Quantifying the Response Strength of the Southern Stratospheric Polar Vortex to Indian Ocean Warming in Austral Summer

The impact of tropical ocean warming on the stratospheric polar vortex has attracted much interest from the climate research community in recent years (e.g., Manzini et al., 2006; Li, 2009, 2010; Cagnazzo and Manzini, 2009; Hu and Pan, 2009; Hu and Fu, 2009; Ineson and Scaife, 2009; Li et al., 2010; Ren, 2012; Ren et al., 2012). This can be attributed to the following factors: (1) recent tropical ocean warming, to a considerable extent, represents the forced signals from anthropogenic greenhouse gases in spite of the minor role of natural variation (Knutson et al., 1999, 2006; Ihara et al., 2008), and understanding its impact is of substantial importance for identifying the role of human activity in inducing stratospheric climate change (Cai and Cowan, 2007); and (2) ocean warming may exert a significant impact on the recovery of stratospheric ozone depletion over the Antarctic through dynamically changing stratospheric thermal states and the resulting climate-chemical feedback. It has been speculated that tropical ocean warming may be an alternative factor responsible for the modest recovery of the Antarctic ozone hole observed in recent years (Li, 2009; Li et al., 2010), which is supplementary to the restriction of ozone consumption materials due to the efficacy of the Montreal Protocol (Perlwitz et al., 2008).

Figure 1.  Modeled responses in southern hemispheric geopotential height (contours; units: gpm) and air temperature (shading; units: K) to IOW in the five AGCMs in austral summer (DJF). Upper panels are for 50 hPa in the lower stratosphere and the lower panels are for 500 hPa in the troposphere. The contour interval in all the panels is 10 gpm.

Due to a large monotonic SST increase relative to its moderate decadal variability (Knutson et al., 1999, 2006), the tropical Indian Ocean is the epicenter for a detectable anthropogenic change. Thus, investigating the role of tropical Indian Ocean Warming (IOW) is a priority in understanding the impact of tropical ocean warming as a whole. Many previous studies have focused on tropospheric circulation, particularly those in the northern hemisphere including the northern polar vortex, the North Atlantic Oscillation/the Northern Annular Mode (e.g., Hoerling et al., 2001, 2004; Li et al., 2006, 2010). However, scientists have recently begun to also pay attention to the southern hemispheric polar stratosphere, since it is relevant to the recovery of the southern polar ozone hole and its future trend. (Li, 2009) and (Li et al., 2010) analyzed outputs of ensemble experiments prescribed with temporally ramped SST anomalies (warming) over the tropical Indian Ocean in five AGCMs (GFDL AM2, NCAR CCM3 and CAM3, NCEP GFS, and ECHAM5; please refer to the Tabel 1 for more details), and found that IOW tends to warm the southern polar stratosphere and weaken the southern polar vortex. The result is qualitatively consistent among the AGCMs (Li, 2009, Fig. 2) (Fig. 1), and implies a side-effect of anthropogenic greenhouse gas emissions; that is, favoring the recovery from ozone depletion (Li et al., 2010).

Figure 2.  The same as Fig. 1 but for the precipitation responses (units: mm month^-1).

However, a considerable difference is seen in the timing and strength of the lower stratospheric response between the models (Fig.1, upper panels) (Li, 2009, Fig.2). Three of the models (CCM3, CAM3, and GFS) show the strongest response during spring to early summer (i.e., October-November-December), whereas AM2 reveals the strongest response for austral winter (June/July). Where the austral summer (December-January-February; DJF) is concerned, two kinds of response strength are seen. The stronger one is from three of the AGCMs (CCM3, CAM3, and GFS) with modeled 50 hPa geopotential height responses of 150-200 gpm, while the weaker one is from the other two AGCMs (AM2 and ECHAM5) with a much weaker magnitude of just 30-50 gpm (Fig. 1). Correspondingly, air temperature response in the polar lower stratosphere is 2.0^°C-3.0^°C in the first three AGCMs, but around 1.0^°C in the last two models (Fig. 1, shading) (Li, 2009, Fig. 2). This means that the strength in the first three models may be as great as 3-5 times in the last two models. However, as to which case represents reality is unclear. Thus, it is difficult to estimate quantitatively what the impact magnitude of IOW on the southern polar vortex and the associated ozone recovery is in the real world.

Physically, a tropical SST anomaly influences extratropical stratospheric circulation through vertical propagation of Rossby waves with a smaller wave number (Charney and Drazin, 1961). This means two possible factors contribute to the above difference. One is the tropospheric signals induced by the SST, and the other is the background flow governing the wave's upward propagation and its deflection and reflection. Only when the background circulation is weak westerly can quasi-stationary Rossby waves propagate upward to reach the stratosphere. The importance of background flow was also revealed by a recent study (Codron, 2005), which showed that the shape and dynamics of the Southern Annular Mode can be modified by even modest changes to the background state. Hence, the model's capability in reproducing the observed background climate is important for understanding modeled responses.

Previous studies (e.g., Peng et al., 1997) on the uncertainty of modeled responses provide implications for the importance of background flow. Generally, modeled atmospheric responses to an external forcing depend strongly on the models' intrinsic variability, and only those realistically reproducing the observed atmospheric intrinsic variability may simulate the observed responses. Further, that external forcing induces an atmospheric response mainly through exciting intrinsic variability modes rather than inducing a new mode (Peng and Robinson, 2001). This implies that the strength of circulation systems and the leading variability patterns in a model's atmosphere may modulate modeled response strength. Thus, a comparison of a model's intrinsic variability may provide implications for our understanding of the modeled responses.

The present study will try to understand the aforementioned modeled stratospheric response difference between AGCMs by analyzing and comparing the models' stratospheric background states and intrinsic variability, with the possible tropospheric impact set aside. The present study focuses only upon austral summer (December-February), because during this season the IOW-induced strongest heating is located to the southern side of the equator (Fig. 2) and tends to induce the strongest tropospheric responses in the southern extratropics along with the climatological seasonal cycle of background SST.

The paper is organized as follows. Section 2 describes the datasets and methods. Section 3 compares the models' mean stratospheric climatological states in the five AGCMs used. Section 4 compares their intrinsic variability, including the standard deviation, the leading variability modes, and the longitudinal thermal transport of a quasi-stationary wave at 100 hPa, which reflects the vertical component of Eliassen-Palm flux (EPF) due to upward propagation of the planetary wave. This is followed by a brief discussion of the explanation for the different response magnitudes in section 5. Finally, a summary is given in section 6.

Monthly circulation variables including geopotential heights, air temperature, and zonal wind from the National Center for Environment Prediction (NCEP)-National Centers for Atmospheric Research (NCAR) Reanalysis (Kalnay et al., 1996) (Rean hereafter) for 1979-2008 are used as an observational dataset. This period is utilized because the data quality in the southern hemisphere is more reasonable due to the use of satellite observations since 1979 (Kistler et al., 2001).

The same model outputs from the five AGCMs as (Li et al., 2010) are used. A general description of these models and the experiments is given in Table 1. Because an idealized temporal ramping of SST was prescribed in the Indian Ocean in all the experiments and a trend response is expected, a detrending process is thus used in all the models' outputs for every single run when yielding model climatology. The model climatology for each season is obtained by averaging the ensemble detrended outputs for all model years. In comparison with the observations, a number of factors such as ozone depletion, volcanoes, and changes in greenhouse gases, which might affect climatology, are not included in the model runs.

Figure 3.  Comparison of modeled climatological zonal-mean geopotential heights in the five AGCMs as well as in the NCEP-NCAR Reanalysis. The dense (light) shading indicates the values greater than 20 700 (less than 18 300). The characters on the x-axis represent the calendar month from January to December (units: gpm).

Figure 4.  Comparison of climatological seasonal evolution of (a) the southern lower stratospheric polar vortex, which is expressed as 50 hPa geopotential heights averaged over the polar cap (65^°S to the south) (units: gpm), and (b) the polar jet expressed by averaged 50 hPa zonal-mean zonal wind over the band spanning 50^°-60^°S (units: m s^-1) in the five AGCMs as well as in the NCEP-NCAR Reanalysis.

Figure 5.  The same as Fig. 3, but for zonal-mean zonal wind. Shallow (dense) shading represents the values less than 0 (greater than 50). Units: m s^-1.

Figure 6.  Comparison of zonal-mean zonal wind in the five AGCMs as well as in the observation in austral summer (DJF). Units: m s^-1.

The strength of the stratospheric polar vortex and associated intrinsic variability, including monthly standard deviation and the leading modes of 50 hPa geopotential heights, are compared first. The EOF analysis is used to determine the leading hemispheric patterns, and is calculated for the simulated and observed 50 hPa heights to the south of 20^°S, respectively. For a single AGCM, the time series for EOF analysis is formed by combining detrended height fields in all the runs of the model.

As mentioned above, the primary physics for tropospheric forcing to influence the polar stratosphere is through yielding the upward propagating planetary waves. The waves reaching the stratosphere may be modulated by the stratospheric basic state through deflection and reflection. The factors influencing the processes include the waves themselves and the background flow the waves reside in. A comparison of quasi-stationary waves and background flow can give indications, considering that quasi-stationary waves account for a considerable part of planetary waves. Thus, the features of quasi-stationary waves including their strength, spatial pattern, and upward propagation (EPF) are also compared.

Figure3 compares seasonal evolution of zonal-mean geopotential heights at 50 hPa in the five AGCMs with the Rean. Obviously, all the AGCMs capture the seasonal evolution of geopotential heights. Lower geopotential heights at high latitudes (to the south of 70^°S), with the strongest polar vortex in late winter (July-August-September; JAS), is consistent with the Rean. Even so, a modest difference between the AGCMs is seen. The modeled polar vortices in three AGCMs (AM2, ECHAM5, and CCM3) are closer to the Rean, while the two other AGCMs (CAM3 and GFS) yield a relatively stronger polar vortex. A similar difference can be seen in air temperatures (not shown). This suggests a correspondence in the strengths of modeled responses to the polar vortices themselves in the models.

To scrutinize this point, Fig. 4a displays a comparison of the polar vortex index, which is defined as averaged geopotential heights at 50 hPa over the polar cap (65^°-90^°S). It shows a systematic bias with a stronger polar vortex in all the models. In comparison, the biases in GFS and CAM3 are most evident, while the bias is relatively smaller in the three other models (AM2, CCM3, and ECHAM5).

Figure 4b illustrates the climatological seasonal evolution of the southern polar jet, which is calculated as 50 hPa zonal-mean zonal wind at the latitudes of 50^°-60^°S. Except for late winter (JAS) during which ECHAM5 and AM2 produce a weaker one, the modeled jets are stronger in the other seasons and models relative to the observed. This illustrates again that the polar vortex in the lower stratosphere is stronger than the observed.

Figure 5 illustrates a comparison of the climatological seasonal evolution of zonal-mean zonal wind. In agreement with the above, most of the AGCMs reproduce the observed seasonality well generally, including the easterly in the tropics and the westerly in the extratropics, except for ECHAM, which has much weaker easterlies in summer (DJF). The tropical easterly achieves a maximum in summer (DJF), while the extratropical westerly achieves a maximum in late winter to spring (August-September). Corresponding to the polar vortex, the midlatitudinal westerly winds in CAM3 and GFS are stronger, while those in other models are relatively closer to the observed.

The above difference in the polar jet is also seen in the vertical profile of zonal-mean zonal wind in austral summer (Fig. 6). All the AGCMs reproduce the observations well, including a strong midlatitudinal tropospheric westerly wind and subtropical stratospheric easterly wind. The models' tropospheric and lower stratospheric westerlies are stronger than the observed, but the tropical stratospheric easterly winds are weaker. In particular, the tropospheric westerly winds in three AGCMs (CAM3, CCM3, and GFS) are even stronger, which correspond to the stronger 500 hPa height responses (Fig. 1).

Figure 7.  Comparison of standard deviation of 50hPa geopotential heights in austral summer in the five AGCMs as well as in the observations. Shading represents the difference of the modeled minus the observed. Units: gpm.

Figure 8.  Comparison of the leading EOF of 50 hPa geopotential heights in austral summer in the five AGCMs as well as in the observations.

Figure 7 displays a comparison of the standard deviation (SDV) of 50 hPa monthly geopotential heights. As observed, the SDV has a maximum over the polar region, representing the strongest variability of geopotential height there. Even so, a significant difference between the AGCMs, or between the AGCMs and the observed, is still visible (Fig. 7, shading). Except for the polar region, the SDV in all the AGCMs is overall smaller than the observed. This illustrates the models' insufficiencies to reproduce the observed variability, except at the polar region. The small amount of variabilities at lower latitudes may be due to the absence of quasi-biennial oscillation (Takahashi, 1999; Scaife et al., 2000; Giorgetta et al., 2002). In the polar region, biases are even more significant but less uniform in sign. Particularly, three AGCMs (CAM3, CCM3, and GFS) bear larger values, while the remaining two (AM2 and ECHAM5) yield values closer to the observed. Thus, the simulated variability of southern polar vortex in the latter two AGCMs is more reasonable.

As mentioned above, modeled responses to external forcing, including both their amplitudes and spatial patterns, are generally determined by the models' leading intrinsic variability patterns. Therefore, a comparison of the models' leading patterns can provide additional implications to explain the modeled responses. Figure 8 presents the first leading EOFs of southern hemispheric (to the south of 20^°S) 50 hPa heights. First, the spatial patterns of the leading EOFs in all the AGCMs resemble the observed, exhibiting a seesaw pattern between mid and high latitudes-the so-called Southern Annular Mode. A substantial resemblance of the EOFs to the observed is further seen from their spatial correlations (Table 2). The rate of explained variance by the leading EOF in the AGCMs is slightly higher than the observed. Despite the resemblance, an evident difference is visible. The amplitude of the EOF over the polar region in three AGCMs (CAM3, CCM3, and GFS) is greater than the observed. This bias corresponds to the larger systematic bias in the polar vortex in these models.

From Fig. 9, it can be seen that all the AGCMs capture the observed quasi-stationary wave well, including the structure with wave number 1 near the polar area. In comparison, AM2 and ECHAM5 reproduce the amplitude more realistically than the three other AGCMs (CAM3, CCM3, and GFS). The latter has a greater amplitude.

Quasi-stationary wave induced meridional heat transport [v*T*] (brackets [ ] indicate zonal mean) reflects the vertical component of EPF, which is a forcing indicator of the upward propagation of tropospheric planetary waves on the stratosphere. Figure 10 displays the seasonal evolution of the climatological mean [v*T*] at 100 hPa. All the five AGCMs reproduce the observed features well generally, including the strongest meridional transport in the mid-high latitudes to the southern polar region (negative values from ∼45^°S to the south) in spring (September-October-November; SON), the strong southward heat transport over the subtropics to the midlatitudes (30^°-45^°S) in the winter half year (April-November), weaker negative values in the subtropics (∼20^°S) through a whole year, and a positive value near the equator. Thus, the basic feature of vertical propagation of planetary waves across the tropopause is captured by all the AGCMs. However, in comparison to the observed, biases in the models are evident. The heat transport to the South Pole in the mid-high latitudes is obviously smaller, and their relative magnitudes and seasonality differ somewhat from the observed. Besides the attribution to the model bias, one candidate explanation for the inconsistency is that a number of factors (like ozone depletion, volcanoes, and changes in greenhouse gases) are not included in the model runs.

The high-latitudinal poleward heat transport in CAM3 looks closer to the observed in austral spring; however, in austral summer (DJF), this term in AM2 and ECHAM5 (rather than CAM3) is closer to the observed (Fig. 10g). This is consistent with the stationary planetary wave activity revealed in Fig. 8. Again, in austral summer AM2 and ECHAM5 show better performance.

Figure 9.  The same as Fig. 8, but for stationary waves at 50 hPa geopotential heights. Units: gpm.

Figure 10.  Comparison of stationary heat transport at 100 hPa in the five AGCMs as well as in the observations: (a-f) seasonal evolution, and (g) mean in the latitudinal band (55^°-65^°S) during austral summer (DJF). Units: K m s^-1.

Figure 11.  Comparison of modeled responses of stationary heat transport at 100 hPa to IOW in the five AGCMs with the observed trend. The observed trend is calculated as the difference of the mean through 1994-2008 minus 1979-1993. Units: K m s^-1.

Polar stratospheric air temperatures are crucially influenced by planetary-scale wave vertical propagation originating from the troposphere (Andrews et al., 1987), since breaking of these waves drives a meridional residual circulation (Brewer-Dobson circulation) in the stratosphere with rising motion in the tropics, poleward flow in the midlatitudes, and descending motion in polar regions. The adiabatic heating of the descending motion maintains higher temperatures over the polar region than that determined by the radiation balance. Thus, one possible cause for the above stratospheric difference is from the troposphere.

Indeed, a significant difference in simulated tropospheric responses can be seen among the AGCMs. At 500 hPa (Fig. 1, lower panels), both AM2 and ECHAM induce a weaker positive geopotential height response over the polar region with a maximum of 20-30 gpm, while the three other models (CAM3, CCM3, and GFS) yield a stronger response with a maximum of 40 gpm. Also, the models with the stronger tropospheric responses (CAM3, CCM3, and GFS) have greater biases in the tropospheric climatological background flow (Fig. 6). For the precipitation response (Fig. 2), GFS exceeds ECHAM by a factor of about two with respect to the maximum in the Indian Ocean. Moreover, GFS simulates a maximum response in precipitation south of the equator (0^°-20^°S), whereas ECHAM reveals maximum changes between 0^° and 20^°N. These differences appear to be strong candidates to explain at least a certain part of the sensitivity in stratospheric polar vortex strength, because rainfall-related latent heating release dominates total diabatic heating and may be a primary factor to drive remote tropospheric responses in the extratropics and beyond.

The tropospheric impact on the lower stroposphere may be seen from the EPF aross the tropopause. Because only monthly data are archived in the model outputs, heat flux induced by quasi-stationary waves, [v*T*], which is equivalent to the vertical component of EPFs, is analyzed here. We start by looking at the differences between the late and early periods-that is, the influence of the IOW-from the perspective that the troposphere exerts an upper impact through analyzing [v*T*].

Figure 11 shows a comparison of the modeled monthly response of [v*T*] at 100 hPa with the observed trend. Significant poleward heat transport exists in the high latitudes (∼60^°S) from austral late summer (JAS) to autumn (March-April-May; MAM), which is consistent overall in all the AGCMs. This explains the consistency in the modeled dynamical warming over the polar region. A comparison with the observed trend shows a consistency except for November. The observed trend in November is primarily induced by the radiative cooling effect due to ozone depletion (e.g., Perlwitz et al., 2008). Thus, the dynamic influence due to the quasi-stationary wave excited by tropical IOW tends to exert an opposite impact to ozone depletion.

It is intriguing that the difference in the troposphere is smaller than in the stratosphere (Fig.1). The ratio of the strongest versus weakest magnitude of tropospheric responses among the models is less than 2.0, much smaller than the value (∼5.0) in the stroposphere. This suggests that the heating-induced tropospheric response difference will be amplified within the stratosphere, implying a greater importance of stratospheric intrinsic variability and stratospheric background flow. The present comparisons demonstrate that two AGCMs (AM2 and ECHAM5) capture observed intrinsic variability and background flow better than the other three (CCM3, CAM3, and GFS). Thus, the modeled responses in the former two models may be more realistic. This means that, in the polar lower stratosphere (∼100 hPa), a 1.0^°C warming may have been dynamically induced by the observed IOW with the same strength during the past 50 years. This forced warming provides a possible explanation for the fact that the modeled cooling trend solely forced by ozone depletion is stronger than the observed (Gillet et al., 2003, Fig. 2; Kindem and Christiansen, 2001, Fig. 2).

A previous multiple-AGCM study suggested that tropical IOW tends to dynamically warm the southern polar stratosphere, thus favoring the recovery of ozone depletion. However, a significant difference in the strength of modeled responses was seen in these AGCMs. The modeled impacts in two of the AGCMs (AM2 and ECHAM5) are much smaller than those in the other three (CCM3, CAM3, and GFS). Determining which among these modeled impacts reflects the case in nature is important not only for understanding the impacts but also for projecting the recovery speed of ozone depletion, because ocean warming is very likely to be forced by anthropogenic greenhouse gas emissions. In the present study, the issue has been investigated by comparing the modeled intrinsic variability in all the AGCMs with the NCEP-NCAR Reanalysis, considering the models with unrealistic intrinsic variability tend to distort the atmospheric responses to external forcing. Both the stratospheric background flow and intrinsic variability were compared with the observed. The results suggest that two of the AGCMs (AM2 and ECHAM5) simulate the observed stratospheric basic state and intrinsic variability more realistically. In view of the larger model-to-model difference in response amplitude in the stratosphere relative to that in the troposphere, it is concluded that the IOW influence on the southern polar stratosphere is closer to those in AM2 and ECHAM5, but weaker than those simulated by CCM3, CAM3, and GFS. This means that the observed IOW during the past 50 years may have dynamically induced a 1.0^°C warming in the polar lowest stratosphere (∼100 hPa), and also implies that AM2 and ECHAM5 are more appropriate for studying stratospheric response to external forcing.

The finding that AM2 and ECHAM5 simulate the stratospheric basic state and intrinsic variability more realistically may be not a coincidence. The top of the atmosphere in the two models is 3 hPa and 1 hPa, respectively (Table 1). They are not only much higher than the three other models (∼10 hPa), but also closer to the upper top of the stratosphere in nature (∼1 hPa). This provides a potential explanation in that these two models have a better expression of the stratosphere.

The present result does not exclude the potential impact of tropospheric signals. Indeed, considerable differences in simulated tropical rainfall and midtropospheric atmospheric circulation responses between the models are seen. Nonetheless, the smaller model-to-model difference ratio in the troposphere argues for the lesser importance of the tropospheric signals.

There are several weaknesses in the present study. First, the physics for the formation of different stratospheric response strengths has not been deeply explored. The focused-upon season (austral summer) is not the strongest as far as the tropospheric-stratospheric coupling is concerned, although IOW is expected to seasonally induce the strongest tropospheric anomalies in the southern extratropics. Second, none of the five AGCMs used includes a detailed representation of the middle atmosphere and only one covers the complete stratosphere. There is evidence that inclusion of the middle atmosphere is important for representation of the mean flow and variability in the upper troposphere/lower stratosphere (e.g., Shindell et al., 1999; Huebener et al., 2007). It is completely unclear how far wave deflection/wave breaking in the stratosphere plays a role in explaining the results as presented here. Hence, the present study is not able toexplain wave interaction between the troposphere and the stratosphere better. Third, none of the models are coupled with an active ocean, while (Copsey et al., 2006) showed the importance of air-sea coupling in the tropical Indian Ocean. How well these AGCMs, when incorporated with a middle atmospheric model and/or coupled with an active ocean, will simulate the observed stratospheric intrinsic variability is unclear. Finally, the dataset used in this research covers the period 1979-2008, which obviously only includes the phase of global warming since 1977/78. It is known that there was a climate shift around the year 2000. Whether there exists a difference between the periods 1979-2000 and 2000-08 for the results achieved here is unclear. All these issues deserve further investigation in the future.