Relaxing the well-mixed greenhouse gas approximation in climate simulations: Consequences for stratospheric climate

Authors

  • C. L. Curry,

    1. School of Earth and Ocean Sciences, University of Victoria, Victoria, British Columbia, Canada
    2. Canadian Centre for Climate Modelling and Analysis, Meteorological Service of Canada, University of Victoria, Victoria, British Columbia, Canada
    Search for more papers by this author
  • N. A. McFarlane,

    1. Canadian Centre for Climate Modelling and Analysis, Meteorological Service of Canada, University of Victoria, Victoria, British Columbia, Canada
    Search for more papers by this author
  • J. F. Scinocca

    1. Canadian Centre for Climate Modelling and Analysis, Meteorological Service of Canada, University of Victoria, Victoria, British Columbia, Canada
    Search for more papers by this author

Abstract

[1] The climatic consequences of relaxing the uniform greenhouse gas (GHG) assumption in the Canadian Centre for Climate Modelling and Analysis atmospheric general circulation model are examined. A simple chemical loss parameterization for nitrous oxide, methane, CFC-11, and CFC-12 is employed that includes stratospheric water vapor production from methane oxidation. Multidecadal mean distributions of these species are obtained that compare reasonably well with UARS satellite observations of the stratosphere. The radiative impact of these changes is a widespread cooling of the stratosphere (with a spatially averaged, annual mean value of 0.6 K), compared to the model with specified uniform GHG distributions. This cooling results from an approximate doubling of the amount of middle to upper stratospheric moisture (as a result of methane oxidation) and exceeds the radiatively induced warming due to decreases in the other GHGs. Annual mean temperature changes of up to +8 K in the upper winter polar stratosphere, by contrast, are dynamically induced because of increases in the residual mean circulation and associated heating.

1. Introduction

[2] Greenhouse gases (GHGs) play a key role in Earth's radiative energy balance, and are thought to be responsible for much of the radiative forcing of climate since the preindustrial era. Their representation in general circulation models of the atmosphere (AGCMs) was initially limited to the dominant species: namely water vapor, carbon dioxide, and ozone. The variety of other radiatively active gases, all present at a concentration far less than that of CO2, were often represented in terms of “equivalent CO2”: that is, as an amount of CO2 estimated to produce the same radiative effect. There have been several critical studies comparing the equivalent CO2 approach with explicit inclusion of the subordinate GHGs [Wang et al., 1991, 1992; Govindasamy et al., 2001], showing that an equivalent global mean radiative forcing may conceal a host of differences, among them the pattern of the forcing, temperature response and radiative heating rate, and aspects of the atmospheric circulation.

[3] Not only the variety of GHGs must be grappled with (including the ever increasing number of anthropogenically generated species), but also their spatial and temporal variability. Early work considered all GHGs (except ozone and water vapor) to be uniformly mixed throughout the atmosphere, a plausible assumption for tropospheric models used primarily for the study of radiative budgets and dynamical properties. However, AGCM domains now routinely extend into the stratosphere, where most GHGs (except CO2) undergo large excursions from their tropospheric values. The dominant sinks of many key species are atmospheric (and often exclusively stratospheric, as for N2O and the CFCs), and the distribution of these sinks is responsible for much of the observed spatiotemporal variation of these gases. Furthermore, most GHG perturbation experiments that incorporate a middle atmosphere suggest that the maximum temperature response to increasing GHGs is expected near the stratopause (between 5 and 0.3 hPa [Akmaev and Fomichev, 2000]). This emphasizes the need to consider the entire troposphere-stratosphere system in climate change detection and attribution efforts.

[4] In this work, we conduct a direct comparison between the effect of assumed uniform versus three-dimensional (3-D) distributions of the main radiatively active GHGs, in the context of fully dynamical AGCM simulations. Some studies have attempted to quantify the radiative impact of GHG spatial heterogeneity, albeit in a less direct manner. Using a one-dimensional radiative-convective model, Ramanathan et al. [1985] carried out GHG perturbation experiments using both uniform and exponentially decreasing vertical profiles for N2O, CH4, and CFCs. For a doubling of N2O and CH4, they found no significant change in the computed surface temperature difference, δTs, between uniform and nonuniform cases, but noted a 15% smaller δTs in the nonuniform case for a 1 ppbv increase in CFC surface concentrations.

[5] Minschwaner et al. [1998] used zonally averaged GHG distributions from the UARS satellite along with a radiative transfer code to compute the net radiative forcing (i.e., the change in net radiative flux across the tropopause following a GHG perturbation, allowing for stratospheric adjustment [Ramaswamy et al., 2001]) of each species compared to its estimated preindustrial forcing. Repeating the calculations for globally uniform concentrations they found negligible differences in this quantity for N2O and CH4. However, in the case of CFC-12 a 10% smaller global mean radiative forcing was inferred in the nonuniform case. Hansen et al. [1997], who examined the effect of uniform versus 3-D distributions of CFC-11 and -12, obtained results similar to both previous studies, and noted considerable sensitivity to the assumed concentration decrease in the vertical. Boville et al. [2001], using a simple chemical loss parameterization for several radiatively active trace gases in the NCAR CCM3 coupled GCM (CSM-1.3), obtained 3-D distributions that reproduced many features of observed GHGs. Indeed, we adopt a similar approach in section 2.2 below. These authors did not explore the radiative or dynamical consequences of these distribution changes, however.

[6] Perhaps the most relevant examination of this issue in the present context was conducted by Freckleton et al. [1998]. Using a 3-D chemical transport model to calculate the distribution of CH4, observed distributions of CFCs, and a narrowband radiative transfer scheme, these authors found that assuming a uniform GHG distribution results in an overestimate of the radiative forcing since the preindustrial period for both CH4 and CFCs. For CH4, the differences were ≲4%, depending on latitude, but the overestimate for CFCs ranged from 5% at the equator to >40% at the poles, where the forcing is smallest. However, Freckleton et al. used the fixed dynamical heating approximation to calculate the radiative forcing due to the perturbations, which suppresses certain dynamical feedbacks present in a full AGCM, and also mentioned that the CH4 distribution may not have reached chemical equilibrium at the time of the radiative calculations.

[7] The destruction of methane occurs mainly via OH oxidation, although reaction with O(1D) contributes roughly equally above the midstratosphere, and photodissociation becomes nonnegligible in the upper stratosphere and mesosphere. Methane oxidation therefore constitutes a significant source of humidity in the stratosphere, accounting for 50% or more of the H2O mixing ratio above ∼30 km according to models [Boville et al., 2001]. Thus a future increase in tropospheric CH4 concentration should cause a corresponding increase in stratospheric humidity, although changes in cross-tropopause transport of H2O also need to be considered [Rosenlof, 2002]. The radiative impact of these putative stratospheric H2O changes has been the focus of many studies. Rind and Lonergan [1995] performed a stratospheric H2O doubling experiment with specified uniform H2O concentrations in both the control and perturbation runs. They found a stratospheric cooling of 2–3 K, an upper tropospheric warming of 0.5 K, and a negligible impact on surface temperature when sea surface temperatures (SSTs) were specified. A small positive δTs was found when a mixed layer ocean was included. Forster and Shine [1999] used a coarse resolution coupled GCM to simulate the effect of a ∼10% uniform increase in stratospheric H2O, meant to approximate the observed increase over the previous 20 years [Rosenlof, 2002]. Their results agreed qualitatively with those of Rind and Lonergan, and also revealed the sensitivity of the results to the exact location and magnitude of the water vapor perturbation [Oinas et al., 2001; Forster and Shine, 2002]. MacKenzie and Harwood [2004] examined the middle atmosphere response to a humidity increase due to CH4 oxidation in a future equilibrium scenario. In addition to the expected radiative response, these authors found a modest strengthening of the stratospheric (residual mean) circulation. These studies and others confirm much earlier work on the expected radiative and dynamical impacts of stratospheric GHG increases, starting with the seminal work of Fels et al. [1980].

[8] In this paper, we investigate the impact of GHG spatial heterogeneity, particularly in the stratosphere, by isolating radiative and dynamical differences between the uniformly mixed and 3-D GHG distributions. In section 2, we describe the model and experimental procedure. In section 3, we present the modeled tracer distributions and compare them to observations. The radiative and dynamical effects of these 3-D distributions are examined in section 4, followed by further discussion and conclusions in section 5.

2. Model Description and Experimental Design

[9] The GCM experiments were performed with the CCCma third generation atmospheric GCM (AGCM3), an improved version of the model described by McFarlane et al. [1992]. AGCM3 employs a spectral transform method to represent the horizontal structure of the main prognostic variables. The spectral representation uses triangular truncation at principle wave number 47, with an associated gaussian physics grid of 96 × 48 points (3.75 × 3.75 degrees). The vertical representation is in terms of a hybrid coordinate η defined on a finite element grid. In terms of η, the ambient pressure is p = p0η + (psp0)[(η − ηT)/(1 − ηT)]3/2, where ηTpT/p0 with pT = 1.06 mbar, p0 = 1013 mbar, and ps the instantaneous surface pressure. In the standard model configuration, there are 31 hybrid vertical levels between 995 mbar (approximately 50 m above the surface at sea level) and 1 mbar (∼48 km), with zero mass flux boundary conditions applied at the upper (0.5 mbar) and lower (1013 mbar) boundaries. In order to better resolve the upward transport of water vapor through the tropopause region, the model vertical levels were altered from the standard configuration. The number of levels in the 50–300 mbar range was increased from 5 to 13, while the number of tropospheric levels was reduced by 4 (with an appropriate rearrangement to keep the layer depth smoothly varying with height). The net result was an increase in the total number of model levels to 35. This rearrangement had a minimal effect on the surface climate, while increasing the minimum mixing ratio of H2O at the tropopause by almost an order of magnitude (thus correcting an existing bias in the current version of the AGCM). In the modified level scheme, the vertical resolution varies from 0.064 km at the surface to 0.70 km in the lower stratosphere to 7.2 km at the model top.

2.1. Improvements in AGCM3

[10] Improvements of AGCM3 over its predecessor include: a multicomponent land surface scheme (CLASS) [Verseghy et al., 1993], a mass-flux-based cumulus cloud parameterization [Zhang and McFarlane, 1995], a revised representation of topography and turbulent transfer at the Earth's surface [Holzer, 1996; Abdella and McFarlane, 1996], an anisotropic orographic gravity wave drag parameterization [Scinocca and McFarlane, 2000], and an improved treatment of radiative transfer (see below). An early version of AGCM3 extended upward into the mesosphere (to 0.0006 mbar) was described in detail by Beagley et al. [1997] and the general features of the tropospheric and lower stratospheric climatology may be found there. The climate of the middle to upper stratosphere, however, differs somewhat because of the placement of the upper boundary at 0.5 mbar in the present model.

[11] Several aspects of passive tracer transport in AGCM3 were explored by Holzer [1999] and Holzer and Boer [2001], but chemically active tracers had not yet been incorporated. The GHG tracer advection is performed spectrally and, because of the absence of explicit sources (section 2.3), the mixing ratios remained smooth and positive at all times. Because of the large dynamic range of specific humidity in the model domain, water vapor was treated separately from the other tracers insofar as a transformed variable was used, defined in terms of the specific humidity q as

equation image

where q0 is a constant [Boer, 1995]. The resulting smoothness of the moisture field comes at a price, however: while spectral advection of the transformed variable s is conservative, the resulting q is in general nonconservative. However, by an appropriate choice of q0, global nonconservation can be kept acceptably low [Merryfield et al., 2003] (available at http://www.cmc.ec.gc.ca/rpn/wgne/), and an additional procedure is employed to ensure strict global conservation at each time step [McFarlane et al., 1992].

[12] The treatment of solar radiative heating in AGCM3 is the same as in AGCM2 [McFarlane et al., 1992], except that the previous two-band spectral partitioning was replaced with a four-band scheme ranging continuously from 0.25 μm to 4.0 μm. This change has little influence on the clear-sky fluxes, but leads to an improved representation of cloud-radiation interactions. AGCM3 also includes the Slingo [1989] parameterization linking cloud optical properties to liquid water content and droplet effective radius. Aerosols are treated as a background field, with geographically dependent type and optical/infrared properties [World Meteorological Organization, 1983].

[13] Improvements were also made to the AGCM2 terrestrial longwave radiation scheme. This scheme is based on the broadband treatment of Morcrette [1991], with 6 spectral bands ranging from 0 to 3000 cm−1 (3.3 μm), and has been calibrated against the results of both narrowband and line-by-line models. Absorption and emission by the principal GHGs, comprising H2O, CO2, O3, N2O, CH4, CFC-11, and CFC-12, is explicitly treated. Significantly for this study, the transmission data for many of these gases were updated, and a new parameterization of the water vapor continuum was used, based on Zhong and Haigh [1995]. This entailed splitting the 500–800 cm−1 and 800–1250 cm−1 bands into four separate intervals, giving an 8-band model for the H2O continuum emission and absorption. This change reduced the longwave cooling rates in the lower troposphere, which were too high in AGCM2. Aerosols and clouds are treated as absorbers in the longwave, using the same band model as for GHGs and the emissivity model of Chýlek et al. [1992].

2.2. Chemical Parameterization

[14] The chemical destruction of N2O, CH4, CFC-11, and CFC-12 is treated in a simplified manner, wherein the time rate of change of each species due to chemical processes is given by

equation image

where X is the volume mixing ratio and the loss rate α (sec−1) is a specified function of latitude, height, and time. For N2O and CH4, the loss rates were extracted as monthly and zonal means from a dynamically equilibrated run of the Canadian Middle Atmosphere Model (CMAM), a vertically extended version of AGCM3 including interactive chemistry [Beagley et al., 1997; de Grandpré et al., 1997, 2000; Fomichev et al., 2002]. In the absence of CMAM-generated loss rates for CFCs, a single vertical loss profile was adopted for these species, representative of equinox at 30°N (from a chemical model by J. A. Logan and M. J. Prather, cited by Holton [1986]). This profile was applied at all latitudes and times. As a test, this approach was first tried for N2O, and resulted in zonal mean mixing ratios that compared well with those obtained using the fully latitudinal and time-dependent rates. This suggests that for species whose only sink (in this case, UV photolysis) maximizes in the upper stratosphere, the vertical loss profile together with the model transport are the dominant influences on the concentration distribution.

[15] Surface concentrations of each GHG are prescribed (section 2.3), and the 3-D distributions evolve in balance with this surface boundary condition. Because of chemical destruction elsewhere in the model domain, there is an associated implied surface source of each constituent. Since CO2 has no significant atmospheric sinks, it is assumed to be well mixed throughout the model domain; this despite the fact that in reality, anthropogenic sources lead to an interhemispheric gradient of a few ppmv near the surface [Gurney et al., 2002]. Since our focus is on the impact of GHG nonuniformity on the stratosphere, this simplification should be justified.

[16] The oxidation of CH4, parameterized by equation (1), ultimately results in a source of water vapor. Model studies using explicit chemistry have found that each molecule of CH4 oxidized yields approximately 1.5 to 2 molecules of H2O, with the conversion factor increasing upward in the stratosphere (where this factor is less than 2, the excess hydrogen is converted into H2 [Le Texier et al., 1988]). In the parameterization used here, we adopt a uniform value of 2 for this conversion factor, in line with other studies using a similar approach [Boville et al., 2001; MacKenzie and Harwood, 2004]. The water vapor created by this process is added to the model specific humidity and thereby incorporated into the global hydrological cycle and the radiative energy budget.

2.3. Experimental Design

[17] Climatological monthly mean sea surface temperature and sea ice extent from the AMIP2 intercomparison, spanning the observational period 1979–1996 (K. E. Taylor et al., AMIP II sea surface temperature and sea ice concentration boundary conditions, 2001, available at http://www-pcmdi.llnl.gov/projects/amip/AMIP2EXPDSN/BCS/amip2bcs.html), are specified as boundary conditions over the oceans. Stratospheric ozone is represented by the zonal mean SUNYA climatology (X.-Z. Liang and W.-C. Wang, Atmospheric ozone climatology for use in general circulation models, 1997, available at http://www-pcmdi.llnl.gov/projects/amip/AMIP2EXPDSN/OZONE/ozone.html), which spans the period October 1984 to November 1989. There is no interannual variability associated with these forcings, nor is there any response of ozone to changes in water vapor, CFCs, or ambient temperature.

[18] The model was spun up from observations and integrated for a total of 25 years, with analysis performed on the last 20 years of the run, unless stated otherwise. The lowest model level concentrations of N2O, CH4, CFC-11, and CFC-12 were set to the values 311 ppbv, 1720 ppbv, 266 pptv, and 522 pptv, respectively. These values represent the deseasonalized, globally averaged, observed surface concentrations in mid-1994 [Ehhalt et al., 2001]. The CO2 concentration was set to its mid-1994 global mean value of 358 ppmv throughout the model domain.

3. Modeled Tracer Distributions

[19] The modeled tracer distributions were compared with climatology derived from Upper Atmosphere Research Satellite (UARS) observations, which span the period October 1991 to present. The N2O and CFC-12 data are from the CLAES instrument, which had only a 16 month lifetime (January 1992 to April 1993), while the CH4 and H2O data are mainly from HALOE (v19, spanning October 1991 to September 2001), augmented in the polar regions by CLAES (for CH4) and MLS (for H2O). Details of how these monthly mean climatologies were constructed are given by the SPARC Data Center Web site and by Randel et al. [1998]. The observations of CFC-12 were presented in a preliminary form by Nightingale et al. [1996], and both CFC-11 and CFC-12 analyzed in more detail by Roche et al. [1998]. Since a validated climatology is not yet available for CFC-11, we omit a discussion of its spatial distribution in what follows, noting only that it is qualitatively very similar to that of CFC-12.

[20] As seen in Figures 1, 2, and 3, the model simulates many key features of the observed seasonal, zonal mean N2O, CH4, and CFC-12 distributions: the latitude-dependent decrease of mixing ratio with height (due to chemical destruction), the bulging isopleths in the tropics (reflecting the mean tropical ascent and extratropical descent of the Brewer-Dobson circulation [Brewer, 1949; Dobson, 1956]), and the steeper falloff of mixing ratios in the upper polar winter stratosphere than in the summer hemisphere (due to subsidence within the polar vortex). However, while the broad midlatitude plateau in Southern Hemisphere (SH) winter is also well captured by the model, it is not very evident in Northern Hemisphere (NH) winter. This midlatitude “surf zone” is indicative of quasi-isentropic mixing by planetary waves originating in the troposphere that steepen and break in the stratosphere (see section 4.5.2 and Andrews et al. [1987]). This process is highly sensitive to the sign and magnitude of the zonal wind u; thus it is important to note that the model underestimates u (by up to 50%) in the region between the tropospheric and stratospheric jets in NH winter (i.e., poleward of 30°N, 10–100 mbar; Beagley et al. [1997, Figure 3]). One consequence of the weak winds is an enhancement of the amplitude of the upward propagating waves, which leads to dissipation and mixing at lower altitudes (100–50 mbar (16–21 km)) than occurs in the real stratosphere (50–5 mbar (21–38 km); e.g., Figure 1, middle right panel). Conversely, the absence of a zonal wind bias in the model SH lower stratosphere means that most wave breaking occurs at higher altitude, consistent with the presence of the midlatitude plateau (e.g., Figure 1, top and middle left panels).

Figure 1.

Zonal mean N2O concentrations (ppbv). (top) Model 20-year mean JJA (left panel) and DJF (right panel). (middle) UARS observations for 1992 JJA (left panel) and 1992–1993 DJF (right panel). (bottom) Difference, model – UARS (ppbv), in JJA (left panel) and DJF (right panel). Positive differences are shaded.

Figure 2.

Same as Figure 1, but for CH4. Note that the UARS observations are averaged over a 10-year period in this case.

Figure 3.

Same as Figure 1, but for CFC-12. All concentrations are in pptv.

[21] Overall, the species concentrations compare favorably with other models of similar complexity [Boville et al., 2001; MacKenzie and Harwood, 2004], although there are considerable local differences as seen in the difference plots (bottom panels of Figures 13). The overall agreement of the model CH4 distribution with UARS appears to meet and even exceed that found in the models of Steil et al. [2003] and Al-Saadi et al. [2004], both of which are AGCMs with explicit chemistry.

[22] The model zonal mean water vapor distribution is shown in Figure 4. The increase in H2O concentration with height above the tropopause, due to CH4 oxidation, is evident with a discernable minimum in the 70–100 mbar range. The exact location of the minimum is seasonally dependent, being higher and more pronounced in JJA than in DJF, in agreement with the HALOE observations. Compared to HALOE, however, the model H2O mixing ratio is generally too low, by up to 1.6 ppmv depending on location. The largest discrepancy ∼40% occurs at 50–70 mbar, just above the tropical tropopause. The same deficiency has appeared elsewhere, both in other AGCMs using parameterized CH4 oxidation [Boville et al., 2001; MacKenzie and Harwood, 2004] and in models including explicit chemistry [e.g., de Grandpré et al., 1997].

Figure 4.

Same as Figure 2, but for H2O. All concentrations are in ppmv, and in the top two panels the highest contour level is equal to 2 × 104 ppmv.

[23] In the SH polar night (Figure 4, top and middle left panels), the model is able to qualitatively reproduce the characteristic dehydration feature at the lower edge of the polar westerly jet, caused by downward transport of drier air from upper levels. The excessive dehydration that afflicts some models with a “cold pole” in this region (see the model intercomparison of Austin et al. [2003] and MacKenzie and Harwood [2004]) does not occur in AGCM3, which has a “warm pole” bias of up to +14 K at 10 mbar compared to NMC climatology, possibly caused by excessive orographic gravity wave drag in these regions [Scinocca, 2003; Scinocca and McFarlane, 2000]. This implies that the saturation vapor pressure is reached less often than in the real stratosphere.

[24] The geographic structure of the water vapor field at 100 mbar is compared with HALOE observations (1993–2002 means, obtained courtesy of M. Park) in Figure 5. Since the model has a systematic low bias at this level, we use different shading to highlight regions of minimum H2O (in January) and maximum H2O (in July) in the two cases. In January (Figures 5a and 5b), the model simulates the morphology of the observed field reasonably well. In particular, a minimum in H2O is seen over southeast Asia and the tropical western Pacific in both HALOE and AGCM3, coinciding with the minimum climatological tropopause temperature (not shown; see following paragraph). In July, maxima over the South Asian and North American summer monsoon regions are also evident (Figures 5c and 5d), although the Asian maximum is distributed over a wider range of longitudes in the model. In addition, the mid-Atlantic maximum seen in the HALOE data at 15°N, 30°W is not captured by the model. These results show that the observed north-south asymmetry in upper tropospheric/lower stratospheric water vapor [Rosenlof et al., 1997] is reasonably well simulated. In order to compare the model dynamic range to that seen in observations, we show north-south transects through the H2O minima at 120°E longitude in January and through the H2O maxima at 60°E in July in Figures 5e and 5f. Figure 5e shows that apart from the 1–1.6 ppmv low bias, the model H2O mixing ratio tracks the HALOE January minimum profile quite closely, reproducing each of the local extrema seen in the data. The agreement in July at 60°E is somewhat poorer, with the model overestimating the mixing ratio gradient near 30°N, while underestimating the mixing ratio at most other latitudes.

Figure 5.

Geographic structure of water vapor mixing ratio at 100 mbar for (a) model 20-year January mean, (b) HALOE 1993–2002 January mean, (c) model 20-year July mean, and (d) HALOE 1993–2002 July mean. Note that the contour intervals and cutoffs for shading are different in each figure: In Figure 5a, contour values below 1.9 ppmv are shaded; in Figure 5b, values below 2.9 ppmv are shaded; in Figure 5c, values above 3.4 ppmv are shaded; and in Figure 5d, values above 4.4 ppmv are shaded. (e) North-south transect of Figure 5a (solid line) and Figure 5b (dashed line) at 120°E. (f) North-south transect of Figure 5c(solid line) and Figure 5d (dashed line) at 60°E.

[25] An important factor in the simulation of lower stratospheric H2O is the ability of the model to approximate the observed tropical tropopause temperature, which constitutes a first-order control on the upward transport of H2O to the stratosphere [Mote et al., 1996]. Moreover, the bulk of the moist air entering the stratosphere is supplied by strong convection over the western Pacific warm pool in DJF [Fueglistaler et al., 2005]. On the 100 mbar surface, one finds that the model 20-year DJF mean temperature minimum (186 K) and relative humidity maximum (82%) coincide over a broad region centered on 180°E and the equator (not shown; however, see Figure 5 for the H2O distribution at 100 mbar). Compared to the observed temperatures in this region (ERA-40 1979–2001 DJF mean [Fueglistaler et al., 2005]), the model is biased low by ≃4 K. This bias, coupled with the high mean relative humidity over the region, is strongly suggestive of excessive dehydration of upwelling air in the model. A rough estimate of the temperature sensitivity of the H2O mixing ratio may be obtained from standard expressions for the saturation vapor pressure over ice, ei (dependent upon T), and H2O mixing ratio (a function of ei and ambient pressure [cf. Bohren and Albrecht, 1998]). For the temperature bias and relative humidity cited above, the resulting H2O deficit is 1.5 ppmv, consistent with the values seen in the lower stratosphere in Figures 4 and 5.

[26] Figure 6 shows the annual cycle of the zonal mean H2O mixing ratio as a function of latitude at 100 mbar, the level at which much of the vertical and meridional transport of tracers occurs [Randel et al., 2001]. The HALOE/MLS data show a clear H2O peak in the NH subtropics in August–September, followed around 5 months later by opposing maxima at northern and southern high latitudes (Figure 6a). This behavior is also seen in the model (Figure 6b), although the high-latitude maxima occur earlier (in November–December) and are not as localized in time. Despite the low bias in model H2O, the amplitude of the seasonal cycle at this height and at NH midlatitudes is ∼2 ppmv in both HALOE/MLS and AGCM3.

Figure 6.

Latitude-time (January–December) variation of water vapor mixing ratio (ppmv) on the 100 mbar level. (a) HALOE/MLS data, 1991–2001; (b) model 20-year mean. In Figure 6a, the shaded region indicates values below 3.6 ppmv; in Figure 6b, the shaded region indicates values below 2.4 ppmv.

4. Effects of the Three-Dimensional Structure of Greenhouse Gases on Climate

4.1. General Considerations

[27] In this section, we consider the response of the stratosphere to changes in the concentration of a generic GHG using a simple illustrative model. The annual mean, equilibrium stratosphere is characterized as being in approximate balance between shortwave heating (largely independent of temperature), longwave heating/cooling (strongly temperature dependent) and dynamical heating/cooling, namely,

equation image

where ∂T/∂t is the temperature tendency, assumed to be nearly zero in the annual mean, and QSW, QLW, and QDYN are the shortwave, longwave, and dynamical heating rates (K d−1), respectively.

[28] The equilibrium response to a small change δX in the local concentration of the GHG is

equation image

As discussed by Fels et al. [1980] and Ramanathan et al. [1983], the annual mean response δT to nonozone perturbations δX is dictated largely by longwave radiative balance (δQLW ≈ 0), throughout the tropical stratosphere, while dynamical effects become important in the polar stratosphere and the tropical tropopause regions [Andrews et al., 1987]. The longwave radiative response to δX has two contributions: (1) the change in longwave cooling due to δX at fixed T (denoted by δQ*LW), and (2) the change due to the alteration of T itself. That is,

equation image

The subscript zero indicates that all quantities except the variable of differentiation are held constant, and trad ≡ −(∂QLW/∂T)0−1 is the radiative relaxation time for infrared cooling. For a single spectral band (here taken to be the 15 μm band of CO2, which dominates the longwave cooling in the stratosphere), and in the cool-to-space approximation (which holds sufficiently far from the lower boundary and in the absence of strong curvature of the absorption profile [Salby, 1996]), we have

equation image

where Bν(T) is the Planck function, equation image the band-integrated transmission function, X is the zonal mean equilibrium GHG concentration at pressure p, θ ≡ hν/kB, and f is an approximately temperature-independent function describing the exchange of photons between pressure level p and space [Fels, 1982]. From the definition of trad, we obtain

equation image

For CO2, θ = 960 K and f increases approximately linearly with increasing height z in the stratosphere [Fels, 1982]. Since the function T2 exp (960/T) decreases by a factor of 2.2 between the tropopause (≃190 K) and stratopause (≃270 K), equation (6) shows that trad is a decreasing function of height. It now follows from equations (3)(6) that

equation image

[29] The pressure (or height) dependence of the perturbation δX is particularly important for determining both the sign and magnitude of δT. For an exponentially decreasing GHG concentration profile, the zenith optical depth is given by τ = βLWXp, where βLW is a constant proportional to the band strength [Fels et al., 1980]. The band-integrated transmission function is then equation image = exp(−βLWXp) and the calculation of δQ*LW according to equations (4) and (5) yields

equation image

[30] Now turning to the shortwave heating, since O3 and CO2 are fixed in all of the experiments discussed in this work, only changes in water vapor affect QSW. The latter is given by

equation image

where S0 is the top of atmosphere solar intensity, μ0 the cosine of the solar zenith angle, and βSW is proportional to the band strength. Since H2O is optically thin in the stratosphere, this expression simplifies to

equation image

where equation image0 is the annual mean incident solar intensity.

[31] The above equations apply throughout the stratosphere, at the stated level of approximation. If we focus on the tropics, which is near radiative equilibrium in the annual mean, then δQDYN can be neglected to first order. Then from equations (7)(9), we obtain the temperature change due to radiative effects alone:

equation image

[32] Inspection of equation (10) now admits the following conclusions. First, one sees that for a given δX the response is largest at low temperature, reflecting the increased radiative capacity of cooler regions (e.g., the tropical tropopause). Note that for δX nonzero in the stratosphere only, the latitudinal variation in δT is expected to be minor compared to the vertical variation, at least in the annual mean, because of the weak latitudinal gradient in stratospheric T. Second, at low pressures (i.e., in the upper stratosphere), δQ*LW is large and negative (positive) for δX > 0 (<0), while for p > (βLWX)−1, δQ*LW becomes small and positive (negative) for δX > 0 (<0). In the event that δQSW can be neglected relative to δQLW (an assumption that we validate a posteriori using model results in the following section), δT is of the same sign as δQ*LW, and thus one expects δT < 0 in the upper stratosphere and δT > 0 in the lower stratosphere/upper troposphere. This change in the sign of δT is a robust characteristic of previous studies of the stratospheric response to GHG concentration change [Manabe and Wetherald, 1975; Sigmond et al., 2004].

[33] Further radiative effects also come into play. The increased downwelling longwave emission from a GHG-warmed region in the lower stratosphere is expected to warm the troposphere, although the magnitude of this effect varies among models. In this case a positive feedback may also be present: a warmer tropopause results in an increase in lower stratospheric moisture (through the elevation of the local air temperature and relative humidity), producing more absorption and re-emission.

[34] The radiative effects of all GHG changes and the relative contributions of each species are obtained in a series of GCM experiments, as summarized in Table 1. The model configuration including all of the changes to the tracers discussed in section 3 is referred to as run “ALL,” while the various perturbation experiments are described in the individual sections below. Unless stated otherwise, the model results appearing in the remainder of the paper are 20-year means.

Table 1. Experimental Design and Identifiers for the Various Model Runs
Run IDN2O Sink?CH4 Sink?CFC Sink?H2O Source?
UNInononono
DRYyesyesnono
WETyesyesnoyes
ALLyesyesyesyes

4.2. Stratospheric Water Vapor Experiment

[35] In this experiment, we isolate the impact of H2O production by CH4 oxidation on the model radiative energy balance. A model version similar to ALL, except with no sink of CFCs (hereafter referred to as experiment “WET”) was compared with an identical version having, in addition, no H2O production from CH4 (but still including CH4 destruction; hereafter called experiment “DRY”).

[36] Figure 7a shows the 20-year mean difference in the zonal mean H2O mixing ratio between the two runs. This difference increases upward from near zero at the tropopause to a peak value of 3.6 ppmv at the 5 mbar level in the polar stratosphere. The difference is almost entirely due to H2O production from CH4 oxidation, with a small contribution from transport differences between the two runs (driven by the slightly altered radiative balance). Both positive and negative differences, on the order of a few hundred ppmv, are seen in the troposphere (not shown). These are not statistically significant, and arise from interannual variability in the tropospheric water cycle.

Figure 7.

Twenty-year zonal mean differences in the stratospheric water vapor experiment (WET–DRY): (a) annual mean H2O mixing ratio (ppmv), (b) annual mean temperature, (c) January mean temperature, and (d) July mean temperature. Temperatures are in K. Shading in Figures 7b, 7c, and 7d indicates points at which the difference is significant at the 95% confidence level. In Figure 7a, only levels above 250 mbar are shown; see text for details.

[37] Figures 7b, 7c, and 7d show the 20-year mean annual, January, and July zonal mean temperature difference, equation image, between the two runs. The shaded regions in the figures indicate points at which the difference between the models is significant at the 95% confidence level, as determined by an approximate difference of means test [von Storch and Zwiers, 1999]. Since this test is susceptible to errors resulting from temporal and spatial autocorrelation, we conducted a more stringent test taking this possibility into account, the details of which are described in Appendix A. The more stringent test confirms that the differences in equation image as indicated by the 20-year means are indeed significant in the indicated regions.

[38] Figure 7b shows that throughout most of the stratosphere, in situ H2O production results in cooler temperatures by −0.6 K on average (the global mean value of equation image between 100 and 1 mbar), with the annual mean difference exceeding −2 K in the south polar upper stratosphere. The pattern of equation image strongly resembles the difference field of the mixing ratio itself, suggesting that local radiative processes dominate the temperature response, except in the extratropical upper stratosphere where some hemispheric asymmetry is apparent. The largest differences are seen in January, where equation image = +3.7 K at the north polar stratopause (Figure 7c). While not significant at the 95% level, these local maxima are suggestive of differences in extratropical wave activity, an issue discussed later in section 4.5.2.

[39] Figure 7 shows several small regions of significant positive equation image in the upper troposphere (150–300 mbar). While the significance deteriorates under the more stringent test, a response of this type has been observed in previous GCM and radiative-convective model calculations where larger H2O and/or CO2 increases near the tropopause were imposed [Ramanathan et al., 1985; Rind and Lonergan, 1995; Forster and Shine, 1999; Sigmond et al., 2004]. Here, the annual mean equation image exceeds 0.5 K in limited regions of the extratropics and the January mean exceeds 1.2 K at the north polar tropopause, while the global annual mean on the 200 mbar surface is 0.19 ± 0.23 K. No discernable impact on the surface temperature Ts was seen in this experiment, which is not surprising given that the same fixed SST climatology was used in all runs. In similar experiments where SSTs are allowed to respond to surface air temperature changes, the troposphere and surface exhibit a perceptible warming [e.g., Rind and Lonergan, 1995].

[40] The changes in longwave and shortwave emission are displayed in Figure 8. Shortwave heating has increased throughout the stratosphere, with statistically significant change in the extratropical lower and middle stratosphere. Since O3 and CO2 are prescribed, this increase can be directly attributed to changes in the H2O distribution. The extratropical maxima in equation imageSW are consistent with the largest increases in water vapor there (on a given pressure surface; Figure 7a). Nevertheless, the peak increase of 0.004 K d−1 at 15 mbar is still much smaller than the unperturbed value of ≈2 K d−1. The longwave change exceeds the shortwave over most of the stratosphere, in accord with the discussion of section 4.1: e.g., in the upper tropical stratosphere, the longwave anomaly of +0.01–0.05 K d−1 is an order of magnitude larger than the shortwave heating change. Since, in the unperturbed state, there is longwave cooling everywhere except in the lowermost tropical stratosphere, this indicates a net decrease in longwave cooling over most of the tropics, consistent with the decreased temperatures there (Figure 7b). No significant changes in cloudiness were observed in this experiment, and differences in the top of atmosphere, annual and zonal mean shortwave and longwave cloud forcing were much smaller than their clear-sky counterparts at all latitudes.

Figure 8.

Twenty-year annual mean, zonal mean heating rates (K d−1) for experiment WET-DRY: (a) shortwave and (b) longwave. Shading indicates 95% significance.

[41] In Figure 8b, a net increase in longwave cooling is seen in the extratropics, despite the lower temperatures in run WET. As we shall see in section 4.5.2, this is due to the model dynamical response to the radiative perturbation in the tropics. Further discussion of these matters is deferred to that section, where the model response to the entire suite of GHG perturbations, including dynamical adjustments, is considered.

4.3. Methane and Nitrous Oxide Experiment

[42] In this experiment, we study the radiative impact of the nonuniform distributions of N2O and CH4. Model version DRY was compared with a version having globally uniform GHGs, i.e., no chemical destruction (hereafter referred to as model “UNI”). In this experiment, the change in concentration δX is always negative, because of the activation of the chemical sink in run DRY.

[43] Figure 9 shows the 20-year annual, January, and July mean equation image for this experiment. In contrast to the results of the water vapor experiment, equation image is positive over most of the stratosphere, with an average annual mean value of 0.4 K, exceeding 1 K in the upper south polar stratosphere. Negative differences appear in the lower stratosphere, tropopause, and midstratosphere polar regions, but these are not significant at the 95% level. Nevertheless, the sign of these changes is consistent with the discussion of section 4.1, given that the tracer concentration change δX is negative (i.e., δX < 0 and p > [βLWX]−1 ⇒ δQ*LW < 0 and δT < 0). In the lower stratosphere, a decrease in GHG concentration reduces the absorption and re-emission of upwelling surface-troposphere longwave emission, and so has a small local cooling effect. On the other hand, because of the strong infrared opacity of stratospheric CO2, H2O, and O3, this effect diminishes with increasing height above the tropopause [Ramanathan et al., 1985]. In the upper stratosphere, fewer GHG absorbers means less cooling to space, thus producing the warm anomaly seen in Figure 9.

Figure 9.

Zonal mean temperature differences (K) in the nonuniform N2O and CH4 experiment (DRY–UNI). Twenty-year means: (top) annual, (middle) January, and (bottom) July. Shading indicates 95% significance.

4.4. CFC Experiment

[44] In this experiment, sinks of CFCs are introduced and the results compared with the previous experiments. Model configuration ALL was compared with version WET which has uniform CFCs throughout the model domain. However, the differences arising from the nonuniform distributions of CFC-11 and CFC-12 are not statistically significant in the current GCM experimental arrangement using 20-year means. In an effort to detect these small differences at the prescribed significance level, we extended both model runs by 20 years, thereby obtaining 40-year means of quantities of interest.

[45] Figure 10 shows the 40-year zonal and annual mean distribution of equation image for this experiment. A formally significant temperature difference is found only in the SH midstratosphere, where a small cooling anomaly ∼−0.1 K is present, surrounded by a more extended region of cooling (at <95% statistical significance). This latitudinally asymmetric δT pattern was also seen in experiment DRY–UNI, but in that case was situated at lower altitude, in the upper troposphere-tropopause region (Figure 9, top). The regions of significant difference shown in the figure do not pass the more stringent statistical test (Appendix A), indicating that the response is difficult to distinguish from the internal variability of the AGCM. The corresponding differences for January and July (not shown) are similarly diminutive. Since the relative decrease in GHG concentration above the tropopause is in fact larger in this case than in the previous experiment, the lack of response must be due to differences between the radiative properties of CFCs versus N2O and CH4.

Figure 10.

Forty-year annual and zonal mean temperature difference (K) in the nonuniform CFC experiment (ALL–WET). Shading indicates 95% significance.

[46] The fact that the strongest absorption bands of CFCs are located in the atmospheric window, between 9 and 12 μm, leads to a distinct radiative response that has been examined in several previous studies, usually employing the fixed dynamical heating assumption [Ramanathan et al., 1987; Forster et al., 1997; Forster and Joshi, 2005]. In these studies, spatially uniform CFC increases tended to induce warming, not just in the troposphere, but also at the tropopause and midstratosphere (up to ∼35 km (7 mbar) in the tropics in the calculations of Forster and Joshi [2005]). A modest cooling occurred only in the upper stratosphere. This is much like O3, which provides heating up to ∼10 mbar via the 9.6 μm band, but opposite to CO2, N2O, CH4, and H2O, all of which have a cooling effect at the tropopause and above.

[47] These interspecies differences are evidently due to the fact that upwelling longwave radiation from the troposphere is absorbed and reradiated by CFCs without much coupling to the surrounding stratosphere, because of the lack of spectral overlap of the CFC bands with those of other species (with the exception of the weak H2O continuum). So while CFCs are efficient at local heating via the angular redistribution of photons, they do not act as major coolants for the stratosphere. This is consistent with the fact that a uniform distribution of CFCs contributes nearly an order of magnitude less radiative cooling in the stratosphere than N2O and CH4 combined [Zhong et al., 1993]. Thus, for an equivalent concentration decrease applied in the stratosphere only, the temperature response of CFCs in the lower to middle stratosphere should be negative, switching sign only above ∼7 mbar), and small. These expectations are borne out in Figure 10.

[48] As a final check that the radiation code in AGCM3 was adequately representing the CFCs, we conducted a third, 20-year experiment in which CFCs were removed from the model entirely. The pattern of equation image between this experiment and run UNI (not shown) displays statistically significant changes at the tropical tropopause (negative, up to −0.26 K) and in the NH lower stratosphere (negative, up to −0.72 K). The coupled version of AGCM3 has a (surface) climate sensitivity in the tropics ∼1 K/Wm−2 [Boer and Yu, 2003]. Taking this as an upper limit to the climate sensitivity of AGCM3, we estimate a surface radiative forcing of ≳0.25 Wm−2 for the change in CFCs since the preindustrial period. This lower bound lies significantly below both the estimate of Hauglustaine et al. [1994], who obtained 0.4 Wm−2 using a 2-D chemical-dynamical model, and the IPCC TAR global mean value of 0.34 Wm−2 [Ramaswamy et al., 2001], indicating that AGCM3 may underestimate somewhat the direct longwave effect of CFC changes.

4.5. Combined GHG Experiment

[49] In this experiment, we assess the cumulative effect of the GHG changes discussed individually above, by comparing runs ALL and UNI. Since this is the case of greatest interest, we discuss radiative and dynamical effects separately below.

4.5.1. Radiative Response

[50] Figure 11 shows the 20-year annual, January, and July mean zonal mean temperature difference equation image between runs ALL and UNI. Throughout most of the stratosphere the 3-D GHG model is cooler than the uniform model, with the same global mean 〈δT〉 = −0.6 K as in the stratospheric water vapor experiment.

Figure 11.

Zonal mean temperature difference (K) for experiment ALL–UNI. Twenty-year means: (top) annual, (middle) January, and (bottom) July. Shading indicates 95% significance.

[51] The spatial pattern of the temperature change reflects both the additional stratospheric water vapor due to CH4 oxidation and the “missing” N2O, CH4, and CFCs due to the presence of chemical sinks. As seen by comparison with Figure 7, the sign and pattern of equation image indicate that the oxidation of CH4 to H2O is the dominant effect on the temperature response. The remaining changes (i.e., sinks of N2O, CH4, and CFCs) produce a downward shift of the cooling layer: the maximum negative equation image now occurs at ∼10 mbar. The changes are statistically significant except in the upper polar stratosphere, where there exists a high degree of interannual variability. As in the water vapor experiment, the largest differences are seen in January at the north polar stratopause (middle panel of Figure 11), where equation image > +8 K.

[52] Figure 11 (top) shows a significant warm anomaly (max. equation image = +0.24 K) in the tropopause region, but in contrast to experiment WET–DRY, it is now confined almost entirely to the tropics. As will be shown in section 5, the effects of the individual GHG changes are roughly additive, so that the upper tropospheric/lower stratospheric cool anomaly of experiment DRY–UNI offsets somewhat the warm anomaly of experiment WET–DRY, particularly in the extratropics. Despite this, the region of the warm anomaly now extends deeper into the troposphere than in the previous experiment. Figure 11 is similar to Figure 17 of Schwarzkopf and Ramaswamy [1999], who investigated the radiative impact of including uniform N2O, CH4, halocarbons and the water vapor continuum in a GCM where none of these effects were present previously. As would be expected, these authors found a more significant impact on the troposphere than occurs in the present study, where the GHG perturbations are confined to the stratosphere. However, the magnitude and sign of the annual mean temperature response in the tropical stratosphere is about the same in the two experiments.

[53] Figure 12 shows the spatial pattern of the difference in the vertically averaged, mass-weighted, annual mean temperature between 100 and 1 mbar, δequation image, for each of the experiments (omitting ALL–WET, for which there is no significant change). In Figure 12a, δequation image for experiment ALL–UNI is seen to be negative everywhere, with the largest differences (∼−2 K) in the polar regions and the smallest differences (∼−0.1 K) in the tropics. Near the south pole, δequation image is almost 40% larger (in absolute value) than near the north pole, giving a distinct north-south asymmetry in this field. This is also evident in Figure 11, where equation image is seen to be much larger in the Arctic upper stratosphere in winter than in Antarctic winter, giving a positive annual mean equation image in the Arctic upper stratosphere. Figure 12a also shows a pronounced dipolar pattern in the NH extratropics, with a ∼1 K larger cooling anomaly over Asia compared to North America. Figure 12b shows the spatial structure of δequation image in experiment DRY–UNI. Consistent with the results presented in Figure 9, δequation image is positive everywhere except in the polar regions, with a peak value of 0.4 K in NH midlatitudes.

Figure 12.

Twenty-year annual mean differences for three of the experiments, vertically averaged between 100 and 1 mbar. The first three plots are vertically averaged temperature differences for experiments (a) ALL-UNI, (b) DRY–UNI, and (c) WET–DRY. (d) Vertically averaged difference in H2O mixing ratio from experiment WET–DRY.

[54] The corresponding quantity for experiment WET–DRY is shown in Figure 12c, which has a very similar appearance to Figure 12a, except it lacks the strong north-south asymmetry in δequation image. This outlines the dominant role of water vapor in these perturbation experiments. This δequation image may be compared with that of the H2O concentration difference itself, δequation imageimage shown in Figure 12d. While the two distributions are broadly similar with respect to their overall zonal symmetry and the equator-to-pole correspondence between δequation image and δequation imageimage there is more structure in the former field than can be attributed to local H2O cooling alone. We shall turn to this issue next, where the dynamical response of the model to the full suite of GHG changes is examined.

4.5.2. Dynamical Response

[55] Here we examine features of the spatial distribution of δT that cannot be explained by radiative changes alone. These are: first, the decrease in ∣equation image∣ with height above 10 mbar in regions away from the winter pole seen in experiment ALL–UNI; and second, the high-latitude annual mean hemispheric asymmetry in δT seen in all four GCM experiments.

[56] Figure 13 shows the globally averaged contribution of each of the terms in equation (3) in experiment ALL–UNI. equation imageδ(∂T/∂t)equation image is very near zero, as expected. The figure shows that the balance is mainly between equation imageδQDYNequation image (calculated as the residual of ∂T/∂t and δQRAD = δQLW + δQSW) and equation imageδQLW〉, except between ∼20 and 30 km height, where 〈δQSW〉 exceeds 〈δQDYN〉, because of increased solar heating from H2O (section 4.2). Comparing Figure 13 with the annual mean latitude-height distribution of equation imageDYN (not shown) shows that the upper stratosphere exhibits dynamical warming in the global mean, with extratropical warming exceeding tropical cooling. Conversely, the small but statistically significant dynamical cooling seen in the extratropical SH leads to net dynamical cooling below about 17 km. The maximum equation imageDYN is seen in the Arctic upper stratosphere in January (not shown), and is nearly an order of magnitude larger than its SH counterpart in July, indicating a much stronger dynamical response in NH winter. While planetary wave breaking is known to be stronger in the NH because of the larger surface source there, the difference in the model dynamical response to the radiatively perturbed stratosphere requires further investigation. The pattern of equation imageLW (not shown) bears a strong resemblance to equation imageLW in the water vapor experiment (Figure 8b), and is basically the inverse of equation imageDYN, as would be expected from the global mean results in Figure 13.

Figure 13.

Twenty-year annual mean, globally averaged heating rates (K d−1) for experiment ALL-UNI: longwave (solid line), shortwave (short-dashed line), dynamical (dotted line), and total (long-dashed line).

[57] Interactions between middle atmosphere radiation and dynamics are well illustrated in the transformed Eulerian mean (TEM) perspective of Andrews et al. [1987]. Under steady state, quasi-geostrophic conditions, the TEM thermodynamic energy, zonal wind, and continuity equations read

equation image

where ρ0(z) is the standard atmosphere density, f0 the Coriolis parameter, N the buoyancy frequency, trad the radiative relaxation time, H the scale height, and R the gas constant. Tr denotes the temperature that would obtain if all dynamical heating and cooling processes were suppressed, while v* and w* are the meridional and vertical components, respectively, of the residual mean wind. The quantity ρ0−1equation image · F + equation image is the net zonal acceleration due to friction, consisting of both resolved and unresolved components. The Eliassen–Palm flux F is nonzero whenever model-resolved (e.g., planetary) waves are present, while equation image represents all unresolved zonal forces (e.g., gravity wave drag). Equations (11) show how departures from purely radiative conditions result in a residual meridional circulation driven by sources and sinks of wave energy. In particular, divergence of F (∇ · F > 0) indicates a source of (resolved) waves, while convergence (∇ · F < 0) is a sign of wave transience and dissipation.

[58] That such dissipation is occurring in the model can be seen in Figure 14, which shows 20-year means of the January zonal wind and E-P flux divergence (actually ∇ · F/[ρ0a cos ϕ] in the model, a: Earth radius), along with the corresponding differenced quantities from experiment ALL–UNI. As expected, regions of ∇ · F < 0 are coincident with the (westerly) polar night jet in the extratropical NH. The convergence pattern seen in Figure 14c is not unlike the observationally derived total zonal forcing seen in Figure 7 of Rosenlof [1995], even though the model-parameterized subgrid-scale forcing (equation image) has not been diagnosed. The differenced equation image field (Figure 14b) shows a significantly slower westerly jet in the NH in run ALL, coincident with a 20–40% increase in the E-P flux convergence there (Figure 14d). The corresponding differences for July (not shown) reveal no statistically significant difference in the E-P flux between experiments, although there is a modest increase in equation image (∼2 ms−1) in the NH midstratosphere.

Figure 14.

Twenty-year mean January zonal mean zonal wind U (m s−1) for runs (a) ALL and (b) ALL-UNI. Twenty-year mean January zonal mean E-P flux divergence equation image · F/(ρ0a cos ϕ) (m s−1 d−1) for runs (c) ALL and (d) ALL-UNI. Only the region above 150 mbar is shown. Shading in Figures 14b and 14d indicates 95% significance.

[59] The residual mean circulation may be examined more directly by calculating the stream function Ψ from the TEM meridional velocity equation image* using [Andrews et al., 1987]

equation image

The results for run ALL in January and July are shown in Figures 15a and 15c. The model reproduces the characteristic two-cell structure derived from observations, with upwelling in the tropics and downwelling in the extratropics, and a much stronger circulation in the winter hemisphere [Rosenlof, 1995]. Tropical upwelling is stronger in the NH than in the SH, and the summer circulation is weaker in July than in January, again in accord with observations. Figures 15b and 15d show the differenced stream function δΨ for experiment ALL–UNI. In January, the circulation is enhanced throughout the NH, with a maximum in midlatitudes near 70 mbar, while the SH circulation is largely unchanged. Most of the NH changes in Ψ are statistically significant. In July, however, the circulation change is more evenly distributed over the two hemispheres, with statistically significant changes only in the NH extratropical lower stratosphere, where δΨ reinforces the equilibrium circulation pattern. In general, we may conclude that the GHG changes have led to an overall strengthening of the stratospheric meridional circulation.

Figure 15.

Twenty-year mean zonal mean residual mean streamfunction Ψ (106 kg s−1) for runs (a) ALL, January; (b) ALL-UNI, January; (c) ALL, July; and (d) ALL-UNI, July. Only the region above 150 mbar is shown. Shading in Figures 15b and 15d indicates 95% significance.

[60] These results may be compared with previous simulations that examined circulation changes due to water vapor increases in the stratosphere only. Rind and Lonergan [1995], who applied a uniform doubling of H2O concentration (from 3 to 6 ppmv) above 100 mbar, found a similar pattern of dynamical warming in the upper stratosphere to that seen in our Figure 11, accompanied by an annual mean E-P flux convergence increase of up to 10% above 10 mbar and a residual circulation increase of ∼5%. MacKenzie and Harwood [2004], who examined the middle atmosphere response to a nonuniform, but smaller, humidity increase than in the present study (up to 1.5 ppmv H2O above 10 mbar), noted an asymmetric NH versus SH dynamical response, as diagnosed by an increase in the DJF eddy heat flux (presumably equation image or equation image) in the upper stratosphere at 60°N, paired with a decrease in the JJA eddy heat flux at 60°S. They did not report changes in equation image · F and Ψ directly, however, precluding a more detailed comparison with our results. Finally, while not a H2O perturbation experiment, Sigmond et al. [2004] conducted a GCM simulation in which CO2 was uniformly doubled in the mesosphere only (between ∼200 and 0.01 mbar), reporting the same diagnostics as discussed above (for DJF only). They found equation image > 0 in both hemispheres, a small δ(∇ · F) < 0 between 5 and 1 mbar in the NH, and a more varied δΨ response, with localized increases in the lower stratosphere and a patchwork of increases and decreases elsewhere. Insofar as H2O and CO2 may be treated as generic GHGs, these results, along with the others summarized above, are seen to be qualitatively consistent with the dynamical changes observed in the present experiment.

5. Discussion and Implications for Model-Simulated Climate

[61] Figure 16 shows the global mean vertical profiles of the temperature change in all four GCM experiments. The 〈δT〉 profile for experiments WET–DRY and ALL–UNI is significant at the 95% level above 120 mbar (∼15 km), while the experiment DRY–UNI profile is significant above 10 mbar (∼32 km) and the experiment ALL–WET profile is not significant at this level, but is displayed for illustrative purposes. As might be expected for the modest changes in GHG abundance examined here, the ALL–UNI 〈δT〉 curve is very nearly the sum of the other three curves (shown by a dotted line in the figure). Thus the global mean temperature responses to the various changes discussed here are essentially additive, a result that is consistent with more detailed studies of climate response to a variety of forcings [Boer and Yu, 2003]. However, it is also true that the detailed spatial temperature response is not strictly additive, especially in regions where dynamical feedbacks are important (section 4.5.2).

Figure 16.

Twenty-year global mean temperature difference (K) in the four GHG perturbation experiments, as a function of height (km). The dotted curve, nearly coincident with the solid curve, is the sum of the other curves.

[62] The magnitude of 〈δT〉 obtained for experiment ALL–UNI (–0.4 to –1.0 K) may be compared with the results of other studies comparing GHG representations in an AGCM. For example, in the Govindasamy et al. [2001] comparison of explicit non-CO2 GHGs and equivalent CO2 treatments (both spatially uniform), a range of 〈δT〉 = 0.3 K at 18 km to 1.8 K at 40 km was obtained, with the explicit GHG model always warmer. The differences found in this work are thus of comparable magnitude. Taken together, these results imply that switching from an equivalent CO2 representation to one with explicit, 3-D GHGs would prompt a smaller net change in 〈δT〉, because of a partial cancellation of these two effects.

[63] It is also of interest to ask whether the 3-D treatment of greenhouse tracers leads to improved agreement between the model climatology and observations. For the latter we use the NMC upper air climatology for 1979–1988, which extends up to 10 mbar, and focus on temperature only. Comparison of the model UNI and model ALL equation image fields with the NMC climatology (not shown) reveals a modest (∼1 K) improvement in the January peak anomaly in the NH polar stratosphere, and a ∼2 K reduction in the July peak anomaly in the SH upper polar stratosphere, upward of ∼50 mbar. However, additional cooling from elevated H2O in the upper tropical stratosphere in July leads to a modest increase (∼1 K) in the anomaly there. The fact that the improved agreement is seen mostly in the polar stratosphere indicates that the dynamical consequences of the GHG changes have had a beneficial effect.

[64] As the climatic effects of changes to the GHG distributions revealed by the above AGCM experiments are modest, it is worth examining their likely impact in a slightly broader context. Two key aspects of the climate response were suppressed in these experiments. First, the fixing of SSTs and sea ice distributions to climatological mean values damps the tropospheric response to the net increase in downward longwave radiation from the stratosphere. Although the expected surface temperature response upon the inclusion of the ocean and sea ice response is small, it is not negligible [Rind and Lonergan, 1995]. This response will be explored in future coupled model simulations including an ocean component.

[65] Second, similar fixing of the ozone distribution suppresses an important chemical-radiative feedback. Chemical transport model studies of the interaction between H2O, CFCs, polar stratospheric cloud (PSC) amount, temperature, and ozone have established a chain of interdependent processes. In broad outline, PSC formation in the extratropical lower stratosphere is enhanced under conditions of increased humidity and cooling. PSCs constitute an active site for the chemical synthesis of chlorine radicals which destroy O3 via the Clx catalytic cycle, particularly during austral spring in the Antarctic vortex. The O3 loss rate is modulated both by PSC amount and the prior availability of Clx radicals, the latter largely provided by the photolysis of CFCs. MacKenzie and Harwood [2004] showed that a relatively simple parameterization of the PSC formation threshold [Cariolle and Deque, 1986] along with an O3 destruction rate scaling with CFC loading can qualitatively reproduce the observed winter-to-spring evolution of O3 concentration in the subpolar stratosphere. Hence this may represent a promising approach whereby the simplified chemical parameterization used in the present study can be used to capture this important feedback process.

[66] In summary, we have examined the climatic consequences of relaxing the uniform GHG assumption, a simplification still widely used in AGCMs. The inclusion of a simple chemical loss parameterization leads to fairly realistic stratospheric distributions of N2O, CH4, CFC-11 and -12, and H2O. The model dynamics is able to reproduce gross aspects of the observed seasonal cycles of these tracers in the stratosphere. The radiative consequences of the distribution changes were examined in detail, and it was shown that cooling (∼0.6 K) due to increased stratospheric water vapor from CH4 oxidation exceeds the radiatively induced warming resulting from decreases in the other GHGs. We believe that this is the first study to have simultaneously considered these countervailing effects. An exceptional region is the upper extratropical stratosphere, where dynamical heating due to increased wave activity in winter leads to strong (≳5 K) warming in this region. The net dynamical effect is a slight but significant increase in the strength of the stratospheric residual mean circulation. Taken together, these results instill confidence that the model representation of these important GHGs is satisfactory both for further refinement of radiative-chemical processes and for coupled model equilibrium and transient climate simulations.

Appendix A:: Inferring Field Significance From Local Statistical Tests

[67] As described by von Storch and Zwiers [1999], use of a purely local test, such as Student's t test, to infer global significance is problematic. This is due to the intrinsic spatial and temporal coherence of physical fields in a GCM: properties of nearby grid points are not independent of each other, nor are successively time-averaged conditions at a single point. Thus it may be the case that for a sample drawn from numerous randomized simulations of the same model configuration, regions of significance calculated from pooled means might even exceed in area those found in the present comparison of two or more different model configurations.

[68] In the present context, where the 95% significance region lies largely in the stratosphere, successive annual means of temperature, for example, are likely to be both spatially and temporally correlated [Zwiers, 1990]. Although the model is subject to a fixed boundary condition at the surface, serial autocorrelation is a real possibility in the stratosphere, where the atmospheric “memory” can be on the order of several years. In an attempt to minimize these effects, we constructed a 10-year subsample from the 20 annual means (i.e., every other year) for each run. This sample of 20 annual means (10-year subsamples from two distinct model runs), was then subjected to a spatial correlation test, along the lines of Livezey and Chen [1983]. These authors applied a Monte Carlo method to a sample of white noise fields to generate a distribution of the fractional coverage of erroneous 95% significance regions. From this distribution, they noted the critical fraction κ above which laid 5% of all regions deemed significant. This critical fraction was then compared to that obtained from the usual t test conducted on the fields of interest.

[69] The only difference from the above in the procedure we followed was that our samples were constructed from the individual model annual means themselves (instead of white noise), which better reflects the model variability. This was done using bootstrap resampling on the 20 annual mean files (10 alternate years from each model configuration), with the number of realizations chosen as 2000. The results of this procedure for the various experiments described in the text appear in Table A1. It can be seen that while the area of the significance region changes under subsampling, it does not always decrease (compare experiment ALL–WET). The relevant issue for the assessment of significance is then whether the fractional coverage under subsampling exceeds the calculated equation image, which it does for each of the annual mean equation image fields listed in Table A1, except the CFC experiment ALL–WET.

Table A1. Results of the Spatial Correlation Test on Annual Mean equation image Fieldsa
Experimentfsig (20-Year)fsig (10-Year)κ
  • a

    Here, fsig, fraction of zonal mean cross section that fulfills 95% significance; κ, critical rejection rate.

  • b

    The fsig values are for 40- and 20-year annual means in this experiment.

WET-DRY0.5080.4440.187
DRY-UNI0.2980.2260.164
ALL-WETb0.0420.1240.151
ALL-UNI0.5030.4270.179

Acknowledgments

[70] We thank Stephen Beagley and David Plummer for providing the CMAM loss rates, Steve Lambert and Mijeong Park for assistance with the UARS data, Gerd Folberth, Jiangnan Li, and Francis Zwiers for helpful discussions, and George Boer for his detailed comments on the manuscript. The comments of two anonymous reviewers helped improve and streamline the paper. CC is funded by the Canadian Foundation for Climate and Atmospheric Sciences (CFCAS) as part of the Canadian Global Coupled Model (CGC3M) research network.

Ancillary