Southern-Hemisphere high-latitude stratospheric warming revisit

Previous studies showed significant stratospheric warming at the Southern-Hemisphere (SH) high latitudes in September and October over 1979–2006. The warming trend center was located over the Southern Ocean poleward of the Western Pacific in September, with a maximum trend of about 2.8 K/decade. The warming trends in October showed a dipole pattern, with the warming center over the Ross and Amundsen Sea, and the maximum warming trend is about 2.6 K/decade. In the present study, we revisit the problem of the SH stratospheric warming in the recent decade. It is found that the SH high-latitude stratosphere continued warming in September and October over 2007–2017, but with very different spatial patterns. Multiple linear regression demonstrates that ozone increases play an important role in the SH high-latitude stratospheric warming in September and November, while the changes in the Brewer-Dobson circulation contributes little to the warming. This is different from the situation over 1979–2006 when the SH high-latitude stratospheric warming was mainly caused by the strengthening of the Brewer-Dobson circulation and the eastward shift of the warming center. Simulations forced with observed ozone changes over 2007–2017 shows warming trends, suggesting that the observed warming trends over 2007–2017 are at least partly due to ozone recovery. The warming trends due to ozone recovery have important implications for stratospheric, tropospheric and surface climates on SH.


Introduction
It is well known that the ozone layer experienced depletion in the last quarter of the twentieth century. Especially, severe ozone depletion was found in the Antarctic stratosphere in austral spring, i.e., the so-called Antarctic Ozone Hole. Associated with ozone depletion, the Antarctic stratosphere demonstrated strong cooling trends in austral spring and summer from the late 1970s to the late 1990s (Randel and Wu 1999;Solomon 1999; Thompson and Solomon 2002). It is thought that the cooling trend is mainly caused by the radiative effect of severe ozone depletion (Ramaswamy et al. 2001), and that increasing greenhouse gases also partly contributed to the cooling trends (Langematz et al. 2003) due to their radiatively cooling effects in the stratosphere.
In contrast to the cooling trend, Johanson and Fu (2007) and Hu and Fu (2009) found warming trends over a large portion of the Southern-Hemisphere (SH) high-latitude stratosphere in September and October, and the warming and cooling trends demonstrate a wavenumber 1 pattern. Their results, derived from satellite Microwave Sounding Unit (MSU) observations, show the largest warming of 7-8 °C in the September and October over the period of 1979-2006. Warming trends are also found at all stratospheric levels in the NCEP/NCAR reanalysis, and the maximum warming is about 11 °C at 30 and 20 hPa. In fact, warming signals can also be identified in earlier works (Ramaswamy et al. 1996;Randel and Wu 1999). However, little attention was paid to the warming trends because all these studies had concentrated on ozone-induced stratospheric cooling.
It is known that polar stratospheric temperatures are determined by both radiative and dynamical processes (Andrews et al. 1987;Hu and Tung 2002). The radiative effects of both ozone depletion and increasing greenhouse 1 3 gases lead to stratospheric cooling. Then, the warming trends are likely a result of wave-driven dynamic heating. Indeed, Hu and Fu (2009) demonstrated that the SH high-latitude stratospheric warming is caused by enhanced wave-driven dynamic heating. That is, increasing wave activity caused the strengthening of the Brewer-Dobson circulation (BDC), which consequently leads to stronger downward motion in the polar region and enhanced adiabatic heating in SH high latitudes. They showed that the stratospheric warming trends are closely related to sea surface temperature (SST) warming in the tropical Pacific, and their simulations forced by observed SST demonstrated that increasing wave activity in the SH stratosphere is caused by SST warming. The results by Hu and Fu (2009) were further confirmed by Lin et al. (2009), and Fu et al. (2010, 2015. Especially, they showed how ozone-induced radiative cooling and wavedriven dynamic heating cancel each other and lead to the wavenumber one pattern of stratospheric temperature trends over SH high latitudes. Moreover, the maximum covariance analysis by Lin et al. (2012) showed that the La Nina-like and the central-Pacific El Nino-like SST anomalies result in stronger stratospheric planetary wave activity and phase shifts of the stratospheric stationary waves.
As pointed out by Hu and Fu (2009), the warming trends may have important effects on the recovery of the Antarctic ozone hole. First, the warming may reduce the formation of the polar stratospheric clouds, which would consequently reduce ozone depletion in the Antarctic stratosphere. Second, increasing wave activity may cause weakening of the polar vortex or the polar night jet, which would lead to more high-ozone air mixed into the polar region. Both effects would benefit the recovery of the ozone hole.
The most recent report of ozone assessments has pointed out that the ozone hole is recovering (WMO 2018), as a result of the Montreal Protocol. Observations showed that Antarctic ozone has started increasing in the twentyfirst century (Salby et al. 2011;Solomon et al. 2016). A recent study showed satellite observational results of significant recovery of O 3 in the lower, middle and upper stratosphere and total column ozone in the Southern-Hemisphere middle and high latitudes (Wespes et al. 2019). It was suggested that zonalmean warming trends in the Antarctic lower stratosphere are associated with ozone recovery, based on results from the reanalysis and simulations by chemistry-climate models (Maycock et al. 2018;Randel et al. 2017;Solomon et al. 2017), although the model studies showed a large spread of trends in polar stratospheric temperatures (Maycock et al. 2018;Thompson et al. 2012). However, Philipona et al. (2018) found that the Antarctic lower stratosphere remains cooling in the twentyfirst century, using four radiosonde stations on the Antarctic. These results imply that there exist zonally non-uniform distributions of temperature trends in the Antarctic lower stratosphere.
The purpose of the present study is to revisit the warming trends in the SH high-latitude stratosphere in the recent decade, using up-to-date satellite observations. Especially, we focus on the spatial patterns of temperature trends. We also preform simulations, with stratospheric ozone forcing, to demonstrate how stratospheric ozone changes in the recent decade generate stratospheric temperature trends.

Data
To investigate changes of temperatures in the lower stratosphere (TLS), we use the Advanced Microwave Sounding Units (AMSU) lower stratospheric channel brightness temperature data version 4.0 from 1979 to 2017 (Mears and Wentz 2009). Monthly-mean temperature data, with 2.5° × 2.5° horizontal resolution, is used in this study (data available at http://www.remss .com/measu remen ts/upper -air-tempe ratur e/). Here, TLS is a combination of MSU channel 4 and AMSU channel 9. The weighting function of TLS has peak sensitivities over approximate range of 150-40 hPa (13-22 km) (Young et al. 2011), which can well capture the lower stratospheric condition. The monthly gridded global dataset from the Center for Satellite Applications and Research (STAR) for the National Oceanic and Atmospheric Administration (NOAA) merged the Stratospheric Sounding Unit (SSU) and AMSU-A version 3.0, with resolution of 2.5° × 2.5°, is used to calculate temperature trends in the middle and upper stratosphere (Zou and Qian 2016). The weighting functions of these channels for temperatures in the mid-stratosphere (TMS), upper-stratosphere (TUS), and top-stratosphere (TTS) peak at approximately 10, 3, and 1.5 hPa, respectively. The data is available at ftp://ftp.star.nesdis.noaa.gov/pub/ smcd/emb/mscat/data/SSU/SSU_v3.0/.
Monthly mean total column ozone data is from the Multi Sensor Re-analysis version 2 (MSR2) over 1979(Van Der et al. 2015a. The MSR2 is constructed using all available satellite observations, surface Brewer and Dobson observations, with a data assimilation model. It has horizontal resolution of 0.5° × 0.5°. The unit of total column ozone is the Dobson Unit (DU). It is available at http://www.temis .nl/proto cols/O3glo bal.html.
ERA-Interim daily reanalysis produced by the European Centre for Medium Range Weather Forecasts (ECMWF) (Dee et al. 2011) is used to calculate eddy heat fluxes. We use the monthly mean data with spatial resolution of 1.5° × 1.5° from 1979 to 2017. The data is available at http://apps.ecmwf .int/datas ets/data/inter im-full-moda/levty pe=sfc/).

GCM experiments
The model used in the present study is the Whole Atmosphere Community Climate Model (WACCM) (Marsh et al. 2013), which is compiled from the National Center for Atmospheric Research (NCAR) Community Earth System Model version 1.2 (CESM1.2). The specified chemistry WACCM (SC-WACCM) is used to specify ozone concentrations in the atmosphere (Smith et al. 2014). The SC-WACCM has horizontal resolution of 1.9° × 2.5° and 66 vertical levels from the ground to 4.5 × 10 −6 hPa.
To study the effect of ozone recovery on SH stratospheric temperatures in the recent decade, we perform two simulations with different ozone prescriptions, while greenhouse gases and SSTs are fixed at 2000. In the control experiment, monthly mean ozone at year 2000 is prescribed (case F_2000_WACCM_SC), and the model is run for 30 years. In the ozone recovery experiment (O3R), output from the control experiment is used as initial conditions, the net ozone change over 2007-2017 is added to the column ozone of year 2000, and the simulation is also run for 30 years. Here, the recovery of total column ozone is obtained from the linear trend over 2007-2017, which is multiplied by 10 years (shown in Fig. 3). The vertical distributions of total column ozone recovery are obtained from the vertical profile used in the reference simulation (REF-B2) of WACCM in the Chemistry-Climate Model Validation (CCMVal-2) (Eyring et al. 2005(Eyring et al. , 2010. The effects of ozone recovery on stratospheric temperatures can be characterized by the difference between the last 15-year averages of the O 3 R simulation and the control simulation. Figure 1 shows spatial distributions of SH TLS trends in austral spring for 1979-2006 (upper panels), and 2007-2017 (lower panels), derived from satellite observations. The warming trends in upper panels are almost the same as that in Hu and Fu (2009) although they are from different MSU/ AMSU datasets. Warming trends in September are located over the sector of western Pacific Ocean (Fig. 1a). Significant warming trends in October are located over the Amundsen Sea and Ross Sea (Fig. 1b), with significant cooling trends over the sector of the Indian Ocean. In November (Fig. 1c), there are only significant cooling trends over the southern Indian Ocean. Warming trends in September and October are about 2.8 and 2.6 K/decade, respectively. Trends in all the 3 months show a zonal wavenumber one or dipole pattern. To demonstrate temperature changes at higher stratospheric levels, we plot trends in TMS, TUS, and TTS over 2007-2017 in Fig. 2. In September, warming trends decrease with altitude, and the area of warming trends becomes smaller with increasing altitude. The spatial patterns of warming trends show westward and poleward tilting. At the same time, cooling trends become stronger and dominant with increasing altitude. In October, all these levels show dominant cooling trends, in contrast to the warming trends at the lower stratosphere. In November, all levels demonstrate warming trends over the sector of western Pacific Ocean. The cooling trends are in the polar region in the upper and top stratosphere, in contrast to the warming trends in the polar region in the lower stratosphere. The vertical structures of temperature trends are consistent with the "mirrored shape" between the lower and upper stratosphere of zonalmean Antarctic stratospheric temperature changes (Solomon et al. 2017).

Stratospheric temperature and ozone changes
The above results demonstrate that warming in the SH high-latitude lower-stratosphere continued in the recent decade. An important question is what caused the warming trends in the recent decade. Previous works have attributed the warming over 1979-2006 to increasing wave-driven dynamic heating in the SH stratosphere (Hu and Fu 2009;Lin et al. 2009). It is known that Antarctic ozone started increasing since the late 1990s. Whether increasing ozone contributed to the warming trends in the recent decade is the major question that we want to address in the present study. In the following, we first show ozone changes in the recent decade before we perform attribution analysis.
The upper panels in Fig. 3 show spatial distributions of trends in total column ozone over 1979-2006, derived from MSR2. Severe ozone depletion in the Antarctic stratosphere can be clearly seen in Fig. 3a-c. The maximum ozone decreases are about 42, 55, and 40 DU/decade in September, October, and November, respectively. It is the radiative cooling effect of ozone depletion that contributed to the cooling trends in the Antarctic stratosphere (Fig. 1a-c). The lower panels show ozone increases over 2007-2017. The maximum ozone increase is about 66 and 103 DU over the 10 years in September and November, respectively (Fig. 2d, f). It is important to note that the spatial patterns of the ozone trends well resemble those of the temperature trends for 2007-2017 ( Fig. 1d-f). The spatial correlation coefficients between the changes in ozone and lower stratospheric temperature from 45° S to 90° S are 0.91, 0.89, and 0.99 in September, October, and November, respectively, indicating that the warming trends over 2007-2017 might largely be related to the increase in ozone.

Attribution of the TLS trend
In this section, we perform attribution analysis to reveal what caused the warming trends in recent decade. Following Lin et al. (2009), we use the method of multiple linear regression to attribute the respective contributions of wave-driven dynamic heating, ozone changes, and wave-phase shifts to temperature changes. In Lin et al. (2009), they defined the ozone index as the area-weighted spatial-mean total ozone over 45° S poleward to represent the ozone effect on highlatitude stratospheric warming. Here, to better characterize the zonally non-uniform distribution of ozone changes, we define the ozone index for each grid point, which is slightly different from the ozone index in Lin et al. (2009). The highlatitude stratosphere is heated adiabatically by the descending branch of BDC. To denote the strength of the BDC, an "eddy-heat flux index" is defined as the area-weighted are statistically significant at the 95% confidence level (student t test). The dotted contour lines enclose regions where the trends are significant at the 90% confidence level. The unit is K/decade three-month-mean (including the previous 2 months) eddy heat flux at 150 hPa over 45°-90° S calculated from the ERA Interim reanalysis. In addition to these, the zonal shift of the wavenumber-1 phase temperature pattern alters the zonal distribution of temperature trends and thus leads to temperature changes, although it doesn't affect the zonalmean temperature. Thus, we define a "phase index" as the longitude of the largest temperature averaged over 50º-70º S in the lower stratosphere.  These results indicate that ozone increase and the phase shift of the wavenumber-1 pattern have the major contribution to the TLS trend, while eddy heat flux has little contribution. We will further address this using multiple linear regressions later. Figure 5 shows the time series of area-weighted TLS over the warming area enclosed by the contour of 2 K/decade in October over 2007-2017, the eddy-heat flux index, the area-weighted spatial-mean ozone over the warming area enclosed by the contour of 2 K/decade in October over 2007-2017, and the phase index in October. TLS shows increase over 2007-2017, and its linear trend is about 2.6 K over the recent decade, about half of that in September. The eddy-heat flux index also shows increasing over the recent decade, but the trend is statistically insignificant. Ozone change is about 22 DU over 2007-2017. The phase index shows a positive trend, about 11° of eastward shift over the decade. The correlation coefficients between TLS and the eddy-heat flux index, TLS and ozone, and TLS and the phase index are 0.46, 0.97, and − 0.41, respectively. These results suggest that both BDC strengthening and ozone increase may contribute to the warming trends at the SH high latitudes in October. Figure 6. shows the time series of these variables in November. TLS shows a significant warming trend over 2007-2017, with a value of about 9.5 K over the recent decade. It is about twice larger than that in September. The linear trend of eddy-heat fluxes, about 0.49 Km/s, is weak The period is too short to calculate statistical significance for ozone trends over [2007][2008][2009][2010][2011][2012][2013][2014][2015][2016][2017] and insignificant. The increase of total column ozone is about 70 DU over 2007-2017. The linear trend of ozone is statistically significant at the 90% confidence level. The phase index shows an insignificant negative trend, about 46° of westward shift over the decade. The correlation coefficients between TLS and the eddy-heat flux index, TLS and ozone, and TLS and the phase index are 0.64, 0.90, and 0.39, respectively. These results suggest that the significant increase of ozone might be the major contribution to the warming trends at the SH high latitudes in November.
Using the method of multiple linear regression, Lin et al. (2009) attributed TLS warming over 1979-2006 to the BDC strengthening, phase shift of the temperature wavenumber-1 pattern, as well as their cancellation with ozone depletion. Especially, they found that the BDC strengthening played the major role in causing the warming in the SH high-latitude stratosphere over 1979-2006. Following Lin et al. (2009, we carry out multiple linear regression of gridded monthly mean TLS upon the eddyheat flux index, ozone index, phase-shift index, and the residual term. The contributions to the TLS trend are then obtained by multiplying the regression coefficients by the linear trend in the corresponding indexes. Figure 7 shows the multiple regression results for September, October, and November over 2007-2017. In September, ozone increase has the largest contribution to the observed TLS warming trend, and the spatial pattern of ozone changes can well capture that of the observed TLS trends. In contrast, both eddy heat flux and phase shift have minor contributions to the TLS trend pattern. The residual term is relatively small. This is consistent with the time series in Fig. 4. In October, it is also ozone changes that have the major contribution to the observed TLS trend pattern, while phase changes have a weak negative contribution. For November, it is the same that ozone increase over the polar region has the major contribution to the polar warming, and that the contributions of the other two terms are negotiable. Overall, the observed TLS trends over 2007-2017 can be well explained by the ozone changes.

Simulation results
To further verify the contribution of ozone changes to the observed temperature trends in the recent decade, we carry out simulations as described in Sect. 2. Simulated temperature responses to observed ozone changes over 2007-2017 are shown in Fig. 8. For September (left panels of Fig. 8), the warming trend pattern and its location are close to that in Figs. 1d and 2 (left panels). The warming trends are weaker than observations. The warming trend patterns also show westward and poleward tilting with increasing altitudes, similar to that in Figs. 1d and 2 (left panels). The middle panels of Fig. 8 show weak temperature responses to ozone recovery in October. The simulated warming trends in TLS are rather weak. At higher stratospheric levels, temperature trends become negative. They are consistent with the observational results in Fig. 1e and the middle panels of Fig. 2, but with weaker trends. For November (right panels), simulated TLS trends largely resemble that in Fig. 1f. Temperature trends at higher stratospheric levels also resemble observations in the right panels of Fig. 2. The simulated maximum warming trends in September and November are about 2 and 2.6 K, respectively. In summary, the simulation results demonstrate consistent results with observations, especially in September and November in general, although the simulation cannot reproduce the magnitudes of the observed warming trends.

Discussion and conclusions
In this study, we have revisited the lower stratospheric warming over SH high latitudes in austral spring found by Johanson and Fu (2007) and Hu and Fu (2009). Updated satellite observations demonstrate that the warming continued in the lower stratosphere in the recent decade, especially in September and November. The maximum warming trends in September and November are up to 7 and 13 K over 2007-2017, respectively. However, the spatial patterns of temperature trends are largely different from that over 1979-2006. The wavenumber 1 pattern in September shifted by about 90° in longitude compared to that over 1979-2006, and it shifted by about 180° in October. In November, warming trends are mainly in the lower stratospheric polar region.
Satellite observations show vertically tilting structures of temperature trends. The warming trends in September title westward and poleward with the increasing altitude. In October, the lower stratospheric shows warming, while the middle and upper stratosphere show cooling trends. For November, large warming trends are found in the lower Antarctic stratosphere, while the polar region becomes cooling with increasing altitudes.
Multiple linear regression shows that ozone recovery in the Antarctic stratosphere plays an important role in causing the observed warming trends over 2007-2017. This is different from that over 1979-2006 when the strengthening of the BDC was the major factor in causing the high-latitude warming (Hu and Fu 2009;Lin et al. 2009). The time series of eddy heat fluxes shows that wave activity had little increase in the recent decade, as shown in Figs. 4 and 5, while the recovery of Antarctic ozone became the major factor that contributed to the observed warming.
Our simulations forced by ozone recovery can reasonably reproduce the spatial patterns of temperature trends, especially in September and November. The simulation results also generate the observed vertically tilting structures of temperature trends at different SH high-latitude stratospheric levels. However, the magnitudes of simulated temperature trends are weaker than observations, the simulation result could not well reproduce the warming trends in October. Thus, the simulation results here suggest that at least part of the high-latitude stratospheric warming trends over 2007-2017 are due to ozone recovery.
Note that it requires further studies to confirm the contribution of ozone recovery on the observed warming trends, An important question is why planetary wave activity in the SH stratosphere has no longer increased in the recent decade. According to the analysis by Hu and Fu (2009) and Lin et al. (2012), stratospheric wave activity over 1979-2006 was enhanced because of SST warming over the tropical western Pacific Ocean. If it is true, the result here implies that SST over the tropical western Pacific has not been increased over 2007-2017, which is verified from observation. One possibility is that the Pacific Decadal Oscillation (PDO) has switched from the warm phase to the cool phase since the late 1990s. Whether it is the case requires future analysis and simulation studies.
It is worth noting that ozone recovery and the associated Antarctic stratospheric warming have important implications not only for stratospheric climate but also for tropospheric and surface climates. Recent works have showed that ozone recovery would lead to retreat of Antarctic sea ice, due to dynamic or radiative effects (Bitz and Polvani 2012;Xia et al. 2019). It would also cause the SH Hadley cell narrowed Tao et al. 2016). These all would cause consequent surface climate changes on SH. Top panels: September, middle panels: October, and bottom panels: November. From left to right, the plots are observed TLS trends, TLS trends attributable to the BDC as represented by the eddy heat flux index, TLS trends attributable to ozone index, TLS trends attributable to phase shifts, summation of the three terms, and the residual term, i.e., the difference between the observed trends and summation. The unit is K/decade tively. From left to right, the panels are for September, October, and November, respectively. Color interval is 0.5 K Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/.