Journal of Geophysical Research: Atmospheres

Changes in various branches of the Brewer–Dobson circulation from an ensemble of chemistry climate models

Authors


Corresponding author: P. Lin, Department of Atmospheric Sciences, University of Washington, 408 ATG Bldg., 4000 15th Ave NE, Seattle, WA 98195, USA. (plin@atmos.washington.edu)

Abstract

[1] We analyzed the changes of simulated Brewer–Dobson circulation (BDC) for 1960–2099 from 12 chemistry climate models participating the Chemistry-Climate Model Validation activity phase 2 (CCMVal-2). We decomposed the BDC into transition, shallow, and deep branches with vertical extent of 100–70, 70–30, and above 30 hPa, respectively. Models consistently simulate the acceleration in all three BDC branches over 140 years, but the acceleration rate of the deep branches is much smaller. The acceleration rate of the transition and shallow branches in general shows weak seasonal or hemispheric dependence and increases with time, consistent with the continuous and homogeneous increase of greenhouse gas concentrations. The trend magnitudes of shallow and transition branches differ from model to model, which are found to be associated with the simulated changes in subtropical jets and tropical upper tropospheric temperature. The acceleration of the deep branch is also a response to the increase of greenhouse gas concentrations but is modulated by the changes in ozone concentrations. The effect of ozone changes is particularly prominent in the southern deep branch during austral summer: almost all models simulated strong significant acceleration during the ozone depletion era, weak deceleration during the ozone recovery era, and near-zero trends during the stable ozone era. However, the ozone effect is less evident in other seasons and in other branches.

1 Introduction

[2] The Brewer–Dobson circulation (BDC) is the slow overturning circulation of the stratosphere with an ascending branch in the tropics and descending branches in the extratropical region of each hemisphere (e.g., [Holton et al., 1995]). The strength and width of the BDC vary with season, with the stronger cell occurring in the winter hemisphere and the upwelling regions shifting toward the summer hemisphere [Rosenlof, 1995]. The BDC moves air into and out of the stratosphere and thus sets the mean age or the residence time of air in the stratosphere [Hall & Plumb, 1994]. In particular, the BDC transports ozone and other chemically/radiatively important trace gases from the tropics to polar regions and thus influences the formation and the recovery of the ozone hole [e.g., Solomon, 1999; Austin and Wilson, 2006]. The BDC also affects the stratospheric concentrations of trace gases that have tropospheric origins by setting the rate at which these species enter the stratosphere in the tropics (e.g., [Randel et al., 2006]). Some of these stratospheric species (such as ozone and water vapor) can contribute significantly to the radiation balance of the troposphere and the surface (e.g., [Forster & Shine, 1997; Solomon et al., 2010]).

[3] Because of its importance to both stratospheric and tropospheric climate and stratospheric ozone chemistry, many efforts have been made to simulate the BDC, especially its change under anthropogenic forcing. A stronger BDC in the future as greenhouse gas (GHG) concentrations rise is consistently predicted by many general circulation models (GCMs) (e.g., [Rind et al., 1990; Butchart et al., 2006]), simple GCMs with idealized physics (e.g., [Eichelberger & Hartmann, 2005]), and by many recent coupled chemistry climate models (CCMs) (e.g., [Butchart et al., 2010; SPARC CCMVal, 2010]). Several studies also find BDC changes in response to changes in ozone concentrations (e.g., [Li et al., 2008; Oman et al., 2009; McLandress et al., 2010]). Detailed mechanisms of how these forcings drive the BDC changes are not fully understood yet and vary among models [SPARC CCMVal, 2010].

[4] The relative importance of these two anthropogenic forcings have been discussed in many previous studies (e.g., [Shindell & Schmidt, 2004; Oman et al., 2009; Morgenstern et al., 2010; Polvani et al., 2011; McLandress et al., 2010; McLandress et al., 2011]). Oman et al., [2009] suggested ozone depletion is the dominant forcing for the stratospheric BDC changes over the past few decades in their simulations, but ozone and GHG forcings are more comparable in simulations done by McLandress et al. [2010]. Several individual model studies suggest that the lower part of the BDC would accelerate more in the future (e.g., [Garcia & Randel, 2008; Garny et al., 2011]).

[5] In this paper, we analyzed the simulated BDC from 12 recent CCMs from past to future with changes in both GHGs and ozone concentrations. Most previous studies have focused on changes in the overall strength of the BDC (e.g., [Butchart et al., 2006; SPARC CCMVal, 2010]). Here we investigated the detailed spatial and temporal structures of changes in the BDC from these simulations. In particular, we divided the BDC into different branches based on its vertical extent and hemispheric location and examined their changes in the periods of 1961–2000, 2011–2050, and 2058–2098. We found that the different BDC branches are distinct in their temporal evolution from past to future, and the distinction is consistent among different models. We show that these spatial and temporal structures of the BDC can serve as the “fingerprints” of the forcings and thus are helpful to distinguish and understand the agents driving these changes. In the following, section 2 describes the data and methods used in this analysis, section 3 presents the results, section 4 discusses the mechanisms driving the BDC changes, and section 5 gives the summary and conclusions.

2 Data and Methods

2.1 Model and Reanalysis Data

[6] We examined the model output from 12 CCMs participating in the Chemistry Climate Model Validation activity phase 2 (CCMVal-2, [Eyring et al., 2005; SPARC CCMVal, 2010]). These are the state-of-the-art CCMs that are employed to predict ozone recovery in the recent ozone assessment [WMO, 2011]. Detailed representations of stratospheric dynamical, radiative, and chemical processes are included in these models but vary greatly from model to model.

[7] We analyzed simulations from the REF-B2 scenario, which is a transient run from past to future under anthropogenic forcing. The GHG concentrations in this scenario are taken from observations before 2000 and from A1B scenario defined in IPCC [2000] afterward, and the ozone-depleting substances are adjusted from halogen emission scenario A1 defined in WMO [2007] to include the phaseout of halogen emissions in the 21st century. Except for CMAM, none of these models were run with a coupled ocean; instead, they prescribe sea surface temperature (SST) and sea ice concentrations (SIC) from IPCC-AR4 20th century and A1B simulations. These models also produce simulations under the REF-B1 scenario, in which all forcings and SST/SIC are taken from the observations. We found these REF-B1 simulations to be in general similar to their REF-B2 counterparts in our analysis, which is also shown in previous studies [SPARC CCMVal, 2010]. More details of these models and scenarios can be found in the SPARC CCMVal [2010], the study by Morgenstern et al. [2010], and references therein.

[8] The REF-B2 simulations in most of these models cover the period of 1960–2098 with two exceptions: UMUKCA-METO ends its REF-B2 simulation in 2084, and GEOSCCM REF-B2 simulation starts in 2000. For GEOSCCM, we used results from the REF-B1 simulations for 1960–1999. No appreciable discontinuity can be found in the combined time series. We calculated trends for three 40-year periods: 1961–2000, 2011–2050, 2059–2098, which roughly correspond to the ozone depletion, ozone recovery, and stable ozone periods, respectively. Five models have multiple ensembles (Four models have three ensembles, and one has two ensembles). The ensemble mean for these models is used here. Using one ensemble of each model yields very similar results.

[9] We used ERA–interim reanalysis data [Dee et al., 2011] to check the simulated BDC climatology. Seviour et al. [2012] found that the BDC is well represented in this reanalysis data.

2.2 Defining Different BDC Branches

[10] The Lagrangian-mean meridional mass transport by the BDC can be approximated by the transformed Eulerian mean (TEM) stream function [Andrews et al., 1987]. Figure 1 illustrates the annual mean TEM stream function from these models, and the seasonal mean TEM stream functions are plotted in Figure 2. The strength of the BDC is estimated by mass flux across a boundary level calculated from the TEM velocity. We set three boundary levels and define different BDC branches by whether they can transport air parcels across a certain boundary level. Previous studies (e.g., [Butchart et al., 2006; Butchart et al., 2010; SPARC CCMVal, 2010]) employed the upward mass fluxes across 70 hPa to represent the overall strength of the BDC. 70 hPa is the upper boundary of the tropical tropopause layer [Fu et al., 2007] and roughly the lower boundary of the “overworld” [Hoskins, 1991; Holton et al., 1995; Rosenlof et al., 1997]. Several studies have also considered mass fluxes below 70 hPa (e.g., [Garny et al., 2011]). We thus also examined the mass fluxes across 100 hPa, which is roughly the level of the mean tropical tropopause. The layer just above tropical tropopause is referred to as the “tropically controlled transition region” by Rosenlof et al. [1997]. The layer between 100 and 70 hPa is the upper part of the transition layer between the stratosphere and the troposphere referred to as the “tropical tropopause layer” [Fueglistaler et al., 2009]. Here we refer to the branch that can extend beyond 70 hPa as the “stratospheric branch” and the branch that passes across 100 hPa but not 70 hPa as the “transition branch.”

Figure 1.

Climatology of annual mean TEM stream function from the multimodel mean for 1980–2009 in units of 108 kg/s. Positive values for clockwise turning, and negative values for counterclockwise turning. Various branches of the Brewer–Dobson circulation defined in the text are represented by various colors.

Figure 2.

Climatology of seasonal mean TEM stream function from the multimodel mean for 1980–2009 in units of 108 kg/s.

[11] The stratospheric BDC branch is further divided into deep and shallow branches with the boundary level of 30 hPa. We chose 30 hPa to be the boundary so that the shallow and the deep branches transport roughly equal amounts of mass. This definition is consistent with that of Birner and Bönisch [2011], who defined the shallow and deep BDC branches based on transit time along trajectories. Our results, however, are not qualitatively sensitive to the choice of the boundary levels. The definition of various BDC branches is summarized in Figure 1.

[12] We use the TEM vertical velocity w* archived in the CCMVal-2 database and derive w* from the 6-hourly ERA-Interim reanalysis data. Vertical mass flux is then calculated by horizontally integrating w* weighted by air density and cosine of latitude. We calculated the upward and downward mass fluxes separately. The upward mass fluxes in these simulations are largely confined to the lower latitudes and are closely (but not exactly) balanced by the summation of the downward mass fluxes in each hemispheres. The strengths of the transition branch, shallow branch, and deep branch are estimated by the difference between the upward (or downward) mass fluxes across 100 and 70 hPa, between 70 and 30 hPa, and mass flux across 30 hPa, respectively. The northern and southern branches are separated by calculating the downward mass fluxes in each hemisphere.

3 Results

[13] Based on the aforementioned definitions, we calculated mass fluxes associated with various BDC branches from each model. Table 1 lists the annual mean mass fluxes transported by each BDC branch from multimodel mean (MMM) and ERA interim reanalysis averaged over 1980–2009. The reanalysis-derived mass fluxes across 70 hPa agree with previous estimation by Rosenlof [1995] based on stratospheric analysis datasets and radiative heating rate calculation. All BDC branches from MMM agree well with the reanalysis, although the simulated mass fluxes are consistently slightly smaller than reanalysis. The relative strength of each branch from the models matches that from the reanalysis very well. Half of the mass ascending through 100 hPa is transported poleward by the transition branches before reaching 70 hPa. Then half of the mass ascending through 70 hPa is transported by the shallow branch, and the remaining half is transported by the deep branch. Northern branches are stronger than their counterparts in the Southern Hemisphere (SH).

Table 1. Climatology of annual mean mass fluxes (109kg/s) transported by different branches of the Brewer–Dobson circulation for 1980–2009 from ERA interim reanalyses (OBS) and multimodel mean (MMM)
 Total (100 hPa- )Transition (100–70 hPa)Stratospheric (70 hPa-)Shallow (70–30 hPa)Deep (30 hPa- )
TTNSTTNSTNS
  1. “T” stands for upwelling in the tropics, “N” for downwelling in the Northern Hemisphere, and “S” for downwelling in the Southern Hemisphere.

OBS12.776.193.202.996.593.311.861.443.281.731.54
MMM10.865.022.832.375.832.871.641.392.961.501.50

[14] Figure 3 compares annual and seasonal mean strength of the BDC stratospheric branches in each model, MMM, and ERA interim reanalysis. Both shallow and deep branches are stronger in winter and weaker in summer. The MMM climatology matches the reanalysis very well in terms of the magnitude and the seasonality of each branch. Individual models all capture the correct seasonality. For most models, the simulated annual mean strength of the BDC stratospheric branch is between 70% and 120% of the reanalysis. Larger model spread is found in seasonal mean strength, but in general, the model simulations agree well with the reanalysis.

Figure 3.

Climatology of mass fluxes transported by four stratospheric branches of the Brewer–Dobson circulation from ERA interim reanalysis (OBS) and 12 CCM simulations for 1980–2009 for (a) annual mean, (b) December–January–February, (c) March–April–May, (d) June–July–August, and (d) September–October–November. The results from the multimodel mean (MMM) of the 12 CCMs are also shown.

[15] Figure 4 compares annual and seasonal mean strength of the BDC transition branch in models and reanalysis. Again, the MMM climatology shows excellent agreement with the reanalysis. The models show larger spread of the strength in the transition branch than the stratospheric branch, with a few cases where the simulated transition branch is less than 50% of the reanalysis. UMUKCA-METO and UMUKCA-UCAM (which are based on the same underlying climate model) show very weak transition branch year-around.

Figure 4.

As in Figure 3, except for the transition branches of the Brewer–Dobson circulation.

[16] The temporal evolution of all BDC branches from past to future is examined in the following sections.

3.1 Stratospheric branches

[17] Figure 5 shows the time series of annual mean mass flux anomalies transported by stratospheric BDC and its deep and shallow components for each model. The MMM time series are also shown. The steady strengthening of the BDC over the period 1960–2099 is consistently seen in all models, which is also shown in previous studies [Butchart et al., 2006; SPARC CCMVal, 2010]. As shown in the figure, the strengthening of the shallow branches is the main contributor to the overall stratospheric BDC trends. The trends of the shallow branch also dominate the intermodel spread of the BDC trends. On the other hand, these models also consistently simulate a strengthening in the deep branch, but the change is much smaller than the shallow branch.

Figure 5.

Annual mean mass fluxes anomalies with respect to 1980–2009 climatology for shallow (red lines), deep (blue lines), and stratospheric (black lines) branches of the Brewer–Dobson circulation. Thin lines are for each CCM, and thick lines are for the multimodel mean.

[18] We calculated the seasonal and annual mean trends of stratospheric BDC branches for 1961–2000, 2011–2050, and 2059–2098. These trends are shown in Figure 6. The relative trends with respect to the climatology of 1980–2008 from the MMM are listed in Table 2. Significant strengthening trends are seen in all seasons, in all three periods, and in both shallow and deep branches for MMM. However, shallow branches in general have stronger trends than their counterparts in deep branches, and more of them from individual models pass the statistical significance test. The annual mean stratospheric BDC on average is strengthened by 1.1 × 108kgs− 1 or 1.8% per decade for the entire period (1960–2099), which is consistent with previous model assessments [Butchart et al., 2006; Butchart et al., 2010; SPARC CCMVal, 2010]. Shallow branch contributes 0.8 × 108kgs− 1 per decade to the strengthening, and deep branch contributes 0.3 × 108kgs− 1 per decade. In the relative sense, shallow and deep branches accelerate by 2.7% and 1.0% per decade, respectively. Comparing the MMM trends in the three periods (Figure 6 and Table 2), larger trends in the BDC stratospheric branch are found in the later period for annual mean and in each season except for DJF. The acceleration rate of the MMM shallow branch increases with time in all seasons, which is consistent with the continuous increase of GHG concentration. Two models (Niwa-SOCOL and SOCOL) show much stronger acceleration than others in shallow branch for 2059–2098 in DJF and MAM. The MMM trends of the shallow branch excluding these two outliers are still stronger in the future than in the past (not shown). The MMM deep branch, on the other hand, does not show stronger acceleration in the future than in the past. In DJF, the trends of the deep branch are considerably weaker in the future than in the past.

Figure 6.

Annual and seasonal mean trends in the mass fluxes transported by (a) total stratospheric BDC, (b) shallow branch, and (c) deep branch in units of 108kgs− 1 per decade. Trends are calculated for three periods: 1961–2000 (left), 2011–2050 (middle), and 2059–2098 (right). Trends from each model are indicated by black circle when they are significantly different from zero at 95% confidence level and by red cross when they are not significant. Black diamond and error bar indicate multimodel mean trends and their 95% confidence intervals. The uncertainty of the multimodel mean trends is estimated from the multimodel mean time series.

Table 2. Relative trends (% per decade) of the mass fluxes transported by different stratospheric branches of the Brewer–Dobson circulation with respect to the 1980–2009 climatology from the multimodel mean
  AnnualDJFMAMJJASON
  1. Trends that are not significant at 95% confidence level are marked by underline. Statistical significance is estimated from the multimodel mean time series.

Total stratospheric1961–20001.52.31.31.31.0
2011–20501.91.92.22.21.5
2059–20982.52.43.12.42.3
Shallow1961–20001.82.31.71.81.1
2011–20503.03.13.23.42.3
2059–20983.93.74.44.13.5
Deep1961–20001.32.30.80.90.9
2011–20500.90.71.01.20.9
2059–20981.21.01.60.91.3
N Shallow1961–20001.61.51.82.71.2
2011–20503.13.43.63.82.1
2059–20983.83.65.25.22.5
S Shallow1961–20001.64.11.31.40.4
2011–20502.62.02.63.02.6
2059–20983.83.43.53.65.4
N Deep1961–20001.01.11.60.30.8
2011–20501.41.40.91.81.4
2059–20981.71.33.03.01.4
S Deep1961–20001.610.30.52.01.1
2011–20500.6–2.91.31.00.2
2059–20980.70.40.90.61.0

[19] In the 1961–2000 period, both shallow and deep branches show stronger trends in DJF than other seasons. The strengthening of the deep branch in DJF in the past is especially prominent. As a result, the total stratospheric BDC trend in DJF is about twice as large as those in other seasons, for both absolute and relative trends. The BDC trends show less seasonal dependence in the future.

[20] We further disaggregated the shallow and the deep BDC into the northern and the southern branches, and their trends are shown in Figure 7. Table 2 summarizes the relative trends in each of the BDC stratospheric branches from the MMM. Most deep branch trends are much smaller than their shallow branch counterparts. On average, both northern and southern shallow branches accelerate by ∼ 0.4 × 108kgs− 1 or ∼ 2.6% per decade for 1960–2099. For the deep branch, the northern branch accelerates by 0.2 × 108kgs− 1 or 1.3% per decade, and the southern branch accelerates by 0.1 × 108kgs− 1 or 0.8% per decade.

Figure 7.

As in Figure 6, except for (a) northern shallow branch, (b) southern shallow branch, (c) northern deep branch, and (d) southern deep branch.

[21] For both northern and southern shallow branches, (Figures 7a and 7b), MMM trends are stronger in the future except in DJF for the southern shallow branch. Larger spread among models and more nonsignificant trends are found in winter and spring, which is consistent with larger interannual variability in those seasons. The intermodel spread is especially large for the northern shallow branch in DJF for 2059–2098, where trends from two models (Niwa-SOCOL and SOCOL) are more than twice as large as the rest, and one model (ULAQ) shows significant deceleration. The shallow branch acceleration shows weak seasonal dependence, with slightly stronger acceleration in winter. However, the strongest relative trends usually are not in winter because the climatological mass fluxes are much stronger in winter (see Table 2).

[22] For the deep branch, the most drastic changes occur in the southern branch in DJF (Figure 7d). All but one model show strong and significant strengthening in the past. These strengthening trends on average are about 0.5 × 108kgs− 1 or 10% per decade, with one exceeding 25% per decade (per model basis). However, none of the models produce significant acceleration trends in DJF in the future. Ten of the 12 models predict deceleration in the southern deep branch for 2011–2050, four of which are statistically significant. For 2059–2098, all but one model predict nonsignificant trends. While the summer southern deep branch contributes little to the overall BDC transport in the climatology, its shift in trend from the past to future is so large that the overall stratospheric BDC trend in DJF is noticeably weaker in the future (Figure 6a). Beyond austral summer, the deep branch shows larger spread of trends in winter and spring seasons, similar to those in the shallow branch.

3.2 Transition Branches

[23] Figure 8 shows the time series of annual mean mass flux anomalies for the BDC transition branch. The stratospheric branch is also plotted for comparison. As with the stratospheric branch, strengthening of the transition branch over the 140 years from 1960 to 2099 is consistently seen in every model. The acceleration rate of the transition branch is similar to those of the stratospheric branches during the first half of the 140 years, and surpasses the stratospheric branches in the second half. One model (CMAM) shows very strong acceleration of its transition branch from 2030, which brings its transition branch at the end of the 21st century to about twice the strength in 1960.

Figure 8.

As in Figure 5, except for transition (green lines) and stratospheric (black lines) branches of the Brewer–Dobson circulation.

[24] Overall, the models predict an acceleration rate of 1.5 × 108kgs− 1 or 3.0% per decade for the transition branch for 1960–2099, of which, 0.7 × 108kgs− 1 per decade is contributed by the Northern Hemisphere (NH), and 0.8 × 108kgs− 1 per decade by the SH. The trends in the transition branch for 1961–2000, 2011–2050, and 2059–2098 are shown in Figure 9. The corresponding MMM relative trends are listed in Table 3. The MMM trends of the transition branch all indicate significant acceleration except for the northern branch during 2059–2098 in MAM. The acceleration rate of the transition branch roughly doubles that of the shallow branch. However, the spatial and temporal distribution of the acceleration of the transition branch is very similar to the shallow branch (Figures 7a and 7b vs. Figures 9b and 9c). As with the shallow branch, stronger acceleration of the transition branch is generally found in the future rather than in the past. The absolute increase of mass flux transported by the transition branch is fairly equal among the four seasons, with a slightly larger MMM value and larger model spread in winter. The model spread is especially large for the northern branch in winter for the last 40 years of the 21st century. The relative acceleration rate of the transition branch is strongest in boreal summer and austral spring.

Figure 9.

As in Figure 6,except for (a) total transition branch, (b) northern transition branch, and (c) southern transition branch.

Table 3. As in Table 2, except for the transition branches of the Brewer–Dobson circulation
 AnnualDJFMAMJJASON
Total transition1961–20001.61.31.52.11.8
2011–20503.33.42.63.93.5
2059–20983.93.42.35.15.4
N transition1960–20001.51.11.33.01.3
2011–20502.63.01.54.32.2
2059–20982.82.20.36.04.0
S transition1960–20001.71.41.51.62.4
2011–20503.63.63.03.45.0
2059–20984.65.53.54.16.7

4 Discussion

[25] The “fingerprints” of GHGs and ozone in driving the BDC changes can be seen in the spatial and temporal distributions of BDC trends. Generally speaking, the transition, shallow, and deep branches show continuous acceleration from past to future, which is consistent with the continuous increase of GHGs concentrations. The GHGs “fingerprint” is especially apparent in the shallow and transition branches, since most of them show an increase of acceleration rate with time with weak seasonal or hemispheric dependence. Changes in ozone concentrations, on the other hand, are not homogeneous in space and time. They are stronger in the SH and in the spring season and show opposite trends in the past and in the future. The effect of this spatial and temporal behavior would be embedded in the resulting circulation changes. The ozone's effect is most perceptible in the southern deep branch during DJF, which shows strong acceleration as ozone sharply decreases, weak deceleration as ozone slowly recovers, and near-zero trends as ozone stabilizes. Hints of ozone's effect are also seen in the southern shallow branch during DJF and in the northern deep branch during MAM, which show reduction of trends in the future. This is consistent with previous studies [Li et al., 2008; Garcia and Randel, 2008; Oman et al., 2009] suggesting ozone depletion leads to BDC acceleration and vice versa. Recently, McLandress et al. [2010] suggested ozone depletion may cause a BDC deceleration in SON (but acceleration in DJF), which may explain the insignificant trends of the southern shallow branch seen in almost all models in SON in the past. However, the deceleration induced by ozone depletion in SON is not evident in the southern deep branch. No ozone “fingerprint” can be identified in the transition branch.

[26] Ozone and GHGs impact the BDC through different pathways. Many efforts have been made to understand the detailed mechanisms driving the BDC changes (e.g., [Rind et al., 1990; Eichelberger & Hartmann, 2005; Li et al., 2008]). Several recent works [Garcia & Randel, 2008; Shepherd & McLandress, 2011] suggest that the increase in GHGs leads to stronger upper tropospheric warming, which strengthens the upper flanks of the subtropical jets, leading to an upward shift in the critical layer. As a result, more waves can penetrate the tropopause and dissipate in the lower stratosphere. According to the “downward control” principle [Haynes et al., 1991], the vertical mass flux across a certain level is determined by wave dissipation above it. The BDC strengthening from this mechanism is therefore expected to be largely confined to the lower stratosphere since the anomalous wave dissipation is confined to the lower stratosphere where the subtropical jets intensify. As shown in Figure 10, consistent strengthening and upward expansion of the subtropical jets are seen among different models in both hemispheres in all three periods considered. This agrees with the consistent strengthening of the BDC shallow and transition branches in these models.

Figure 10.

Annual mean zonal wind trends (color contours) and climatology (black contours) for (a) 1961–2000, (b) 2011–2050, and (c) 2059–2098 from multimodel mean. The trends are in units of ms− 1/decade with contour interval 0.15ms− 1/decade. The statistical significance of the multimodel mean trends against the intermodel spread is estimated by Student's t test at 99% confidence level, and significant trends are indicated by filled contours. The zonal wind climatology are plotted with contour interval 10ms− 1. For clarity, only westerlies are plotted.

[27] Furthermore, models that show stronger acceleration in the BDC shallow and transition branches are also the models with stronger acceleration of zonal wind in the subtropical lower stratosphere. Figure 11 shows the trend in the combined BDC shallow and transition branch (i.e., between 100 and 30 hPa) for 1960–2099 from each model versus the zonal wind acceleration at the turnaround latitude averaged over 100–30 hPa. A high correlation is found between the two with a correlation coefficient of 0.84. The two outliers (Niwa-SOCOL and SOCOL) showing the strongest acceleration in the shallow branch in 2059–2098 (Figure 6) also show stronger zonal wind acceleration in the subtropical lower stratosphere than other models (not shown). The zonal wind acceleration in the subtropical lower stratosphere, in turn, is found to be highly correlated with the warming in the tropical upper troposphere as shown in Figure 12. This is because the strong warming in the tropical upper troposphere increases the meridional temperature gradient at the subtropical upper troposphere and accelerates the zonal wind above through thermal wind balance. As shown in the figure, all but one model lay closely along the linear fitting line. In the outlier model (ULAQ), the strong warming in the upper troposphere is found over a much wider region extending to ∼ 50. Thus, the meridional temperature gradient and zonal wind in the subtropics do not increase much despite the strong warming in this model. (ULAQ is an outlier in regard to several other stratospheric dynamical variables [Butchart et al., 2010].) Overall, the correlation between the tropical upper tropospheric warming and the zonal wind acceleration is 0.81 (0.93 excluding ULAQ).

Figure 11.

Scatterplot of the trends in mass fluxes for the combined BDC shallow and transition branch versus the zonal wind trends at the “turnaround” latitude averaged over 100–30 hPa. The trends are calculated for 1960–2099. The turnaround latitude is where w* change signs from positive in the tropics to negative in the extratropics. We defined the turnaround latitude in the w* climatology. It varies with model and level but is usually around 30 in both hemispheres between 100 and 30 hPa.

Figure 12.

Scatter plot of the zonal wind trend at the turnaround latitude averaged over 100–30 hPa versus the temperature trend at 150 hPa averaged over 20S − 20N. The trends are calculated for 1960–2099. The turnaround latitude is defined in Figure 11.

[28] Recent studies [Deckert & Dameris, 2008; Garny et al., 2011] suggest that warmer tropical SST under increasing GHGs can excite more waves and leads to stronger tropical upwelling. These convection-excited waves are largely trapped in the tropical upper troposphere, so the resulting meridional circulation anomaly is expected to be shallow and narrow, with both ascending and descending anomalies locating in the tropics [Garny et al., 2011]. Thus, the net effect of these tropical waves is to redistribute the upwelling within the tropics, but it has little impact on the total upwelling averaged over the tropics (see also the discussion in Shepherd & McLandress, [2011]). Since our analysis only considered the integrated mass fluxes over a wide region, much of this type of response would be masked.

[29] Ozone depletion and recovery, on the other hand, would directly affect the strength of the polar night jets, which further affect the propagation of planetary waves in midlatitude/high latitude. Because the dissipation of these large-scale waves originating in midlatitudes/high latitudes usually occurs at higher altitudes, it would be expected to affect the BDC deep branch. A moderately strong westerly is favorable for the vertical propagation of the planetary waves and hence stronger BDC, while easterly and very strong westerly prohibit wave propagation [Charney & Drazin, 1961]. The ozone-induced westerly anomalies are therefore efficient in enhancing wave propagation and enhancing BDC only when the background westerlies are relatively weak. As shown in Figure 10, strong acceleration of the circumpolar westerlies is seen in the SH stratosphere during the ozone depletion era, which is largely confined in austral spring and summer (not shown). However, the strong ozone effect in southern deep branch is only found in DJF because that is when the simulated background westerlies are weak.

[30] In contrast, the simulated westerly polar night jets are relatively strong in SON. Wave propagation is therefore not very sensitive to stronger westerlies. If the background polar vortex is too strong, the westerly anomalies induced by ozone depletion may lead to reduced wave propagation. Thus, a BDC deceleration is expected in this case as suggested by McLandress et al. [2010]. This ozone depletion–induced BDC deceleration in spring is hinted in the southern shallow branch in our analysis, where near-zero trends are the result from a cancellation between GHG increases and ozone depletion but is not seen in the deep branch.

[31] In the NH, the polar night jets are weaker than those in the SH, and the ozone effect would be expected to be strong in boreal spring. The MMM NH polar jets in MAM accelerate in the past and decelerate in the future (not shown). This is consistent with the reduction of acceleration rate in the northern deep branch from past to future in MAM. Overall speaking, the ozone effect is weaker in the NH than in the SH.

[32] While the results presented in this paper appear to be robust among models, one needs to keep in mind that there could be biases in the model simulations compared to reality. Several studies [Fu et al., 2010; Ueyama & Wallace, 2010] inferred the BDC changes over the past few decades from lower stratospheric temperature measurements and found anticorrelation between temperature variations in the tropics and polar regions on interannual and decadal scales. Young et al. [2011] further showed that the anticorrelation exist throughout the stratosphere. These temperature signals suggest that changes in the BDC deep branch may be the main players. Observational analyses [Lin et al., 2009; Fu et al., 2010; Young et al., 2012] have also shown that the strongest BDC acceleration occurs in austral spring but changes little in summer, which does not agree with the model simulations analyzed here. This discrepancy in seasonality may relate to a well-known deficit of these models: they tend to produce a too-strong and persistent polar vortex in the SH (e.g., [SPARC CCMVal, 2010; Butchart et al., 2011; Gerber et al., 2010]).

[33] Recently, studies suggested the current general circulation models overestimate the tropical upper tropospheric warming in response to GHGs increase [Fu et al., 2011; Po-Chedley & Fu, 2012]. If this were true, then the strengthening in the BDC shallow and transition branches, which dominates the overall simulated BDC strengthening, would also be overestimated in the model simulations.

5 Summary and Conclusions

[34] In this paper, we analyzed the simulated BDC from 12 CCMVal-2 CCMs from 1960 to the end of the 21st century. The BDC is divided into transition, stratospheric shallow, and deep branches based on its vertical extent and further into northern and southern branches. For air mass ascending through 100 hPa, about half the mass is transported poleward by the transition branch, and the shallow and deep branches each transport one quarter. Overall strengthening is seen in all three branches over the 140 years, but contribution from the deep branch is much less than one quarter. Most BDC acceleration is found in shallow and transition branches, which distributes fairly evenly among the four seasons and between the two hemispheres.

[35] We compared the BDC trends in each branch over the past few decades, during the early 21st century and near the end of the 21st century, and found continuous BDC acceleration except in the southern deep branch during austral summer. Models consistently simulated strong acceleration of the austral summer deep cell in the past, but deceleration during the early 21st century, and near-zero trends near the end of the 21st century. Most of the models show stronger acceleration rate of the transition and shallow branches in the future than in the past.

[36] The spatial and temporal distribution of the BDC trends shows the distinct fingerprints of ozone and GHG forcing. Ozone's influence on the BDC is manifested in the southern deep branch during austral summer and is also hinted in the northern and southern deep branches in spring and the southern shallow branch in spring and summer. Beyond these specific branches in these specific seasons, the BDC changes are dominated by the GHGs-related continuous acceleration.

[37] While most of the results presented here are robust across models, relatively large intermodel spread is seen in the trends associated with the shallow and transition branches, especially in the NH and in winter. Part of the intermodel spread can be explained by the differences of simulated wind changes, which in turn are largely caused by warming in the tropical upper troposphere. Furthermore, some of the results that are robust across models might not necessarily agree with observations. Further research is required to reconcile the model-predicted BDC strengthening with observations.

Acknowledgments

[38] We thank Profs. J. M. Wallace and D. L. Hartmann for useful discussions. We acknowledge the modeling groups for making their simulations available for this analysis, the Chemistry-Climate Model Validation (CCMVal) Activity for WCRP's (World Climate Research Programme) SPARC (Stratospheric Processes and their Role in Climate) project for organizing and coordinating the model data analysis activity, and the British Atmospheric Data Center (BADC) for collecting and archiving the CCMVal model output. European Centre for Medium-Range Weather Forecasts (ECMWF) ERA-Interim data used in this study have been obtained from the ECMWF data server. This work is supported by NASA grants NNX09AH73G and NNX11AE544. It is also in part supported by the National Basic Research Program of China (2010CB428604) and the National Science Foundation of China under grant 41275070.