Understanding the Time Scales of the Tropospheric Circulation Response to Abrupt CO2 Forcing in the Southern Hemisphere: Seasonality and the Role of the Stratosphere (original) (raw)

1. Introduction

Global climate models project that, in response to increasing greenhouse gases in Earth’s atmosphere, the planet’s general circulation will undergo robust changes, including a poleward expansion and weakening of the Hadley circulation (Held and Soden 2006; Lu et al. 2007; Frierson et al. 2007), a poleward shift in the subtropical dry zones (Lu et al. 2007; Johanson and Fu 2009; Scheff and Frierson 2012), and a poleward shift in the midlatitude eddy-driven jet streams (Kushner et al. 2001; Barnes and Polvani 2013; Simpson et al. 2014), most notably in the Southern Hemisphere (SH). While the sign of these circulation changes is consistent across most present-day global climate models—those that participated in phase 5 of the Coupled Model Intercomparison Project (CMIP5)—there remains little consensus on the mechanisms responsible for the atmospheric circulation response (e.g., Vallis et al. 2015).

To gain better insight into the mechanisms responsible for the atmospheric circulation response to increasing greenhouse gases, a number of recent studies have begun to explore the time scales of the circulation response in global climate models. One common way to do this is to partition the circulation response into two components: 1) a fast time-scale response associated with the direct radiative forcing of increasing CO2 and 2) a slower time-scale response associated with warming sea surface temperatures (SSTs). Studies employing this methodology have concluded that the bulk of the atmospheric circulation response to increasing CO2 arises from the warming sea surface temperatures (Stephenson and Held 1993; Deser and Phillips 2009; Staten et al. 2012; Grise and Polvani 2014a), but the direct radiative effect of the atmospheric CO2 increase itself can also lead to rapid changes in tropical and subtropical precipitation (Bony et al. 2013; He and Soden 2017), poleward shifts of the midlatitude jets (Grise and Polvani 2014a), and changes in the strength of tropical overturning circulations (Merlis 2015; He and Soden 2015; Shaw and Voigt 2015).

Most studies on the atmospheric circulation response to increasing greenhouse gases have focused on the slower time-scale component of the response, which has been linked dynamically to warming global surface temperatures via changes in tropospheric static stability (Frierson et al. 2007; Lu et al. 2010; Kang and Lu 2012), tropopause height (Lorenz and DeWeaver 2007), upper tropospheric–lower stratospheric meridional temperature gradients (Butler et al. 2010; Arblaster et al. 2011; Wilcox et al. 2012; Harvey et al. 2014; Gerber and Son 2014), and surface meridional temperature gradients (Brayshaw et al. 2008; Ring and Plumb 2008; Butler et al. 2010; Chen et al. 2010) [see also recent review by Vallis et al. (2015)]. Relatively few studies, however, have examined the fast time-scale component of the circulation response. In the tropics, the rapid circulation response has been attributed to horizontal (Merlis 2015) and vertical (Bony et al. 2013) gradients in the direct radiative effects of CO2, as well as to land–sea temperature contrast, which initially strengthens as landmasses warm more rapidly than oceans and promotes a fast time-scale strengthening of tropical monsoon circulations (Shaw and Voigt 2015; He and Soden 2017). In the extratropics, using a single climate model, Wu et al. (2012, 2013a) found evidence that the atmospheric circulation response initiates at stratospheric levels within the first month after an instantaneous doubling of CO2 and then descends to tropospheric levels several months later (see also Staten et al. 2014). This initial stratospheric circulation response is driven by the direct radiative effects of CO2 in the stratosphere, which then alters the propagation of Rossby waves in the stratosphere and upper troposphere, further changing the stratospheric winds and promoting the circulation response to couple down to tropospheric levels within a few months.

In this study, we build on these previous results and examine the time scales of the SH atmospheric circulation response to increasing atmospheric CO2 in CMIP5 models. We focus on the SH for three reasons: 1) apart from the single-model studies of Wu et al. (2012, 2013a) and Staten et al. (2014), previous studies on the time scales of the circulation response have focused on the tropics alone, 2) the effects of land–sea temperature contrast on the circulation response are much less pronounced in the SH than in the Northern Hemisphere, thus simplifying the problem, and 3) the circulation responses in the SH extratropics are much more robust across models and straightforward to interpret. Our results provide evidence that, with increasing atmospheric CO2 concentrations, certain key aspects of the SH atmospheric circulation response equilibrate substantially faster than the global-mean surface temperature.

2. Methodology

To conduct our analysis, we examine monthly mean output from CMIP5 global climate models (Taylor et al. 2012). CMIP5 data are freely available from the Program for Climate Model Diagnosis and Intercomparison at Lawrence Livermore National Laboratory and from the Centre for Environmental Data Analysis. Following Grise and Polvani (2016), we restrict our analysis to the 23 models with values of equilibrium climate sensitivity defined in Forster et al. (2013) (see list of models in Table 1).

Table 1.

Listing of the CMIP5 models used in this study. Models with AMIP, AMIP 4 × CO2, and AMIP future runs are denoted with asterisks. (Expansions of acronyms are available online at http://www.ametsoc.org/PubsAcronymList.)

Table 1.

Table 1.

For each model, we examine three different forcing scenarios: 1) preindustrial control (200+ yr runs of unforced variability), 2) abrupt 4 × CO2 (150-yr runs, in which atmospheric CO2 concentrations are abruptly quadrupled from preindustrial levels at the beginning of the run), and 3) 1% yr−1 CO2 (140-yr runs, in which atmospheric CO2 concentrations are increased by 1% per year from preindustrial levels). For each scenario, we use one ensemble member (“r1i1p1”) per model. Following Grise and Polvani (2016), we define the preindustrial control climatology as the average of all available years of the preindustrial control run, and we estimate the 4 × CO2 climatology using the average of years 101–150 from the abrupt 4 × CO2 run. We exclude results from the 1% yr−1 CO2 scenario of two models (GFDL-ESM2G and GFDL-ESM2M), as they stop increasing CO2 after 70 yr instead of 140 yr. We calculate each model’s response to CO2 forcing as the difference between its perturbation run (abrupt 4 × CO2 or 1% yr−1 CO2) and its preindustrial control climatology and then average the individual model responses together to find the multimodel-mean response.

Following Grise and Polvani (2014a), for a subset of models, we also examine three 30-yr atmosphere-only scenarios with prescribed SSTs and sea ice concentrations: 1) AMIP, 2) AMIP 4 × CO2, and 3) AMIP future. These three scenarios are available from 9 of the 23 models used in this study (see models denoted by asterisks in Table 1). The AMIP scenario uses historical radiative forcings and prescribes observed time-varying SSTs and sea ice from the 1979–2008 period. The AMIP 4 × CO2 scenario uses the same SSTs and sea ice as the AMIP scenario but quadruples atmospheric CO2 concentrations. The AMIP future scenario uses the same radiative forcings and sea ice as the AMIP scenario but augments the AMIP SSTs with a patterned SST anomaly based on the CMIP3 multimodel-mean SST response to 4 × CO2 (see http://cfmip.metoffice.com/CMIP5.html for further details).

Using these scenarios, we estimate the direct radiative contribution of quadrupling atmospheric CO2 using the difference in the climatologies of the AMIP 4 × CO2 and AMIP runs, and we estimate the indirect (SST warming) contribution of quadrupling atmospheric CO2 using the difference in the climatologies of the AMIP future and AMIP runs. However, we note that, in CMIP5 models, these direct and indirect contributions of 4 × CO2 forcing are not strictly additive to the total response. The total response to 4 × CO2 forcing (as measured by the difference of the preindustrial and abrupt 4 × CO2 scenarios) is calculated by quadrupling CO2 from preindustrial levels (~285 ppm), whereas the direct radiative component of the 4 × CO2 forcing is calculated by quadrupling CO2 from historical levels (~360 ppm). Furthermore, to calculate the indirect (SST) component of the 4 × CO2 forcing, the SST warming pattern applied to all models is the same and is normalized to a 4-K global-mean value, whereas the SSTs for the total response (from the abrupt 4 × CO2 scenario) vary widely according to model climate sensitivity and, in the multimodel mean, only result in a ~3.4-K global-mean value. Sea ice concentration changes are also not included in either the direct or indirect components of the response. Further discussion of this methodology is provided in Grise and Polvani (2014a) (see also Deser and Phillips 2009).

Following Grise and Polvani (2016), we utilize three metrics of the SH zonal mean atmospheric circulation (see their Fig. 1):

  1. Latitude of the midlatitude eddy-driven jet (_ϕ_u850): latitude of the maximum value of zonal mean zonal wind at 850 hPa
  2. Poleward edge of the subtropical dry zone (ϕ _P_−_E_=0): latitude where the zonal mean precipitation minus evaporation changes sign between net evaporation in the subtropics and net precipitation at midlatitudes
  3. Poleward boundary of the Hadley circulation (_ϕ_Ψ500): latitude where the 500-hPa mean meridional mass streamfunction changes sign between the thermally direct overturning circulation in the tropics and the thermally indirect overturning circulation at midlatitudes

To calculate these metrics, a first guess of the latitude of each metric is estimated using the nearest latitude gridpoint i in each model. A quadratic (for _ϕ_u850) or linear (for ϕ _P−E_=0 and _ϕ_Ψ500) function is then fit to the three latitude grid points (i − 1, i, and i + 1) surrounding this first-guess estimate, and the function is used to refine the value of the metric to 0.01° latitude resolution.

3. Results

To begin, we show in Fig. 1 the annual-mean, CMIP5 multimodel-mean response of the three SH circulation metrics defined above to an abrupt quadrupling of atmospheric CO2, together with the evolution of the global-mean surface temperature. Examining the circulation response to an abrupt change in CO2 is useful because, in more realistic scenarios where CO2 concentrations are increased more slowly, it is more difficult to untangle and distinguish the varying time scales of the circulation response. Figure 1 shows that the annual-mean response of SH ϕ _P−E_=0 to an abrupt quadrupling of CO2 closely follows that of the global-mean surface temperature (cf. green and black lines in Fig. 1; see also scatterplot at left of Fig. 1). In contrast, the annual-mean responses of SH _ϕ_u850 and _ϕ_Ψ500 do not follow the global-mean surface temperature curve. Instead, the SH midlatitude jet and Hadley cell edge initially shift poleward at a greater rate than the global-mean surface temperature warms. Specifically, 90% of the poleward shift in _ϕ_u850 and _ϕ_Ψ500 occurs in the first 7 yr of the response. By comparison, 90% of the poleward shift in ϕ _P-E_=0 occurs after 39 yr, and 90% of the final (year-101–150 average) global-mean surface temperature response occurs after 65 yr.

Fig. 1.

Fig. 1.

Fig. 1.

(left) Scatterplots of annual-mean, multimodel-mean responses of SH (top) _ϕ_Ψ500, (middle) ϕ _P−E_=0, and (bottom) _ϕ_u850 to abrupt 4 × CO2 forcing with the coinciding global-mean surface temperature response. Each circle represents 1 yr of the 150-yr abrupt 4 × CO2 scenario. (right) Time series of the annual-mean, multimodel-mean responses of global-mean surface temperature (black, K), SH _ϕ_Ψ500 (red), ϕ _P−E_=0 (green), and _ϕ_u850 (blue) to abrupt 4 × CO2 forcing.

Citation: Journal of Climate 30, 21; 10.1175/JCLI-D-16-0849.1

The varying time scales of the circulation response shown in Fig. 1 not only are a characteristic of the multimodel mean but also occur in all individual CMIP5 models examined in this study. Figure 2 reproduces the results from Fig. 1 but shows the spread across all individual model responses. Here, to better visualize the time scales of the circulation response in each model, we plot each of the model responses in terms of the fraction of its final (year-101–150 average) response. Although these circulation metrics have large interannual variability, it is clear that, during the first ~40–60 yr of the abrupt 4 × CO2 scenario, the intermodel spread in the _ϕ_u850 responses (Fig. 2a, blue lines) and _ϕ_Ψ500 responses (Fig. 2b, red lines) lies above the intermodel spread in the global-mean surface temperature responses (Fig. 2, black lines), whereas the intermodel spread in the ϕ _P−E_=0 responses (Fig. 2c, green lines) falls around the intermodel spread in the global-mean surface temperature responses. This behavior is particularly striking during the first 20 yr after CO2 quadrupling (Fig. 2, right panels). In fact, in every single model, _ϕ_u850 and _ϕ_Ψ500 are adjusting to their final values on a faster time scale than the global-mean surface temperature (see difference column in Fig. 2, right). Hence, the behavior shown in Fig. 1 is pervasive across CMIP5 models.

Fig. 2.

Fig. 2.

Fig. 2.

(left) Time series of individual CMIP5 model annual-mean responses of SH (a) _ϕ_u850, (b) _ϕ_Ψ500, and (c) ϕ _P−E_=0 to abrupt 4 × CO2 forcing with the coinciding global-mean surface temperature responses (black lines in each panel). Individual model time series have been smoothed with an 11-yr running mean for plotting purposes. The responses are plotted in terms of the fraction of the total response to 4 × CO2 (defined as the average over years 101–150). (right) Average of the responses over the first 20 yr of the abrupt 4 × CO2 scenario in each model (denoted by ×) and the multimodel mean (denoted by •). The difference column shows the difference between the fractional response of each metric in each model and the fractional response of the global-mean surface temperature in that model.

Citation: Journal of Climate 30, 21; 10.1175/JCLI-D-16-0849.1

The results in Figs. 1 and 2 raise a number of questions:

  1. Why do some features of the SH tropospheric circulation (_ϕ_u850 and _ϕ_Ψ500) respond faster to CO2 forcing than the global-mean surface temperature?
  2. Do SH _ϕ_u850 and SH _ϕ_Ψ500 respond faster than the global-mean surface temperature during all seasons or only during certain seasons?
  3. Why is the response of SH ϕ _P−E_=0 to CO2 forcing delayed compared to that of SH _ϕ_u850 and SH _ϕ_Ψ500?
  4. Are the results shown in Figs. 1 and 2 relevant for more realistic scenarios in which atmospheric CO2 concentrations increase slowly over time?

We address each of these questions individually in the following four subsections.

a. Fast time-scale response of the SH tropospheric circulation

In this subsection, we address why the responses of SH _ϕ_u850 and _ϕ_Ψ500 to abrupt 4 × CO2 forcing are faster than that of the global-mean surface temperature. For brevity, we focus on the results for SH _ϕ_u850 but note that the results for SH _ϕ_Ψ500 yield very similar conclusions.

First, in Fig. 3a, we review the annual-mean, multimodel-mean response of the SH zonal mean zonal wind to 4 × CO2 forcing (as averaged over years 101–150 of the abrupt 4 × CO2 scenario). As documented in many previous studies, the response is characterized by a poleward shift in the tropospheric midlatitude westerly jet and a strengthening of winds in the subtropical and midlatitude stratosphere (e.g., Fig. 12.19 of Collins et al. 2014). However, Fig. 3b reveals that much of this response is established within the first 20 yr after an abrupt quadrupling of atmospheric CO2, which we refer to hereafter as the “rapid response.”

Fig. 3.

Fig. 3.

Fig. 3.

Annual-mean, multimodel-mean response of (a)–(c) zonal-mean zonal wind (m s−1) and (d)–(f) zonal-mean temperature (K) to 4 × CO2 forcing: (a),(d) response averaged over years 101–150 of the abrupt 4 × CO2 run, (b),(e) response averaged over years 1–20 of the abrupt 4 × CO2 run, and (c),(f) number of years after which the response to abrupt 4 × CO2 forcing reaches 90% of its final value [as shown in panels (a) and (d)]. Stippling indicates where the response is 95% statistically significant via Student’s t test. Black contours in (a) and (b) show the preindustrial control climatology (contour interval: 10 m s−1; solid: positive; dashed: negative; zero contour: not shown). Values in (c) are plotted only for the significant values in (a).

Citation: Journal of Climate 30, 21; 10.1175/JCLI-D-16-0849.1

The rapid responses of the zonal wind in the troposphere and polar stratosphere (poleward of 60°S) appear nearly identical to their final responses (cf. Fig. 3b with Fig. 3a), even though the global-mean surface temperature response is only, on average, 63% of its final value during this period (Fig. 2, right). As further evidence of this, in Fig. 3c, we plot the number of years required for the zonal mean zonal wind response to reach 90% of its final value following an abrupt quadrupling of atmospheric CO2. The majority of the tropospheric circulation response occurs very rapidly (within ~5 yr after CO2 quadrupling), but note also that the circulation response at the poleward flank of the stratospheric jet occurs equally as fast. By contrast, the stratospheric wind response on the equatorward side of the stratospheric jet evolves on a much slower time scale (~30+ yr), more consistent with that of ϕ _P−E_=0 and the global-mean surface temperature (Fig. 1).

To understand the varying time scales of the zonal mean zonal wind response, we partition the response into two components: 1) a component due to the direct radiative forcing of 4 × CO2 and 2) a component due to the associated warming SSTs. The results are shown in the top row of Fig. 4. Recall that the model runs necessary for this decomposition are only available for 9 models used in this study (see Table 1), but fortunately, this subset of models is able to well capture the multimodel-mean zonal mean zonal wind response (cf. Fig. 4a to Fig. 3a). The results in Fig. 4 reveal that the bulk of the zonal mean zonal wind response to 4 × CO2 forcing can be explained through the effects of increasing SSTs alone (Fig. 4c) but that the direct radiative forcing of increasing atmospheric CO2 is sufficient to alter the strength and position of the extratropical stratospheric jet and to shift the tropospheric jet poleward (Fig. 4b; see also Grise and Polvani 2014a). It is unlikely to be a coincidence that the features of the zonal mean zonal wind response with the fastest time scales (Fig. 3c) are the same features most impacted by the direct radiative forcing of CO2 (Fig. 4b), suggesting that radiative processes help to initiate the faster circulation responses in these regions.

Fig. 4.

Fig. 4.

Fig. 4.

As in the left column of Fig. 3, but for (a),(d) the total response to 4 × CO2 forcing (years 101–150 of abrupt 4 × CO2 scenario − piControl), (b),(e) the contribution from atmospheric CO2 forcing alone (AMIP 4 × CO2 − AMIP), and (c),(f) the contribution from increasing SSTs alone (AMIP future − AMIP). Results are shown for the nine models with AMIP, AMIP 4 × CO2, and AMIP future runs (see Table 1). Stippling indicates regions where eight or more models agree on the sign of the response.

Citation: Journal of Climate 30, 21; 10.1175/JCLI-D-16-0849.1

In the bottom row of Figs. 3 and 4, we repeat the analysis from the top row of Figs. 3 and 4 but for zonal mean temperature. The results reveal that temperatures at stratospheric levels rapidly equilibrate to the radiative forcing of increased CO2, whereas temperatures at tropospheric levels evolve on the much slower time scale of the global-mean surface temperature, consistent with the warming of the global oceans (cf. Fig. 3e with Fig. 3d, and Fig. 4f with Fig. 4e). Consequently, the fast time scale of the tropospheric circulation response (as shown in Fig. 1 and Fig. 3, top row) is suggestive of driving by temperatures at stratospheric levels rather than those at tropospheric levels (cf. Fig. 3c with Fig. 3f).

The fast time-scale effects of the direct radiative forcing of CO2 and the slower time-scale effects of increasing global surface temperatures both act to shift the tropospheric circulation poleward (Figs. 4b,c), but there is little commonality between their zonal mean temperature responses (Figs. 4e,f). Interestingly, one common feature is a cooling of the lowermost stratosphere. Previous studies have emphasized the importance of the upper troposphere–lower stratosphere meridional temperature gradient in governing the tropospheric circulation response to anthropogenic forcing (Lorenz and DeWeaver 2007; Butler et al. 2010; Arblaster et al. 2011; Wang et al. 2012; Wilcox et al. 2012; Gerber and Son 2014). Consistent with these previous studies, we find that CMIP5 models with greater tropical upper-tropospheric warming (150–200 hPa; 0°–30°S) and/or greater polar lower-stratospheric cooling (150–200 hPa; 65°–90°S) in response to 4 × CO2 forcing (as averaged over years 101–150 of the abrupt 4 × CO2 scenario) exhibit greater poleward shifts in the annual mean position of SH _ϕ_u850 (see correlation values listed in Table 2). Similar results can be derived for the annual mean positions of SH _ϕ_Ψ500 and ϕ _P−E_=0 (not shown). However, in the case of SH _ϕ_u850, only the correlations with the polar lower-stratospheric cooling response are significant during all four seasons (Table 2), a point to which we will return in the next subsection.

Table 2.

Correlations between the poleward shift of SH _ϕ_u850 in response to 4 × CO2 forcing and various temperature responses to 4 × CO2 forcing in CMIP5 models. The responses to 4 × CO2 forcing are calculated using averages over years 101–150 of the abrupt 4 × CO2 scenario. Bold values indicate correlation coefficients that are statistically significant at the 95% confidence level.

Table 2.

Table 2.

To summarize, large facets of the tropospheric circulation response to abrupt 4 × CO2 forcing evolve within the first 20 yr, even though, on average, only 63% of global surface temperature warming has occurred during this period (Fig. 2, right). Although the majority of the tropospheric circulation response can be attributed to rising global surface temperatures (Fig. 4c), the features of the tropospheric circulation that respond the fastest to abrupt CO2 forcing are very similar to those driven by the direct radiative effects of CO2 (cf. Fig. 3b with Fig. 4b). The meridional temperature gradient in the upper troposphere–lower stratosphere (which can be affected by both radiative processes and SST warming) appears to be key in explaining the intermodel variance of the tropospheric circulation response (Table 2). These linkages among the tropospheric circulation response, radiative processes, global surface temperature warming, and upper troposphere–lower stratosphere temperatures are explored further in the next subsection, where we examine the seasonality of the responses.

b. Seasonality

In this subsection, we address the seasonality of the SH tropospheric circulation response to abrupt 4 × CO2 forcing. It is important to note that the abrupt 4 × CO2 scenario in CMIP5 models is initialized at the beginning of the calendar year, so we caution readers from focusing on the seasonality of the first year of the response, as it depends on the time of year when the forcing is imposed (see Wu et al. 2013a).

In Fig. 5, we show the responses of SH _ϕ_u850, _ϕ_Ψ500, and ϕ _P−E_=0 to abrupt 4 × CO2 forcing for all four seasons: December–February (DJF), March–May (MAM), June–August (JJA), and September–October (SON). Consistent with Fig. 1, the poleward shift of ϕ _P−E_=0 closely follows the global-mean surface temperature response during all four seasons (Fig. 5c). In contrast, the rapid responses of _ϕ_u850 and _ϕ_Ψ500 to abrupt 4 × CO2 forcing (Fig. 1, blue and red lines) appear to arise primarily from JJA and SON; during DJF and MAM, the responses of _ϕ_u850 and _ϕ_Ψ500 instead more closely follow the global-mean surface temperature response (Figs. 5a,b). The seasonality of the circulation responses can be better visualized by examining averages over the first 20 yr after CO2 quadrupling (Fig. 5, right). SH _ϕ_u850 adjusts to its final value on a faster time scale during JJASON in all but one model (Fig. 5a, right), and SH _ϕ_Ψ500 adjusts to its final value on a faster time scale during JJASON in all but three models (Fig. 5b, right). In contrast, only 65% of models (15 out of 23) show SH ϕ _P−E_=0 adjusting to its final value on a faster time scale during JJASON (Fig. 5c, right).

Fig. 5.

Fig. 5.

Fig. 5.

(left) Seasonal-mean multimodel-mean responses to abrupt 4 × CO2 forcing: (a) SH _ϕ_u850, (b) SH _ϕ_Ψ500, and (c) SH ϕ _P−E_=0. The annual-mean, multimodel-mean, global-mean surface temperature response is the black line in each panel. The time series have been low-pass filtered for plotting purposes. (right) As in the right column of Fig. 2, but for averages over the first 20 yr of the abrupt 4 × CO2 scenario in each model for the DJFMAM and JJASON seasons; the small upward-pointing red and gray arrows at the top of the top panel indicate outliers lying outside the axes bounds. Fractional responses are only plotted for those models whose final responses exceed 0.5° latitude.

Citation: Journal of Climate 30, 21; 10.1175/JCLI-D-16-0849.1

Based on the correlations in Table 2, one might speculate that the seasonality in the time scales of the circulation responses shown in Fig. 5 reflects seasonality in the time scales of temperatures in the upper troposphere and lower stratosphere. To explore this hypothesis, in Fig. 6, we show the multimodel-mean response of tropical upper-tropospheric temperatures and polar lower-stratospheric temperatures to an abrupt quadrupling of atmospheric CO2 for all four seasons. During all seasons, the tropical upper-tropospheric temperatures evolve on the time scale of the global-mean surface temperature, consistent with the moist adiabatic adjustment of tropical tropospheric lapse rates to surface warming (Fig. 6a). In contrast, the response of polar lower-stratospheric temperatures exhibits distinct seasonality. During DJF and MAM, polar lower-stratospheric temperatures cool on a time scale similar to that of the global-mean surface temperature rise (Fig. 6b, compare blue and green lines to black line). However, during JJA and SON, polar lower-stratospheric temperatures maintain an approximately constant level of cooling (1–2 K) throughout the duration of the 150-yr run. We note that this behavior is robust across models: in all but one model, polar lower-stratospheric temperatures adjust to their final values on a faster time scale during JJASON (Fig. 6b, right).

Fig. 6.

Fig. 6.

Fig. 6.

As in Fig. 5, but for the responses of (a) tropical upper-tropospheric temperatures (150–200 hPa; 0°–30°S), (b) polar lower-stratospheric temperatures (150–200 hPa; 65°–90°S), and (c) polar tropopause pressure (65°–90°S) determined using the method of Reichler et al. (2003). In the left panels, the axes have been reversed for the global-mean surface temperature response in (b) and the tropopause pressure response in (c). In the right panel of (b), fractional responses are only plotted for those models whose final polar lower-stratospheric cooling responses exceed 0.5 K.

Citation: Journal of Climate 30, 21; 10.1175/JCLI-D-16-0849.1

The seasonality of the polar lower-stratospheric temperature response can be explained by the climatology of the vertical temperature profile at SH high latitudes. During summer months (DJF), temperatures increase with height above the polar tropopause owing to the effects of ozone heating (Fig. 7a). Because the temperature lapse rate changes sign at the tropopause, if the tropopause temperature remains fixed, an increase in tropopause height will, by definition, be associated with cooling at lower-stratospheric levels (Fig. 7a; Lorenz and DeWeaver 2007; see also Fig. 11 of Vallis et al. 2015). In response to abrupt 4 × CO2 forcing, the tropopause height increases as tropospheric temperatures warm (Fig. 6c), so lower-stratospheric cooling associated with tropopause height increases will evolve on a time scale similar to that of the global-mean surface temperature (Fig. 6b, blue and green lines; see also longer time scales in the lower stratosphere in Fig. 3f). Evidence for this effect can also be seen by partitioning the DJF zonal mean temperature response to 4 × CO2 forcing into the components associated with the direct radiative forcing of CO2 and increasing SSTs (Fig. 8, top row). The results confirm that the majority of the polar lower-stratospheric cooling response during DJF can be attributed to the warming of surface temperatures (Fig. 8c) rather than to the direct radiative effects of increased CO2 (Fig. 8b).

Fig. 7.

Fig. 7.

Fig. 7.

Multimodel-mean vertical profiles of temperature averaged over the 65°–90°S latitude band: preindustrial control (blue), abrupt 4 × CO2 scenario (averaged over years 1–20, red), and abrupt 4 × CO2 scenario (averaged over years 101–150, black) for (a) DJF and (b) JJA.

Citation: Journal of Climate 30, 21; 10.1175/JCLI-D-16-0849.1

During winter months (JJA), the tropopause is nearly absent at SH high latitudes, as temperatures continually decrease with height into the stratosphere owing to the lack of ozone heating (Fig. 7b). Because there is little change in lapse rate across the tropopause, the increasing tropopause height during these seasons (Fig. 6c, orange and red lines) does not strongly project on polar lower-stratospheric temperatures (cf. red and black lines in Fig. 7b). Consequently, in contrast to DJF, the majority of the polar lower-stratospheric cooling response during JJA can be attributed to the direct radiative effects of CO2 (Fig. 8e) rather than to warming surface temperatures (Fig. 8f). In fact, the warming of the SSTs actually contributes to a slight warming of the polar lower stratosphere during JJA and SON (Fig. 8f; see also Figs. 6b and 7b). As a result, the polar lower-stratospheric cooling response during these seasons is weaker and is dominated by the fast time-scale effects of radiative processes.

The seasonally varying time scales of the SH _ϕ_u850 and _ϕ_Ψ500 responses (as shown in Figs. 5a,b) are consistent with driving by the seasonally varying polar lower-stratospheric temperature responses (as shown in Fig. 6b). To provide support for this hypothesis, in Table 3, we partition the poleward shift of SH _ϕ_u850 in response to 4 × CO2 forcing into the components associated with the direct radiative effects of CO2 and increasing SSTs, keeping in mind that the two components are not required to be additive to the total response (see section 2). Consistent with Fig. 4c, the poleward shift in _ϕ_u850 is dominated by the effects of SST warming, but note that the impact of the SSTs on _ϕ_u850 is substantially smaller during JJA and SON. During these seasons, the warming of the SSTs does not contribute to cooling in the polar lower stratosphere and hence does not as strongly enhance the meridional temperature gradient in the upper troposphere–lower stratosphere (cf. Figs. 8c and 8f). Consequently, the direct radiative effects of increasing CO2 play a greater relative role in driving the poleward shift in _ϕ_u850 during JJA and SON, consistent with the faster time scales of the polar lower-stratospheric temperature response (Fig. 6b) and _ϕ_u850 and _ϕ_Ψ500 responses (Figs. 5a,b) during these seasons.

Table 3.

Multimodel-mean poleward shift of SH _ϕ_u850 in response to 4 × CO2 forcing: the top row shows the total response to 4 × CO2 forcing (years 101–150 of abrupt 4 × CO2 scenario − piControl), the middle row shows the contribution from atmospheric CO2 forcing alone (AMIP 4 × CO2 − AMIP), and the bottom row shows the contribution from increasing SSTs alone (AMIP future − AMIP). The 95% confidence bounds are provided. Results are shown for the nine models with AMIP, AMIP 4 × CO2, and AMIP future runs (see Table 1).

Table 3.

Table 3.

The relationship between the polar lower-stratospheric temperature response and the tropospheric circulation response is also supported by examining the intermodel spread of CMIP5 models. As shown in Table 2, models with greater polar lower-stratospheric cooling in response to 4 × CO2 forcing (as averaged over years 101–150 of the abrupt 4 × CO2 run) exhibit greater poleward shifts in SH _ϕ_u850 during all seasons (see also Fig. 11 of Lorenz and DeWeaver 2007), whereas models with greater upper-tropospheric warming and global-mean surface temperature warming only exhibit greater poleward shifts in SH _ϕ_u850 during DJF and MAM (see also Grise and Polvani 2014b, 2016). The correlations between the tropospheric circulation response and the polar lower-stratospheric temperature response during DJF and MAM are easily explained by the intermodel spread in climate sensitivity, as the polar lower-stratospheric temperature response during these seasons is largely driven by rising global surface temperatures (Figs. 7a and 8c) and thus is significantly correlated with a model’s climate sensitivity (r = −0.51 for DJF and r = −0.42 for MAM). The correlations between the SH _ϕ_u850 response and the polar lower-stratospheric temperature response during JJA and SON are more surprising, given that the polar lower-stratospheric temperature response during these seasons is not significantly correlated with a model’s climate sensitivity (r = 0.17 for JJA and r = −0.18 for SON). For interested readers, we provide further discussion in the appendix of the factors unrelated to climate sensitivity that may be important in governing the intermodel spread in the SH _ϕ_u850 response during JJA and SON. Finally, we note that evidence for the role of polar lower-stratospheric temperatures in the SH _ϕ_Ψ500 response is much less convincing than for SH _ϕ_u850, as the intermodel spread in the SH _ϕ_Ψ500 response is more strongly correlated with climate sensitivity than with the polar lower-stratospheric temperature response during all seasons (see Grise and Polvani 2014b, 2016).

To summarize, in response to abrupt 4 × CO2 forcing, we have shown that SH _ϕ_u850 and _ϕ_Ψ500 respond on a faster time scale than the global-mean surface temperature (Fig. 1) and that this faster time-scale response arises primarily from the JJA and SON seasons (Fig. 5). The seasonally varying time scales of the tropospheric circulation response are consistent with driving by temperatures in the polar lower stratosphere, which are more strongly influenced by fast time-scale radiative processes during JJA and SON (Figs. 6b and 8). Numerous studies have linked variability in SH polar stratospheric temperatures to variability in the SH tropospheric circulation: in the case of interannual variability during austral spring (Thompson and Wallace 2000; Thompson et al. 2005), in the case of stratospheric ozone depletion during both austral summer (Thompson and Solomon 2002; Polvani et al. 2011; Thompson et al. 2011) and austral fall (Ivy et al. 2017), and in idealized model experiments (e.g., Polvani and Kushner 2002; Williams 2006; Butler et al. 2010). Our results add to this preponderance of evidence by suggesting that polar stratospheric temperatures may also be relevant in interpreting some aspects of the tropospheric circulation response to CO2 forcing [see also Wu et al. (2013a) and Staten et al. (2014)]. However, it is important to emphasize that our results here simply show correlation and not causality. The causal mechanisms linking polar stratospheric temperature variability to tropospheric circulation variability remain elusive despite over a decade of community efforts to understand them (see discussion in section 5 of Garfinkel et al. 2013).

c. Delayed response of the SH subtropical dry zone edge

Next, we address why the SH ϕ _P−E_=0 response to increasing atmospheric CO2 appears delayed compared to the SH _ϕ_u850 and _ϕ_Ψ500 responses (Fig. 1). This is somewhat surprising given the strong covariability between the poleward edges of the Hadley circulation and subtropical dry zones noted in previous studies (Lu et al. 2007; Polvani et al. 2011; Kang and Polvani 2011; Quan et al. 2014; Solomon et al. 2016). To understand the SH ϕ _P−E_=0 response, we decompose the response of the zonal mean precipitation minus evaporation (P − E) field into three components, following Seager et al. (2010) and Wu et al. (2013b):

e1

e1

e2

e2

e3

e3

e4

e4

In Eqs. (1)–(4), the symbols are defined as follows: angled brackets denote zonal means, overbars denote monthly means, primes denote deviations from the monthly mean, a is Earth’s radius, g is the gravitational acceleration, ρ w is the density of water, p s is surface pressure, p is pressure, ϕ is latitude, υ is meridional wind, and q is specific humidity. The term δ(TH) [Eq. (2)] represents changes in P − E due to changes in thermodynamics (specific humidity) with the mean circulation field fixed, the term δ(MCD) [Eq. (3)] represents changes in P − E due to changes in the mean circulation with the specific humidity field fixed, and the term δ(TE) [Eq. (4)] represents changes in P − E due to transient eddy moisture flux convergence. Following Wu et al. (2013b), we have neglected nonlinear and surface terms in Eq. (1), which are assumed to be small (see Seager and Naik 2012).

Figure 9 shows the decomposition of the annual-mean, multimodel mean SH ϕ _P−E_=0 response to an abrupt quadrupling of atmospheric CO2 (as shown in Fig. 1, green) using Eq. (1). The thermodynamic term δ(TH) does little to change ϕ _P−E_=0; the increases in moisture in a warmer climate simply amplify the climatological P − E field in a wet-get-wetter, dry-get-drier response (e.g., Allen and Ingram 2002; Held and Soden 2006). The mean circulation term δ(MCD) leads to a rapid ~0.8° poleward shift in ϕ _P−E_=0 within a decade of the abrupt quadrupling of CO2, consistent with the time scale of the responses of SH _ϕ_u850 and _ϕ_Ψ500 shown in Fig. 1. The mean circulation term adjusts to its final value on a faster time scale than the global-mean surface temperature in all but two models (Fig. 9, right).

Fig. 9.

Fig. 9.

Fig. 9.

(left) As in the right panel of Fig. 1, but for the decomposition of the total ϕ _P−E_=0 response (green) into the thermodynamic (red), mean circulation (blue), and residual (purple; total − thermodynamic − mean circulation) components defined in Eq. (1). (right) As in the right panels of Fig. 2, but for the surface temperature, residual, and MCD terms. The difference column shows the difference between the fractional response of the residual and MCD terms and the fractional response of the global-mean surface temperature in that model.

Citation: Journal of Climate 30, 21; 10.1175/JCLI-D-16-0849.1

The transient eddy moisture convergence term δTE cannot be directly calculated for most models, as only five models provide the necessary daily data required to calculate it. Instead, we estimate this term using the residual between the total P − E response [left-hand side of Eq. (1)] and the δ(TH) and δ(MCD) terms. The resulting residual ϕ _P−E_=0 response (Fig. 9, purple line) evolves on a very similar time scale as the total ϕ _P−E_=0 response (Fig. 9, green line) and the global-mean surface temperature response (Fig. 9, black line; see also scatterplot in right panel), suggesting that eddy moisture fluxes are key to the delayed response of the subtropical dry zone edge relative to the response of the jet and Hadley cell edge. This is supported by a direct calculation of the δ(TE) term using the available models with daily data, which show that the majority (~55%) of the residual ϕ _P−E_=0 response (Fig. 9, purple line) can be attributed to eddy moisture fluxes (not shown). The effects of the transient eddy moisture flux convergence on SH ϕ _P−E_=0 increase throughout the duration of the abrupt 4 × CO2 scenario as specific humidity levels in the troposphere rise with the global-mean surface temperature.

Examining interannual variability in the control runs of CMIP5 models, we find that interannual variability in SH ϕ _P−E_=0 is significantly correlated with that of SH _ϕ_Ψ500, as noted by previous studies. On average, SH ϕ _P−E_=0 shifts poleward by 0.6°–0.7° for every 1° shift in _ϕ_Ψ500 (see also Fig. 6 of Polvani et al. 2011), although the exact ratio varies by model (from 0.4:1 to 1:1). This ratio is consistent with the fact that, during the first ~10 yr following abrupt CO2 quadrupling, the response of SH ϕ _P−E_=0 is ~60%–70% of that of _ϕ_Ψ500 (cf. green and red lines in Fig. 1). Hence, the SH subtropical dry zone edge does indeed shift poleward consistently with the rapid time-scale expansion of the SH Hadley cell.

However, in contrast to interannual variability when mean circulation changes dominate SH ϕ _P−E_=0 variability, when atmospheric CO2 is increased, SH ϕ _P−E_=0 subsequently undergoes an additional poleward shift on the time scale of the global-mean surface temperature rise, consistent with driving by eddy moisture fluxes in a warming climate (Fig. 9, purple line). As a result, by the end of the abrupt 4 × CO2 scenario, the magnitudes of the multimodel-mean responses of ϕ _P−E_=0 and _ϕ_Ψ500 are nearly identical (Fig. 1; see also Fig. 2 of Lu et al. 2007) and no longer follow the ~0.7:1.0 ratio found from interannual variability. Again, we note that these ratios reflect the multimodel mean, but in all but one model, the ratio between the poleward shifts in SH ϕ _P−E_=0 and _ϕ_Ψ500 increases throughout the duration of the abrupt 4 × CO2 scenario (not shown).

d. Relevance of results to more realistic CO2 forcings

Finally, one may question whether the results in this study are relevant for more realistic future scenarios, in which atmospheric CO2 concentrations increase progressively over time. To address this issue, in Fig. 10, we have repeated our analysis from Fig. 1 using the 1% yr−1 CO2 increase scenario. Here, as in Fig. 2, we have plotted each of the responses in terms of the fraction of their total response to 4 × CO2 forcing. At first glance, all three circulation metrics appear to increase monotonically with global-mean surface temperature. However, on closer inspection, one can see that _ϕ_u850 and _ϕ_Ψ500 are adjusting to their final values at a slightly faster rate than ϕ _P−E_=0 and the global-mean surface temperature, consistent with the results from Fig. 1 (i.e., the blue and red lines are consistently above the green and black lines in Fig. 10). By the last 20 yr of the 1% yr−1 CO2 scenario, _ϕ_u850 and _ϕ_Ψ500 have achieved a larger fraction of their total 4 × CO2 response than the global-mean surface temperature in 16 out of 21 models (Fig. 10, right) and ϕ _P−E_=0 in 15 out of 21 models (not shown). Hence, there is some limited evidence that, even in slowly increasing CO2 scenarios, some aspects of the atmospheric circulation appear to equilibrate faster than the global-mean surface temperature. To our knowledge, this has not been noted before in the literature.

Fig. 10.

Fig. 10.

Fig. 10.

(left) Time series of the annual-mean, multimodel-mean responses of SH _ϕ_Ψ500 (red), ϕ _P−E_=0 (green), and _ϕ_u850 (blue) and global-mean surface temperature (black) to 1% yr−1 increases in CO2 from preindustrial levels. The responses are plotted in terms of the fraction of the total response to 4 × CO2 (defined as the average over years 101–150 shown in the right panel of Fig. 1). The time series have been low-pass filtered for plotting purposes. (right) As in the right panels of Fig. 2, but for the last 20 yr of the 1% yr−1 CO2 scenario.

Citation: Journal of Climate 30, 21; 10.1175/JCLI-D-16-0849.1

4. Conclusions

In this study, we have examined the time scales of the SH tropospheric circulation response to increasing atmospheric CO2 concentrations. We found that key elements of the circulation response, including the poleward shift of the midlatitude jet and Hadley cell edge, evolve on a faster time scale than the global-mean surface temperature, particularly during the JJA and SON seasons (Figs. 1 and 5). The time scales of the SH tropospheric circulation response appear closely linked to the meridional temperature gradient in the upper troposphere–lower stratosphere and, in particular, to the temperatures in the polar lower stratosphere.

During all seasons, warming global surface temperatures contribute to large warming in the tropical upper troposphere, consistent with the moist adiabatic adjustment of tropical tropospheric lapse rates to surface warming (Figs. 4f and 6a). During DJF and MAM, warming global surface temperatures also contribute to substantial cooling in the SH polar lower stratosphere, consistent with the rising of the tropopause height (Figs. 6b and 8c). Consequently, during DJF and MAM, warming global surface temperatures drive the time scales of the meridional temperature gradient in the upper troposphere–lower stratosphere and thus the SH tropospheric circulation response.

However, during JJA and SON, warming global surface temperatures contribute to very slight warming in the SH polar lower stratosphere (Figs. 6b and 8f), as there is little difference between the climatological polar tropospheric and stratospheric lapse rates during these seasons (Fig. 7b). As a result, the SH polar lower-stratospheric temperature response is dominated by the cooling influence of the direct radiative effects of increased CO2 (Fig. 8e), and hence SH polar lower-stratospheric temperatures reach their equilibrium levels on a much faster time scale than the global-mean surface temperature during these seasons (Fig. 6b, orange and red lines). Consequently, during JJA and SON, the fast time-scale effects of the direct radiative forcing of CO2 have a stronger influence on the meridional temperature gradient in the upper troposphere–lower stratosphere and thus on the SH tropospheric circulation response (see also Table 3).

Gerber and Son (2014) previously identified polar lower-stratospheric cooling as a key factor in driving the tropospheric circulation response to stratospheric ozone depletion, and we show here that it is also a key factor in understanding the tropospheric circulation response to increasing greenhouse gas concentrations (Table 2; see also Fig. 11 of Lorenz and DeWeaver 2007). Our paper adds to the emerging evidence that the rapid adjustment of stratospheric temperatures to the radiative forcing of CO2 may affect the tropospheric circulation on much faster time scales than that expected from rising global-mean surface temperatures (Wu et al. 2013a; Staten et al. 2014). Furthermore, our paper highlights how other elements of the SH tropospheric circulation response, notably the poleward shift in the subtropical dry zones, are supplemented by eddy moisture fluxes and hence evolve on slower time scales that are more in line with that of the global-mean surface temperature (Fig. 9). Consequently, some circulation metrics may be more appropriate than others in detecting emerging anthropogenic influences in the atmospheric general circulation.

Acknowledgments

We thank Isaac Held, Karen Shell, Geoff Vallis, Darryn Waugh, and two anonymous reviewers for helpful comments. All data used in this paper are freely available through the Earth System Grid Federation (https://pcmdi.llnl.gov/search/cmip5/) as described in Taylor et al. (2012). We acknowledge the WCRP’s Working Group on Coupled Modelling, which is responsible for CMIP, and we thank the climate modeling groups (listed in Table 1) for producing and making available their model output. For CMIP, the U.S. Department of Energy’s PCMDI provides coordinating support and led development of software infrastructure in partnership with the Global Organization for Earth System Science Portals. This material is based upon work supported by the National Science Foundation under Grant AGS-1522829. L.M.P. is also supported by a NSF grant to Columbia University.

APPENDIX

Understanding the Intermodel Spread in the Wintertime SH ϕu850 Response to 4 × CO2 Forcing

In Table 2, we show that the intermodel spread in the SH _ϕ_u850 response to 4 × CO2 forcing is significantly correlated with the polar lower-stratospheric temperature response during all seasons. In section 3b, we explain that the correlations between the SH _ϕ_u850 response and polar lower-stratospheric temperature response during DJF and MAM can be expected from the intermodel spread in equilibrium climate sensitivity. In this appendix, we explain that the correlations between the SH _ϕ_u850 response and polar lower-stratospheric temperature response during JJA and SON can be expected from the intermodel spread in the control climatologies of the models.

To demonstrate this, Fig. A1 shows the composites of the preindustrial control temperature profiles averaged over SH high latitudes for two subsets of CMIP5 models: 1) the 10 models with the largest polar lower-stratospheric (150–200 hPa; 65°–90°S) cooling responses to 4 × CO2 forcing (plotted in blue) and 2) the 10 models with the smallest polar lower-stratospheric cooling responses to 4 × CO2 forcing (plotted in red). These subsets are defined separately for each season. There is little difference in the climatological temperature profiles between the two subsets of models in DJF (Fig. A1a), but in JJA, the models that cool more strongly in response to 4 × CO2 forcing (Fig. A1b, blue) have warmer climatological temperatures in the polar stratosphere. This result is unlikely to occur by chance. If we randomly select two subsets of 10 models, the difference in climatological polar stratospheric temperatures seen in Fig. A1b only occurs in ~0.5% of random selections (based on a Monte Carlo test of 1000 random selections).

Fig. A1.

Fig. A1.

Fig. A1.

As in Figs. 7a and 7b, but for preindustrial control temperature profiles averaged over the 10 CMIP5 models examined in this study with the largest polar lower-stratospheric (150–200 hPa; 65°–90°S) cooling responses to 4 × CO2 forcing (blue) (after years 101–150 of the abrupt 4 × CO2 run) and the 10 CMIP5 models with the smallest polar lower-stratospheric cooling responses (red). The subsets are defined separately for each season.

Citation: Journal of Climate 30, 21; 10.1175/JCLI-D-16-0849.1

The key difference between the two subsets of models in Fig. A1b arises from the lapse rate of the temperature profile near the tropopause (note the kink in the blue line immediately above 250 hPa and the absence of a kink in the red line): models with the largest climatological lapse rates in the wintertime stratosphere (as measured by the temperature difference over the 70–250-hPa layer) have the smallest cooling responses in the polar lower stratosphere in response to 4 × CO2 forcing (r = 0.78 for JJA and r = 0.64 for SON). As discussed by Vallis et al. (2015), if the tropopause temperature remains fixed, the climatological lapse rate is important for understanding the effects of an increase in tropopause height on stratospheric temperatures (see their Fig. 11). When the lapse rate changes sign at the tropopause (such as during the DJF season; Fig. 7a), an increase in the tropopause height will result in cooling at lower-stratospheric levels (see also Fig. 8c). However, the more positive the lapse rate becomes above the tropopause (i.e., the less temperatures increase with height or the more temperatures decrease with height), the less important this effect becomes (e.g., cf. DJF and JJA seasons in Figs. 7 and 8). Hence, all else being equal, models with the most positive climatological lapse rates above the tropopause (i.e., with the largest temperatures decreases with height; Fig. A1b, red) might be expected to have the smallest cooling responses in the polar lower stratosphere, in agreement with the results shown in Fig. A1b.

The results in Table 2 reveal that the poleward shift in SH _ϕ_u850 in response to 4 × CO2 forcing is closely linked to the polar lower-stratospheric temperature response, and the results in Fig. A1 reveal that the polar lower-stratospheric temperature response is closely linked to the preindustrial control temperature profile during winter months. Therefore, poleward shifts in SH _ϕ_u850 in response to 4 × CO2 forcing are significantly correlated with the preindustrial control climatological lapse rates in the polar lower stratosphere (65°–90°S, 70–250 hPa) during the JJA (r = −0.56) and SON (r = −0.61) seasons; that is, the models with the largest poleward jet shifts during winter months have climatological polar stratospheric temperatures that decrease the least with height above the tropopause. This result is in agreement with the relationship between the models’ wintertime control climatologies and SH _ϕ_u850 responses documented in previous studies (e.g., Grise and Polvani 2014b; Simpson and Polvani 2016).

REFERENCES