Sensitivity of the southern hemisphere tropospheric jet response to Antarctic ozone depletion: prescribed versus interactive chemistry

. Southern hemisphere lower stratospheric ozone depletion has been shown to lead to a poleward shift of the tropospheric jet stream during austral summer, inﬂuencing surface atmosphere and ocean conditions, such as surface temperatures and sea ice extent. The characteristics of stratospheric and tropospheric responses to ozone depletion, however, differ largely among climate models depending on the representation of ozone in the models. The most accurate way to represent ozone in a model is to calculate it interactively. However, due to computational costs, 5 in particular for long–term coupled ocean-atmosphere model integrations, the more common way is to prescribe ozone from observations or calculated model ﬁelds. Here, we investigate the difference between an interactive and a speciﬁed chemistry version of the same atmospheric model in a fully–coupled setup using a 9–member chemistry–climate model ensemble. In the speciﬁed chemistry version of the model the ozone ﬁelds are prescribed using the output from the interactive chemistry model version. In contrast to earlier studies, we use daily–resolved ozone ﬁelds in the speciﬁed chemistry simulations to achieve a 10 better comparability between the ozone forcing with and without interactive chemistry. We ﬁnd that although the short–wave heating rate trend in response to ozone depletion is the same in the different chemistry settings, the interactive chemistry ensemble shows a stronger trend in polar cap stratospheric temperatures (by about 0.7 K decade − 1 ) and circumpolar stratospheric zonal mean zonal winds (by about 1.6 ms − 1 decade − 1 ) as compared to the speciﬁed chemistry ensemble. This difference between interactive and speciﬁed chemistry in the stratospheric response to ozone depletion also affects the tropospheric response, 15 namely the poleward shift of the tropospheric jet stream. We attribute part of these differences to the missing representation of feedbacks between chemistry and dynamics in the speciﬁed chemistry ensemble, which affect the dynamical heating rates, and part of it to the lack of spatial asymmetries in the prescribed ozone ﬁelds. This effect is investigated using a sensitivity ensemble that was forced by a three–dimensional instead of a two–dimensional ozone ﬁeld. This study emphasizes the value of interactive chemistry for the representation of the southern hemisphere tropospheric jet 20 response to ozone depletion and infers that for periods with strong ozone variability (trends) the details of the ozone forcing can be crucial for representing southern hemispheric climate variability. depletion the future ozone recovery.

our experiments are carried out based on historical forcing conditions for 1955 to 2005 and on the representative concentration pathway 8.5 (RCP8.5) for the period of 2006 to 2013. Hence, the simulations cover a 58-year period that covers the period in which catalytic ozone depletion started and before ozone recovery becomes important. The external forcings are mostly based on the CMIP5 recommendations: GHG and ODS concentrations (Meinshausen et al., 2011), as well as volcanic aerosol concentrations (Tilmes et al., 2009). However, for the spectral solar irradiances and the geomagnetic activity as proxy forcing for energetic particle effects the more recently published CMIP6 forcing was applied (Matthes et al., 2017).

Observational Data
To verify our modeled temperature trend, we compare it with observational temperature data from the Integrated Global Radiosonde Archive, version 1 (IGRA) from the National Centers for Environmental Information (NCEI) of the National Oceanic and Atmospheric Administration (NOAA). The earliest data records in IGRA go back to 1905. However, time records as well as the temporal and vertical resolution differs between the stations included in this archive (Durre et al., 2006). The IGRA data 205 used in this study covers 17 different height levels from the surface up to 10 hPa and only a selected time period from 1969 to 1998 is considered. It has to be noted that the spatial distribution of the IGRA stations is rather sparse in the SH, especially over higher latitudes. However, there is a good agreement of the maximum negative temperature trend between the IGRA data and estimates from other radiosonde products Young et al. (2013, see Table 2).

210
Our analysis focuses on the evaluation of climatologies and linear trends, in particular for the SH ozone trend and its impact on other climate variables. The climatologies and trends are calculated from the ensemble average of all nine ensemble members.
Climatological differences between the simulations with and without interactive chemistry are displayed as: Chem ON minus Chem OFF for the time period 1955 -2013 to illustrate the effects of interactive chemistry. Statistical significance at the 95% level is tested using a two-sample t-test. Furthermore, we consider lead-lag correlations between ozone at 50 hPa and polar cap 215 dynamical heating rates at each level for the same time period for each ensemble member separately. Before the correlation coefficients are calculated a slowly-varying climatology (Gerber et al., 2010) is removed from the data to avoid correlating trends. Afterwards, the ensemble mean of the correlation coefficients is calculated. Statistical significance at the 95 % level is chosen to be given for each point in which all individual ensemble members reach a p-value ≤ 0.05.
The SH trends for polar cap temperature, heating rates, and zonal mean zonal wind (60 • -70 • S) are calculated for the period 220 of 1969 -1998, which is marked by a strong ozone decline in October in the SH lower polar stratosphere (Fig. 4a). We restrict the trend analysis to this period for a better comparison to earlier model and observational studies (e.g., Calvo et al., 2012;Young et al., 2013;Calvo et al., 2017). To determine the statistical significance of the linear trend differences, a new time series is produced by taking the difference between the time series of the ensemble means. This approach reduces noise levels by subtracting the variability of the individual time series and favors the identification of real trend differences (Santer et al., 225 2000). The trend significance is estimated using the commonly used Mann-Kendall test at a confidence level of 95%.
To address the impact of interactive chemistry on inter-annual variability, the timescale of the Southern Annular Mode (SAM) is evaluated for Chem ON, Chem OFF, and Chem OFF 3D following the procedure of Simpson et al. (2011) and Ivanciu et al. (in prep). The SAM index used for this calculation is determined for each ensemble member separately and follows the definition by Gerber et al. (2010) using the first EOF of daily zonal mean geopotential height, which is previously adjusted by 230 removing the global mean and a slowly varying climatology to remove variability on decadal timescales. For the calculation of the SAM timescale, the autocorrelation function of each SAM index is calculated and smoothed. Then the e-folding timescale is estimated by using a least squares fit to an exponential curve up to a lag of 50 days to the smoothed autocorrelation function (Simpson et al., 2011, and Ivanciu et al., in prep).
3 The impact of stratospheric chemistry on southern hemispheric climate and trends 235 It was shown in Haase and Matthes (2019) (in the following referred to as HM19) that including interactive chemistry leads to a stronger and a colder polar stratospheric vortex on both hemispheres. The differences between the interactive and specified chemistry simulations were shown to be largest during mid-winter and in spring when ozone chemistry gets important. These results were based on only one model realization per experiment. Here, an ensemble of 9 realizations per experiment is used to evaluate the impact of interactive chemistry on the SH climatology and trend. In a first step the climatological difference 240 between Chem ON and Chem OFF is analysed for the whole model period . Figure 1 shows the seasonal evolution of zonal mean zonal wind at 10 hPa and zonal mean temperature at 30 hPa similar to Figure 2 in HM19. It shows that the main results presented in HM19 are reproduced by the 9-member ensemble. Including interactive chemistry leads to a stronger PNJ (Fig. 1a) and a colder polar stratospheric vortex (Fig. 1b) on both hemispheres. The significance of this difference is larger as compared to HM19, while the amplitudes of the differences are smaller. This is an expected feature from taking the 245 average over 9 ensemble members compared to only considering a single realization since averaging reduces the imprints of natural variability; the forced signal is therefore easier to detect. In the Chem ON ensemble, a significantly stronger PNJ is apparent from September until April in the NH, and from September to December in the SH (Fig. 1a). The months that show the largest differences between the interactive and specified chemistry ensemble agree well with HM19: January and March in the NH and October to November in the SH. The impact of interactive chemistry on lower stratospheric temperatures is even 250 more significant showing a cooler polar lower stratosphere covering almost the whole year (with the exception of January and February in the SH) and a warmer lower stratosphere between 40 • S and 40 • N (Fig. 1b). This result is consistent with a weaker shallow branch of the BDC in the model experiment with interactive chemistry discussed in HM19 (see also Suppl. Fig. S3).

Stratospheric mean state
The climatological differences for the SH polar stratosphere are depicted in Figure 2. Although the ozone and short-wave 255 (SW) heating climatologies are almost identical between Chem ON and Chem OFF (not shown), there is still a difference in the climatology of the polar cap temperatures that is also imprinted in the strength of the PNJ ( Fig. 2a and b). The temperature difference is characterized by lower values in the lower and middle stratosphere from May until November, with maximum differences in September and October, followed by higher values peaking in December (Fig 2b). This pattern compares well 8 https://doi.org/10.5194/acp-2020-441 Preprint. Discussion started: 3 June 2020 c Author(s) 2020. CC BY 4.0 License. with the one found by Neely et al. (2014) but shows a higher statistical significance, also covering the stratospheric levels above 260 30 hPa during all seasons. As mentioned earlier, this temperature difference is also reflected in the strength of the PNJ (Fig.   2a), which is stronger in Chem ON, especially during November and December when the strength of the PNJ normally starts to decrease. Following HM19, long-wave (LW) and dynamical heating rate climatologies are considered to investigate the polar cap temperature difference between Chem ON and Chem OFF ( Fig. 2c and d) in more detail. In agreement with the findings of HM19 for the NH, Figure 2d shows that also on the SH, the dynamical heating rates are responsible for the temperature 265 differences between Chem ON and Chem OFF, whereas the LW heating rates tend to damp the temperature tendencies caused by the dynamics (Fig. 2c).
The impact of the dynamics onto the mean state of the stratosphere suggests that similar feedbacks as compared to the NH can be expected also for the SH. Figure 3 shows a lag correlation between polar cap ozone at 50 hPa and the dynamical heating rates with ozone leading the dynamics by 15 days following the procedure in HM19. The climatological zonal mean 270 zonal wind for values ≤ 20 ms −1 is also depicted (contours). The negative correlation between ozone and dynamical heating rates represents the negative feedback discussed earlier: under weaker westerly wind background conditions, ozone depletion and the associated radiative cooling lead to a westerly acceleration in the lower stratosphere that enhances upward wave propagation and dissipation which eventually leads to an earlier break-down of the stratospheric polar vortex. This feedback is also apparent in the Chem OFF simulation but it is weaker in amplitude. This is different compared to the findings for the NH, 275 where the negative feedback was not found for the specified chemistry version of the model. We suppose that the presence of this correlation in Chem OFF is due to the fact that a part of the negative feedback is included in the prescribed ozone field, which is characterized by a strong negative trend in ozone (see following section), which dominates ozone variability on the SH. Apart from the negative feedback also the positive feedback during stronger westerly background winds can be detected in Figure 3 in the lowermost stratosphere.

280
Figures 1 and 2 showed that including interactive chemistry leads to a stronger PNJ and a colder polar stratospheric vortex, especially towards the end of the vortex lifetime. The differences between Chem ON and Chem OFF are mainly due to differences in dynamical heating, which we attribute at least partly to the representation of chemical-dynamical interactions (feedbacks). Especially, the dominant negative feedback, which starts in November in the upper stratosphere and peaks in January in the lower stratosphere is stronger in Chem ON contributing to the enhanced dynamical heating in this ensemble, 285 which also starts in November (Fig. 2). Since this period is strongly influenced by ozone depletion, we also expect an impact of chemical-dynamical interactions onto the stratospheric and tropospheric trends associated with ozone depletion. This will be the focus for the remainder of our analysis.

Stratospheric trends
Figure 4a exemplarily depicts the temporal evolution of ozone mixing ratios at 50 hPa in October, which represents the max-290 imum ozone depletion in our model simulations (Fig. 4b). The ozone trend in Chem ON agrees well among the different ensemble members (gray lines in Fig. 4a), starting of with a weak negative trend from 1955 to the late 1960s, followed by a strong negative trend, which levels off in the mid-1990s. To address the model's response to SH ozone depletion the period of facilitate comparisons to earlier studies using the WACCM model or observational data (Table 2).
Due to the ozone depletion from 1969-1998, which reaches its maximum of about -0.9 ppmv decade −1 in the lower stratosphere during October (Fig. 4b) a decrease in polar lower stratospheric temperatures can be observed (Fig. 5). In Chem ON the negative temperature trend maximizes with -6.6 K decade −1 in December at about 90 hPa (Fig. 5a) and is therefore stronger and delayed by one month compared to observations, which show a maximum trend of -4.0 K decade −1 during November at about 100 hPa (IGRA; Fig. 5b and Table 2). The overestimated temperature trend in Chem ON, however, is quite common in 300 CCMs (e.g., Eyring et al., 2010;Young et al., 2013). It compares well to the published WACCM4 trend (see Calvo et al., 2012, 2017, and Table 2) but is larger than the trend found in WACCM-CCMI (Calvo et al., 2017). The reduction of the trend found in WACCM4 compared to WACCM-CCMI can be explained by a reduction of the cold pole bias in the model (Calvo et al., 2017).
Although a reduction of the cold pole bias was also achieved in our WACCM version by implementing a few changes to the model code (see Methods and Supplement for details), the trend is not significantly weaker compared to the original WACCM4 305 version analyzed in Calvo et al. (2012). In agreement with the overestimated temperature trend, also the ozone trend is with -0.9 ppmv decade −1 (Fig. 4b) rather in agreement with WACCM4 than with WACCM-CCMI (Calvo et al., 2017). This indicates that the reduction of the cold pole bias implemented here, is not sufficient to reproduce the WACCM-CCMI trend. However, the comparison between our ensembles of Chem ON and Chem OFF simulations is still very suitable to address the question of how important ozone feedbacks are for the stratospheric and tropospheric circulation.

310
The negative temperature trend due to ozone depletion is followed by a positive temperature trend at altitudes above 30 hPa in the model and observational data (Fig. 5). This positive temperature trend coincides with a positive ozone trend (Fig. 4b). The ozone trend, however is not the only contributor to this temperature trend pattern. Figure 6 depicts the different heating rate trends, that combine to the temperature trend pattern. The SW heating rate trend (Fig. 6a) resembles the trend in ozone (Fig.   4b) during the time of the year when solar radiation is available at such high latitudes. It explains the negative temperature 315 trend in the lower stratosphere and parts of the positive temperature trend following it in the upper stratosphere. However, also long-wave (LW, Fig. 6b) and dynamical (DYN, Fig. 6c) heating rate trends contribute to the temperature trend. Especially, the dynamical heating rate trend is decisive for that part of the temperature trend pattern that can not be explained by the SW heating trend. There is a strong positive trend in dynamical heating starting in November in the upper stratosphere and propagating down to about 100 hPa in January. This positive trend can be explained by a stronger descent of air masses 320 through an increase in the residual meridional circulation, i.e. a strengthening of the BDC, during the ozone depletion period (e.g., Keeble et al., 2014). This dynamical response is due to the negative feedback (compare Fig. 3)  , which is also apparent in Figure 3. The LW heating rate trend mostly damps the signals from the SW and dynamical heating rate trends (Fig. 6).
Is this feedback loop at all represented in Chem OFF? Figures 7a and b show the 1969-1998 temperature trend for Chem OFF and the difference of the trend between Chem ON and Chem OFF. By construction, the polar cap ozone trend is the same between the two ensembles; and so is the trend in SW heating rates (not shown). Nevertheless, with a maximum of -5.9 330 K decade −1 in November, the maximum temperature trend in Chem OFF is weaker compared to Chem ON (-6.6 K decade −1 ) and occurs earlier, which is indicative of a weaker representation of the positive chemical-dynamical feedback (positive correlation in Fig. 3) when ozone is prescribed rather than calculated interactively. This gets clearer when the difference between Chem ON and Chem OFF is considered (Fig. 7b): The largest differences occur in December and January, and are characterized by a longer lasting cooling trend in Chem ON in the lower stratosphere (positive feedback) as well as by a stronger warming 335 trend starting in December in the upper stratosphere reaching down into the lowermost stratosphere in February and March (negative feedback). These differences can mainly be attributed to stronger trends in the dynamical heating rates in Chem ON (Figs. 7c and d), which is due to the better representation of feedbacks between chemistry and dynamics in the fully-coupled chemistry model setup.
In accordance to the findings of HM19, the stronger dynamical warming can be explained by negative feedbacks between ozone 340 chemistry and model dynamics. During weak westerly winds, an unusually low ozone concentration can lead to enhanced upward planetary wave propagation by extending the lifetime of the westerly wind regime, which enhances a descent over polar latitudes resulting in an additional adiabatic warming. Apart from the negative feedback, which was found to be apparent also in the NH, a positive feedback can be detected in the SH ozone depletion period during stronger westerly background winds.
It is statistically significant only for 7 out of 9 members (not shown) and restricted to the lowermost stratosphere. This positive 345 correlation could explain the stronger dynamical cooling in Chem ON compared to Chem OFF (Fig. 7). It has to be noted that Chem OFF is able to represent this feedback pattern to a certain extent (Fig. 3) because of the strong ozone signal that the model is forced with (parts of the feedback can be considered to be included in the prescribed ozone fields).
The negative temperature trend in the lower polar stratosphere increases the meridional temperature gradient and leads to a strengthening of the PNJ, especially towards the end of the polar vortex lifetime as discussed before. Figures 8a and b show 350 that the maximum trend in zonal mean zonal wind in the PNJ region is stronger in Chem ON (9.2 ms −1 decade −1 ) compared to Chem OFF (7.8 ms −1 decade −1 ). The largest differences in the zonal mean zonal wind trend can found in the middle stratosphere during December (Fig. 8c), which supports our earlier argumentation about the characteristics of chemical-dynamical feedbacks in the two ensembles. Namely, that an extension of the vortex lifetime, which is stronger in Chem ON compared to Chem OFF, favors the occurrence of the negative feedback. However, in both ensembles a significant zonal mean zonal wind 355 trend can be found also at the surface from November through February, which will be investigated in the following with a focus on the austral summer season (DJF). Figure 9 shows the 1969-1998 trend for zonal mean zonal wind with latitude and height (color shading) along with the climatological wind over the same period (contours) for DJF. It is evident that the strengthening of the PNJ is connected also to 360 a strengthening of the tropospheric jet and its poleward displacement in agreement with earlier studies (e.g., Thompson and Solomon, 2002;Son et al., 2008;Eyring et al., 2013). In Chem OFF, the strengthening of the poleward flank of the tropospheric