The Role of Ozone Depletion in the Lack of Cooling in the Antarctic Upper Stratosphere during Austral Winter

Observations and model simulations show substantial cooling in the global stratosphere since 1979, especially in the Antarctic due to the strong radiative effect of severe ozone depletion in austral spring and summer between the late 1970s and the late 1990s (Randel and Wu, 1999; Solomon, 1999; Ramaswamy et al., 2001; Thompson and Solomon, 2002; Hu, 2006). However, most studies only focus on the Antarctic stratospheric temperature changes in the lower stratosphere during austral spring (e.g., Sekiguchi, 1986; Randel and Wu, 1999; Gillett and Thompson, 2003; Hu and Fu, 2009; Ivy et al., 2016). Less attention has been paid to the Antarctic upper stratospheric temperature. In contrast to the cooling in the lower stratosphere, some studies recently have shown that the Antarctic upper stratosphere has experienced a lack of cooling since the late 1970s (Randel et al., 2016). Using measurements from the Stratospheric Sounding Unit (SSU), Randel et al. (2016) pointed out that the temperature trends in the Antarctic upper stratosphere are near zero, with significant cooling at other latitudes over 1979–97.

Ball et al. (2016) pointed out that the upper stratospheric temperature has shown significant responses to changes in the upper deep branch of the Brewer–Dobson circulation (BDC) that lead to adiabatic heating, or cooling. The BDC is mainly driven by the midlatitude upward-propagating planetary and gravity waves through changing westward momentum as they break there (Haynes et al., 1991; Holton et al., 1995; Butchart, 2014). Wave forcing and the mean state of the flow could interact with each other (Charney and Drazin, 1961; Holton and Mass, 1976). Changes in wave forcing could also affect the spatial distribution of ozone by adjusting the strength of the BDC to change the ozone transport (e.g., Brewer, 1949; Dobson, 1956; Rind et al., 2002; García-Herrera et al., 2006; Jiang et al., 2007; Weber et al., 2011). Sridharan et al. (2012) and Nath and Sridharan (2015) showed that the dynamical perturbations at mid-to-high latitudes can directly influence the changes in ozone and temperature. Ozone can also in turn influence the stratospheric circulation (e.g., Hu and Tung, 2003; Nathan and Cordero, 2007; Hu et al., 2015; Zhang et al., 2020, 2022). For example, Hu et al. (2015) pointed out that stratospheric ozone depletion (recovery) leads to weakening (strengthening) of the Arctic polar vortex, and cooling of the Arctic polar vortex was found to be dynamically induced via modulation of planetary wave activity by stratospheric ozone increases during boreal winter. Several modeling studies also found that the BDC becomes stronger (weaker) in response to ozone depletion (recovery) (e.g., McLandress et al., 2010; Polvani et al., 2011, 2018, 2019; Lin and Fu, 2013). The aim of this paper is to gain a better understanding of the coupling between temperature, ozone, and the BDC in the upper stratosphere.

Ozone-depleting substances (ODSs) and greenhouse gases (GHGs) are considered to be two major factors responsible for global stratospheric temperature changes (Stolarski et al., 2010). Previous studies have suggested a strong decreasing trend in Antarctic stratospheric ozone during the late 20th century and a dramatic increasing trend in global GHGs since the industrial revolution (Randel and Wu, 1999; Thompson and Solomon, 2002; Gillett et al., 2011; Polvani et al., 2011; Checa-Garcia et al., 2016). The ozone depletion and GHGs contribute significantly to the observed cooling in the stratosphere due to their strong radiative cooling effect (Ramaswamy et al., 1996, 2001; Shindell and Schmidt, 2004). Therefore, the lack of cooling in the Antarctic upper stratosphere is likely due to dynamical processes. There is a strong link between ozone and stratospheric temperature, showing negative dynamical feedbacks in the polar upper stratosphere (Albers and Nathan, 2013), and the feedbacks associated with ozone and planetary waves are height dependent. Moreover, planetary waves are also strongly modulated by GHGs, with stronger planetary waves forcing from the troposphere into the stratosphere when GHG concentrations increase (Butchart and Scaife, 2001; Kodera et al., 2008; Bell et al., 2010). Correctly modeling the dynamical response to ODSs and GHGs is essential to understand the processes that drive the temperature changes. Herein, we investigate the role of changes in ODSs and GHGs in the Antarctic upper stratospheric temperature, using simulations from the Chemistry–Climate Model Validation Activity Phase 2 (CCMVal-2) and Chemistry–Climate Model Initiative (CCMI) projects (Eyring et al., 2008, 2013), as well as the Coupled Model Intercomparison Project Phase 6 (CMIP6; Boer et al., 2016).

In the present study, the 1979–2014 monthly mean temperature from the SSU channel 3 (SSU3) on board the National Oceanic and Atmospheric Administration (NOAA) operational satellites is obtained from the NOAA/STAR website (http://www.star.nesdis.noaa.gov/smcd/emb/mscat/) [see Zou and Qian (2016) for more details on this dataset]. The 1961–2014 monthly mean temperature, ozone, and zonal wind are from 12 Chemistry–Climate Model (CCM) simulations (whose output is available for reference and two sensitivity simulations), except that the ozone output is unavailable for the NIWA-UKCA model from the CCMI dataset. The output of six simulations is obtained from the CCMVal-2 (Eyring et al., 2006, 2010; Morgenstern et al., 2010), and the other six simulations are from the CCMI (Eyring et al., 2013; Morgenstern et al., 2017). These models are summarized in Table 1. Three different forcing scenarios are considered here: “REF-B2” for CCMVal-2 and “REF-C2” for CCMI in which the reference models are forced by time-varying concentrations of ODSs and GHGs (All forcings); “SCN-B2b” for CCMVal-2 and “SEN-C2-fODS” for CCMI with fixed ODSs at 1960 levels; “SCN-B2c” for CCMVal-2 and “SEN-C2-fGHG” for CCMI with fixed GHGs at 1960 levels. More details can be found in Morgenstern et al. (2010, 2017). Similar to the above models, three types of simulations from CMIP6 are also employed: historical simulations and two new historical simulations with stratospheric ozone-only and well-mixed GHG-only forcing. Details are shown in Table 2.

Other meteorological datasets, including observed ozone from the Stratospheric Water and Ozone Satellite Homogenized (SWOOSH) datasets (Davis et al., 2016), as well as monthly mean temperature, 6-hourly temperature, and zonal wind from the ECMWF fifth-generation reanalysis, ERA5 (Hersbach and Dee, 2016; Li et al., 2020), and the Japanese 55-year Reanalysis, JRA55 (Ebita et al., 2011; Kobayashi and Iwasaki, 2016), are employed in this paper. The meridional eddy heat flux is obtained from the zonal mean of the product of the temperature and the eddy component of the zonal wind based on 6-hourly datasets. The vertically integrated temperature trends calculated in this study are based on the SSU3 weighting functions. Details of the weighting functions can be found in Randel et al. (2016).

The divergence of the Eliassen–Palm (EP) flux is a key diagnostic for wave–mean-flow interaction and can be used to describe the resolved wave breaking (Andrews and McIntyre, 1976, 1978; Andrews et al., 1987). The EP flux is given as (see Andrews et al., 1987):

where $u$, $v,$ and $w$ are the zonal, meridional, and vertical velocity, respectively. $\theta $ is the potential temperature. $a$ is the radius of the Earth.$ \text{ }{\rho }_{0} $ is the air density. $f$ is the Coriolis parameter. $H$ is the density scale height taken as 7 km. ${P_0}$=1000 hPa, and P is pressure. Subscripts $z$ and $\varphi $ denote derivatives with respect to pressure and latitude, respectively. An overbar represents the zonal mean, and a prime symbol denotes departures from the zonal mean. The divergence of the EP flux calculated in this paper is based on 6-hourly datasets.

The thermal wind equation indicates that the horizontal temperature gradients are closely associated with vertical wind shear:

where R is the specific gas constant for dry air and y is northward distance.

The refractive index (n) of planetary waves developed by Andrews et al. (1987) based on the form of Matsuno (1970) is applied in this study and was expanded by Hu and Tung (2002):

where

is the meridional gradient of the zonal mean potential vorticity. Here, $k$, $N$, and $ \Omega $ denote the zonal wave number (ZWN hereafter), buoyancy frequency, and Earth rotation frequency, respectively. Expansion of the third term on the right-hand side of Eq. (7) yields:

Nekola and White (1999) proposed a method which can be used to calculate the difference in slope between two datasets [e.g., (x1, y1) and (x2, y2)]. Although it was initially built to calculate the difference in slope between the regression lines of distance decay plots, it could be extended to anything else. First, the y values of the two datasets compared were rescaled to a common mean. Then, the y value for each pair along with the corresponding x was randomly reassigned to one of the two datasets. After this randomization had been carried out, linear regressions were applied to determine the slope of the y function for each of the randomized datasets, and the absolute value of the difference between the two slopes was determined. This procedure was repeated 999 times. The difference between the slopes of the original datasets was also calculated and compared with the distribution of the differences between the slopes of the 999 randomized datasets in order to determine its significance level. Details of this method can be seen in Nekola and White (1999), Steinitz et al. (2005), and Steinitz et al. (2006).

Changes in the Antarctic upper stratospheric temperature are good indicators for the chemistry–climate coupling in the southern upper stratosphere. In the late 20th century, the increased GHGs and decreased stratospheric ozone are considered to be significantly related to the strong cooling in the global stratosphere due to their strong radiative effect. However, using SSU3 observations for 1979–97, Randel et al. (2016) found that the annual mean temperature trends in the upper stratosphere are near zero at southern high latitudes, but there is strong cooling at other latitudes, indicating a significant difference in temperature trends between southern high and low latitudes. To further investigate this problem, the latitudinal distribution of linear trends for zonal mean temperature for each season are examined (Fig. 1). It is found that the difference in temperature trends between southern high and low latitudes is most significant in austral winter (Fig. 1c). However, there is no significant difference in temperature trends between northern high and low latitudes during austral summer (Fig. 1a). Compared with Fig. 1a, in austral autumn and spring, the temperature trends in the Antarctic upper stratosphere are relatively weaker (Figs. 1b, 1d) and may be influenced by lower stratospheric ozone changes and the associated dynamical effects. Therefore, this study focuses on the austral winter temperature trends in the Antarctic upper stratosphere. Using Solar Backscatter Ultraviolet (SBUV) data, Hood et al. (1993) estimated the altitude, latitude, and seasonal dependences of stratospheric ozone trend. Results show that the maximum negative trend occurs in the southern upper stratosphere during austral winter, making it easier to isolate the causality of the temperature trend there. Considering that ozone depletion and increasing GHGs contribute significantly to stratospheric cooling due to their strong radiative cooling effect, why is there a lack of cooling in the Antarctic upper stratosphere during austral winter? The topic of this paper mainly focuses on the lack of cooling in the Antarctic upper stratosphere in austral winter. Temperature trends in other seasons and in the northern hemisphere will be discussed in the final section.

Figure 1.  Latitude profiles of linear trends for monthly zonal mean SSU3 temperature in austral (a) summer, (b) autumn, (c) winter, and (d) spring for 1979–97. Thick black solid lines indicate that the trend at each latitude in the range of 90°–30°S is significantly different from the trend at the equator of 0° at the 95% confidence level. For the specific method of significance test between the two trends, please refer to section 2.3.

What is responsible for the lack of cooling in the Antarctic upper stratosphere in austral winter? Zonal mean temperature trends in austral winter derived from three types of simulations are shown in Fig. 2. An obvious difference in temperature trends between southern high and low latitudes can be found in the ensemble mean with all forcings and with fixed GHGs, indicating that the lack of cooling in the upper stratosphere at southern high latitudes may be the result of increased ODSs during 1979–97 (Fig. 2a). Note that 12 chemistry–climate models from the CCMVal-2 and CCMI datasets are analyzed because only these models have all reference and sensitivity simulations available. To test the representativity of these models, we compared the results from the 12 reference simulations with those from all 34 reference simulations obtained from CCMVal-2 and CCMI simulations (Table 3). Results from the 12 members are similar to those from the 34 members, suggesting that the former is a good representation of the latter. In addition, the results from two reanalysis datasets (based on the ERA5 and JRA55) are also in good agreement with the results from satellite observations, with a distinct lack of upper stratospheric cooling in the Antarctic. However, the peak from using the two reanalysis datasets is near 60°S and is more equatorward than the peak from using the satellite observations, which may be attributed to too few observations in the upper Antarctic stratosphere. To address potential sensitivity due to the short time period, results from three types of simulations from CCMs and CMIP6 for a longer period of 1961–2014 are shown in Figs. 2c–f, and they are in good agreement with the results for 1979–97. Considering the consistency of the results from CCMs and CMIP6, and our focus on chemical–dynamic processes, an analysis based on the CCMs is given in the following sections.

Figure 2.  (a) Same as Fig. 1c. The grey line shows the temperature trends derived from NOAA. The brown, green, dashed orange, and dashed blue lines represent the ensemble-mean trends of CCMs. The brown line represents the ensemble-mean temperature trends of CCMs (Table 1, 12 simulations) with all forcings. The green line is the same as the brown line, but for 34 simulations (Table 3). Dashed orange and dashed blue lines represent the ensemble mean with fixed ODSs and GHGs (Table 1, 12 simulations), respectively. Solid pink and purple lines are from reanalysis data of ERA5 and JRA55 datasets, respectively. (b) is the same as (a), but for 1998–2014. (c) and (d) are similar to (a) and (b), but for the ensemble mean (Table 1, 12 members) of the longer periods of 1961–97 and 1998–2014. (e) and (f) are the same as (c) and (d), but for the ensemble mean of CMIP6 simulations (Table 2, 26 members).

Latitude–height cross sections of ensemble-mean linear trends in the stratospheric temperature are presented in Fig. 3. Significant differences in temperature trends between southern high latitudes and low latitudes can be seen in simulations with all forcings for 1961–97 (Fig. 3a). This may be attributed to the ODSs forcing (Fig. 3e), which produces a significant contrast of temperature trends between the two regions. Furthermore, we quantitatively investigate the temperature trend difference between the southern high (60°–90°S) and low latitudes (0°–30°S). The difference in temperature trends between the two regions exceeds 0.32 K (10 yr)^–1 in both the reference and fixed-GHG simulations, with only about –0.03 K (10 yr)^–1 in the fixed-ODSs simulations. Moreover, the trend difference between the two sensitivity simulations is significantly different (at the 99% confidence level) in the first period (1961–97). Temperature trends from CMIP6 simulations (Fig. 4) are similar to the results from CCMs (Fig. 3), showing the lack of cooling in the Antarctic upper stratosphere with ozone forcings during 1961–97. These findings suggest that the lack of cooling in the Antarctic upper stratosphere is mainly induced by ozone depletion due to increased ODSs. During 1998–2014, the Antarctic warming may be a statistical artifact due to the short length of this period, and there is no significant warming for a longer period of 1998–2099.

Figure 3.  Latitude–height cross sections of ensemble-mean linear trends for zonal mean temperature [K (10 yr)^–1] in austral winter over (a, c, e) 1961–97 and (b, d, f) 1998–2014. (a) and (b) are for the reference ensemble mean. (c, d) and (e, f) are the same as (a, b), but for the ensemble mean with fixed ODSs and GHGs, respectively. Detailed simulations can be seen in Table 1. The dotted areas indicate that the trends are statistically significant at 95% confidence levels.

Figure 4.  Same as Fig. 3, but for the ensemble mean of CMIP6 simulations (Table 2, 26 members).

The above results suggest that the increased ODSs are the main contributor to the lack of cooling in the Antarctic upper stratosphere during 1961–97. However, this is opposite to the direct radiative cooling related to ozone depletion during that period (Figs. 5 and 6). Thus, the lack of cooling in the Antarctic upper stratosphere may be related to dynamical processes with more upward planetary waves into stratosphere indirectly induced by increased ODSs.

Figure 5.  Latitude–height cross sections of ensemble-mean linear trends for zonal mean ozone [10⁻⁷ mol mol^–1 (10 yr)^–1] in austral winter for 1961–97 (left panel) and 1998–2014 (right panel). (a) and (b) are for the reference ensemble mean. (c, d) and (e, f) are the same as (a, b), but for the ensemble mean with fixed ODSs and GHGs, respectively. Detailed simulations can be seen in Table 1. The dotted areas indicate that the trends are statistically significant at 95% confidence levels.

Figure 6.  Latitude–height cross sections of linear trends for observed zonal mean ozone [ppmv (10 yr)^–1] in austral winter for (a) 1984–97 and (b) 1998–2014. The dataset is from SWOOSH. The dotted areas indicate that the trends are statistically significant at 95% confidence levels.

Ozone depletion induced by an increase in ODSs can cool the middle and high latitudes via the radiative effect, which would cause a larger meridional temperature gradient between high and low latitudes. By the thermal wind balance, the midlatitude baroclinic zone of the atmosphere is associated with a planetary-scale meridional temperature gradient between the equator and the pole. Zonal wind trends for the three groups of simulations are presented in Fig. 7. For reference and fixed-GHG simulations, significant vertical shear is shown in the midlatitude zonal winds for 1961–97 in Figs. 7a and 7e, consistent with the meridional gradient in the temperature trends displayed in Figs. 3a and 3e and that in the ozone trends shown in Figs. 5a and 5e.

Figure 7.  As in Fig. 5, but for zonal mean wind [m s^–1 (10 yr)^–1].

Results from reanalysis datasets are also examined. Signals of a significant lack of cooling are seen in the Antarctic upper stratosphere during austral winter for 1979–97 (Fig. 8a), consistent with the observations (Fig. 2). The eddy heat flux is a useful proxy for the vertical component of EP flux (Andrews et al., 1987) and can represent wave energy propagating into the stratosphere (Newman et al., 2001; Eyring et al., 2005). Figure 8c shows that the lack of cooling in the Antarctic upper stratosphere is accompanied by a negative trend in the eddy heat flux, indicating more planetary waves propagating into the southern stratosphere. It is in good agreement with the results shown in Fig. 8e. Subsequently, more planetary waves break up in the upper stratosphere at southern high latitudes during austral winter for 1979–97 (Fig. 8g). In contrast, the eddy heat flux trends and EP flux convergence trends afterward did not change significantly.

Figure 8.  (a) Same as Fig. 3, but for the average of ERA5 and JRA55. (c) Latitude–height cross sections of linear trends for meridional eddy heat flux [K m s^–1 (10 yr)^–1] in austral winter for 1979–97. (e) The climatology (black line, K m s^–1) and linear trends [red line, K m s^–1(10 yr)^–1] of vertically integrated meridional eddy heat flux in austral winter for 1979–97. (g) Latitude–height cross sections of linear trends for EP Flux divergence [m s^–1 d^–1 (10 yr)^–1] in austral winter for 1979–97. (b, d, f, h) Same as (a, c, e, g), but for 1998–2014.

Previous studies have found that stratospheric ozone depletion may change the temperature gradient or potential vorticity gradient, which are closely related to the wave refractive index (e.g., Hu et al., 2015; Wang et al., 2021). Figure 9 shows wave refractive index trends during austral winter. As discussed by Matsuno (1970), it is expected that planetary waves of wave number k are able to propagate in regions where $ {n_k}^2 $>0 and are refracted from regions where $ {n_k}^2 $<0, and the larger the $ {n_k}^2 $ in a region, the easier it is for planetary waves to propagate there. Our results show that in austral winter, planetary waves have a great chance to propagate upward from the lower stratosphere to the upper stratosphere at southern middle and high latitudes for zonal wave 1 (ZWN1) during 1979–97 (Fig. 9a). However, the waves are more likely to propagate to the Antarctic middle stratosphere during 1998–2014 (Fig. 9b). Therefore, dynamical warming is more plausible in the Antarctic upper stratosphere during 1979–97. Similar conclusions could be drawn for zonal waves 2 and 3 (ZWN2 and ZWN3). The above conclusions suggest that the lack of cooling in the Antarctic upper stratosphere is mainly attributed to dynamical processes.

Figure 9.  (a) Latitude–height cross sections of linear trends for refractive index squared (a²n²) for ZWN1 (first row) planetary waves (the average of ERA5 and JRA55) in austral winter for 1979–97. (c) and (e) are the same as (a), but for ZWN2 and ZWN3, respectively. (b, d, f) are the same as (a, c, e), but for 1998–2014.

Using observations and multi-model ensemble data from CCMs and CMIP6, we evaluated upper stratospheric temperature trends. Observations show that in the late 20th century, Antarctic upper stratospheric temperature trends exhibit distinct seasonal variation with an obvious lack of cooling in austral winter, which can be reproduced by chemistry–climate models.

ODSs and GHGs are two major contributors to variations in the stratospheric temperature. To further distinguish their contributions, three sets of integrations are investigated in this paper. Our results suggest that the lack of cooling in the Antarctic upper stratosphere is mainly dynamically driven by increased ODSs in the late 20th century. Ozone depletion related to increased ODSs could strengthen the meridional temperature gradient in the middle latitudes and allow more planetary waves into the upper stratosphere, which in turn could weaken zonal winds through the breaking of planetary waves. This would intensify the meridional residual circulation (Fu et al., 2019), resulting in enhanced adiabatic heating in the Antarctic upper stratosphere. These changes suggest that the lack of cooling in the Antarctic upper stratosphere is mainly due to enhanced wave-driven dynamical heating.

Unlike the Antarctic temperature trends in austral winter, no significant lack of cooling is found in the Arctic upper stratosphere in austral summer. This hemispheric asymmetry is probably due to large internal variability in the northern polar stratosphere, as these models simulate a temperature gradient that is similar to the simulated Antarctic counterpart (now shown) but is missing in observations (Fig. 1a) and reanalysis datasets. In austral autumn and spring, it is difficult to separate the contributions from lower and upper stratospheric ozone for the observed lack of cooling in the polar upper stratosphere (Figs. 1b, 1d). In contrast, in austral winter, it is easier to isolate the causality of the temperature trend there.

Although the reanalysis data can reproduce the lack of cooling in the Antarctic and its related dynamical mechanism, some issues need to be noted. The quality of the two reanalysis datasets at southern high latitudes is relatively poor. In addition to the latitude of peak warming being a little too far northward compared to the satellite observation, we have also found large residuals in the balance of the thermodynamic equation. This indicates that the quality of the reanalysis data in the upper Antarctic stratosphere needs to be further improved. Improved observations in the upper Antarctic stratosphere are called for in the future.

Acknowledgements. This project was supported by Grant Nos. 41875047 and 91837206 from the National Natural Science Foundation of China (NSFC) and Grant No. JIH2308007 from Fudan University. We thank NOAA for producing SSU satellite temperature datasets, SWOOSH for the ozone dataset, ECMWF and JMA for meteorological fields from ERA5 and JRA55, the WCRP SPARC Chemistry–Climate Model Validation (CCMVal) Activity, the SPARC/IGAC Chemistry–Climate Model Initiative (CCMI), CMIP6, DAMIP, the participating modelling groups, and the ESGF centres (see details on the CMIP Panel website at http://www.wcrp-climate.org/index.php/wgcm-cmip/about-cmip) for organizing and coordinating the model data analysis activity, as well as the British Atmospheric Data Centre (BADC) for collecting and archiving the model output. We also acknowledge Dr. William RANDEL from NCAR and Dr. Lorenzo POLVANI from Columbia University for providing helpful comments for this study.