
Taken together, these results highlight the importance of using interactive ozone-chemistry under contemporary stratospheric conditions (Y2000chem) given that (a) a tight coupling between temperature, zonal wind and ozone emerges in the LS compared to the 4 × CO 2 chem and the 1 × CO 2 chem experiments (figures 4(b), (d) and S2-S5), and (b) colder and more variable and extreme LS temperatures emerge compared to the Y2000nochem experiments. Our results also illustrate that anthropogenic ODS emissions driving the polar ozone loss are not only resulting in a surface UV increase but also substantially affect and modulate the springtime temperature variability. These findings, together with recent evidence highlighting the importance of Arctic ozone for temperature extremes (Rieder et al 2019), the determination of the FSW (Friedel et al 2022b), stratosphere-troposphere coupling and NH climate anomalies (Friedel et al 2022a), and large-scale climate modes (Xie et al 2016) motivate the wider application of high-top models and inclusion of interactive ozone chemistry in future multi-model activities such as CMIP.
In closing, we note that for future work, exploring the agreement of sensitivity experiments across multimodel ensembles (CCMI, CCMVal) with the results presented here would be of interest. That said, given that multi-model intercomparisons are performed using transient forcings, we wish to suggest a suite of additional time slice experiments to the CCMI modelling teams considering different GHG and ODS levels, allowing for a clean comparison of the modelled responses in mean state and variability. Further, an alternative line for future work could be to explore the differences in mean state and variability in model simulations with interactive chemistry vs. simulations that apply linearized ozone schemes.
Recent work in this respect shows promising results for the representation of the ozone response to CO 2 (e.g. Meraner et al 2020), but aspects of polar cap variability and implications for temperature remain unexplored and should be subject of future work.

Data availability statement
The data that support the findings of this study are available upon reasonable request from the authors.