The impact of SF6 sinks on age of air climatologies and trends

Mean age of air (AoA) is a common diagnostic for the strength of the stratospheric overturning circulation in both climate models and observations. AoA climatologies and its trends over the recent decades of model simulations and proxies derived from observations of long-lived tracers do not agree. Satellite observations show much older air than climate models and while most models compute a clear decrease of AoA over the last decades, a thirty-year timeseries from measurements shows a 5 statistically non-significant positive trend. Measurement-based AoA derivations are often based on observations of the trace gas SF6, a fairly long-lived gas with a near-linear increase of emissions during the recent decades. However, SF6 has chemical sinks in the mesosphere, which are not considered in most model studies. In this study, we explicitly compute the chemical SF6 sinks based on chemical processes in the global chemistry-climate model EMAC. We show that good agreement of stratospheric AoA in EMAC and MIPAS is reached through the inclusion of chemical SF6 sinks, as those lead to a strong increase of the 10 stratospheric AoA and thereby to a better agreement with MIPAS satellite observations. Remaining larger differences in high latitudes are addressed and possible reasons are discussed. Subsequently, we demonstrate that also the AoA trends are strongly influenced by the chemical SF6 sinks. Under consideration of the SF6 sinks, the AoA trends over the recent decades reverse sign from negative to positive. We conduct sensitivity simulations which reveal that this sign reversal results neither from trends of the stratospheric circulation strength, nor from changes in the strength of the SF6 sinks. We illustrate that even a constant 15 SF6 destruction rate causes a positive trend in the derived AoA, since the amount of depleted SF6 scales with the increasing SF6 abundance itself. In our simulations, this effect overcompensates the impact of the accelerating stratospheric circulation which naturally decreases AoA. Although various sources of uncertainties cannot be quantified in detail in this study, our results suggest that the inclusion of SF6 depletion in models has the potential to reconcile the AoA trends of models and observations.


20
The Brewer-Dobson circulation (BDC) describes the stratospheric transport circulation, consisting of the mean overturning circulation of air ascending in the tropical pipe, moving poleward and descending in the extratropics (Brewer, 1949;Dobson and Massey, 1956), as well as isentropic mixing. A good measure to diagnose this transport circulation is the age of stratospheric AoA (see e.g. SPARC, 2010;Dietmüller et al., 2018) and the AoA trend over recent decades even differs in sign between observations and models. While most climate models show a clear decrease of AoA over time (see e.g. Butchart and Scaife, 2001;Garcia et al., 2011;Eichinger et al., 2019), consistent with the simulated acceleration of the BDC in the course of climate change (see e.g. Garcia and Randel, 2008), the time series of the observations presented in the studies by Engel  Ray et al. (2014) and Engel et al. (2017) show a (statistically non-significant) positive trend. This discrepancy has been addressed in numerous studies: Garcia et al. (2011) showed that due to the concave growth rate of tropospheric SF 6 concentrations, the AoA trends derived from an SF 6 tracer are smaller than the trends derived from a synthetic, linearly growing AoA tracer (also after accounting for the nonlinear growth rates of SF 6 ). They noted that the sparse sampling of in situ observations can be the reason for the above mentioned trend discrepancies. Birner and  as well as Bönisch et al. 45 (2011) argued that differences in the changes between the deep and the shallow BDC branch can possibly explain these. Ploeger et al. (2015) showed that the residual circulation transit time cannot explain the AoA trends, and that the integrated effect of mixing (which is coupled to residual circulation changes, see Garny et al., 2014) is crucial. Moreover, Stiller et al. (2017) could explain a hemispheric asymmetry by a shift of subtropical transport barriers. However, a comprehensive explanation for the trend differences between models and observations is still missing. 50 SF 6 sinks lead to older apparent AoA, as well as shorter lifetimes. Leedham Elvidge et al. (2018) evaluated AoA from several tracers including SF 6 and found clear differences between these, which indicated a shorter SF 6 lifetime than previously assumed. The strongest chemical SF 6 removal reactions take place in the mesosphere, the most important removal processes are electron attachment and UV-photolysis, but these processes have not yet been precisely quantified. Ravishankara et al. (1993) estimated an SF 6 lifetime of 3200 years and Reddmann et al. (2001) found a lifetime between 400 and 10000 years, depending 55 on the assumed loss reactions and electron density. A more recent model study by Kovács et al. (2017) reported a mean SF 6 lifetime of 1278 years and Ray et al. (2017) provided a range between 580 and 1400 years based on in situ measurements in the stratospheric polar vortex. The most recent study of Kouznetsov et al. (2020) shows an SF 6 lifetime ranging between 600 and 2900 years. Due to these uncertainties and the complex computation of the chemical reactions, most model studies do not consider any SF 6 sinks for the calculation of AoA from SF 6 mixing ratios. This can explain why most climate models generally 60 show younger stratospheric air than observations, in particular within the polar vortices (e.g. Haenel et al., 2015;Ray et al., 2017).
In the present study, we apply the chemistry climate model EMAC (ECHAM MESSy Atmospheric Chemistry, Jöckel et al., 2010;Jöckel et al., 2016). It uses the second version of the Modular Earth Submodel System (MESSy2) to link multiinstitutional computer codes. In our simulations, we employed the MESSy submodul "SF6" which explicitly calculates SF6 65 sinks based on physical processes (based on Reddmann et al., 2001), rather than on crude parameterisations. In Sect. 2 we describe the EMAC model and the SF6 submodel as well as the observational data we use for comparison. Sect. 3 contains a comparison of the EMAC climatologies with MIPAS data, a comparison of the EMAC trends with MIPAS and balloon borne measurements and an analysis of the results of two sensitivity simulations. In Sect. 4, we discuss the results and provide some concluding remarks. lower stratosphere region (UTLS) is 500-600 m. In the standard reference setup, we use the basic EMAC modules for dynamics, radiation, clouds, and diagnostics (AEROPT, CLOUD, CLOUDOPT, CVTRANS, E5VDIFF, GWAVE, ORBIT, OROGW,   PTRAC, QBO, RAD, SURFACE, TNUDGE, TROPOP, VAXTRA, refer to Jöckel et al., 2005;Jöckel et al., 2010, for details on these submodels). Additionally we included the new submodel SF6.

80
The submodel SF6 is used to calculate the lifetime of SF 6 by explicitly accounting for the sinks of SF 6 in the mesosphere. The calculation method for this is based on the reaction scheme of Reddmann et al. (2001). The most important reaction involved in the chemical degradation of SF 6 , namely electron attachment, is included in the SF 6 submodel. The configuration of the submodel allows for a simple exponential profile for the electron field and a more complex field based on Brasseur and Solomon (1986), whereas in the present study we use the latter option. It depends on altitude, latitude, solar zenith angle, air density and 85 day of year (see Fig. S1 in Supplementary Information). In contrast to Reddmann et al. (2001), UV-photolysis of SF 6 is not included in the submodel. The loss rate by photolysis is several orders of magnitude weaker than that of electron attachment up to altitudes of about 100 km (see e.g. Fig. 9 in Totterdill et al., 2015) and is therefore not relevant for the focus of our study.
Further reactions considered are photodetachment of SF − 6 (Datskos et al., 1995), destruction of SF − 6 by atomic hydrogen, destructive Electron Attachment and Secondary Reactions recovery reaction hydrogen chloride and ozone (Huey et al., 1995), stabilisation of excited SF − 6 by collisions and autodetachment of SF − 6 . An 90 overview is provided in Table 1. Reddmann et al. (2001) used climatological profiles for the forementioned gases, while in our submodel channel objects (see Jöckel et al., 2016) are used. Such channel objects can be calculated in other submodules (e.g. interactive chemistry), prescribed as external time series (in this study) or just be simple climatologies. The autodetachment rate can be chosen in the namelist and was set to 10 6 s −1 (see Reddmann et al., 2001). For the relative importance of the contributions of the various reactions see Fig. S1 in the Supplementary Information.

Simulation Setup
Other than the SF6 submodel, no interactive chemistry is activated in the simulations for this study. The reactant species involved in the SF 6 chemistry and the radiatively active gases (CO 2 , CH 4 , N 2 O, O 3 ) are transiently prescribed from the ESCiMo RC1-base-07 simulation (see Jöckel et al., 2016) as monthly and zonal means. Moreover, we prescribe the Hadley Centre Sea Ice and Sea Surface Temperature data set (HADISST), the CCMI-1 volcanic aerosol dataset (for its effect on 100 infrared radiative heating, see Arfeuille et al., 2013;Morgenstern et al., 2017) and QBO nudging (see Jöckel et al., 2016).
Our simulations range from 1950 to 2011, whereas at least the first ten years have to be considered as a spin-up period. The projection simulation runs from 1950 to 2100 with the SF 6 reactant species and GHGs prescribed from the ESCiMo RC2-base-04 simulation (see Jöckel et al., 2016)  namelist settings (see Jöckel et al., 2005). A summary of the simulations used in this study can be found in Table 2.

Satellite and in situ data
The MIPAS (Michelson Interferometer for Passive Atmospheric Sounding) instrument on Envisat (Environmental Satellite) allowed for the retrieval of SF 6 by measuring the thermal emission in the mid-infrared, while orbiting the Earth sunsynchronously 14 times a day. This high-resolution Fourier transform spectrometer measured at the atmospheric limb and 120 provided data for SF 6 retrievals in full spectral resolution from 2002 to 2004 and in reduced resolution from 2005 to 2012 between 6 and 40 km of altitude (Stiller et al., 2012;Haenel et al., 2015). In this study, a newer version of the MIPAS dataset existing as of 2019 will be shown, whereby new SF 6 absorption cross-sections have been used for the SF 6 retrieval Harrison, 2020). Except for the newer absorption cross-sections for SF 6 and the accounting for a trichlorofluoromethane (CFC11) band in the vicinity of the SF 6 signature, the SF 6 retrieval and conversion into AoA was done according 125 to the description by Haenel et al. (2015). In particular, the Level-1b data version is still V5. Engel et al. (2009) collected available air samples of SF 6 and CO 2 from balloon-borne cryogenic whole air-samplers flown during 27 balloon flights, with data up to 43 km, and re-analysed these samples in a self-consistent manner. The derived SF 6 data cover the years 1975 to 2005 (with a gap between 1985 and 1994) and covers the mid-latitudes between 32°N and 51°N.
Since AoA profiles from midlatitudes above appromimately 25 km or 30 hPa turned out to be constant over altitude, a mean measurements. In this paper, only the AoA data points derived from SF 6 are used. Engel et al. (2017) extended the initial dataset from Engel et al. (2009) to 2016, however the AoA here is derived from CO 2 measurements and thus also excluded.
135 Table 2. Overview of simulations undertaken in this study

Simulation Details
Reference (

Analysis method
The basic concepts for the calculation of mean AoA are introduced in Hall and Plumb (1994). In the case of a tracer with a linear increasing lower boundary condition, AoA can be determined by the time lag between the mixing ratio at a given point in the atmosphere and the same mixing ratio of the reference time series. As for any realistic tracer, SF 6 does not exhibit perfectly linear growth, for which adjustments in the AoA calculation are needed. This study follows the calculation method employed For the derivation of AoA from MIPAS SF 6 observations, Stiller et al. (2008Stiller et al. ( , 2012 and Haenel et al. (2015) used a slightly smoothed version of the global mean of SF 6 surface measurements as the reference time series instead of SF 6 at the stratospheric entry point, which is not available from observations (see e.g. Dlugokencky, 2020). The non-linearity of the reference curve was considered by its convolution with an idealized age spectrum parameterized as a function of the mean age within an 150 iterative approach. For more details, see Stiller et al. (2012) and Haenel et al. (2015).
In our simulations the AoA calculations are applied to a total of four tracers, which can be organised into two groups. The first assumes a strict linear growth of SF 6 , producing a linear reference curve, while the second considers a realistic growth of SF 6 based on observed emissions, creating a non-linear SF 6 reference curve. Technically, these "emissions" are realised via lower boundary conditions in our simulations. As previously mentioned, SF 6 undergoes chemical degradation predominantly 155 in the mesosphere. Consequently, the absence or presence of mesospheric sinks is additionally considered, resulting in a total of four tracers tr(WS, SF 6 ), tr(NS, SF 6 ), tr(WS, lin), and tr(NS, lin). The labelling of these depends on the chemistry involved (with sinks: "WS", without (no) sinks: "NS") and the growth assumed (linear: "lin", non-linear: "SF 6 "), and follows the pattern tr(CHEMISTRY, GROWTH). When referring to simulations with a specific tracer, the labelling will follow the notation SIMULATION(CHEMISTRY, GROWTH), and similarily we use the following notation for AoA inferred from the tracer In order to evaluate the SF 6 mixing ratios simulated of the EMAC model, we first analyse the four tracers in the SD simulation in comparison to observational data. Fig.1a depicts the modelled SF 6 vertical profile climatologies in comparison with MIPAS The tracers with non-linear growth in the SD simulation show smaller tropospheric SF 6 mixing ratios than the linear tracers ( Fig.1a and b). This can be explained by the two different growth scenarios of SF 6 and the prescribed lower boundary conditions 175 (see Fig. S2 in Supplementary Information). The sinks do not have a considerable effect in the troposphere, hence the effect of the SF 6 sinks becomes only noticeable higher up. Furthermore, the effect of the SF 6 depletion becomes increasingly evident with altitude. This is portrayed in the growing differences with altitude between tr(WS, lin) and tr(NS, lin) and the non-linear equivalent. The differences particularly increase for the tracers with linear emissions, as these exhibit higher SF 6 mixing ratios and hence experience greater SF 6 depletion than those with non-linear boundary conditions. Due to the small turnaround times 180 for air in the middle atmosphere, the tracers without sinks exhibit a very low decrease of the SF 6 mixing ratios with altitude. Fig.1a shows that the EMAC simulated non-linear SF 6 is within the observed range of MIPAS SF 6 . Below 30 km, MIPAS SF 6 mixing ratios are smaller, with a near-constant offset of approximately 0.5 pmol/mol up to 20 km. Above 30 km, MIPAS SF 6 shows larger mixing ratios than EMAC. This means that EMAC SF 6 (SD(WS, SF 6 )) shows a larger decrease with altitude than MIPAS SF 6 , suggesting that the sinks in EMAC are too strong. Another explanation could be too strong vertical mixing in EMAC. However, the EMAC SF 6 lies within the MIPAS uncertainty range throughout the atmosphere. The standard deviation increase with height in MIPAS SF 6 can be attributed to the decrease of the SF 6 signal with height, which leads to an increase in noise-error of SF 6 . Additionally, the natural variability of SF 6 itself, as well as the evolution of SF 6 over time, contribute to the increasing standard deviation in the MIPAS SF 6 profile. The increase in standard deviation with height can also be seen in the EMAC SF 6 profiles, particularly in the tracers tr(WS, SF 6 ) and tr(WS, lin). However, it is by far not as large as in the MIPAS 190 data because the simulations have no measurement error and possibly show a smaller natural variabilty than the observations.
The balloon flight SF 6 profile (Ray et al., 2014) in Fig.1b largely resembles the profile of the realistically modelled tracer tr(WS, SF 6 ). Below 25 km, the modelled SF 6 profile shows a constant high bias of around 0.3 pmol/mol, presumably due to the lower boundary conditions used. Larger discrepancies can be seen above 25 km altitude, with higher mixing ratios of the modelled SF 6 . As the data presented in Fig.1b are only for a specific day and region, the particular meteorological situation 195 can be crucial for the comparison.

SF 6 lifetimes
The atmospheric lifetime of SF 6 can be used as an indicator for the accuracy of the SF 6 degradation scheme. We calculate   comparison with the previously published MIPAS data (Stiller et al., 2012;Haenel et al., 2015), the EMAC AoA was actually too young, i.e. the MIPAS AoA was much older in the previous versions in the polar regions. The spectroscopic data used for the SF 6 retrieval in MIPAS cause a rather large bias (that has now been corrected by improved spectroscopy). The new spectroscopic data lead to considerably younger AoA in the middle to upper stratosphere. There are, however, good reasons 245 to believe that the most recent MIPAS data are improved compared to the previous ones: the spectroscopic data used are far better characterized than the previous ones (Harrison, 2020), and the new AoA data from MIPAS agree significantly better with independent measurements than the previous version, in particular at higher altitudes . A detailed analysis of this will be provided in an upcoming paper on the new MIPAS retrieval. On the other hand, free-running EMAC simulations generally have a too weak Antarctic polar vortex (see Jöckel et al., 2016), which is, however, stronger than that of the reference processes of the chemical SF 6 removal can be revised and/or parameterised differently. Here, we showed that SF 6 sinks have 260 the potential to resolve the differences between simulated and observed climatologies of AoA and that EMAC AoA lies within the uncertainties of MIPAS AoA throughout the atmosphere. We therefore consider our simulations suitable for studying the temporal evolution of AoA.

Apparent age of air trends
In this section, we analyse the EMAC AoA trends and compare them with observation-based AoA trends. Fig. 4 shows the 265 AoA time series and linear regressions from the REF and the SD simulations as well as from MIPAS observations (Stiller et al., 2020, and paper in preparation) and from the SF 6 measurements by Engel et al. (2009). As the latter were collected from balloon flights in the Northern Hemisphere mid-latitudes at around 30 km altitude, the EMAC and MIPAS data are also taken from that height and averaged over 30°N to 50°N for consistency. between 30°N and 50°N, for the realistic tracer tr(WS, SF 6 ). The trend calculation follows that of Haenel et al. (2015).
The tracers without SF 6 sinks lead to negative AoA trends, which are consistent with the simulated acceleration of the BDC in the course of climate change (e.g. Garcia and Randel, 2008). Positive AoA trends are obtained for all tracers that take SF 6  (Engel et al., 2009) and from MIPAS observations (Stiller et al., 2020, and paper in preparation) are shown in red and dark-blue, respectively.  Following the trend calculation used in Haenel et al. (2015), the REF(WS, SF 6 ) and SD(WS, SF 6 ) time series show that EMAC AoA bears a good resemblence to that of the new MIPAS retrieval, with an AoA trend of 0.22 ± 0.12 and 0.50 ± 0.13 285 years per decade, respectively. Consistent with the trend calculation of the MIPAS data, the variability due to the QBO is considered by a respective term in the multivariate linear analysis. However, this measure induces only small differences in the EMAC trend calculations (see Table S1 in Supplementary Information for further details). Note also that the rather short period

Explanations for apparent Age of Air trends
In this section, we will analyse the EMAC apparent AoA trends, in particular the sign change of the trend when SF 6 sinks are  The increasing variability and trend of AoA in the last two decades, seen in the simulations with mesospheric SF 6 chemistry, can be attributed at first order to the SF 6 depletion reactions (see Fig.5). In particular, the effect of mesospheric SF 6 sinks 325 is stronger with higher SF 6 mixing ratios. Subsequent downward transport of SF 6 depleted air into the vortex and in-mixing thereof into lower latitudes after the vortex breakup results in apparent older AoA. This can explain that the annual variability increases over time due to the increase in SF 6 mixing ratios.
In the following, we will show that SF 6 sinks with constant destruction rates lead to a positive trend in AoA. The aforementioned link between the positive AoA trend and mesospheric SF 6 depletive chemistry, based on Hall and Waugh (1998), is 330 illustrated below and follows the mathematical formulations put forward by Hall and Plumb (1994) and Schoeberl et al. (2000).
We consider a tracer χ(t) experiencing relative loss e −k·t with constant loss rate k over time t. We denote the mixing ratio of reference SF 6 as χ o (t) with a constant linear growth rate χ oo and write At any location we can then write the concentration as 335 where the Green's function is written as G(τ ) and the concentration lag time as τ . Upon insertion of (1) into (2) we can express the latter as The first integral corresponds to the Laplace transform of the Green function G(τ ), denoted asG(k), and the second integral 340 the differential thereof, and so G(τ ) is assumed to be constant, i.e. this is for a constant circulation strength. The tracer without sinks is referred to as the passive tracer, and (using the subscript p for passive) the concentration χ p (t) can be expressed as a function of AoA, shown here as Γ, namely: When rearranging (5), we can write the AoA inferred from the passive tracer as Differentiating this with respect to time shows that no trend is found when considering the passive tracer, since χ p (t) = χ oo ·t+a where a is constant, and so When considering an active tracer (i.e. with mesospheric sinks), the apparent age Γ s can be expressed as The subscript s denotes the active tracer. In combination with (4) one can thus write whereG(k = 0) = 1 whereasG(k → ∞) = 0. The former case yields no trend in the case excluding mesospheric SF 6 sinks, as seen in the time slice simulation with the idealised tracer (see TS2000(NS, lin) in Table 3).
In the case of a passive tracer (i.e. a tracer without sinks), k = 0 andG(k = 0) = 1. In other words, the AoA trend is zero 360 in the absence of a trend in transport (constant G). This is the case in the TS2000 time slice simulation for the passive tracers (see TS2000(NS, lin) in Table 3). In the reference simulation this translates to a negative trend, consistent with the modelled acceleration of the BDC (e.g. Eichinger et al., 2019;Butchart et al., 2011). For sinks of finite strength, G(k) lies between 0 and 1, so that with consideration of the mesospheric SF 6 degradation scheme in the simulations, a positive trend in apparent AoA is found. In our simulations, this effect overcompensates the impact of the BDC acceleration. It is important to note that 365 the above is valid for a linearly increasing tracer, and that variable growth rates will modify the influence of SF 6 on apparent AoA.

Summary and conclusions
Disagreements with regards to stratospheric AoA and particularly its trends between model simulations and observations have been raising many questions by scientists for more than a decade now. AoA from observations is mostly older than AoA from 370 model simulations and models simulate a decrease in AoA over recent decades, whereas trend estimates from observational data report a non-significant positive trend. To make another step towards understanding the reasons for these discrepancies, we here study the impact of mesospheric SF 6 sinks on AoA climatologies and trends using the EMAC (Jöckel et al., 2010;Jöckel et al., 2016) model with the SF6 submodel (Reddmann et al., 2001) allowing for explicit calculation of SF 6 sinks.
The EMAC SF 6 mixing ratio profiles show good agreement with balloon-borne measurements as well as with satellite 375 observations. Some of the differences between the model and observations are within the uncertainty range of the observations. However, reasons for the quantitative differences in the high latitudes and altitudes can also be found in deficiencies in the representation of the SF 6 sinks in the model or in the dynamics simulated by the model. The EMAC reference simulation yields a global stratospheric SF 6 lifetime of 2100 years, varying between 2500 years and 1900 years for the simulation period. This value lies within the range of 600-2900 years provided by the model study of Kouznetsov et al. (2020) and below the value 3200 380 years calculated by Ravishankara et al. (1993). Kovács et al. (2017) and Ray et al. (2017) recently found somewhat lower SF 6 lifetimes of 1278 years and 850 years, respectively. Although this shows that large uncertainties still exist in determining the SF 6 lifetime, these results also confirm that the EMAC SF 6 depletion mechanisms are reasonable. In our transient simulations (REF and PRO), the SF 6 lifetimes vary by about 25 % following the abundances of the reactant species, basically resembling the pattern of the stratospheric ozone concentration. This behaviour, however, may be a result of the fact that several effects that potentially influence SF 6 lifetime variability are not implemented in full detail.
The inclusion of SF 6 sinks translates into apparent older stratospheric air. When SF 6 sinks are enabled, EMAC AoA therefore compares better with MIPAS satellite observations. In the tropics, overall good agreement can be found, but the results also indicate that EMAC does simulate a too broad, less isolated tropical pipe. In polar regions, however, EMAC AoA is higher than MIPAS AoA. In comparison with the previously published MIPAS data (Stiller et al., 2012;Haenel et al., 2015), the EMAC 390 AoA in the polar regions was actually too low. More research on both models and observations is necessary to resolve these remaining discrepancies.
Without SF 6 sinks, EMAC shows a negative AoA trend over 1965-2011. This is consistent with the simulated acceleration of the Brewer-Dobson circulation resulting from climate change (see e.g. Garcia and Randel, 2008;Butchart et al., 2011;Eichinger et al., 2019). The inclusion of chemical SF 6 sinks leads to positive AoA trends in our simulations, which in turn is 395 consistent with the positive AoA trend derived from MIPAS in the Northern Hemisphere (Haenel et al., 2015). Moreover, the SF 6 sinks help to improve the agreement of our model results with the AoA derived from the balloon-borne in situ measurements by Engel et al. (2009), from which a (non-significant) positive AoA trend was obtained. However, this only accounts for SF 6 -derived AoA, our results cannot help explain the positive trend of the CO 2 -derived AoA in Engel et al. (2009Engel et al. ( , 2017. Furthermore, it has recently been shown that the AoA trend derived in Engel et al. (2009Engel et al. ( , 2017 was likely overestimated due 400 to non-ideal parameter choices in the calculation of AoA (Fritsch et al., 2019). Our sensitivity studies show that the positive AoA trends are neither a result of climate change, nor of changes in the substances involved in SF 6 depletion. The SF 6 sinks themselves are the reason for the increase in apparent AoA. The reason for that is the temporally increasing influence of the chemical SF 6 sinks on AoA. In our simulations, this effect overcompensates the effect of the simulated acceleration of the stratospheric circulation leading to a net increase of AoA. Due to various sources of uncertainties, this result bears quantitative 405 leeway and has to be assessed in finer detail. But nevertheless, for now we can conclude that SF 6 sinks have the potential to explain the long-lasting AoA trend discrepancies between models and observations.