Observations show that stratospheric water vapor (SWV) concentrations increased by ~30% between 1980 and 2000. SWV has also been projected to increase by up to a factor of 2 over the twenty-first century. Trends in SWV impact stratospheric temperatures, which may lead to changes in the stratospheric circulation. Perturbations in temperature and wind in the stratosphere have been shown to influence the extratropical tropospheric circulation. This study investigates the response to a uniform doubling in SWV from 3 to 6 ppmv in a comprehensive stratosphere-resolving atmospheric GCM. The increase in SWV causes stratospheric cooling with a maximum amplitude of 5–6 K in the polar lower stratosphere and 2–3 K in the tropical lower stratosphere. The zonal wind on the upper flanks of the subtropical jets is more westerly by up to ~5 m s−1. Changes in resolved wave drag in the stratosphere result in an increase in the strength of tropical upwelling associated with the Brewer–Dobson circulation of ~10% throughout the year. In the troposphere, the increase in SWV causes significant meridional dipole changes in the midlatitude zonal-mean zonal wind of up to 2.8 m s−1 at 850 hPa, which are largest in boreal winter in both hemispheres. This suggests a more poleward storm track under uniformly increased stratospheric water vapor. The circulation changes in both the stratosphere and troposphere are almost entirely due to the increase in SWV at pressures greater than 50 hPa. The results show that long-term trends in SWV may impact stratospheric temperatures and wind, the strength of the Brewer–Dobson circulation, and extratropical surface climate.
There is continued interest in the impact of stratospheric water vapor (SWV) trends on climate (Forster and Shine 1999, 2002; Smith et al. 2001; Solomon et al. 2010). This has been motivated by the observed increase in SWV of ~30% over the late twentieth century (Scherer et al. 2008; Hurst et al. 2011) and the rapid and persistent decrease of ~15% after 2000 (Randel et al. 2006; Rosenlof and Reid 2008). Furthermore, SWV has been projected to increase by up to a factor of 2 over the coming century in response to increasing long-lived greenhouse gas (LLGHG) concentrations (Gettelman et al. 2010). However, the projections of future SWV trends are highly uncertain. The Chemistry–Climate Model Validation Activity (CCMVal-2) multimodel mean change in tropical lower stratospheric water vapor by 2100 is 1.3 ppmv for a medium LLGHG emissions (REF-B2) scenario, but the changes in individual models range from 0.5 to 4 ppmv. It is therefore important to explore the impacts of changes in SWV on climate.
Trends in SWV may arise via changes in the transport of water vapor into the stratosphere, which occurs predominantly in the tropics, and/or via changes in the production of SWV from the oxidation of methane. It is worth noting that SWV can therefore have a role in climate feedbacks (e.g., in response to changes in the strength of tropical convection; Rosenlof and Reid 2008), climate forcings [e.g., in response to volcanic eruptions (Joshi and Shine 2003) and changes in the production of SWV via methane oxidation (Rohs et al. 2006)], and modes of “unforced” internal climate variability [e.g., the quasi-biennial oscillation (QBO; Randel et al. 1998), the El Niño–Southern Oscillation (ENSO; Considine et al. 2001) and variations in the strength of the Brewer–Dobson circulation (Randel et al. 2006)].
Most of the existing literature on climate impacts has focused on the radiative forcing associated with SWV trends (Forster and Shine 1999, 2002; Smith et al. 2001; Solomon et al. 2010). However, there is now compelling evidence that the stratosphere and troposphere act as a two-way dynamically coupled system, such that changes in stratospheric wind and temperature have an impact on the extratropical tropospheric circulation [for an overview, see, e.g., Gerber et al. (2012)]. This study therefore considers an additional mechanism—specifically, that trends in SWV will cause changes in stratospheric temperature and wind, which may impact the tropospheric circulation.
There has been considerable interest in how the stratosphere influences the tropospheric response to climate forcings and processes, most notably increases in LLGHGs (e.g., Scaife et al. 2012), stratospheric ozone depletion and recovery (e.g., Son et al. 2008; Polvani et al. 2011; McLandress et al. 2011), ENSO (Cagnazzo and Manzini 2009; Ineson and Scaife 2009), and the 11-yr solar cycle (Haigh et al. 2005). However, relatively little attention has been paid to the impact of SWV trends on the coupled stratosphere–troposphere system. Rind and Lonergan (1995) conducted an experiment in which SWV was uniformly doubled from 3 to 6 ppm in a middle atmosphere general circulation model (GCM). Although they found evidence of there being changes in both the stratospheric and tropospheric circulation in response to the increase in SWV, the model’s radiation code did not correctly simulate the vertical structure of the stratospheric temperature response to SWV (Oinas et al. 2001). In their model [the Goddard Institute for Space Studies Global Climate-Middle Atmosphere Model (GISS GCMAM)], the maximum stratospheric cooling occurred in the upper stratosphere, whereas in more complete radiation codes the maximum is in the lower stratosphere (e.g., Forster and Shine 2002). Furthermore, their GCM had, by modern standards, a relatively poor horizontal (8° × 10°) and vertical (23 layers) resolution. MacKenzie and Harwood (2004) examined the impact of SWV changes due to projected future methane trends on the stratosphere. They found that the associated increase in SWV modified the stratospheric circulation response due to LLGHGs and aerosols alone, particularly over the Northern Hemisphere winter pole. However, partly because the effect of methane oxidation on SWV is largest in the upper stratosphere, the lower stratospheric temperature changes in their experiment were small.
The lower stratosphere is a key region for stratosphere–troposphere interaction, and is also where a given change in SWV has the greatest impact on temperature (e.g., Maycock et al. 2011). Joshi et al. (2006) investigated the dynamical impact of an increase in SWV using an atmospheric GCM with considerably higher horizontal resolution (1.875° × 1.25°) than used by Rind and Lonergan (1995), but with a relatively low model lid at ~40 km. A 20% increase in SWV was imposed in the model as an estimate of the 1980–2000 trend. This perturbation resulted in changes in the large-scale extratropical tropospheric circulation that were consistent with a more positive northern annular mode (NAM) index (Thompson and Wallace 2000). However, it was not clear whether the change in SWV in the upper or lower stratosphere had the greatest impact on the troposphere; this point is addressed later in this paper. In a more recent study, Tandon et al. (2011) used a simplified GCM to investigate the tropospheric response to idealized stratospheric heating perturbations with a structure resembling the cooling pattern due to a uniform increase in SWV. The response in their model consisted of a poleward shift in the tropospheric eddy-driven jet and a widening and weakening of the Hadley cell. The results of these studies suggest that trends in SWV may be an important driver of changes in stratospheric and tropospheric temperature and wind. The aim of this study is to investigate for the first time the circulation response to SWV perturbations in a comprehensive stratosphere-resolving atmospheric GCM. The following questions will be addressed:
What is the impact of uniformly doubling SWV on the stratosphere?
Is there evidence of changes in the tropospheric circulation in response to increasing SWV?
Are uniform changes in SWV in the upper or lower stratosphere more important for driving circulation changes in the troposphere?
We use the vertically extended atmospheric component of the Hadley Centre Global Environmental Model version 1 (HadGAM1) described by Hardiman et al. (2010) and Osprey et al. (2010). The model is similar to the core of the Met Office’s Unified model that participated in the coupled chemistry–climate model validation exercise (CCMVal-2) (UMUKCA-METO). There are 60 levels in the vertical domain and the model is run at N48 horizontal resolution (2.5° × 3.75°). The model is forced at the lower boundary over ocean grid points by time-varying monthly-mean sea surface temperatures (SSTs) and sea ice taken from the Hadley Centre Global Sea Ice and Sea Surface Temperature (HadISST; Rayner et al. 2003) dataset. The land surface is allowed to evolve self-consistently. Ozone is imposed as a zonal-mean monthly-mean climatology created from merged satellite datasets covering the period 1979–2003 [see Dall’Amico et al. (2010) for further details]. Observed time-varying concentrations of carbon dioxide and methane for the period 1980–2002 are included, as well as fixed concentrations of nitrous oxide and chlorofluorocarbon (CFC)-11 and -12 at 1993 values. The model does not explicitly include anthropogenic or volcanic aerosols or a solar cycle. However, some component of the effect of such processes will be indirectly included in the simulations through the use of observed SSTs and sea ice.
The model includes additional modifications compared to the version described by Hardiman et al. (2010), which improve the representation of surface exchange processes at coastal grid points (Ackerley et al. 2012). These modifications negatively impact on the climatology and variability of the Northern Hemisphere winter stratospheric polar vortex. Specifically, they result in a weaker stratospheric westerly jet and a relatively high frequency of major sudden stratospheric warmings (SSWs) (1.2 yr−1) compared to the 40-yr European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-40) dataset (0.66 yr−1; Charlton and Polvani 2007). However, the simulated climatology and variability lie within the spread of the current generation of stratosphere-resolving GCMs (Butchart et al. 2010a).
HadGAM1 uses the Edwards and Slingo (1996) radiative transfer scheme (ESRTM), updated to use the correlated-k method to calculate transmittances for absorbing species (Cusack et al. 1999). There has been considerable debate in the literature about the suitability of broadband radiation codes for representing the radiative effects of trends in SWV (e.g., Oinas et al. 2001). Codes based on the ESRTM have been shown to consistently overestimate the radiative forcing and instantaneous change in heating rate for SWV perturbations by up to ~50% (e.g., Maycock and Shine 2012). However, the qualitative structure of the stratospheric temperature response to SWV in the ESRTM is consistent with more detailed radiation codes (Forster and Shine 2002; Maycock et al. 2011). This study will therefore provide an upper estimate of the magnitude of the response to a given SWV perturbation.
The experiments consist of artificially modifying the water vapor field input to the model’s radiation code. The water vapor is only altered above the tropopause, which is diagnosed using the WMO (1957) lapse-rate definition at each model time step. Three sets of SWV perturbation experiments are presented. For each case, the reference simulation, referred to as CTL, has a prescribed uniform fixed SWV distribution of 3 ppmv. The first perturbation experiment, UNI, consists of a uniform fixed doubling in SWV to 6 ppmv. The other experiments are designed to test whether changes in SWV in the lower stratosphere (LS) or upper stratosphere (US) have a greater impact on the atmospheric circulation. The prescribed SWV distributions in these experiments as a function of pressure, p, are as follows:
Each experiment consists of a three-member ensemble run for 23 years from January 1980 to December 2002. Unless otherwise stated, all figures in section 3 should be assumed to show results for 23 × 3 = 69 yr of data.
Since the configuration of the model used here includes prescribed SSTs, the effect of the radiative forcing due to perturbing SWV is not fully captured in the simulations (see, e.g., Forster and Shine 2002). The difference in the global-mean annual-mean surface temperature between the UNI and CTL experiments is negligible (~0.02 K), whereas with interactive SSTs the change may be expected to be around 1.4 K, assuming a stratosphere-adjusted radiative forcing of 1.2 W m−2 (calculated offline using the ESRTM) and a climate sensitivity of 1.15 K (W m−2)−1 for HadGEM1 (Solomon et al. 2007). This study therefore focuses on the atmospheric circulation response to SWV perturbations in the absence of changes in SSTs. The impact of including changes in SSTs on the response to SWV will be considered in a follow-up study. The effects of changing SWV on ozone concentrations either via changes in temperature, circulation, or chemical processes are also neglected.
a. The response to uniformly doubling SWV
1) Stratospheric changes
Figure 1 shows from left to right: the seasonal-mean zonal-mean temperature (, in K) climatology of the CTL run, the difference in between the UNI and CTL experiments, and the difference calculated offline using the fixed dynamical heating (FDH) method for the same uniform 3 to 6 ppmv change in SWV. The top panels show seasonal averages for June to August (JJA) and the bottom panels for December–February (DJF). Briefly, the FDH method provides an estimate of the equilibrium stratospheric temperature change due to a given radiative perturbation under the assumption that there is no change in the contribution of dynamical processes to the net heating rate [see, e.g., Maycock et al. (2011) for further details]. The differences between the center and right-hand panels of Fig. 1 therefore indicate where dynamical processes are acting to modify the purely radiatively determined temperature response. The hatching in the center panels shows where the difference between the UNI and CTL experiments is not statistically significant at the 95% confidence level using a two-tailed Student’s t test.
The climatology of the model shows a warm bias in the polar lower stratosphere (~50 hPa) of ~5 K in both the northern and southern winter hemispheres compared to ERA-40 reanalysis data (Uppala et al. 2005). The warm bias in the SH in JJA is similar to that reported by Hardiman et al. (2010). However, the bias in the NH in DJF does not appear in Hardiman et al. (2010), so it appears largely attributable to the additional modifications to the model described in section 2.
The change in stratospheric due to doubling SWV in the GCM is similar to that calculated using the FDH method, both in terms of magnitude and structure. The largest cooling is in the lower stratosphere and generally decreases in magnitude with increasing height. In addition, the cooling at ~100 hPa is around a factor of 2 larger in the extratropics compared to in the tropics. However, there are also notable differences between the responses. In the NH in DJF, the FDH calculation predicts a cooling throughout the polar stratosphere. However, in the GCM there is weaker cooling and even a slight warming (<1 K, not statistically significant at 95% level) between 1 and 10 hPa over the polar cap. This indicates that dynamical processes are acting to oppose the radiative effect of increasing SWV. Conversely, in the SH the cooling over the polar cap is ~1–2 K larger than in the FDH calculations in both seasons. The role of dynamical processes in modifying the radiatively determined stratospheric response to the SWV perturbation is therefore markedly different in the two hemispheres.
Figure 2 shows the zonal-mean zonal wind (; m s−1) for the UNI and CTL experiments. The left-hand panels show the climatology of the CTL experiment and the right-hand panels show the differences between the UNI and CTL experiments. The hatching indicates where the differences are not statistically significant at the 95% confidence level. In approximate thermal wind balance with the changes in horizontal temperature gradient in the lowermost stratosphere, the zonal flow is more westerly on the upper flanks of the subtropical jets in both hemispheres and in both seasons. This feature closely resembles the response to increases in LLGHGs in GCMs (e.g., Lorenz and DeWeaver 2007). For LLGHGs, this is largely driven by enhanced tropical upper tropospheric warming associated with the lapse-rate feedback to increases in near-surface temperature (Wu et al. 2011), since the impact of carbon dioxide perturbations on extratropical lower stratospheric temperatures is quite small (e.g., Fels et al. 1980). By contrast, the anticipated surface warming associated with the increase in SWV (see section 2) is smaller than the extratropical lower stratospheric cooling, and therefore the latter effect is likely to be the main driver of the lower stratospheric response to changes in SWV.
Figure 2 also shows that significant changes in occur in the mid and upper stratosphere in the UNI experiment, particularly in the winter hemispheres. In the SH in JJA, the stratospheric westerly jet is strengthened by ~7 m s−1 at 1 hPa and 40°S. The largest difference in a given month is 14 m s−1 at 1 hPa and 60°S in September (not shown). The increase in the strength of the SH polar vortex is associated with a delay in the date of the final warming by ~13 days (~1 standard deviation of the model’s interannual variability), which marks the springtime breakup of the vortex as measured by the date of the final reversal of to easterly at 60° and 10 hPa (e.g., Butchart et al. 2006). Conversely, in the NH in DJF there is a deceleration of ~5 m s−1 on the poleward side of the jet core at pressures less than 10 hPa, but no detectable change in the average final warming date. The increase in SWV therefore results in a stronger and more persistent stratospheric jet in the SH, but a weaker upper stratospheric jet in the NH with no robust change in vortex breakup date.
Having looked at the changes in the mean flow, we will now examine the response of eddies to the increase in SWV. The breaking of upward propagating planetary-scale Rossby waves makes an important contribution to the pseudomomentum budget of the winter stratosphere and leads to variations in the zonal-mean flow. The divergence of the Eliassen–Palm flux (EP flux) can be used as a measure of the zonal force per unit mass due to resolved wave breaking (e.g., Andrews et al. 1987). Figure 3 shows the EP flux divergence (∇ · F; m s−1 day−1) in the same layout as Fig. 2. In the climatology, more negative values are found in the midstratosphere in the NH in DJF (~−2 m s−1 day−1) than in the SH in JJA (~−1 m s−1 day−1), which indicates that more planetary waves are generated in the NH than in the SH. In the UNI experiment, the EP flux convergence in the NH in DJF increases in magnitude by 10%–15% in the mid and upper stratosphere. Conversely, in the SH in JJA the magnitude of ∇ · F decreases by ~25%. The opposite sign of the change in upper stratospheric EP flux divergence in the winter hemispheres may be related to differences in the background climatology, since the vertical wind profile is important for the propagation and refraction of planetary waves traveling from the troposphere into the stratosphere (Charney and Drazin 1961). In the SH, there is a significant decrease in the meridional eddy heat flux at 100 hPa integrated over 40°–80°S (not shown), which is approximately proportional to the vertical component of the EP flux and can be used as a measure of the amount of wave activity entering the lower stratosphere (e.g., Polvani and Waugh 2004). In the NH, there is no significant change in at 100 hPa across the extratropics, which suggests that the change in EP flux divergence may be instead related to changes in the refraction and breaking of waves within the stratosphere.
In the transformed Eulerian mean (TEM) framework (Andrews et al. 1987), it is clear that the effect of an increase in wave drag in the upper stratosphere will be twofold: first, it will directly impact , leading to a deceleration of the flow, and second it will induce an anomalous poleward residual circulation that will cause adiabatic warming over the polar cap, thereby maintaining thermal wind balance with the changes in zonal wind. The latter effect can for example explain the dynamically induced warming over the NH polar cap in the lower center panel of Fig. 1. In the NH, the effects of planetary wave breaking can lead to a temporary breakdown of the wintertime stratospheric polar vortex. Such events are known as major sudden stratospheric warmings and are commonly diagnosed by a reversal of the westerly jet at 60°N and 10 hPa to easterly (Charlton and Polvani 2007). However, despite the increase in stratospheric wave drag in the NH in DJF, there is no statistically significant change in the frequency of major SSWs in the UNI experiment. The response shown in Fig. 2 therefore reflects a change in the mean state and not the extreme variability of the NH polar vortex.
As noted above, wave drag in the stratosphere also drives a residual mean meridional circulation, more commonly known as the Brewer–Dobson circulation (BDC; e.g., Plumb 2002). This circulation consists of upward motion in the tropics, poleward flow in the stratosphere, and downwelling in the extratropics. The changes in EP flux divergence shown in Fig. 3 will be linked to changes in the residual circulation. The net vertical mass flux F across a pressure surface in a latitude band φ1 < φ < φ2 can be used as a measure of the strength of the residual circulation, and is calculated as follows:
where a is the radius of the earth, ρ0 is the local density, and φ is latitude. Here, is the residual vertical velocity in the TEM framework, which is defined as (Andrews et al. 1987)
where the overbars represent a zonal-mean quantity, primes represent deviations from the zonal-mean, and subscripts represent a partial derivative; also, Ψ is the residual mass streamfunction and the other terms are as defined in Andrews et al. (1987).
Given that the residual circulation obeys the continuity equation, the total upward mass flux in the tropics, F↑ (kg s−1), can be calculated by summing the net downward mass fluxes, F↓, in both hemispheres. These are calculated following the method of Rosenlof (1995). For each month, the latitude at which at 70 hPa changes from positive to negative (i.e., from upward to downward motion) is identified; this is denoted the turnaround latitude, φturn. In January, φturn was calculated to be 29°N and 44°S in the NH and SH, respectively. In July the corresponding values are 49°N and 29°S. Under the assumption that F↓ is zero at the pole, Eq. (1) can be integrated from the pole (φ = ±90°) to the respective φturn in each hemisphere to give F↓,NH and F↓,SH. The tropical upwelling mass flux, F↑, is then simply
The solid lines in Fig. 4 show the seasonal cycle of F↑ at 70 hPa for the CTL and UNI experiments. In the CTL experiment, the upwelling in the tropics is ~40% stronger in boreal winter compared to in austral winter, which is consistent with there being more planetary wave breaking in the NH stratosphere. In the UNI experiment, there is an increase in F↑ of ~10% throughout the year compared to the CTL experiment. The dashed and the dash-dotted lines in Fig. 4 show the mass flux contributions from F↓,NH and F↓,SH respectively for each experiment. From August to November, the change in F↓,SH is small and the increase in F↓,NH makes the dominant contribution to the change in F↑. From December to March, there is an increase in the downward mass flux in both hemispheres, with the change in F↓,NH being almost double that in F↓,SH. From April to July, the increases in F↓,NH and F↓,SH are approximately equal, and therefore both hemispheres make an important contribution to the change in F↑ in this season.
The strength of the BDC is important for the transport of trace species, including water vapor and ozone, into and out of the stratosphere (e.g., Butchart and Scaife 2001). It is therefore important to understand how the BDC changes in response to different climate forcings and feedbacks, and the role that different atmospheric waves play in driving these changes. A ubiquitous feature of the circulation response to increased LLGHG concentrations is an increase in the strength of the BDC, which is typically largest in DJF (Butchart et al. 2006, 2010b). A recent study by Shepherd and McLandress (2011) has shown that this can largely be explained by an increase in lower stratospheric wave drag associated with a lifting of the critical layers for wave breaking owing to the increased westerlies on the upper flanks of the subtropical jets. The fact that F↓ increases in the summer season in both hemispheres in Fig. 4 indicates that changes in lower stratospheric wave drag must play a role in the strengthening of the BDC in the UNI experiment, since westerly winds (i.e., winter conditions) are required for planetary waves to propagate into the mid and upper stratosphere. This is also evident by examining the seasonal cycle in mass flux higher up in the stratosphere (e.g., at 30 hPa; not shown) which, through downward control arguments (Haynes et al. 1991), will be largely determined by planetary wave drag in the mid and upper stratosphere. At higher altitudes, F↑ only increases in boreal winter in the UNI experiment, due to an increase in F↓,NH, and therefore changes in lower stratospheric wave drag must make a significant contribution to the increase in F↑ evident in Fig. 4; this is therefore consistent with the mechanism described by Shepherd and McLandress (2011).
There has also been considerable debate in the literature regarding the contribution of resolved waves (i.e., EP flux divergence) and parameterized waves (i.e., orographic and nonorographic gravity drag) to changes in the BDC. The study by McLandress and Shepherd (2009) highlighted the importance of changes in both the EP flux divergence and parameterized orographic gravity wave drag (OGWD) close to the turnaround latitudes for a strengthening of the BDC in response to increased LLGHGs in the Canadian Middle Atmosphere Model (CMAM) [see also Li et al. (2008), who find a similar result in the GFDL chemistry–climate model]. In this version of HadGAM1, the contribution of OGWD to the residual circulation at 70 hPa is considerably smaller than in CMAM and the dominant contribution comes from the resolved wave drag (Hardiman et al. 2010). Downward control calculations (not shown) indicate that the change in mass flux in the UNI experiment is also dominated by the change in resolved wave drag, and that the contributions from parameterized wave drag are small. Understanding these differences therefore remains an important area for future research, particularly given the uncertainties in observed gravity wave drag in the stratosphere.
Finally, the above results highlight a possible negative feedback for SWV transport. A uniform increase in SWV leads to a strengthening of the BDC, which would cause a lifting and cooling of the tropical tropopause. This is consistent with the pressure at the tropical tropopause being 5–10 hPa lower in the UNI experiment compared to the CTL experiment throughout the year. To first order, this would act to reduce the amount of water vapor entering the stratosphere, which is again consistent with the changes in SWV simulated by the model in response to the radiatively imposed SWV perturbation.
2) Tropospheric changes
Figure 2 shows that the increase in SWV results in north–south dipole changes in in the midlatitudes in both hemispheres, which correspond to weaker westerlies between ~20° and 45° and stronger westerlies between ~45° and 70°. These changes are largest in DJF in both hemispheres and have a peak-to-peak amplitude of 3.0 and 5.0 m s−1 at 250 hPa, and 1.2 and 2.8 m s−1 at 850 hPa, in the NH and SH, respectively. Similar changes of a smaller magnitude are also evident in JJA. This indicates that there is a poleward shift in the eddy-driven jets, which is a ubiquitous feature of the tropospheric response to changes in the lower stratospheric state (e.g., Kushner and Polvani 2004; Simpson et al. 2009; Butler et al. 2010; Polvani et al. 2011). Such changes in the eddy-driven jets are linked to annular mode variability in the troposphere (Lorenz and Hartmann 2001), with a more poleward jet corresponding to a more positive annular mode index (Thompson and Wallace 2000).
Figure 5 shows the differences in mean sea level pressure (MSLP; hPa) between the UNI and CTL experiments for (left) 20°–90°N and (right) 20°–90°S in (top) JJA and (bottom) DJF seasons. For clarity, the gray shading now indicates where the differences are found to be statistically significant at the 95% confidence level. In both hemispheres, and in both seasons, there is a relatively zonally symmetric decrease in MSLP over the polar cap, and a more zonally asymmetric increase in MSLP in the extratropics. These changes correspond to north–south dipoles over the extratropical ocean basins, which is consistent with the structure of the tropospheric annular modes (Thompson and Wallace 2000). The MSLP dipole over the North Atlantic has a peak-to-peak amplitude of 2 and 6 hPa in JJA and DJF, respectively, which can be compared to an amplitude of ~7–8 hPa associated with a 1 standard deviation departure of the North Atlantic oscillation (NAO) index in DJF (Greatbatch 2000). There is evidence of smaller MSLP changes over the North Pacific in DJF. In the SH, the peak-to-peak amplitude of the MSLP dipole is 4 and 6 hPa in JJA and DJF, respectively. The positive anomalies peak to the southwest of Australia, which coincides with the region of maximum intensity of the SH storm track (e.g., Hoskins and Hodges 2005). The changes in the tropospheric circulation due to the increase in SWV are associated with regional changes in surface temperature and precipitation over North America and western Europe, as well as over the Antarctic continent, which are consistent with annular mode variability found in observations and climate models (not shown; e.g., Marshall et al. 2001; Thompson and Wallace 2000).
The regional nature of the changes in MSLP, particularly in the NH, is indicative of there being changes in the position and strength of the storm tracks. To analyze this more closely, three proxies for synoptic eddy activity are used: u (m s−1) at 850 hPa, the change in the standard deviation (Δσ) of 2–6-day bandpass filtered daily 500 hPa geopotential height1 (Z500; m), and the maximum Eady growth rate, σmax (day−1), which is given by (Eady 1949)
where f is the Coriolis parameter and N is the buoyancy frequency. For the purposes of this study, the terms in Eq. (4) are vertically averaged from the surface to 200 hPa, following the method of Scaife et al. (2012). Figure 6 shows polar stereographic plots of the UNI − CTL differences in the three diagnostics from top to bottom for (left) 20°–90°N and (right) 20°–90°S. Note that the Δσ of Z500 is calculated for the 22 complete winters from the first ensemble member only. Also note that the gray shading indicates where the differences are significant at the 95% confidence level. Data are only shown for DJF, since this is when the largest MSLP responses are observed. However, qualitatively similar differences are found in JJA.
Figures 6a and 6b show that there is a strengthening of the westerly low-level zonal wind of up to ~2 m s−1 on the poleward side of the storm tracks between 45° and 60°N and 55° and 65°S in the NH and SH, respectively, and a corresponding smaller decrease at lower latitudes. The low-level zonal wind changes over the North Atlantic peak downstream of the maximum baroclinic zone and extend into the Mediterranean and eastern Europe.
The climatology of the standard deviation of 2–6-day bandpass filtered Z500 shows maxima over the extratropical ocean basins, which are collocated with the baroclinic storm tracks (not shown). The Δσ of Z500 in Figs. 6c and 6d also show dipole changes, with increased variance on the poleward sides of the storm tracks and reduced variance at lower latitudes. This is consistent with the synoptic eddy activity being located further poleward on average.
Finally, the differences in σmax are shown in Figs. 6e and 6f. In the North Atlantic, the differences in σmax peak slightly upstream of the largest changes in the low-level zonal wind. Interestingly, there is also an increase in σmax on the western side of the North Pacific basin, which is not evident in the other diagnostics. The change in σmax is predominantly due to a change in the vertical wind shear across the troposphere, , as opposed to changes in the static stability, which is consistent with the response to increased LLGHG concentrations in stratosphere-resolving GCMs (Scaife et al. 2012). However, other studies have found that the static stability makes a larger contribution to changes in the Eady growth rate under increased LLGHGs (e.g., Lu et al. 2008). This apparent contrast may be partly related to the use of fixed SSTs in the simulations presented in this study, which constrains the changes in tropospheric static stability, as well as to the fact that the terms in Eq. (4) are averaged over the whole troposphere instead of only the lower troposphere as in Lu et al. (2008).
The relationship between the position of the tropospheric eddy-driven jet and the extratropical lower stratospheric temperature in each hemisphere is now considered, following a similar method to Polvani et al. (2011). In Fig. 7, the seasonal-mean latitude of the midlatitude jet at 850 hPa is plotted against the polar cap average (|φ| > 60°) (K) at 100 hPa. To capture the main storm track regions, the average is calculated over the North Atlantic sector (60°W–20°E) in the NH and over the zonal mean in the SH. Figure 7 shows scatterplots for (left) the NH and (right) the SH in (top) JJA and (bottom) DJF. The crosses show individual years for the CTL experiment and the diamonds show individual years for the UNI experiment; the large symbols denote the averages over all years for each experiment.
There is a large amount of scatter in the points, particularly in the winter seasons. The correlation between the two variables within each of the datasets is quite weak (|r| < 0.4). In all cases, the difference in the polar cap at 100 hPa between the UNI and CTL experiments is statistically significant at the 95% confidence level. The statistical significance of the difference in the average jet latitude (Δφ) between the two experiments has been tested using a Monte Carlo random sampling method of the merged datasets (e.g., Wilks 2006). In the cases where the difference in the average jet latitude is found to be significant at the 95% confidence level, the value of Δφ is shown on the plot.
The largest difference in Δφ is in the NH in DJF, with a poleward shift of 2.8°. In this case, the mean difference in is −4 K. In the NH in JJA, the difference in is ~ of that in DJF. The corresponding Δφ is also smaller, with a poleward shift of 1.3°. The extra coupling in DJF is interesting and would be consistent with the mechanism of interactions with storms due to changes in upper-level shear, since this is stronger in winter. In the SH, there is a poleward shift in the jet latitude of 1.8° in DJF. However, despite the difference in polar cap being similar in JJA, there is no significant difference in the jet latitude. This suggests that the underlying troposphere is in some sense more coupled with the extratropical lower stratosphere in DJF than in JJA. This is consistent with the decorrelation time scale of the southern annular mode (SAM) being longer in austral summer than in austral winter (Baldwin et al. 2003). In JJA, the SH eddy-driven jet is located further poleward and is relatively separated from the strong subtropical jet in JJA, creating a “split” jet in the upper-level zonal wind (e.g., Hoskins and Hodges 2005). Conversely, in DJF the subtropical jet is relatively weak and the jet is predominantly eddy-driven. There is therefore a seasonal cycle in the background tropospheric winds, which may influence the interaction and feedbacks between eddies and the mean flow (e.g., Barnes and Hartmann 2010). Furthermore, the stratospheric winds play an important role in determining the tropospheric dynamical time scales in austral spring, with variability in the breakdown of the polar vortex causing an increase in the SAM time scale (Simpson et al. 2011). The seasonal cycle in the tropospheric SAM time scales may affect the amplitude of the response to a given forcing at different times of the year.
To summarize, the overall poleward shift in the eddy-driven jets in both hemispheres in response to the increase in SWV agrees with the response to stratospheric polar cooling in simplified GCMs (e.g., Simpson et al. 2009; Tandon et al. 2011). However, there are distinct seasonal variations in the tropospheric response in both hemispheres, which we have been able to evaluate through the use of a more complex GCM that includes a seasonal cycle.
b. The impact of SWV changes in the upper versus lower stratosphere
The results so far have shown that a uniform doubling in SWV causes circulation changes in both the stratosphere and troposphere. This raises the question as to whether changes in SWV in the upper or lower stratosphere are more important for driving changes in the circulation. For example, it would be useful to know in which regions it is most important for GCMs to correctly simulate SWV, and also whether SWV trends due to changes in transport or changes in stratospheric chemistry are likely to have a bigger dynamical impact.
Figure 8 shows the DJF seasonal-mean differences in (top) (K) and (bottom) (m s−1) compared to the CTL experiment for the LS and US experiments in the left and center panels, respectively. The right-hand panels show the sum of the differences in the two experiments [i.e., (US + LS) − UNI]. The hatching in the left and center panels indicates where the response is not statistically significant at the 95% confidence level. It is not possible to conduct a significance test on the (US + LS) − UNI differences, and therefore shading is not included in these panels. As expected, the strong cooling in the extratropical lower stratosphere is retained in the LS experiment. In the US experiment, there is a more homogeneous cooling of ~1–2 K at pressures less than ~50 hPa. The right-hand panels show that the responses in the LS and US experiments combine to give a similar response to that in the UNI experiment (see Fig. 1). However, the warming over the NH winter pole in DJF between 1 and 10 hPa is ~1–2 K greater than in the UNI experiment, which suggests that the upper stratospheric cooling acts to reduce this dynamical feature of the response. Figure 8e shows that there is almost no change in in the US experiment in either the stratosphere or troposphere. To first order, this is because the cooling in the upper stratosphere has a weak horizontal gradient. Conversely, in the LS experiment, the differences in in the stratosphere and troposphere resemble those in the UNI experiment, both in terms of magnitude and structure.
The vertical residual mass flux in the tropics, F↑, at 70 hPa in the LS and US experiments is plotted in Fig. 4. The vertical mass fluxes in the LS and US experiments almost exactly match those in the UNI and CTL experiments, respectively. There is virtually no change in the mass flux in the US experiment, and almost all of the increase in the UNI experiment is reproduced in the LS experiment. This suggests that the changes in stratospheric wave drag, and therefore the residual circulation, in the UNI experiment are almost entirely due to the changes in temperature and wind associated with the increase in SWV in the lower stratosphere.
The MSLP response in the LS experiment is also very similar to the UNI experiment (not shown). Conversely, no statistically significant changes in MSLP occur in the US experiment. Water vapor trends in the lower stratosphere (e.g., due to changes in tropical lower stratospheric transport) are therefore likely to be more important for the global circulation than equivalent trends in the upper stratosphere (e.g., due to methane oxidation).
This study has investigated the stratospheric and tropospheric circulation response to a uniform doubling in SWV from 3 to 6 ppmv in a comprehensive stratosphere-resolving atmospheric GCM. This represents an idealized SWV perturbation that is more akin to the effects of a long-term trend, such as those simulated for the future period by the CCMVal-2 models under increased long-lived greenhouse gas concentrations (Gettelman et al. 2010), than to the variations found in the observational record (e.g., Hurst et al. 2011). A future paper will consider the impact of changes in SWV derived from past observations (A. C. Maycock et al. 2012, unpublished manuscript).
A uniform increase in SWV causes stratospheric cooling, which is largest in the extratropical lower stratosphere throughout the year. This causes a more westerly zonal-mean zonal wind on the upper flanks of the subtropical jets. In the Northern Hemisphere (NH), there is an increase in the resolved wave drag in the mid and upper stratosphere from September to April, and a weaker westerly jet at pressures less than 10 hPa. Conversely, in the Southern Hemisphere (SH) there is a decrease in upper-stratospheric wave drag of ~25% from July to September, an increase in the strength of the stratospheric jet by up to 14 m s−1, and a delay in the final warming by ~13 days. This is similar to the impact of Antarctic ozone depletion on the stratosphere from November to February (e.g., Thompson et al. 2011; Polvani et al. 2011). There is an increase of ~10% in the vertical mass flux in the tropical lower stratosphere associated with the residual circulation throughout the year. This is mainly related to an increase in wave drag in the suptropical lower stratosphere in both hemispheres, and is consistent with studies of the residual circulation response to increases in long-lived greenhouse gas (LLGHG) concentrations (e.g., Butchart et al. 2006; Shepherd and McLandress 2011). In boreal winter, the increase in the residual circulation is also partly due to the increase resolved wave drag in the mid and upper stratosphere in the NH. The increase in the residual circulation is largely due to changes in the resolved wave drag, with the parameterized orographic and nonorographic gravity wave drag playing only a minor role. This is in contrast to other studies that show a significant role for both resolved and parameterized wave drag in the response to increased LLGHGs (e.g., Li et al. 2008; McLandress and Shepherd 2009). Understanding such differences between models remains an area of future research interest. The changes in wave drag and the related increase in the residual circulation occur almost exclusively in response to the increase in SWV in the lower stratosphere (p > 50 hPa).
The changes in the lower stratospheric state impact the large-scale extratropical tropospheric circulation. There are north–south dipole changes in mean sea level pressure over the extratropical ocean basins, which are concurrent with a poleward shift in the eddy-driven jets and a more positive annular mode index. The largest changes in the tropospheric circulation occur from December to February in both hemispheres. The changes in the tropospheric circulation are also almost entirely due to the increase in SWV in the lower stratosphere (p > 50 hPa). A uniform change in SWV in the upper stratosphere (p < 50 hPa) has virtually no impact on the atmospheric circulation in our model. The results of this study represent the climate response to an SWV perturbation in the absence of changes in sea surface temperatures (SSTs). The effect of coupling to SSTs is likely to be more important for SWV than for changes in other stratospheric constituents, such as ozone, for which the global-mean radiative forcing is quite small. A follow-up study will therefore consider the role of SST changes in the circulation response to SWV perturbations.
The tropospheric circulation changes due to uniformly increasing SWV are consistent with those found by Tandon et al. (2011). However, the use of a more comprehensive GCM with a well-resolved stratosphere has highlighted important changes in the stratospheric circulation which will not be captured in a simplified GCM without topography and in the absence of an annual cycle, such as that used by Tandon et al. (2011). These different experimental approaches should therefore be considered mutually complementary, and demonstrate that the nature of the tropospheric response appears robust across GCMs of varying complexity and differing configurations.
The results of this study indicate that trends in SWV may be an important driver of extratropical climate variability on interannual to multidecadal time scales. For example, it has been shown that the positive trend in the North Atlantic oscillation (NAO) from 1960 to 2000 can be almost entirely explained by the observed trend in lower stratospheric wind over this period (Scaife et al. 2005). The observed increase in SWV over this period may therefore have contributed to the NAO trend (Joshi et al. 2006). Furthermore, simulations under increasing LLGHG concentrations show increases in SWV of up to a factor of 2 over the twenty-first century (Gettelman et al. 2010). Trends in SWV should therefore be considered alongside LLGHGs and stratospheric ozone recovery as an important mechanism for driving temperature and circulation changes in both the troposphere and stratosphere in future climate projections. However, since the limited evidence available suggests past variations in SWV are generally not well reproduced in models (e.g., Oman et al. 2008), the sign and magnitude of future trends remains a source of uncertainty for future climate variability and change.
A. C. Maycock was supported by a NERC PhD studentship and a CASE award from the UK Met Office. M. M. Joshi was funded by NCAS-climate. A. A. Scaife was supported by the Joint DECC/Defra Met Office Hadley Centre Climate Programme (GA01101). The authors would like to thank the detailed comments from the three anonymous reviewers which greatly helped to improve the manuscript.