Externally forced symmetric warming in the Arctic and Antarctic during the second half of the twentieth century

Abstract

In recent decades, the two polar regions have exhibited strikingly different changes, with much greater warming in the Arctic than the Antarctic. However, the warming asymmetry between the two polar regions is quite small during the second half of the twentieth century. By using a multi-member ensemble of simulations with the Community Earth System Model, this study investigates the relative contributions of greenhouse gases, aerosol, and ozone forcings to the responses of Arctic and Antarctic surface temperature during 1955–2000. Results show that both the greenhouse gases- and aerosols-induced changes are greater in the Arctic than in the Antarctic, yet they are opposite and act to balance each other, leaving a limited warming in the Arctic and hence a small bipolar asymmetry. Using a radiative kernel, feedback analysis reveals that both greenhouse gases and aerosol forcings influence the polar surface temperature through albedo feedback related to sea ice changes and lapse rate feedback related to strong surface temperature inversion. The ozone forcing can hardly excite any surface temperature changes over the polar regions even in the Antarctic with the strongest ozone depletion, which is due to a cancellation between the cooling effect from radiative forcing and cloud radiative feedback, and the warming effect from lapse rate feedback and enhanced atmospheric heat transport from lower latitudes.

Introduction

During the past two decades, a robust feature in both observed and modeled surface temperature changes is the enhanced warming over the Arctic, known as the famous Arctic amplification (Holland and Bitz 2003; Serreze and Francis 2006; Serreze and Barry 2011; Cohen et al. 2014; Stuecker et al. 2018). By contrast, the warming in the Antarctic is relatively modest and is largely spatially heterogeneous (Doran et al. 2002; Li et al. 2021). In particular, the Antarctic Peninsula is found to be a region with one of the largest warming trends since the 1950s, whereas the Southern Ocean and the eastern Antarctic have experienced little warming, and even cooling, over recent decades (Thompson et al. 2011; Armour et al. 2016; Oliva et al. 2017). This asymmetric surface temperature change between the two polar regions has global climatic, ecological, and social impacts, and thus its underlying mechanisms have been investigated as a hotspot (Turner et al. 2005; Masson-Delmotte et al. 2006; Hall 2010).

However, the bipolar asymmetry only becomes significant in the recent decades (Fig. 1a). During the second half of the twentieth century, the observed surface temperature changes in the two polar regions are in nearly equal magnitude, exhibiting a symmetric warming feature. It is unclear whether the absence of the bipolar warming asymmetry during this period was rooted in externally forced response or internally generated variability. A recent paper by England (2021) pointed out that the detrended surface temperature variabilities in the Arctic and the Antarctic since the 1920s are determined by external forcings and Pacific decadal variability, respectively. However, they mainly analyzed the multi-decadal oscillation around a linear trend since the 1920s, while in this paper we focus on the bipolar symmetric warming trend during 1955–2000 and the related maintaining mechanisms.

Fig. 1

Time series of surface temperature change (K) over the Arctic (red) and the Antarctic (blue) in a Berkeley Earth’s primary product and b CESM-LENS. The temperature change is relative to a monthly climatological mean obtained by averaging over 1951–1980. The thick solid and dashed lines represent linear trends from 1955 to 2000 and from 2000 to 2020, respectively

For the surface temperature changes in the Arctic, many previous observational and modeling studies ascribed an important role to the effects of increasing greenhouse gases (GHG) and the related radiative feedbacks (Holland and Bitz 2003; Stroeve et al. 2012; Zhang et al. 2018; Liu et al. 2018, 2020). Overall, the GHG forcing can contribute about 2.43 ${\text{W}}\;{\text{m}}^{ - 2}$ to the direct radiative forcing of global temperature change (Houghton et al. 2001). In addition, the GHG-induced warming is further amplified by albedo feedback related to sea ice melting and lapse rate feedback associated with the vertical structure of the atmospheric warming (Budyko 1969; Alexeev et al. 2005; Cai 2005; Screen and Simmonds 2010; Pithan and Mauritsen 2014; Burt et al. 2016; Goosse et al. 2018; Stuecker et al. 2018). However, GHG concentration has shown a rapid increase since the 1950s (Additional file 1: Fig. S1), then why did the accelerated Arctic warming occur in recent decades? It is reasonable to speculate that other anthropogenic forcings and/or natural variability might contribute. For instance, increased anthropogenic aerosol (AER) forcing is found to have mediated the GHG-induced warming in the Arctic (Emori et al. 1999; Myhre et al. 2014; Bellouin et al. 2020; Cohen et al. 2020; England et al. 2021). In addition, some AER, like black carbon, can effectively reduce the surface albedo through deposition on snow and ice and hence play a warming effect in the Arctic (Flanner et al. 2007; He et al. 2014; Qian et al. 2015). As for the ozone forcing, although the ozone depletion (OD) is confined to the Southern Hemisphere, increased Ozone-depleting substances are observed in the Northern Hemisphere, which traps more outgoing longwave radiation and thus contributes positively to the Arctic warming (Ramanathan et al. 1987; Shine 1991; Marshall et al. 2014; Virgin and Smith 2019; Polvani et al. 2020).

For the Antarctic, on the one hand, the warming pattern is likely modulated by atmospheric circulation associated with internally generated variability (Vaughan et al. 2003; Mayewski et al. 2009; Steig et al. 2009; England, 2021; Tewari et al. 2021). For example, the rapid warming over West Antarctic in recent decades is found to be related to the multi-decadal sea surface temperature variability in the North Atlantic and tropical Pacific (Ding et al. 2011; Wang et al. 2020; Eayrs et al. 2021). On the other hand, the surface temperature changes in the Antarctic are also affected by external forcings, especially the anthropogenic emissions of ozone-depleting substances. For example, the OD in the Antarctic lower stratosphere is found to be responsible for the poleward shift of the Southern Hemisphere atmospheric circulation during the austral summer over the second half of the twentieth century (Thompson et al. 2000; Polvani et al. 2011; Fogt and Marshall 2020), which further contributes to the rapid warming over the Antarctic Peninsula through advecting warm air from lower latitudes and enhancing ocean upwelling that transports warmer water from the subsurface to the surface (Solomon et al. 1986; Turner et al. 2005; Monaghan et al. 2008; Joughin et al. 2012; Pritchard et al. 2012; Li et al. 2014; Wang et al. 2020). However, by analyzing climate model outputs, Polvani et al. (2021) found that the Southern Annular Mode change due to the OD has limited influence on the sea ice expansion around Antarctica in recent decades.

Since the externally excited climate response is tangled with the internally generated low-frequency variability, it is extremely challenging to identify the relative roles of different external forcings in shaping the climate change in the polar regions, not to mention their contributions to the symmetric responses between the Arctic and the Antarctic during the second half of the twentieth century. One possible approach to help us better understand the problem is to employ large ensembles of the same model initialized with the different conditions but perturbed by the same external forcing. With ensemble member mean, we can separate the externally forced response from the internally generated variability to a large extent.

By employing the Community Earth System Model large ensemble (Kay et al. 2015), this study performs an analysis on the relative roles of three major climate drivers (i.e., GHG, AER, and OD) on the radiative forcing and related feedbacks over the polar climate, as well as the bipolar symmetric warming during the second half of the twentieth century. As is shown later, we find that the OD forcing, although being strong over the Antarctic, can hardly excite any surface temperature changes over the polar regions due to the cancellation between the changes in lapse rate feedback, the cloud feedback, and the atmospheric heat transport. The GHG forcing indeed favors the generation of Arctic amplification with greater warming in the Arctic than the Antarctic, however, the AER forcing induces a greater cooling in the Arctic and substantially mitigates the bipolar asymmetry. “Data and methods” section leads on to “Results” section that presents the polar surface temperature changes induced by individual forcing as well as the underlying mechanisms, followed by “Summary and discussion”.

Data and methods

CESM-LENS

We examine the large ensemble set of historical simulations of the National Center for Atmospheric Research’s Community Earth System Model version 1 (CESM-LENS; Kay et al. 2015). CESM-LENS consists of coupled atmosphere, land, ocean and sea ice models. The atmospheric component is the Community Atmosphere Model version 5 (CAM5; Kay et al. 2015) with 30 vertical levels and 1.25^o × 0.94° horizontal resolution. The land component, Community Land Model version 4 (CLM4; Lawrence et al. 2012), has the same horizontal resolution as CAM5. The ocean component, the Parallel Ocean Program version 2 (POP2; Danabasoglu et al. 2012), has a nominal 1° horizontal resolution with meridional grid spacing decreasing to 0.3° near the equator. Vertically, it has 60 unevenly spaced vertical layers with the highest resolution of 10 m near the surface. The sea ice component, the Community Ice CodE (CICE; Hunke et al. 2010), has the same horizontal grid as POP2.

We analyze five simulation ensembles in CESM-LENS. To enable the comparison with observations, we first use the historical ensemble consisting of 42 realizations spanning from 1920 to 2020. The three “fixed” ensembles follow the same simulation protocol as the historical ensemble, except that one forcing agent (GHG, AER, or OD) is fixed at a constant level (1920 levels for GHG and AER, 1955 level for OD; England et al. 2016; Lenaerts et al. 2018; Deser et al. 2020). Note that the GHG and AER ensembles consist of 20 members, and the OD ensemble includes 8 members. In the following analysis, we select the same number of the historical ensemble as the single forcing ensemble for the sake of equality. The Arctic and Antarctic regions are defined as 60° N–90° N and 60° S–90° S, respectively.

For any variable of interest, we first subtract individual forcing-fixed ensemble from the historical ensemble and average over 1955–2000. Then, the response to the individual forcing is obtained with the average further subtracting a base value, which is defined as a mean over 1945–1965 for the GHG- and AER-fixed ensembles, and 1955 for the OD-fixed ensemble (since this set of experiments start from 1955). Note that the base value has been tested with different time periods and the results are essentially the same.

Radiative kernels

The energy balance of an atmospheric column in an equilibrium state is among the top-of-atmosphere (TOA) heat flux ($F^{\prime}_{{{\text{TOA}}}}$), the surface heat flux ($F^{\prime}_{{\text{s}}}$), and the convergence of the vertically integrated atmospheric heat transport ($\nabla \cdot F^{\prime}$):

$$0 = F^{\prime}_{{{\text{TOA}}}} + F^{\prime}_{{\text{s}}} + \nabla \cdot F^{\prime},$$

(1)

where prime indicates the difference between two simulations. Changes in atmospheric energy storage are very small and set to zero in this equation. To evaluate the contributions from various physical processes, we employ the radiative kernel method (Soden et al. 2008) to decompose the TOA radiative anomalies ($F^{\prime}_{{{\text{TOA}}}}$) into contributions from different radiative feedbacks:

$$F^{\prime}_{{{\text{TOA}}}} = R_{{\text{F}}} + R^{\prime}_{{{\text{PLK}}}} + R^{\prime}_{{{\text{LPS}}}} + R^{\prime}_{{{\text{ALB}}}} + R^{\prime}_{{{\text{WV}}}} + R^{\prime}_{{{\text{CLD}}}} ,$$

(2)

where $R_{{\text{F}}}$ represents the instantaneous radiative forcings, with the GHG forcing being obtained from GISS modelE (Hansen et al. 2007), and the AER and OD forcings from ACCMIP (Lamarque et al. 2013), respectively. The rest of the terms in Eq. (2) are the Planck feedback $\left( {R^{\prime}_{{{\text{PLK}}}} } \right)$, lapse rate feedback $\left( {R^{\prime}_{{{\text{LPS}}}} } \right)$, albedo feedback $\left( {R^{\prime}_{{{\text{ALB}}}} } \right)$, water vapor feedback $\left( {R^{\prime}_{{{\text{WV}}}} } \right)$, and cloud feedback $\left( {R^{\prime}_{{{\text{CLD}}}} } \right)$. Following Liu et al. (2018), the cloud radiative feedback is estimated as a residual of the other terms in Eq. (2) to close the energy budget. These feedbacks are adopted from the radiative kernels derived from the Rapid Radiative Transfer Model (RRTM) proposed by Huang et al. (2017).

Results

Surface temperature

Comparing Fig. 1b to a, the ensemble mean of the historical simulations in CESM-LENS can faithfully capture the observed warming trends during 1955–2000 in both polar regions, indicating that the symmetric warming feature is largely determined by external forcings rather than internal variabilities. Thus, we first examine the surface temperature changes in the historical simulations and the contributions from each individual forcing during the second half of the twentieth century (Fig. 2).

Fig. 2

Surface temperature changes (K) in the a, b historical simulation and the contributions from c, d GHG, e, f AER, and g, h OD over the two polar regions during 1955–2000 relative to 1945–1965 for GHG and AER and 1955 for OD. The area-weighted surface temperature anomalies are labeled in the corresponding panels

The area-weighted surface temperature response in the Arctic reaches ~ 0.27 K (Fig. 2a), which is more than twice of the global mean surface temperature warming (~ 0.10 K). While the amplified warming in the Arctic is entirely attributed to the GHG forcing (~ 0.73 K; Fig. 2c), AER acts to reduce it to a large extent (~ − 0.56 K; Fig. 2e). In contrast, the OD forcing has a very limited impact on the Arctic surface temperature (~ − 0.06 K; Fig. 2g).

Similar to the situation in the Arctic, the surface temperature change in the Antarctic is also characterized by an enhanced warming (~ 0.28 K; Fig. 2b), and the warming appears to be larger in the Indian sector of the Southern Ocean. In terms of the GHG-induced surface temperature response, the warming in the Antarctic (~ 0.40 K; Fig. 2d) is only about half of that in the Arctic, clearly exhibiting an Arctic amplification feature. However, different from the strong cancellation between GHG- and AER-induced responses in the Arctic, the AER-induced cooling in the Antarctic (~ − 0.15 K; Fig. 2f) is very weak and cannot offset the GHG-induced warming. In addition, it is somewhat striking to see the small contribution from OD (~ 0.03 K; Fig. 2h), since previous studies have argued that the OD forcing can exert strong influence on the Southern Hemisphere troposphere and surface temperature changes since 1950s (Solomon 1990; Randel and Wu 1999; Cionni et al. 2011; Thompson et al. 2011; Previdi and Polvani 2017; Xia et al. 2020). We will show evidence later that this small surface temperature change induced by OD is due to a cancellation between the cooling effect from radiative forcing as well as cloud radiative feedback and the warming effect from lapse rate feedback as well as atmospheric heat transport.

Radiative forcings

The area-mean radiative forcings ($R_{{\text{F}}}$) over the two polar regions due to GHG, AER and OD are presented in Fig. 3. Because of its well-mixed nature, the GHG-induced $R_{{\text{F}}}$ is positive and similar in magnitude between the Arctic (~ 0.82 ${\text{W}}\;{\text{m}}^{ - 2}$) and Antarctic (~ 0.71 ${\text{W}}\;{\text{m}}^{ - 2}$). The $R_{{\text{F}}}$ due to AER forcing is much smaller in magnitude than that to GHG forcing, and it is much larger in the Arctic (~ − 0.22 ${\text{W}}\;{\text{m}}^{ - 2}$) than in the Antarctic (~ − 0.09 ${\text{W}}\;{\text{m}}^{ - 2}$). The OD-induced $R_{{\text{F}}}$ also appears to be asymmetric over the two polar regions, and it is quite negative in the Antarctic (~ − 0.38 ${\text{W}}\;{\text{m}}^{ - 2}$) because of the reduced downwelling longwave radiation there (Ramanathan and Dickinson 1979; Ramaswamy et al. 1992; de F. Forster and Shine 1997; Chen et al. 2007; Shindell et al. 2013; Chiodo et al. 2017; Virgin and Smith 2019).

Fig. 3

Surface temperature changes (K) and contributions to the TOA heat flux (${\text{W}}\;{\text{m}}^{ - 2}$) from radiative forcing ($R_{{\text{F}}}$), feedbacks of Planck $\left( {R^{\prime}_{{{\text{PLK}}}} } \right)$, lapse rate ($\left( {R^{\prime}_{{{\text{LPS}}}} } \right)$, albedo $\left( {R^{\prime}_{{{\text{ALB}}}} } \right)$, water vapor $\left( {R^{\prime}_{{{\text{WV}}}} } \right)$ and cloud $\left( {R^{\prime}_{{{\text{CLD}}}} } \right)$ as well as atmospheric heat convergence ($\nabla \cdot F^{\prime}$) and surface heat flux ($F^{\prime}_{{\text{s}}}$) in GHG (red bars), AER (blue bars), and OD (yellow bars) over the a Arctic (60°–90° S) and b Antarctic (60°–90° N). The error bars denote the standard deviations of values in CESM-LENS members. Note the y-axis scales in a and b are different

Kernels decomposition

To understand how GHG and AER radiative forcings contribute to the symmetric surface temperature changes during the second half of the twentieth century in both polar regions, and why the OD forcing is less effective in causing the surface temperature change in the Antarctic, we decompose the TOA heat flux into radiative feedbacks based on Eq. 2.

As shown in Fig. 3, in both polar regions, the warming response to GHG forcing is mainly contributed by radiative forcing, lapse rate and albedo feedbacks. The water vapor feedback also makes a contribution, but it is quite feeble because of the dry atmosphere there (Payne et al. 2015; Held and Soden 2000). The Planck feedback is necessarily the opposite to the temperature change and is the major damping term to the warming response. The cloud radiative feedback and surface heat fluxes also act to suppress the Arctic warming but to a small magnitude. The contributions from these processes due to AER are overall opposite to those due to GHG. The above results are consistent with previous studies (Soden and Held 2006; Shell et al. 2008; Soden et al. 2008; Pithan and Mauritsen 2014; Stuecker et al. 2018).

Regarding the situation in OD, its impact occurs mainly in the Antarctic. In particular, the convergence of the vertically integrated atmospheric heat transport ($\nabla \cdot F^{\prime}$) appears to be a major contributor and works to transport more energy into the Antarctic, which is quite different from its feebleness in both GHG and AER. In addition, the lapse rate feedback is another major warming process in the Antarctic. However, the warming induced by atmospheric heat transport and lapse rate feedback is largely balanced by the cooling due to cloud radiative feedback and exerted radiative forcing, resulting in a small surface temperature change in the Antarctic.

The albedo feedback is largely proportional to sea ice cover changes and the related mechanism is quite straightforward, and thus not the focus in the following analysis. Apart from the albedo feedback, the lapse rate and cloud feedbacks, as well as atmospheric heat convergence, make significant contributions to the energy balance, especially in the Antarctic. We next examine physical mechanisms related to these processes, with a focus on the compensation relationship among them in OD.

Lapse rate feedback

The lapse rate feedback is closely related to the uneven change of vertical temperature profile (Manabe and Wetherald 1975; Wetherald and Manabe 1988). The coupling mechanisms between the surface and the upper atmosphere are quite different between the high-latitude and tropical regions, i.e., the coupling is through radiation in the former while it is through deep convection in the latter. Therefore, the lapse rate feedback is usually positive in the high latitudes and negative in the tropics (Hansen et al. 1984; Graversen et al. 2014; Pithan and Mauritsen 2014). Specifically, the GHG-forced inversion layer in the polar regions results in a larger warming in the near-surface air than in the upper troposphere (Fig. 4d, f), which leads to the return of longwave radiation and further amplifies the surface warming (Fig. 4a). Meanwhile, the deep convection in the tropics results in warmer upper troposphere than the surface (Fig. 4e), and thus the lapse rate feedback is negative there. The pattern of lapse rate feedback in AER is almost a mirror image of that in GHG in the troposphere (Fig. 4b).

Fig. 4

Spatial pattern of lapse rate feedback contribution to the TOA heat flux (${\text{W}}\;{\text{m}}^{ - 2}$) due to a GHG, b AER and c OD. Vertical temperature changes (K) in the atmosphere due to GHG (red), AER (blue), and OD (yellow) in the d Antarctic (60°–90° S), e Tropics (25° S–25° N), and f Arctic (60°–90° N)

However, the lapse rate feedback due to the OD forcing is positive in both high-latitude and tropical regions (Fig. 4c), a result quite different from those due to the GHG and AER forcings. Shown in Fig. 4d–f are the vertical temperature change profiles in the atmosphere under the three forcings, respectively. Deviation from the vertical uniform temperature profile will result in lapse rate feedback. It is clear that the strongest warming due to GHG and cooling due to AER occur at the lower (upper) troposphere in the polar regions (tropics), leading to positive (negative) lapse rate feedback there (Fig. 4a, b). In contrast, the strongest temperature changes due to OD occur at the bottom of the stratosphere as well as the upper troposphere, especially in the Antarctic and equatorial regions with strong OD forcing, which reduces the outward emission of longwave radiation and leads to positive lapse rate feedback there. Therefore, the lapse rate feedback due to OD is mainly determined by the atmospheric temperature structure of the stratospheric cooling, which is different from that due to GHG and AER that is controlled by the surface–troposphere coupling.

Cloud radiative feedback

Over the Antarctic, the contribution from cloud radiative feedback is quite strong, especially in response to the GHG and OD forcings. We next examine the vertical temperature and cloud fraction changes in the atmosphere (Fig. 5). Both the GHG- and OD-induced atmospheric temperature changes are featured with cooling in the lower stratosphere and warming (or weak cooling) in the upper troposphere, which increases instability and local relative humidity and thus enhances the formation of high-level clouds there (Fig. 5d, f; Hansen et al. 1997; Hodnebrog et al. 2014; Xia et al. 2018). Increasing clouds reflect more solar insolation back to space in summer and thus have negative feedback on the surface temperature change. On the other hand, the increasing clouds emit more longwave radiation to the surface and thus has a warming effect (Zelinka et al. 2012). Overall, the shortwave cooling effect dominates, causing a negative cloud radiative feedback in GHG and OD over the Antarctic (Fig. 3b).

Fig. 5

Changes of zonal mean temperature (K) and cloud fraction (%) due to a, d GHG; b, e AER; and c, f OD forcings in the Southern Hemisphere

Atmospheric heat convergence

For the Arctic region, the contribution from atmospheric energy transport ($\nabla \cdot F^{\prime}$ in Fig. 3a) is negligible compared to those from other major feedbacks in all three forcing scenarios, indicating that the Arctic surface temperature change is dominated by local feedbacks. For the Antarctic region, in contrast, the atmospheric heat convergence ($\nabla \cdot F^{\prime}$ in Fig. 3b) induced by OD plays a critical role and acts to mediate the negative instantaneous radiative forcing.

To examine the relationship between changes in atmospheric energy transport and local energy balance over the Antarctic due to OD, we regress the anomalous TOA net heat flux $\left( {F^{\prime}_{{{\text{TOA}}}} } \right)$ against the anomalous atmospheric heat convergence ($\nabla \cdot F^{\prime}$ in Fig. 6a) from 1955 to 2000. It is found that they are significantly negatively correlated with a correlation coefficient being − 0.81 at the 99% confidence level, suggesting that the atmosphere works to transport more energy from lower latitudes to the Antarctic to mediate the TOA energy imbalance exerted by the OD forcing.

Fig. 6

Linear regression of a OD-induced TOA heat flux ($F^{\prime}_{{{\text{TOA}}}}$) and its components of b shortwave ($F^{\prime}_{{{\text{SW}}}}$) and (c) longwave ($F^{\prime}_{{{\text{LW}}}}$) onto atmospheric energy convergence ($\nabla \cdot F^{\prime}$) in the Antarctic (60°–90° S) during 1955–2000. Lines are a least-squares linear fit to the data, r is the correlation coefficient, and all of the flux variables are filtered with a 12-month running mean. d Zonal mean wind climatology from the historical simulation (contours) and its change due to OD (color) in the Southern Hemisphere

A decomposition of $F^{\prime}_{{{\text{TOA}}}}$ into longwave ($F^{\prime}_{{{\text{LW}}}}$) and shortwave ($F^{\prime}_{{{\text{SW}}}}$) components finds that the strong negative correlation between $F^{\prime}_{{{\text{TOA}}}}$ and $\nabla \cdot F^{\prime}$ is largely due to its longwave component (Fig. 6c). In particular, the OD forcing causes a strong cooling in the lower stratosphere (Fig. 5c), which in turn leads to a reduction in outward emitted longwave radiation. The reduced downwelling longwave radiation further cools the troposphere and stratosphere in the Antarctic, enlarging the meridional temperature gradient. Since the synoptic wave fluxes of heat are essentially diffusive, the increased meridional near-surface temperature gradient triggers a more complex dynamical effect (Held 1999; Grise et al. 2009). This effect on the fluxes of heat in the lower troposphere further causes a heat flux convergence in the Antarctic (Haynes et al. 1991; Song and Robinson 2004; Thompson et al. 2006). In addition, this poleward wave fluxes of heat can drive a poleward shift and a strengthening of the subpolar jet (Fig. 6d), in agreement with previous studies (Karoly 1990; Lorenz and Hartmann 2001; Thompson et al. 2011; Grise et al. 2013; Previdi and Polvani 2014; Swart et al. 2015).

In contrast, the shortwave component has limited contribution (Fig. 6b) because the high albedo over the Antarctic region reflects most of the downward shortwave back to space. This is further supported by the close-to-zero albedo feedback in the Antarctic (Fig. 3b), indicating that the OD forcing has only a small impact on sea ice changes (not shown).

Summary and discussion

During the second half of the twentieth century when anthropogenic forcings were strong, both the observations and model simulations show a symmetric warming between the Arctic and the Antarctic, which is in sharp contrast to the Arctic amplification and strong bipolar asymmetric warming during the recent decades. While the latter has been extensively studied, this study examines the symmetric warming in the Arctic and Antarctic during 1955–2000 and identify the relative contributions from GHG, AER, and OD by employing the CESM-LENS. We find that the OD forcing has little impact even in the Antarctic where the ozone depletion is the strongest, and the GHG-induced warming in the Arctic is largely mitigated by the AER-induced cooling. Taken together, there appears to be a small warming asymmetry between the Arctic and Antarctic.

We quantitatively partition the surface temperature changes into contributions from various climate feedbacks by using radiative kernels. It is found that GHG induces surface temperature changes mainly through lapse rate feedback and albedo feedback in both polar regions, and the feedbacks are stronger in the Arctic than the Antarctic. The AER-induced cooling in both polar regions also works through the lapse rate and albedo feedbacks, and mitigates largely the warming induced by GHG.

Although the OD forcing induces strong cooling in the lower stratosphere over the Antarctic region, its impact on the local surface temperature is rather small. A further analysis finds that the lapse rate feedback and atmospheric heat transport cause significant warming in the Antarctic, which, however, is cancelled out by the cooling effect from the OD radiative forcing and cloud radiative feedback. The warming effect from the lapse rate feedback and atmospheric heat transport is related to the strong stratosphere cooling, which inhibits the outgoing longwave radiation and results in a negative lapse rate feedback. On the other hand, the decrease in the TOA heat flux due to the reduced outgoing longwave radiation is balanced by atmospheric heat transport from lower latitudes to the Antarctic, leading to a strong atmospheric heat convergence there. Moreover, the stratospheric cooling is also responsible for the strong negative cloud feedback, and causes instability in the upper troposphere and promotes the amount of high cloud, resulting in a surface cooling in the Antarctic.

The results in this study are mainly based on single model large ensembles. To validate the robustness of our results, we further analyze the large ensemble of second Generation Canadian Earth System Model (CanESM-LENS) experiments, which consists of 50 realizations spanning over 1950–2020. Overall, the results from CanESM-LENS are in good agreement with those from CESM-LENS. For example, CanESM-LENS can also capture the symmetric warming between the two polar regions during the second half of the twentieth century (Additional file 1: Fig. S2). The OD forcing can hardly excite any surface temperature changes over the Antarctic region (Additional file 1: Fig. S3). Therefore, this above agreement between the two large ensemble experiments suggests that the major conclusions of our study are not model-dependent.

It should be noted that the complex cloud feedback is estimated as a residual in this analysis, and thus errors and uncertainties associated with other feedbacks and nonlinear processes are all included in the cloud feedback. In addition, the simulations of the fixed ozone are performed with only 8 members, and thus influence from internal variability may still exist to some extent. However, the current results can successfully capture the strong cooling in the lower stratosphere and the poleward shift of the westerly jets that have been revealed by many previous studies, giving us confidence that the results discussed in this study are reliable.