Stratospheric influence on tropospheric climate change in the Northern Hemisphere

Authors


Abstract

[1] The role of the stratosphere in tropospheric climate response to increased concentrations of the greenhouse gases during Northern Hemisphere winter is addressed by performing and analyzing a set of simulations with the atmosphere general circulation model ECHAM5. Attention is paid to the difference in the response to doubled CO2 concentration and associated sea surface temperature and sea ice concentration anomaly between a low-top and a stratosphere-resolving model version. We find a larger decrease of the Arctic sea level pressure in late winter in the low-top model when compared to the stratosphere-resolving one. Such dependence of the response on the representation of the stratosphere is consistent with previous multimodel results, indicating that the difference is likely robust across different models. The different response of the tropospheric circulation may have important climatic consequences; for example, we demonstrate a different precipitation response over Europe in these experiments. The different tropospheric response is shown to originate from different response in the polar stratosphere which is attributable to a stronger Brewer-Dobson circulation response in the stratosphere-resolving model. A decomposition of the Brewer-Dobson circulation response to contributions from resolved and parameterized processes show that both contribute toward the stronger downwelling response in the polar stratosphere in the stratosphere-resolving model. Additional sensitivity experiments reveal that the magnitude of the Arctic sea level pressure response, but not the difference between the stratosphere-resolving and low-top model responses, depends on the magnitude of SST anomaly in the tropical Pacific.

1. Introduction

[2] There is growing evidence that the modeled response of the atmosphere to external forcing is sensitive to the representation of stratospheric processes. It has been shown that global models respond differently to stratospheric ozone changes [Son et al., 2008; Perlwitz et al., 2008; Karpechko et al., 2008; Waugh et al., 2009; Simpkins and Karpechko, 2012], greenhouse gas (GHG) concentration increases [Shindell et al., 1999; Gillett et al., 2003; Sigmond et al., 2008; Scaife et al., 2012], and El Niño variability [Bell et al., 2009; Cagnazzo and Manzini, 2009; Ineson and Scaife, 2009] depending on how well the stratospheric processes are represented. The differences in the response are not restricted to the stratosphere but extend down to the surface, implying that an inadequate representation of the stratosphere may have an adverse effect on the accuracy of climate prediction by the model.

[3] Here, we address the response of the Northern Hemisphere (NH) extratropical atmospheric circulation to GHG concentration increases and analyze the differences in the response between a low-top atmosphere general circulation model (a model whose upper lid is located in the stratosphere at about 10 hPa) and a high-top model (a model whose upper lid is located in the mesosphere at about 0.01 hPa). Previous studies focused on similar questions reported opposite results. For example, Shindell et al. [1999] and Sigmond et al. [2008] showed that the difference in wintertime sea level pressure (SLP) response to increased CO2 concentration between high-top and low-top models resembles at the surface the Northern Annular Mode (NAM), with high-top model simulating a larger shift toward the positive NAM phase and larger poleward displacement of extratropical westerlies and associated storm tracks. Sigmond and Scinocca [2010] argued that the difference between the responses is attributable to the differences in the mean stratospheric circulation between the models. However, recently Scaife et al. [2012] analyzed multimodel ensembles and found that the low-top models systematically simulate a larger NAM-like response corresponding to the positive NAM phase in winter compared to the high-top models. Also, Morgenstern et al. [2010] found a negative NAM response in winter to increased CO2 concentration in an ensemble of stratosphere-resolving chemistry-climate model simulations.

[4] The uncertainty in the NAM response to increased GHG forcing has important climatic implications [e.g., Sigmond and Scinocca, 2010]. Scaife et al. [2012] demonstrated that the different NAM response between the models results in doubling of the predicted increase in extreme winter rainfall over Western and Central Europe in a climate with quadrupled CO2 concentration. Karpechko [2010] studied the Intergovernmental Panel on Climate Change Fourth Assessment Report (IPCC AR4) simulations by the CMIP3 models and demonstrated that the uncertainty in future NAM index trend accounts for a considerable fraction of the intermodel spread in temperature and precipitation trends in some NH regions.

[5] With the approach taken in this work, a direct comparison of a high-top, low-top model pair, we aim to provide further evidence that the response of low-top models can be biased owing to the inadequate representation of the stratosphere, and shed light on what are the stratospheric dynamical processes that are crucial for the difference in the response. At the same time we test the results of a similar study by Scaife et al. [2012] by subjecting the models to a weaker forcing (i.e., doubling the CO2 concentration, not quadrupling it), which is comparable to the forcing expected in some GHG emission scenarios by the end of the 21st century. An additional aspect touched on is the relative role of the contribution of El Niño–like changes in the tropical Pacific SSTs to the NH extratropical circulation response to GHG forcing previously discussed by Yamaguchi and Noda [2006] and Meehl et al. [2007].

2. Model and Experiment

[6] We employ the ECHAM5 atmosphere general circulation model described in the work of Roeckner et al. [2006]. The low-top atmospheric version has 31 vertical levels with the top level located at 10 hPa. This model version coupled with an ocean model was used, in particular, for the IPCC AR4 simulations. The middle atmosphere ECHAM5 version [Manzini et al., 2006], here referred to as “high-top” model, has 47 levels and extends up to 0.01 hPa. The models have the same vertical levels between the surface and 100 hPa and are both used here at horizontal resolution T63. Above 100 hPa, the low-top model has 5 full levels while the high-top model has 21 levels, of which 11 are located between 100 and 10 hPa. Since planetary waves emerging from the troposphere typically break above 10 hPa they can reach the upper lid of the low-top model. In order to prevent a spurious reflection of the waves from the upper lid in the low-top model they are numerically dissipated, which is achieved by increasing horizontal diffusion in the uppermost 5 levels [Roeckner et al., 2006]. Such “sponge layer” is not present in the high-top model. Another difference between the models is that, in addition to the parameterization of orographic gravity waves, which is identically included in both models, the high-top model includes a parameterization of the nonorographic waves [Manzini et al., 2006].

[7] We run both model versions with identical forcing to simulate present climate (1xCO2 experiment) and climate under doubled CO2 concentration (2xCO2 experiment). All experiments are run for 32 years. The first 2 years of each simulation are considered spin-up and discarded, and the results are presented for the remaining 30 years. The 1xCO2 experiment is forced with the observed climatological (1956–2000) monthly mean AMIP2 SST and SIC. Figure 1 shows zonal mean zonal winds and temperatures averaged for December–March for both models. In the troposphere the basic states of the two models are almost identical. The differences are seen in the stratosphere, especially in the Northern Hemisphere where the low-top model simulates a colder and stronger polar vortex. Although a model validation is not the focus of our study, it is of interest to compare the climates simulated by the models with observations. The comparison of the models with observations should be done with caution because the GHG concentrations and the lower boundary conditions used in the simulations were fixed in time rather than varied according to observations and also because the high-top model was not tuned to match the observations as thoroughly as the low-top model. Figure 2 shows vertical profiles of December–March zonal winds at 60°N and temperatures averaged over 70°N–90°N for both models and for ERA-40 reanalysis [Uppala et al., 2005]. The regions are chosen because the differences between the models are large there and because the focus in this work is on the Northern Hemisphere. It is seen that the low-top model has a cold bias (∼10 K), which is absent in the high-top model. The stratospheric winds are underestimated by both models below 30 hPa, the bias being larger in the high-top model. The bias for the high-top model in the stratosphere is about 3–7 m s−1 which is not unusual among the stratosphere-resolving models [Butchart et al., 2011].

Figure 1.

December–March zonal mean (a) zonal winds and (b) temperatures in the 1xCO2 experiment. Contours show the zonal means for the low-top model. Shadings show the differences between the high-top and the low-top models.

Figure 2.

December–March zonal mean (a) zonal winds at 60°N and (b) temperatures averaged over 70°N–90°N. Shown are the results from the low-top and high-top models from the 1xCO2 experiment and the 45 year averages from the ERA-40 reanalysis.

[8] A clear improvement in the high-top model is that this model simulates a more realistic stratospheric variability. Figure 3 shows daily time series of zonal mean zonal winds at 60°N for both models and for ERA-40. It is seen that interannual variability is strongly suppressed in the low-top model at both 10 hPa and 70 hPa levels but is more realistic in the high-top model. In particular, the sudden stratospheric warmings (i.e., periods of easterly winds during winter) are virtually absent in the low-top model. According to F test, the wintertime monthly wind variances in the high-top model are significantly larger (p < 0.1) than those in the low-top model from January to March, but not in December. A similar difference in the wind variability was previously found by Cagnazzo and Manzini [2009] for another pair of low-top and high-top ECHAM5 models with coarser vertical and horizontal resolution than in our models.

Figure 3.

Time series of daily zonal mean zonal wind (meters per second) at 60°N and (a–c) 10 hPa and (d–f) 70 hPa for the low-top model (Figures 3a and 3d), high-top model (Figures 3b and 3e), and ERA-40 reanalysis (Figures 3c and 3f). Black curves mark climatological averages. Dark gray shadings mark ±1 standard deviation range. Light gray shadings mark ranges between daily maxima and minima.

[9] The SST and SIC anomalies for the 2xCO2 experiments are calculated using the outputs of several coupled climate model simulations from the 20C3M and SRES A1B experiments of the IPCC AR4 as follows. First the monthly anomalies are calculated for each employed model as a difference between the 2090–2099 period of the SRES A1B experiment and the 1980–1989 period of the 20C3M experiment. The atmospheric CO2 concentrations between these periods differ approximately by factor of 2. The monthly mean anomalies are then averaged across the models and added to the monthly mean AMIP2 climatology. The following models are used to construct the SST anomalies: BCCR-BCM2.0, CGCM3.1(T63), CNRM-CM3, CSIRO-Mk3.5, GFDL-CM2.0, GFDL-CM2.1, GISS-AOM, GISS-EH, GISS-ER, INGV-SXG, IPSL-CM4, MIROC3.2(medres), ECHAM5/MPI-OM, MRI-CGCM2.3.2, PCM, and UKMO-HadCM3. Only one simulation from each model is used. Figure 4 shows the annual mean SST anomalies used to construct the boundary conditions for the 2xCO2 experiment. Note an enhanced warming of the central and eastern tropical Pacific compared to the western Pacific. This pattern was referred to as an El Niño–like mean state change in the IPCC AR4 [Meehl et al., 2007] because it is normally associated with El Niño events. The SIC anomalies (not shown) are constructed in the same way except that only the following models are employed: BCCR-BCM2.0, CGCM3.1(T63), CNRM-CM3, CSIRO-Mk3.5, GFDL-CM2.0, GFDL-CM2.1, IPSL-CM4, MIROC3.2(medres), ECHAM5/MPI-OM, MRI-CGCM2.3.2, and UKMO-HadCM3.

Figure 4.

Annual mean SST anomaly used in the 2xCO2 experiment. The black rectangle marks the region in which the SST anomaly was set to zero in the 2xCO2/tP experiment.

[10] The models used in this study are atmosphere only models; however, the inclusion of a coupled ocean is not expected to qualitatively change the results, as demonstrated by Scaife et al. [2012].

[11] We also perform a sensitivity experiment whose purpose is to test the robustness of the results obtained in the main 2xCO2 experiment. For this experiment, referred to as 2xCO2/tP, the SST anomalies of the 2xCO2 experiment are set to zero in the tropical Pacific (20°S, 20°N, 140°E, 280°E; see Figure 4). Thus, the only difference between the 2xCO2 and 2xCO2/tP experiments is the SSTs in the tropical Pacific. The sensitivity of the NH extratropical circulation to tropical Pacific SSTs has been shown before [e.g., Yamaguchi and Noda, 2006; Cagnazzo and Manzini, 2009; Ineson and Scaife, 2009]. As expected, the response of the NH extratropical circulation to the CO2 forcing in 2xCO2/tP differs considerably from that in the 2xCO2 experiment in both models. We use this fact to test the robustness of the differences between the low-top and high-top model responses found in the 2xCO2 experiment.

[12] The averaged difference in a variable (e.g., sea level pressure or zonal wind) between the 2xCO2 experiments and the 1xCO2 experiment is referred to as climate change response, or just response, in this variable. The statistical significance of the climate change response is estimated using standard two-sample, two-sided t test. The significance of the difference between the climate change responses of the high-top and low-top models is estimated as described in Appendix A.

3. Results

3.1. Near-Surface Dynamical Response

[13] Figures 5a and 5b show mean SLP changes between the 1xCO2 and 2xCO2 experiments for the low-top model and the differences between high-top and low-top model responses for January–March. We focus on this season because it is expected that the influence of the stratosphere on tropospheric climate is largest during winter season [Baldwin and Dunkerton, 1999] and because we find that in our experiments the difference between model responses is largest during late winter (see section 3.2). The response in the low-top model shows a reduction of about 3 hPa in SLP over the Arctic and increases in midlatitudes, especially over central and southern Europe, which resembles the Atlantic part of the positive NAM phase (Figure 5a). In the Pacific sector the negative response indicates a deepening of the Aleutian low, a manifestation of the tropospheric teleconnection with the El Niño–like response in the tropical Pacific. The corresponding changes in near-surface zonal winds (Figure 6a) include increased zonal winds in middle and high latitudes, in particular at 40°–65°N over North Atlantic and central and northern Europe with magnitude of about 1.5–2 m s−1. The increased zonal winds over North Atlantic/Europe coincide with decreased winds over northern Africa, indicating a poleward shift of the tropospheric jet stream. Such a shift is typically simulated by the IPCC AR4 models [e.g., Lorenz and DeWeaver, 2007].

Figure 5.

(a and c) January–March sea level pressure responses of the low-top model and (b and d) the differences in the responses between the high-top and low-top models in the 2xCO2 experiment (Figures 5a and 5b) and the 2xCO2/tP experiment (Figures 5c and 5d). Contours are drawn at ±0.75, 1.5, 2.25, and 3.0 and then each 1.5 hPa. Positive (negative) contours are drawn in red (blue). Dark (light) shadings mark responses, or difference in the responses, significant with p < 0.05 (<0.1).

Figure 6.

(a and c) January–March 850 hPa zonal wind responses of the low-top model and (b and d) the differences in the responses between the high-top and low-top models in the 2xCO2 experiment (Figures 6a and 6b) and the 2xCO2/tP experiment (Figures 6c and 6d). Contours are drawn at ±0.75, 1.5, 2.25, and 3.0 and then each 1 m s−1. Positive (negative) contours are drawn in red (blue). Dark (light) shadings mark responses, or difference in the responses, significant with p < 0.05 (<0.1).

[14] When compared to the low-top model, the Arctic SLP decrease in the high-top model is reduced by about 1.5 hPa (Figure 5b). The differences in the zonal wind response are significant in North Atlantic (Figure 6b) where they exceed 2 m s−1. The smaller Arctic SLP decrease in the high-top model is consistent with the results of Scaife et al. [2012] based on a comparison of several independent high-top, low-top model pairs. The SLP response difference in the work of Scaife et al. [2012] exceeds 2 hPa, which is stronger than in our experiment, consistent with the stronger forcing in their study.

[15] Scaife et al. [2012] suggested that the differences between the model responses may have important consequences for regional climate projections, in particular for European climate projections. Figure 7 shows the precipitation response over Europe. In the low-top model the drying southern Europe and moistening northern Europe responses with magnitude of 0.5–0.7 mm d−1 are consistent with the IPCC AR4 projections by the end of the 21st century [Meehl et al., 2007]. In the high-top model such response is significantly reduced or even canceled in some areas. Qualitatively this result is consistent with that in the work of Scaife et al. [2012] but weaker in magnitude, consistent with stronger forcing in their study.

Figure 7.

(a and c) January–March total precipitation responses of the low-top model and (b and d) the differences in the responses between the high-top and low-top models in the 2xCO2 experiment (Figures 7a and 7b) and the 2xCO2/tP experiment (Figures 7c and 7d). Contours are drawn at ±0.1, 0.3, 0.5, and 0.7 mm d−1. Positive (negative) contours are drawn in blue (red). Dark (light) shadings mark responses, or difference in the responses, significant with p < 0.05 (<0.1).

[16] The robustness of the difference in the responses is substantiated by the 2xCO2/tP experiment. In this experiment, the SST anomaly in the tropical Pacific is set to zero and therefore there is no El Niño–like anomaly pattern in the Tropics in this experiment. Overall the SLP and zonal wind responses in the low-top model (Figures 5c and 6c) are more zonally symmetric and clearly resemble the positive NAM phase. The SLP response (Figure 5c) shows an absence of the negative anomaly in the Aleutian low and a noticeably larger (by more than 60%) Arctic SLP decrease in 2xCO2/tP compared to 2xCO2. We use this strong sensitivity of the extratropical dynamical response to the tropical Pacific SSTs, without focusing on its mechanisms, to test the robustness of the difference seen in Figures 5b, 6b, and 7b. Figure 5d shows that, despite considerable differences between the 2xCO2 and 2xCO2/tP experiments, the difference between the high-top and low-top model responses in the Arctic remains similar to that of the 2xCO2 experiment. The significant difference in the North Atlantic jet stream response seen in the 2xCO2 experiment disappears here (Figure 6d), suggesting a sensitivity to the SST anomaly in the tropical Pacific for this region. Consequently, the precipitation response is locally (e.g., in the North Atlantic) reduced (compare Figures 7b and 7d). However, overall the difference between the high-top and low-top model precipitation responses remains similar.

3.2. Stratosphere-Troposphere Coupling

[17] We have shown that the low-top and high-top models simulate different near-surface circulation response to the same CO2 forcing. By construction, the two models resolve differently the stratosphere (see section 2). We now address the question on what are the key stratospheric processes that lead to the differences shown in Figures 57.

[18] The seasonal progressing of the polar temperature and high-latitude zonal wind responses from October to April is illustrated in Figures 8 and 9. Responses of both models include tropospheric warming and stratospheric cooling but the temperature changes in the lower stratosphere are very small (typically less than 0.4 K) except in the low-top model in December–January. In both models predominantly strengthened winds in the lower stratosphere below 20 hPa are present in the 2xCO2 responses. The strengthening is less pronounced in the high-top model from November until February and is accompanied by a weakened wind response at the upper levels. In December–January the stratospheric response in the high-top model has warmer temperatures (up to 3–4 K) and weaker winds (up to 4–6 m s−1) compared to the low-top model response (Figures 8c and 9c). This response is followed by tropospheric differences of the same sign that last over late winter and early spring. The differences in the troposphere are clearly visible in the zonal wind but very small to negligible in the polar temperature. The latter is, likely, because of the strong constraint imposed on tropospheric temperatures by the use of the same SST and SIC anomalies. Such a delayed connection between stratospheric and tropospheric differences in model responses recall the well-known downward propagation of intraseasonal anomalies deduced from observations [Baldwin and Dunkerton, 1999, 2001] and also reproduced by general circulation models (see, among others, Christiansen [2001] and Gerber et al. [2010]). On the basis of these results, we attribute the different tropospheric responses simulated in late winter shown in section 3.1 to the differences in the stratospheric responses simulated 1–2 months earlier.

Figure 8.

Zonal temperature responses averaged over 70°N–90°N in the (a and d) low-top model and (b and e) the high-top model and (c and f) their differences in the 2xCO2 experiment (Figures 8a–8c) and the 2xCO2/tP experiment (Figures 8d–8f). The data are smoothed with 20 day running mean. Hatching indicates responses, or differences in the responses, significant with p < 0.1.

Figure 9.

Same as Figure 8 but for zonal mean wind at 60°N.

[19] The results of the 2xCO2/tP experiment show that the lower-stratosphere cooling and zonal wind responses in these experiments are stronger during winter (Figures 8d, 8e, 9d, and 9e) than those in the 2xCO2 experiment in both models. However, the difference between the high-top and the low-top model response is remarkably similar (Figures 8f and 9f). The downward propagation from the lower stratosphere to the troposphere of the difference between the high-top and low-top model responses is thus robust at least with respect to the strength of the forcing which is changed by prescribing different tropical Pacific SST anomaly.

[20] To show that the polar stratospheric responses in both models are of dynamical origin, we calculate the residual vertical velocity ( inline image), which enters the dynamical heating term in the zonal mean temperature tendency equations [Andrews et al., 1987, equation (3.5.1b)]. Because the results for temperature and wind responses are consistent between 2xCO2 and 2xCO2/tP, in the following we focus only on the 2xCO2 experiment. The inline image responses in the 2xCO2 experiment (Figures 10a and 10b) reveal an increased downwelling (i.e., increased dynamical heating) in the stratosphere in both models. The differences between model responses are shown in Figure 10c. The intensified downwelling in the NH polar stratosphere in high-top models has been discussed in several previous publications [e.g., Sigmond et al., 2004; McLandress and Shepherd, 2009a] and is a manifestation of an intensified Brewer-Dobson (BD) circulation response [Butchart and Scaife, 2001; Butchart et al., 2006]. Figure 10 shows that an increased downwelling in the polar stratosphere comparable to that in the high-top model is also found in the low-top model (note alternating positive and negative differences in Figure 10c). Comparing Figures 8c and 10c one can see a correspondence between the temperature and inline image differences; in particular, the larger stratospheric warming response in the high-top model from late November to January is related to stronger dynamical heating. Note that the dynamical heating difference does not exactly coincide in time with the temperature difference, but leads it. This is expected since inline image is proportional to temperature tendency, not to temperature itself. Thereafter, the difference in the model responses shows an increased downwelling in the high-top model in the troposphere in January–February, consistently with the reported differences in SLP (Figure 5b).

Figure 10.

Same as in Figures 8a8c but for residual vertical velocity inline image response averaged over 70°N–90°N. The values shown in Figure 10c are multiplied by 1.5.

[21] The fact that the increased downwelling response in the polar stratosphere in the high-top model is not larger than that in the low-top model throughout the whole winter season (Figure 10c) poses the question whether the BD circulation is comparably increased in both models. A standard diagnostic for the strength of the BD circulation is upward mass flux at 70 hPa averaged across the Tropics [e.g., Butchart et al., 2006]. Figure 11a shows mass flux response at 70 hPa averaged over 20°S–20°N for the 2xCO2 experiment. Both models simulate an increased BD circulation in response to 2xCO2 throughout the year. It is interesting that, despite its simplified stratospheric dynamics, the low-top model is capable of simulating the increased BD circulation response. The response is, however, weaker than that in the high-top model throughout the year, in particular between December and May, the NH winter season. This result supports the interpretation that the polar differences in the model responses are the consequences of biases in the representation of dynamical stratospheric processes in the low-top model, namely the low-top cold bias and weak variability (section 2).

Figure 11.

(a) Monthly mean tropical upward mass flux response to 2xCO2 at 70 hPa averaged between 20°S and 20°N and (b) monthly mean vertical component of the Eliassen-Palm flux response to 2xCO2 at 100 hPa averaged between 45°N and 75°N. Black (gray) lines show low-top (high-top) model responses. Error bars show the standard error of the mean responses.

[22] The BD circulation can be illustrated using mass stream function, which is defined following McLandress and Shepherd [2009a] as:

display math

where inline image is the residual meridional velocity [Andrews et al., 1987, equation (3.5.1a)], ϕ latitude, p pressure, and g the acceleration due to gravity. Previous studies have demonstrated that increases in both resolved and parameterized stratospheric drags have contributed to the strengthening BD circulation response to increased GHG forcing [e.g., Butchart et al., 2006, 2011]. The mass stream function for the resolved wave drag can be diagnosed using the downward control principle [Haynes et al., 1991]:

display math

where inline image, inline image is the zonal mean drag due to the Eliassen-Palm flux divergence ∇F [Andrews et al., 1987, equation (3.5.3)], a Earth's radius, ρ air density, f the Coriolis parameter, ū the zonal mean zonal wind, and the subscript means the derivative with respect to the corresponding coordinate. Given that not all parameterized drags were saved in the simulations, the mass stream function for the parameterized processes is deduced by the difference between Ψ, the total stream function, and Ψdc, the downward control stream function (a similar approach to deduce the effects of the parameterized drag was used in the work of Manzini et al. [2003]).

[23] The climate change responses of the mass stream functions (i.e., the difference between 2xCO2 and 1xCO2) are shown in Figure 12. The responses are shown for November–December, when the large difference between the stratospheric dynamical responses of both models is found in the polar stratosphere (Figure 10c). The residual vertical velocity inline image is proportional to the horizontal derivative of the stream function and can be calculated following McLandress and Shepherd [2009a] as:

display math

where H is the mean scale height (7 km). A maximum in stream function response separates an upwelling response south of it from a downwelling response north of it. The location of the maximum in Ψ (total stream function) response is height dependent (Figures 12a and 12b), and it roughly corresponds to the climatological position of the turnaround latitude, consistent with strengthening of the BD circulation. The difference in the response (Figure 12c) has a maximum at about 55°N, indicating that the downwelling response in the high-top model is weaker between the turnaround latitude and 55°N and stronger north of 55°N (except in the small area north of 80°N) compared to the low-top model. The stronger downwelling response north of 55°N in the high-top model is consistent with the different response of the polar stratosphere and, consequently, the different tropospheric response between the models.

Figure 12.

November–December mass stream function response (kg m−2 s−1) to 2xCO2 in the Northern Hemisphere in (a, d, g) the low-top model, (b, e, h) the high-top model, and (c, f, i) their differences. Figures 12a–12c show total stream function response, Figures 12d–12f show the response due to resolved waves, and Figures 12g–12i show the response due to parameterized processes. Solid (dotted) contours mark positive (negative) changes; dashed contours mark zero changes. The regions of anomalous downwelling are shaded.

[24] The Ψdc (downward control stream function) response in the high-top model shows a broad maximum centered at 60°N, which is missing in the low-top model response (Figures 12d and 12e). This maximum is due to a stronger cumulative (e.g., vertically integrated from the model tops) planetary wave drag response in the high-top model, and explains the stronger downwelling response in the high-top model north of 65°N (Figure 12f). This increased resolved wave drag in the high-top model is not associated with an increased frequency of major sudden stratospheric warmings (not shown). In addition, Figure 11b shows that in response to 2xCO2 the vertical component of the Eliassen-Palm flux [Andrews et al., 1987, equation (3.5.3)] is comparably increased in both models at 100 hPa. The increased vertical component of the Eliassen-Palm flux indicates that the forcing due the resolved tropospheric waves contributes to the reported strengthening of the BD circulation in both models. However, the difference between the two models is not significant throughout the year and cannot explain the larger resolved wave drag response in the high-top model. Therefore, the different resolved wave drag response must be due to differences in wave dissipation between the models. In summary, Figure 12f shows that the difference in the response of the resolved wave drag can explain the stronger downwelling response in the high-top model north of 65°N but cannot explain the stronger downwelling between 55°N and 65°N. This latter difference is consequently attributed to the effects of parameterized processes in the models (including orographic and nonorographic gravity wave drags, low-top model sponge layer, and numerical dissipation) on the circulation, and is depicted in Figures 12g12i for clarity.

[25] Figures 12g and 12h show a maximum in midlatitudes at about 70 hPa in both models that is due to the orographic gravity wave drag (OGWD) response (not shown), and it is stronger in the high-top model. The high-top model response is also stronger in midlatitudes between 70 and 10 hPa, suggesting that here the contribution of the nonorographic gravity wave drag (NOGWD) becomes important. Note that a NOGWD parameterization is missing in the low-top model. Figure 12i shows the stronger downwelling in the high-top model south of 70°N which cannot be explained by the resolved processes and therefore is inferred to be due to the contribution of the parameterized processes to the BD circulation response. On the basis of these calculations, the different response of the polar stratosphere between the models is related to a stronger BD circulation response in the high-top model, caused by both stronger resolved and parameterized wave drag responses.

4. Discussion and Conclusions

[26] We have compared responses of low-top and high-top atmospheric models to doubled CO2 concentration and corresponding SST and SIC changes. The low-top model (with top at 10 hPa) is representative of the model generation used in the IPCC AR4, and the high-top model (with top at 0.01 hPa) is representative of a few models now participating in the Coupled Model Intercomparison Project phase 5 (CMIP5). While the high-top model allows for explicit simulation of planetary wave–mean flow interactions in the stratosphere and includes an additional nonorographic gravity wave drag contributing to the overall driving of the BD circulation in the stratosphere and mesosphere, the low-top model attempts to include these effects with a sponge layer. The main results obtained are as follows:

[27] 1. The key difference between the dynamical responses of the two models is a smaller decrease of the Arctic SLP during late winter in the high-top model when compared to the low-top model. The difference has been reproduced in several experiments which differed from each other by prescribed SST anomalies in the tropical Pacific. In addition to the two experiments described in this paper we have found similar differences between the high-top model and low-top model responses in experiments with two other SST patterns which have different magnitude of the SST anomalies in the tropical Pacific, further confirming that the different response is robust with respect to different SST anomaly patterns. Thus, the overly strong Arctic SLP decrease in the low-top model may be partly an artifact owing to a poorly represented stratosphere.

[28] 2. The different dynamical responses diagnosed by the Arctic SLP may have significant consequences for climate simulations in the 21st century. In particular, we have found significant differences in the precipitation response over Europe between the models, which are linked to different response of the Atlantic jet stream. The response of the Atlantic jet stream is, however, sensitive to prescribed SST forcing and likely depends on several factors, not yet well understood. Our findings reinforce similar findings by Scaife et al. [2012] but may be more relevant for the climate simulations of the 21st century because Scaife et al. [2012] studied the response to a quadrupled CO2 concentration; that is, to a forcing which is unlikely to be achieved at least during the next century.

[29] 3. We have demonstrated a sensitivity of the NH extratropical circulation response to projected tropical Pacific SST changes. Such sensitivity was previously emphasized by Yamaguchi and Noda [2006] and Meehl et al. [2007] in relation to the North Pacific climate change. They pointed out that, in response to the projected El Niño–like changes in the tropical Pacific, the North Pacific SLP decreases. However, some IPCC AR4 climate models simulate an increased SLP response in this region associated with a NAM-like response. This NAM-like response is likely related to the strengthening temperature gradient between warming tropical troposphere and cooling extratropical stratosphere [e.g., Lu et al., 2008]. The relative importance of these two factors varies from model to model leading to a scatter in the climate change responses in the North Pacific across the models. Our results suggest that not only is the response in the North Pacific sensitive to the tropical Pacific SST changes but the whole NAM-like response is. The El Niño–like warming pattern causes a negative NAM-like response associated with a positive Arctic SLP change, which is consistent with the NAM-like response to El Niño variability demonstrated among others by Bell et al. [2009], Cagnazzo and Manzini [2009], and Ineson and Scaife [2009].

[30] 4. The difference in the high-top and low-top model responses in the Arctic SLP in late winter has been traced to differences in the modeled stratospheric responses in early winter. This delayed downward connection between stratospheric and tropospheric differences in model responses is reminiscent of downward propagation of intraseasonal anomalies deduced from observations [Baldwin and Dunkerton, 1999, 2001]. The origin of the early winter difference in the stratospheric responses in the two models is attributable to a stronger BD circulation response in the high-top model. While it was generally expected that the low-top models are unlikely to properly simulate stratospheric climate change, the specific processes missing in the low-top models were not previously pinned down. Here we have shown that resolved wave drag alone cannot explain the simulated stronger BD circulation response in the high-top model, implying also a significant role of the gravity wave drags.

[31] The increased resolved wave drag in the high-top model is not associated with an increased frequency of the sudden stratospheric warmings, consistent with the results of McLandress and Shepherd [2009b] and Scaife et al. [2012] but inconsistent with the results of Bell et al. [2010]. We note, however, that Bell et al. [2010] used a mixed layer ocean, therefore it is not ruled out that a feedback with a dynamical ocean could actually increase the frequency of sudden stratospheric warmings. The weaker resolved wave drag response in the low-top model is quite likely associated with the excessive numerical dissipation of the waves in the low-top model, which prevent them from reaching the upper levels. The resolved wave drag is stronger in the high-top model already in the control (1xCO2) simulation (not shown); thus it is reasonable to expect that the response of the drag to the same tropospheric forcing change is also stronger in this model. The analysis of the different parameterized drag response suggests that both OGWD and NOGWD contribute toward the stronger BD circulation response in the high-top model. While the reason for the stronger OGWD response is not clear, the NOGWD contribution to the different response is obviously robust since a NOGWD parameterization is missing in the low-top model. Butchart et al. [2011] found that the NOGWD response accounts in average for 7% of the simulated BD circulation response across several stratosphere-resolving models, consistent with our conclusion.

[32] Previous studies showed that, because of a large internal variability, the response of the NH polar stratosphere to increased GHG concentrations is very uncertain, even with respect to the sign of polar temperature changes [Butchart et al., 2000; Austin et al., 2003]. This uncertainty is most probably related to the planetary wave–mean flow interaction. Therefore the details of the stratospheric response, especially the planetary wave response, may differ from model to model. Sigmond and Scinocca [2010] also suggested that the stratospheric response, and thereafter its projection on the troposphere, can be sensitive to the strength of the simulated stratospheric zonal winds in present-day climate modulated in their study by changes in the OGWD scheme. The role of parameterized gravity wave drag in strengthening the BD circulation inferred in this work is consistent with their results. Indeed, the momentum flux deposition from both orographic and nonorographic gravity waves is an important factor in determining the strength of the stratospheric and mesospheric circulations [Manzini and McFarlane 1998].

[33] In summary, we conclude that the introduction of the stratosphere-resolving models can lead to a significant improvement of tropospheric climate predictions, by including in the SLP response the BD circulation changes. This motivates further testing of these findings using transient climate simulations with coupled atmosphere ocean models including a well resolved stratosphere, in particular those which are being prepared in the frame of the ongoing CMIP5 project.

Appendix A

[34] When estimating the difference in climate change response of variable X between two models, the two-sample t test cannot be used because four independent samples are compared. To estimate the significance of the difference in the response we calculate the t statistic as follows:

display math

where the indexes 1 and 2 refer to different models (i.e., high top and low top), indexes p and f refer to 1xCO2 and 2xCO2 experiments, overbars mean averaged values, n is the length of sample (equal between our experiments), and Sp, pooled standard deviation, is calculated as follows:

display math

The fact that the t statistic defined by equation (A1) has a t distribution with (n1p + n1f + n2p + n2f − 4) degrees of freedom under the null hypothesis of equality of the responses can be verified following the procedure given in the work of von Storch and Zwiers [1999, p. 112].

Acknowledgments

[35] A.Y.K. is supported by the Finnish Academy via a postdoctoral fellowship and the SAARA project. A.Y.K. thanks P. Räisänen for help with setting up ECHAM5 simulations and F. Bunzel for help with the construction of SST/SIC boundary conditions. E.M. acknowledges the partial support of the COMBINE project of the European Commission's 7th Framework Programme. We thank Michael Sigmond and three anonymous reviewers for constructive comments. ECMWF is acknowledged for providing ERA-40 data. PCMDI and WGCM are acknowledged for their roles in making available the WCRP CMIP3 multimodel data set. Support of this data set is provided by the Office of Science, U.S. Department of Energy.

Ancillary

Advertisement