An idealized, dry general circulation model is used to examine the response of the tropospheric circulation to thermal forcings that mimic changes in stratospheric water vapor (SWV). It is found that SWV-like cooling in the stratosphere produces a poleward-shifted, strengthened jet and an expanded, weakened Hadley cell. This response is shown to be almost entirely driven by cooling located in the extratropical lower stratosphere; when cooling is limited to the tropical stratosphere, it generates a much weaker and qualitatively opposite response. It is demonstrated that these circulation changes arise independently of any changes in tropopause height, are insensitive to the detailed structure of the forcing function, and are robust to model resolution. The responses are quantitatively of the same order as those due to well-mixed greenhouse gases, suggesting a potentially significant contribution of SWV to past and future changes in the tropospheric circulation.
Water vapor is the dominant greenhouse gas of Earth’s atmosphere, but important questions remain about its future changes and their impact on climate. This is particularly true of stratospheric water vapor (SWV), whose abundance depends on numerous factors—most importantly, methane oxidation and tropical cold-point temperatures (e.g., Fueglistaler et al. 2009)—for which predictions are highly uncertain (e.g., Solomon et al. 2007, chapter 10). Data over select regions have shown a ~0.5 ppmv decade−1 increase in stratospheric water vapor for much of the twentieth century (Rosenlof et al. 2001; Hurst et al. 2011). After a decrease of ~0.4 ppmv between 2001 and 2005 (Randel et al. 2006; Solomon et al. 2010), SWV levels have increased ~0.5 ppmv since 2006 (Hurst et al. 2011).
As for the effects of SWV changes, earlier studies have focused almost exclusively on the radiatively forced temperature response. Using fixed dynamical heating (FDH) and energy balance models, Forster and Shine (2002) have shown that an increase in SWV causes cooling in the stratosphere (enhanced in the extratropical lower stratosphere) and warming at the surface. This agrees with the FDH calculations of Maycock et al. (2011) and R. Portmann (2010, personal communication), and with the general circulation model (GCM) simulations of Oinas et al. (2001) and Smith et al. (2001). That said, the circulation response to SWV change has received very little attention. One study (Joshi et al. 2006), using a comprehensive GCM, has shown that SWV increase is a significant contributor to recent strengthening of the North Atlantic Oscillation.
This leaves open several important questions which we consider in this paper: First, how does SWV change impact the global circulation—both tropical and extratropical? Second, is the tropospheric circulation response driven preferentially by specific regions of stratospheric cooling? Third, how does the circulation response vary with the amount of SWV change? To address these questions, we employ an idealized, dry GCM in which we impose thermal forcings that mimic the effects of SWV. Using such a model affords greater control over the placement and magnitude of the thermal forcing. Thus we develop a clearer picture of not only the global circulation response to SWV but also the precise contributors to that response.
In this study, we use a model consisting of only dry dynamics and highly idealized parameterizations borrowed from Schneider (2004, hereafter S04) and Schneider and Walker (2006, hereafter SW06). In contrast to the more popular Held and Suarez (1994) configuration (e.g., Polvani and Kushner 2002; Haigh et al. 2005; Lorenz and DeWeaver 2007; Butler et al. 2010), our choices produce a tropical climatology that is closer to observations: the tropical tropopause reaches a potential temperature of 370–380 K, and static stability reflects a moist adiabatic lapse rate (Fig. 1a), compared to the 340–350-K tropopause and dry adiabatic lapse rate of Held and Suarez (1994) (Fig. 1b). For complete reproducibility, all model details are provided in the appendix. Suffice it to say, our model produces a hemispherically symmetric mean circulation, resembling equinoctial conditions.
To study SWV changes, we impose a number of thermal forcings, consisting of perturbations of the model’s radiative equilibrium temperature. The amplitude of each perturbation is controlled primarily by a parameter δT, and additional parameters control its latitudinal and vertical structure. In Fig. 2, we provide visual examples, with complete details available in the appendix. The “LS” (i.e., lower stratosphere) forcing (e.g., Fig. 2a) mimics the temperature response due to a uniform increase in SWV, with strong cooling in the extratropical lower stratosphere and weaker cooling in the tropics and higher altitudes (Forster and Shine 2002; Maycock et al. 2011). As a point of reference, the LS-NEG5 case (δT = −5 K) corresponds roughly to a 5 ppmv increase in SWV.
With the “ELS” forcing (e.g., Fig. 2b), we isolate the extratropical portion of the LS perturbation, and with the “TLS” forcing (e.g., Fig. 2c) we isolate the tropical portion. As an additional test, we apply the “LSdp” forcing (e.g., Fig. 2d), in which the LS perturbation is shifted in altitude. When considering specific perturbation amplitudes, we append the above shorthand appropriately; for example, “LS-POS5” indicates a LS forcing integration with nominal amplitude δT = +5 K.
For each of the above forcing functions—and for each choice of δT—we start our model from rest and integrate for 10 000 days, which is sufficient to obtain a statistically stationary response. To compute all climatological fields, we discard the first 300 days as spinup and time average the rest. To obtain the “response” of the model, we subtract the climatology of the control integration, in which δT = 0 (shown in Fig. 1a). Since there is no topography in this model, and all forcings are hemispherically symmetric, all responses should be hemispherically symmetric. Any asymmetry that remains is due to sampling error. Unless otherwise stated, all integrations are performed at spectral resolution T42 with 40 vertical levels. (See appendix for details.)
We first consider the effect of a thermal perturbation that mimics a 5 ppmv increase in SWV, corresponding to our LS-NEG5 forcing. The response is shown in Fig. 3, left column. Because the stratosphere is close to radiative equilibrium, the temperature response (Fig. 3a) shows a strong resemblance to the thermal forcing (Fig. 2a). The cooling in the stratosphere produces a ~30 hPa rise in polar tropopause height (thick dotted contour), with gradually vanishing tropopause changes toward lower latitudes.
Figure 3b shows the zonal wind response to the LS-NEG5 forcing. There is deceleration of ~1.5 m s−1 on the equatorward flank of the jet, accompanied by ~1.5 m s−1 acceleration on its poleward flank, indicating a poleward shift of the jet. Figure 3c shows the changes in the meridional mass streamfunction Ψ (see Peixoto and Oort 1992, section 7.4.3 for the definition). In the northern hemisphere, there is a slight decrease in Ψ within the Hadley cell, a more substantial increase in Ψ at the Hadley cell edge, and a substantial decrease in Ψ near the poleward edge of the Ferrel cell. These features indicate a weakening and widening of the Hadley cell and a poleward migration of the Ferrel cell. The same interpretation applies in the southern hemisphere but with the sign of Ψ reversed.
A natural question arises next: which latitudes of SWV change contribute most to this response? To answer this, we show the ELS-NEG5 integration, in which cooling is confined to the extratropical lower stratosphere (Fig. 3d). In this case, the tropospheric response is nearly indistinguishable from LS-NEG5, showing a poleward shift of the jets (Fig. 3e), expansion and weakening of the Hadley cells, and a poleward shift of the Ferrel cells (Fig. 3f). This suggests that most of the tropospheric response in LS-NEG5 is attributable to cooling in the extratropical lower stratosphere.
To complement ELS-NEG5, we also present the TLS-NEG10 integration, in which cooling is confined to the tropical lower stratosphere (Fig. 3g). To elucidate the tropospheric response, which is weaker than with extratropical forcing, we here increase the perturbation amplitude to δT = −10 K. Notice that with tropical forcing, the response is qualitatively opposite to that of LS-NEG5 and ELS-NEG5, with a slight equatorward shift of the jets (Fig. 3h), contraction of the Hadley cells, and an equatorward shift of the Ferrel cells (Fig. 3i). For lower amplitudes of tropical cooling (not shown), qualitatively similar features result, but the response is very weak and much longer integrations are required to reach equilibrium. These findings confirm that, in LS-NEG5 (Figs. 3a–c), cooling in the tropical stratosphere contributes negligibly to the tropospheric response.
We have extensively explored the space of forcing functions and amplitudes, and Fig. 4 summarizes the results. Each panel shows several curves, each one representing a set of integrations with the same thermal forcing structure but different forcing amplitudes. For stratospheric cooling (δT < 0), both the LS (red circles) and ELS (blue circles) forcings produce poleward-shifted jets (Fig. 4a) and expanded Hadley cells (Fig. 4b). Because of the hemispheric symmetry of our model, expansion of the Hadley cells indicates a poleward shift of the Hadley cell edges, which also indicates poleward migration of the Ferrel cells. These results are also apparent from Fig. 3, discussed above. For δT < 0, the LS and ELS forcings also produce an increase in jet speed (Fig. 4c) and a weakening of the Hadley circulation (Fig. 4d). All these results, one should note, are robust to changes in the model’s vertical resolution. Whether it is doubled to 80 levels (red squares) or halved to 20 levels (red crosses), the results are essentially identical to the standard 40-level case, over the full range of δT.
The TLS integrations, with forcing confined to the tropics, are indicated by the green circles in Fig. 4. TLS shows a δT dependence opposite to that of LS and ELS, as already illustrated in Fig. 3, and the response is comparatively small over the full range of δT. This further confirms that the tropical stratosphere contributes very little to our model’s response. This does not mean, however, that the LS results are simply the linear composite of ELS and TLS. For example, the shift of the jet in the LS-POS10 case (−4.34 ± 0.04°) is actually larger in magnitude than for ELS-POS10 (−3.42 ± 0.04°), while linearity would suggest a smaller shift.
The results of Fig. 4 are nonlinear in another sense: the response to stratospheric warming (δT > 0) is not simply the reverse of the one associated with stratospheric cooling (δT < 0). This nonlinearity is most extreme in the cases of jet speed (Fig. 4c) and Hadley cell strength (Fig. 4d), which even show some nonmonotonicity.
We first offer a few points of comparison with earlier work. Joshi et al. (2006) simulate a SWV change corresponding to roughly a −1-K perturbation of the northern extratropical stratosphere. The magnitude of their zonal wind response is somewhat larger than what we obtain in our LS-NEG2 integration (not shown). Note, however, that Joshi et al. (2006) consider winter averages rather than the equinoxes and that their model includes topography and a polar vortex, which may explain the quantitative difference. The differences are more drastic when we compare with earlier idealized studies. For example, the zonal wind responses in Haigh et al. (2005) and Lorenz and DeWeaver (2007) are greater than ours by factors of 2–4, when considering comparable forcing functions. The reasons for this model-dependence require further study. Nonetheless, our results are in qualitative agreement with earlier work.
Second, as noted earlier, the thermal forcings we have imposed produce substantial changes in tropopause height. One might argue that the tropospheric circulation response we have shown is a simple consequence of tropopause height change, a finding that is well-documented in the literature (e.g., Williams 2006; Lorenz and DeWeaver 2007). To show that tropopause change is not here the primary cause of the tropospheric circulation response, we present an additional series of integrations, labeled LSdp, in which the stratospheric forcing (e.g., Fig. 2d) is identical to the LS forcing (e.g., Fig. 2a) but shifted higher so as not to affect tropopause height. The response of LSdp-NEG5, shown in Fig. 5, is qualitatively almost identical to the one in Figs. 3a–c, with a slightly smaller amplitude, while tropopause height is basically unchanged. We have verified that this behavior holds over the entire range of forcing amplitudes, as indicated by the red triangles in all panels of Fig. 4.
We also note that there are slight structural differences between our analytically constructed thermal forcing and the thermal response to SWV increase shown in Forster and Shine (2002). For instance, SWV-caused cooling decreases more gradually with height than in our LS forcing. One might wonder how sensitive our results are to the details of our forcing function. To address this question, we have conducted a series of integrations, labeled “WS” (i.e., whole stratosphere), in which we impose uniform cooling of the entire stratosphere; that is, we cool every point above the tropopause by the same amount δT, irrespective of height and latitude. The results (Fig. 4, black circles) show that, over the full range of δT, there is little difference between this simplest WS forcing and the original LS forcing (red circles) that was constructed to mimic Forster and Shine (2002). This further establishes the robustness of our results.
Whether one considers our δT = −5 K integrations, or more conservatively the δT = −2 K results (which scale almost linearly), one conclusion remains the same: changes in SWV—in particular extratropical SWV—generate circulation responses that are of the same order as the modeled response to increased well-mixed greenhouse gases. This is clear, for instance, when comparing the tropical widening in Fig. 4b with Fig. 2 of Johanson and Fu (2009). Indeed, this response is small when compared to observed Hadley cell expansion, but models in general have failed to reproduce this recent trend (Seidel et al. 2008; Johanson and Fu 2009), so the dominant contributors to tropical widening remain unclear. As for the extratropical circulation, our results show responses that are significant in the context of both past changes and future projections; compare the zonal wind responses of Figs. 3b and 3e with Figs. 3 and 6 of Lorenz and DeWeaver (2007). Thus SWV is clearly a key component of extratropical circulation change, and it may prove significant to tropical circulation change as well.
We thank Karen Rosenlof and Shuguang Wang for helpful discussions and three anonymous reviewers for constructive feedback. NFT and LMP are supported by a grant from the U.S. National Science Foundation. SMD is supported by NASA Aura Grant 07-ASTR07-0004 and the NOAA ACC program.
We use the spectral dynamical core of the Geophysical Fluid Dynamics Laboratory (GFDL) Flexible Modeling System (FMS), which employs the spectral transform method (Hoskins and Simmons 1975) in the horizontal and the finite-difference method (Simmons and Burridge 1981) in the vertical. The horizontal truncation is T42. The vertical level interfaces, in sigma coordinates, are σi = (i/L)3, i = 0, 1, 2, … , L, where L is an integer. For all integrations, L = 40 unless otherwise stated.
To mimic convective and radiative processes, we employ two linear relaxation terms in the temperature equation,
where TC and τC are the convective equilibrium temperature and time scale, respectively; TR and τR are the radiative equilibrium temperature and time scale, respectively; and T* is our external thermal forcing.
As in SW06, TC is given by
ensures conservation of enthalpy in (A2); Tm is the moist adiabat,
Ts is the surface temperature at longitude–latitude–time (λ, φ, t); Γm = 7 K km−1; p0 = 1000 hPa; and pLNB is the level of neutral buoyancy for ascent from the surface along Tm (e.g., Bohren and Albrecht 1998, section 6.7). Equation (A2) is applicable only when EC ≥ 0. If EC < 0, then convection is inhibited; that is, TC = T in the entire column. The time scale τC is set to 4 h.
As in S04, TR is given by
Tstrat = 200 K, T0 = 260 K, and Δh = 90 K (see contours in Fig. 2). The time scale τR has a vertical and latitudinal dependence given by
where σ = p/p0, τa = 50 d, τs = 7 d, and σb = 0.7 (S04). We have also tested a more realistic configuration in which TR increases and τR decreases in the middle–upper stratosphere. While this slightly affects the magnitudes of our results, there is no qualitative change.
The thermal forcing T* is the only term that we vary among integrations, and it takes the form
where δT controls the forcing amplitude and sp = 0.2 hPa−1. The parameters p1, p2, and w(φ) control the vertical and latitudinal structure of the thermal forcing, and their values are given in Table A1 for each set of integrations. Note, the bracketed term in (A8) is approximately one for p2 < p < p1 and zero elsewhere.