Future changes in the occurrence rates of major stratospheric warmings (MSWs) have recently been identified in chemistry-climate model (CCM) simulations, but without reaching a consensus, potentially due to the competition of different forcings. We examine future variations in the occurrence rates of MSWs in transient and timeslice simulations of the ECHAM/MESSy atmospheric chemistry (EMAC) CCM, with a focus on the individual effect of different external factors. Although no statistically significant variation is found in the decadal-mean frequency of MSWs, a shift of their timing toward midwinter is detected in the future. The strengthening of the polar vortex in early winter is explained by recovering ozone levels following the future decrease in ozone-depleting substances. In midwinter, a stronger dynamical forcing associated with changes in tropical sea surface temperatures will lead to more MSWs, through a similar mechanism that explains the stratospheric response to El Niño-Southern Oscillation (ENSO).
 Major stratospheric sudden warmings (MSWs) are the most abrupt events of boreal wintertime polar stratospheric variability. They consist of a sudden increase in the temperature of the polar stratosphere and the reversal of the wintertime circulation in that region [Labitzke, 1981]. MSWs are initiated by anomalously high levels of tropospheric wave activity entering the stratosphere that leads to the deceleration of the polar vortex [Matsuno, 1971]. Although intimately linked with tropospheric forcing, MSWs also affect the tropospheric circulation for up to two months [e.g., Baldwin and Dunkerton, 2001]. Thus, a better knowledge of MSWs and the involved mechanisms will help to improve the representation of tropospheric circulation in climate models.
 The interest of the scientific community in MSWs has recently increased, in particular, with the aim of improving the skill of future climate change projections. Some studies have already looked for a possible change in the future occurrence rates of MSWs using general circulation models (GCMs) [e.g., Butchart et al., 2000], atmosphere-ocean GCMs [e.g., Bell et al., 2010; Mitchell et al., 2012a], and more lately, chemistry-climate models (CCMs) [e.g., Charlton-Perez et al., 2008; McLandress and Shepherd, 2009; Mitchell et al., 2012b]. These studies have mainly focused on the identification of possible future changes in the mean frequency of MSWs. However, a consensus could not be achieved due to the wide range of results ranging from an increase in the frequency of MSWs to a decrease. The same kind of discrepancies has also been found, when analyzing different CCMs under the same future scenario [SPARC CCMVal, 2010].
 A possible reason for these discrepancies might be the weak amplitude of future changes in the Arctic polar vortex due to the competition of different contributors and the biases of each model to reproduce the related processes [Mitchell et al., 2012a]. The most important forcings of future climate are rises in greenhouse gas (GHG) concentrations and their induced variations in the atmosphere and in sea surface temperatures (SSTs) and sea-ice concentrations (SICs), and the projected decline in emissions of ozone-depleting substances (ODS). An isolation of the impacts of each of these forcings on the boreal polar stratosphere might help to obtain a more precise picture of future changes in the occurrence rates of MSWs.
 This study examines how MSW variability, particularly their seasonality, will change in the future, and how different external forcings individually impact on the occurrence of these events. For that purpose, we use different transient CCM simulations and timeslice runs that represent present-day and future climate. Additional sensitivity runs explore the separate effects of future forcings.
 All simulations used in this study were run with the CCM ECHAM/MESSy Atmospheric Chemistry (EMAC). EMAC is a numerical chemistry and climate simulation system that includes submodels describing tropospheric and middle atmosphere processes [Jöckel et al., 2006] and is based on 5th generation European Centre Hamburg general circulation model (ECHAM5) [Roeckner et al., 2006]. Here we applied EMAC in the T42L39MA-resolution, i.e., with a T42 truncation (~ 2.8° longitude × 2.8° latitude) and 39 levels up to 0.01 hPa.
 We use two transient simulations covering the period 1960 to 2100, SCN-B2d and RCP8.5. Both runs include natural and anthropogenic forcings and natural variability following the specifications by the CCMVal initiative [Eyring et al., 2008] and the RCP scenario [Meinshausen et al., 2011]. SSTs and SICs are prescribed with output from simulations of the atmosphere-ocean GCM ECHAM5 Max Planck Institute Ocean Model (ECHAM5/MPIOM) [Jungclaus et al., 2006]). Future changes in MSW characteristics are assessed by comparing the last 40 winters of SCN-B2d (2060/2061 to 2099/2100, “future”) with the first 40 ones (1960/1961 to 1999/2000, “past”). Results from the SCN-B2d simulation are compared with the extreme climate change scenario of the RCP8.5 run to obtain confidence in the future signals.
 We also employ different timeslice simulations of at least 20 years with the same model which include changes in SSTs/SICs and concentrations in GHGs and ODS for present and projected future climate, but not natural variability. The reference runs (REF2000 and REF2095), representative of the years 2000 and 2095, provide additional robustness to the SCN-B2d results, as their forcings correspond approximately to conditions for the end of the past and future periods of this run. The sensitivity simulations allow us to isolate the influence of each external forcing on MSWs. See Table 1 for a description of each timeslice run.
Table 1. Description of EMAC Timeslice Runs
Number of Years
aFor SEN2 run the SST field is set at REF2095 values between 15°S and 15°N, linearly interpolated from 15° to 30°N/S and left at REF2000 values elsewhere.
 Major stratospheric warmings are identified by the simultaneous reversal of zonal-mean zonal wind at 10 hPa and 60°N and the zonal-mean temperature difference between 60°N and the pole [Labitzke, 1981]. Two events are separated by at least 10 days of consecutive westerly winds, corresponding to the radiative relaxation time scale at 10 hPa. Stratospheric final warmings are ruled out by imposing 10 days of westerly regime before 30 April and after the occurrence of a MSW.
3 Frequency and Seasonality of MSWs
 We first focus on the analysis of the decadal-mean frequency of MSWs. MSWs in the recent past occur in both transient simulations (SCN-B2d and RCP8.5) with a similar frequency as in observations (6.5 and 5.5 events/decade, respectively, in comparison with the observed 6.0 events/decade [Charlton and Polvani, 2007]). In the SCN-B2d run, future MSWs show a slight increase in their frequency with respect to the recent past (7.3 vs. 6.5 events/decade, respectively), but this change is not statistically significant at a 95% confidence level (t-test of Appendix A of Charlton et al. ). A higher but still statistically insignificant increase is also found under extreme climate change conditions (RCP8.5), i.e., from 5.5 events/decade in the past to 8.3 in the future. This result is in agreement with the mentioned weak amplitude of future circulation changes in the Arctic stratosphere and the conclusions of a multimodel study within the CCMVal initiative [SPARC CCMVal., 2010].
 The timeslice simulations show a higher frequency of MSWs than the transient ones (8.5 and 10.3 events/decade in REF2000 and REF2095, respectively), probably due to the absence of the Quasi-Biennial Oscillation (QBO) in the runs and the performance of persistent weak easterlies in the tropical stratosphere instead [Holton and Tan, 1980]. Nevertheless, when comparing REF2000 and REF2095, qualitatively consistent changes to those analyzed between past and future in the transient runs are found.
 Although no significant changes have been detected in the decadal-mean frequency of MSWs, a different seasonality is observed in the future with respect to the past in the transient runs (Figure 1a). In the past, both runs show a higher number of events in early winter (November-December) than observations [Charlton and Polvani, 2007]. This bias has also been noted in other models that have ECHAM5 as atmospheric base model [Charlton et al., 2007] and might be related to an anomalous tropospheric forcing. In contrast, in the future, the signal of climate change appears to be stronger than this tropospheric forcing and a decrease in MSWs in early winter and an increase in mid- and late winter are observed. A χ2 test that examines the independence between the period of time (past and future) and the seasonal distribution of MSWs (early and mid- to late winter) allows us to determine that this seasonal shift of MSWs is statistically significant at a 95% confidence level. However, the increase in MSWs in midwinter in RCP8.5 takes place in February, one month later than in SCN-B2d. This could be associated with a greater radiative cooling due to higher GHGs in the RCP8.5 run, which delays the effect of a stronger dynamical forcing. As will be shown later, the radiative effect of GHGs becomes important in the first half of January, even under a nonextreme climate change scenario (SRES A1b for GHGs).
 In the REF2000 and REF2095 runs the seasonality of MSWs is slightly different from the transient runs, with earlier MSWs in the timeslices, probably due to the persistent tropical easterlies [Gray et al., 2004]. Nevertheless, similar future changes are found between REF2000 and REF2095, in particular, a higher frequency of MSWs in January and February in the future than in the present (Figure 1b). In early winter, the decrease in the number of events is not as clear as in the transient runs, particularly in December, when both reference simulations show approximately the same frequency of events. However, a more thorough analysis has revealed a different distribution of events throughout the month between runs, with a predominance of events in the second half of December in REF2095 and a flat distribution in REF2000 (not shown).
 The discussed changes in the seasonality of MSWs agree well with statistically significant future variations in the climatology of the polar night jet, with a strengthening in early winter and a weakening in mid- and late winter. These variations are seen in both transient runs and in the reference timeslice experiments (Figure 2, top row). The significance of these results and those of section 4 has been assessed with a two-tailed t-test.
4 Factors for Changes in Major Stratospheric Warmings Seasonality
 The above described changes in the polar vortex state and in the occurrence rates of MSWs have been attributed to different future forcings by isolating their individual effects in the sensitivity runs.
 The strengthening of the future polar vortex in early winter appears only when isolating the effect of changes in ODS emissions (Figure 2d). It is consistent with a stronger meridional temperature gradient in the middle stratosphere, resulting from a warming of the tropics and a cooling at high latitudes (Figure 3a). The temperature distribution agrees with a weakening of the Brewer-Dobson circulation (BDC) in November induced by lower ODS emissions (Figure 3b) and higher ozone concentrations in the future. The change in the BDC can be explained by an increase in the vertical stability as a result of a lower stratospheric warming due to the ozone gain (not shown). The increased vertical stability reduces the tropospheric wave energy flux into the stratosphere and leads to a relative weakening of the BDC, with a subsequent cooling of the polar middle stratosphere. This impact of ODS-induced changes in atmospheric state on the planetary wave propagation is also confirmed when analyzing the associated variations in the probability of negative squared refractive index (see Figure S1 of the auxiliary material). Changes in the BDC due to O3-induced variations in the thermal structure of the atmosphere have been reported by Rind et al. .
 Concerning midwinter, the GHG induced SST-changes, in particular those in the tropics, lead to the weakening of the polar vortex (Figures 2g–2i). Although the influence of the SSTs starts at the beginning of January, they are counterbalanced by the GHG radiative effect during the first half of this month (Figure 2f). The changes found in this winter period have a dynamical origin, as justified by the analysis of the extratropical 100 hPa eddy heat flux (a diagnostic of the tropospheric wave activity entering the stratosphere). In the SCN-B2d and REF2095 runs, a stronger future tropospheric forcing appears in January (Figures 4a and 4c). In the RCP8.5 simulation, this increase in wave forcing is also found, but at the beginning of February in agreement with the delayed weakening of the polar vortex in the future compared to the SCN-B2d experiment (Figure 4b). The mechanism explaining the stronger tropospheric wave forcing could be similar to that associated with El Niño events, as future tropical SST changes are characterized by a warming, particularly over the Pacific (not shown). We find a deepening of the Aleutian low and an intensification of the wavenumber-1 stationary wave in January and February [e.g., Ineson and Scaife, 2009] in all our sensitivity runs including future changes in SSTs, even for tropical SST changes only as shown in Figure 4d.
5 Discussion and Conclusions
 In this study, we analyzed the variability of the polar vortex in transient and timeslice simulations of the past and future performed with the EMAC CCM. Although no statistically significant differences in the decadal-mean frequency of MSWs have been detected, a different polar stratospheric response to future changes has been found in early and middle to late winter, with a strengthening of the polar vortex in the first period and a weakening in the second. This different response is reflected in the seasonality of MSWs, with a shift toward more events in midwinter. Our results confirm some previous CCM studies showing a similar change in the distribution of future MSWs and in the seasonal cycle of the polar night jet [e.g., Charlton-Perez et al., 2008; Mitchell et al., 2012b].
 The same signal in future changes in MSWs (i.e., zero change in decadal-mean frequency and seasonal shift) is a robust result in the different transient and timeslice runs of the EMAC CCM. Moreover, the length of runs (at least 20 years) ensures the reliability of the results [Eyring et al., 2008].
 An important result of this study is that we could attribute the origin of the different polar stratospheric responses and the future changes in the MSWs to variations in external forcings:
 Future declining ODS emissions are responsible for the strengthening of the polar vortex and a lower occurrence rate of MSWs in early winter. The strengthening of the polar vortex results from the recovery of stratospheric ozone and a weakening of the BDC that impact on the meridional distribution of the middle stratospheric temperature and thus, on the intensity of the polar vortex, in agreement with Rind et al. .
 Future GHG-induced tropical SST changes primarily explain the increase in the number of MSWs in midwinter. This variation is related to a strengthening of the future tropospheric forcing that appears to be due to a similar mechanism as that involved in the El Niño-Southern Oscillation (ENSO) influence on the polar stratosphere [e.g., Ineson and Scaife, 2009].
 To our knowledge, this is the first sensitivity analysis carried out to explain the origin of future changes in the timing of MSWs and in the seasonal cycle of the wintertime polar stratosphere. Our conclusions are consistent with previous studies that were not focused on the analysis of MSWs. For instance, several studies that explored the effect of CO2 doubling or quadrupling on the polar stratosphere also found a weakening of the polar vortex in midwinter [e.g., Butchart et al., 2000; Bell et al., 2010]. The relevant role of GHG-induced changes in SSTs and, in particular, those in the tropics for the increased tropospheric dynamical forcing has also been highlighted [e.g., Oman et al., 2009].
 Finally, the different response of the polar stratosphere to future changes in early and midwinter should be taken into account in the analyses of the impact of climate change on the polar stratospheric circulation. This opposite future response together with the different impact on the polar stratosphere of external forcings could explain the discrepancies in the projections of mean frequency of MSWs reported by previous CCM studies.
 This work was funded by the “Deutsche Forschungsgemeinschaft” (DFG) within the research programmes SHARP and CAWSES under the grants: LA 1025/13-1; LA 1025/14-1; LA 1025/15-1; LE 1865/1-2; LA 1025/5-3. We would like to thank the North-German Supercomputing Alliance (HLRN) and the ECMWF computing center in Reading for computing time and support. We also thank Scott Osprey and an anonymous reviewer for their comments.
 The Editor thanks Scott Osprey and an anonymous reviewer for their assistance in evaluating this paper.