Shenlin Ci , Png-Chi Hsu , Fei Liu

a Earth System Modeling and Climate Dynamics Research Center, Nanjing University of Information Science and Technology, Nanjing, China

b School of Atmospheric Sciences, Key Laboratory of Tropical Atmosphere-Ocean System Ministry of Education, Sun Yat-Sen University, China

c Southern Marine Science and Engineering Guangdong Laboratory, Zhuhai, China

Keywords:Arctic amplification Antarctic amplification Radiative kernel CMIP6 Global warming

ABSTRACT The climate in polar regions has experienced an obvious warming amplification due to global warming. In this study, the changes in polar amplification are analyzed in response to feedback mechanisms (including Planck,lapse rate, cloud, water vapor, albedo feedback, CO2 radiative forcing, ocean heat uptake, and atmospheric heat transport) under three warming scenarios in CMIP6 —namely, SSP1-2.6, SSP2-4.5, and SSP5-8.5. The results show that, by quantifying the warming contribution of different feedback mechanisms to surface air temperature with the “radiative kernel ”method, Arctic amplification (AA) is stronger than Antarctic amplification (ANA), mostly resulting from the lapse rate feedback, followed by the albedo and Planck feedbacks. Furthermore, ocean heat uptake causes stronger polar warming in winter than in summer. During winter, the lapse rate feedback causes a larger AA than ANA. The intermodel spread for both AA and ANA decrease with increasing strength of global warming from SSP1-2.6 to SSP5-8.5, and the dominant mechanisms are the Planck, lapse rate, albedo, and ocean heat uptake feedbacks. These findings help to enhance our understanding of polar regions’ responses to different strengths of global warming.

1. Introductio n

Earth has been experiencing indisputable global warming under anthropogenic greenhouse gas (GHG) forcing since the industrial revolution ( IPCC, 2014 ). Among the warming regions, the Arctic has been the one with the most dramatic increase in surface air temperature(SAT), showing an increase of about twice the level of the global average ( Graversen et al., 2008 ; Screen and Simmonds, 2010 ; Taylor et al.,2013 ; Goosse et al., 2018 ; Wu et al., 2019 ). During the winters (January to March) of 2016 and 2018, the SAT in the central Arctic was 6 °C above the 1981—2010 average level ( Meredith et al., 2019 ). This phenomenon,which is known as the Arctic Amplification (AA), serves as one of the most significant features of global climate change and will exacerbate global warming and the loss of Arctic sea ice ( Wu et al., 2019 ).

The climate of the polar regions, which has displayed large internal variability, is highly sensitive to changes in climate forcing, such as water vapor, cloud, or other feedback mechanisms ( Goosse et al., 2018 ).The Arctic sea ice has been thinning, with the area of ice accretion over five years declining by 90% between 1979 and 2018 ( Wei et al., 2019 ).Large loss of sea ice in the Antarctic has also been found since 2014( Wang, 2019 ; England et al., 2020 ). Besides, the Antarctic sea-ice cover plummeted by 2 million kmfrom 2014 to 2017, which is almost as much as the loss in the Arctic in the past 40 years ( Wang, 2019 ).

The AA, which is defined as the ratio of the Arctic (north of 60°N)warming to tropical warming, has been found to be caused by several important factors ( Pithan and Mauritsen, 2014 ; Goosse et al., 2018 ;Stuecker et al., 2018 ). Surface albedo feedback is believed to be the uppermost contributor to the AA ( Svante, 1896 ; Manabe and Wetherald, 1975 ; Hall, 2003 ; Holland and Bitz, 2003 ; Mark and Jennifer, 2006 ;Screen and Simmonds, 2010 ; Crook et al., 2011 ; Taylor et al., 2013 ;Pithan and Mauritsen, 2014 ; Virgin and Smith, 2019 ). Large AA is usually associated with significant sea-ice loss ( Dai et al., 2019 ). Moreover,temperature feedback, which has recently been found to be the largest contributor to the AA ( Ramanathan, 1977 ; Pithan and Mauritsen, 2014 ;Virgin and Smith, 2019 ; Previdi et al., 2020 ), can be decomposed into the Planck feedback on the basis of the well-mixed troposphere, and the lapse rate feedback for the vertical structure of warming ( Manabe and Wetherald, 1975 ; Bintanja et al., 2012 ; Stevens et al., 2013 ; Pithan and Mauritsen, 2014 ; Stuecker et al., 2018 ; Zhang et al., 2018 , 2020 ).

The water vapor feedback —a strong feedback under GHG-induced global warming ( Soden, 2005 ; Gordon et al., 2013 ) —is much larger in the tropics than in the Arctic. Therefore, it is a negative feedback for the AA ( Pithan and Mauritsen, 2014 ). The cloud feedback, however, is a complex contributor to the AA because the net cloud feedback depends heavily on the data and methods used and meets various uncertainties( Jennifer et al., 2016 ; Alkama et al., 2020 ). Moreover, the changes in the atmospheric heat transport (AHT) ( Manabe and Wetherald, 1980 ;Graversen et al., 2008 ; Alexeev and Jackson, 2012 ) and ocean heat uptake (OHU) ( Holland and Bitz, 2003 ; Rose et al., 2014 ) can also affect the AA.

To explore the relative roles that these feedbacks have played in causing the AA, the surface and top-of-the-atmosphere (TOA) fluxes from climate models have been studied ( Crook et al., 2011 ; Bintanja and Linden, 2013 ). In this process, the relative temperature change caused by each individual feedback in a warming climate can be identified by calculating the “radiative kernel ”( Soden, 2007 ). The lapse rate feedback has been found to be the largest contributor to the AA in the 4×COexperiment in phase 5 of the Coupled Model Intercomparison Project (CMIP5) ( Pithan and Mauritsen, 2014 ; Stuecker et al., 2018 ;Zhang et al., 2018 ; Previdi et al., 2020 ; Zhang et al., 2020 ), followed by the Planck feedback and albedo feedback ( Goosse et al., 2018 ).

Despite the Antarctic amplification (ANA) being weaker than the AA in the 21st century, it has nevertheless also been identified ( Marc, 2017 ;Goosse et al., 2018 ; Stuecker et al., 2018 ; Smith et al., 2019 ) because the lapse rate feedback is stronger in the Arctic and the negative cloud and OHU feedbacks are larger in the Antarctic ( Goosse et al., 2018 ).

The aims of this study are to explore the contributions of these feedbacks to the AA and ANA under different warming scenarios simulated by recently released CMIP6 (phase 6 of the Coupled Model Intercomparison Project) outputs, and identify in detail the relative importance of each feedback to the change in polar amplification. In Section 2 , the model simulations and formulae used are introduced. In Section 3 , the polar amplifications and associated feedbacks are described. The differences between the warming in the Arctic and in the Antarctic under different global warming scenarios are presented in Section 3 , along with the underlying mechanisms leading to the change in the AA. Conclusions are drawn in Section 4 .

2. Data and methods

2.1. Model simulations

In this study, 15 CMIP6 (r1i1p1f1) climate models are used because only they can provide all the available variables needed here. Detailed information about these models is given in Table S1. To demonstrate the contributions of individual feedbacks to the polar amplifications under different strengths of global warming, monthly data of three scenarios —namely, SSP1-2.6, SSP2-4.5, and SSP5-8.5 —from 2015 to 2100,and the historical run from 1850 to 2014, are used (Table S2).

2.2. Methods

The “radiative kernel ”method proposed by Soden (2007) can make an accurate assessment of flux change as follows:

Radiative feedback can be explained as the warming contribution following the method proposed by Goosse et al. (2018) . For example,in the SSP5-8.5 experiment, the warming contribution from the water vapor feedback can be calculated as + 2.53 K in the Arctic, which means that water vapor helps to promote the Arctic warming by 2.53 K. Then,the change in temperature of individual feedback mechanisms can be quantified as follows:

where λrepresents each single feedback (here, the lapse rate, cloud, water vapor, and albedo feedback), ΔTrepresents the change in surface temperature, and the terms on the right-hand side of the equation represent the warming contributions from each individual feedback mechanism discussed above. F is the radiative forcing, which is calculated using

3. Results

3.1. Significant polar amplifications

The Arctic region (60°—90°N) and the Antarctic region (60°—90°S) all show significant amplification compared with the tropical region (30°S—30°N) (Fig. S1). In the Arctic, the annual average regional temperature increase can reach 7.05 K, which is more than twice the level in the tropical region, whose temperature change is only 3.3 K (Fig. S1). The polar amplification is also significant in both winter and summer, especially in winter, when the average temperature increase in the Arctic can reach 15.6 K.

3.2. Difference between the AA and ANA

The AA is mostly due to the lapse rate feedback ( Fig. 1 (a)). The calculation of contributions from each feedback is independent for each model, and in this way their ensemble mean is obtained. In the Arctic, a greater increase in SAT is needed because it is difficult for the near-surface cold and dense air to mix with the light air, and the radiation is the main coupling mechanism ( Florian, 2015 ). The next factor is the Planck feedback. Due to the warmer tropospheric temperature, the longwave radiation emitted into space will increase. According to the Stephen—Boltzmann theorem, a larger surface warming is needed to off-set the energy imbalance at a colder background (i.e., the Arctic). The surface albedo feedback is also an important factor to the AA associated with the loss of snow and sea ice with high albedo. The surface absorbs more heat for the loss of sea ice with high albedo in the Arctic, which is conducive to the warming. The least positive contribution is from the AHT. The polar heat transfer of the atmosphere is positively correlated with the warming, but its significance is relatively weak. The same feedback mechanism also plays a major positive role in the ANA ( Fig. 1 (b)).To sum up, the temperature feedback, the albedo feedback, and the AHT together act as the main contributors to the polar warming compared with the tropical warming, in which the lapse rate feedback is the dominant contributor. The result is similar to that in CMIP5 for the last 30 years of the 4 ×COexperiment compared with the last 30 years of the control run, and the difference lies in the relative importance of Planck and albedo feedback ( Pithan and Mauritsen, 2014 ; Goosse et al., 2018 ).In CMIP5, the Planck feedback contributes in a greater way to the AA,while the albedo feedback is more significant in the result observed in CMIP6.

Fig. 1 (c) shows the AA vs ANA under the three scenarios. It is obvious that the AA is larger than the ANA in almost all the CMIP6 models —the same as that in the observation ( Mark and Jennifer, 2006 ;Florian, 2015 ). The warming contributions of all the feedback mechanisms to annual polar warming are marked in Fig. 1 (d). The lapse rate,albedo, and Planck feedbacks all play leading roles in causing a larger AA than ANA in all 15 models under the three scenarios (details shown in Fig. S2).

Fig. 1. Different warming contributions of feedback mechanisms and climate forcing to the AA and ANA. (a) Responses for Arctic warming compared with tropical warming averaged in three scenarios (SSP1-2.6, SSP2-4.5, and SSP5-8.5) for the multi-model mean of all 15 studied CMIP6 models. (b) As in (a), except for Antarctic warming. (c) AA vs ANA in the three scenarios compared with the historical run for the 15 CMIP6 models. The black dot denotes the average of all three scenarios for the multi-model mean. The ratio of a larger AA than ANA is 93.33% of all models. (d) As in (a) but for different contributions of each feedback to the AA vs ANA.

3.3. Seasonal evolution of polar amplifications

The winter warming is greater than the summer warming in polar regions ( Fig. 2 (a)). Here, winter in the Arctic is the same as summer in the Antarctic (December—February) and summer in the Arctic is the same as winter in the Antarctic (June—August). In the Arctic ( Fig. 2 (b)),the effects of the OHU, lapse rate feedback, and COradiative forcing play a leading role in making greater winter warming —similar to the discussion in Pithan and Mauritsen (2014) from a TOA perspective in CMIP5 models, showing different relative importance between COforcing and AHT. The OHU, lapse rate feedback, cloud feedback, and COradiative forcing are dominant in increasing the Antarctic winter warming ( Fig. 2 (c)). The OHU is the most important forcing in the polar regions, mainly via the release of heat from the winter ocean ( Bintanja and Linden, 2013 ; Pithan and Mauritsen, 2014 ). The difference between the Arctic and Antarctic warmings in winter is much more obvious than that in summer. A larger Arctic warming in winter is seen in almost all the models under the three warming scenarios ( Fig. 2 (a)). The main feedback mechanism contributing to a larger AA is still the lapse rate feedback ( Fig. 2 (d)) during winter —the same as that in the annual warming( Fig. 1 (d)).

Fig. 2. Temperature change in winter vs summer and the warming contributions of different feedback mechanisms. (a) Winter warming vs summer warming under three scenarios (SSP1-2.6, SSP2-4.5, and SSP5-8.5) for 15 models compared with the historical run. The red dots denote the temperature change in the Arctic, while the dark blue ones are those in the Antarctic. (b) Different warming contributions to winter and summer warming in the Arctic averaged in the three scenarios for the multi-model mean of the 15 CMIP6 models. (c, d) As in (b), except for the warming contribution in the Antarctic and for the warming contributions to the AA and ANA during winter, respectively.

3.4. Changes of the AA and ANA with increasing warming

It is unclear, however, whether there is any change in the diversity of the AA and ANA with the increasing strength of global warming from SSP1-2.6 to SSP5-8.5. For each model and scenario, a two-dimensional value corresponding to the homologous polar amplification and tropical warming (average of the last 30 years of each scenario) can be obtained.Besides, the values of the same model under the three scenarios can be lined up, and 15 lines are obtained to denote how the polar amplification changes with increasing warming, as shown in Fig. 3 (a, b). In the Arctic ( Fig. 3 (a)), more than half of the models show the same variation trend as the multi-model mean, in which the AA decreases with the increasing tropical warming, while the other models show an opposite trend. In the Antarctic ( Fig. 3 (b)), the variation trend also decreases with the increasing tropical warming in most models. The multi-model-mean variation trends decrease with increasing warming in both the Arctic and Antarctic, though the magnitude is much smaller in the Antarctic.The larger the tropical surface warming is, the smaller the intermodel spread can be.

With the small multi-model changes in the ANA with increasing tropical warming taken into consideration, only the AA can be discussed.To determine the effect of each individual feedback mechanism on the AA, we first divide the warming contribution to polar amplification of each factor by the tropical warming under each of the three scenarios first. For the individual factors, we calculate the linear regression coefficients between these values and take the tropical warming under the corresponding scenario as a yardstick. Fig. 3 (c) shows the corresponding contributions of different factors to the AA. The sum of these values corresponds to the linear regression coefficients of the lines of corresponding models in Fig. 3 (a). In Fig. 3 (c), the changes in the contributions of the Planck, lapse rate, and albedo feedbacks are the most significant ones among all factors. The changes in the normalized warming contributions of the Planck, lapse rate, and albedo feedbacks in CMIP6 models with increasing tropical warming under the three scenarios are shown in detail in Fig. S3. The warming contributions of the three feedbacks decrease with increasing warming in almost all models (Fig. S3),leading to a decrease of the AA as well ( Fig. 3 (a)) because the three feedbacks act as the most important promoters of the AA in all models.It can also explain why the diversity of different models gets smaller with increasing warming, because the difference among models is also determined by the difference of the warming contributions of the three feedbacks; and with the increasing warming, the contributions of the three feedbacks decrease and approach each other. Thus, a preliminary conclusion can be drawn that the lapse rate, albedo, and Planck feedbacks dominate the change in the AA with increasing warming in CMIP6 models. In Fig. 3 (c), certain models exhibit opposite variation trends in the same feedback even when calculated with the same method. For example, in the albedo feedback, each model has a different simulation of the change in sea ice from SSP1-2.6 to SSP5-8.5 and can have different variation trends of the warming contribution. Therefore, it is clear that the intermodel spread stems from certain internal factors, which needs more thought and research in future work.

Fig. 3. Changes in the (a) AA and (b) ANA with increasing warming in CMIP6 models and (c) the variation trend in the contribution to the AA with increasing warming. The dashed lines in (a) and (b) represent the models in which the AA or ANA decreases with the tropical warming, while the solid lines are those in which the AA or ANA increases, and the black line is the multi-model mean. Positive values in (c) indicate that the contribution of the corresponding factor intensifies the AA, and vice versa.

4. Summary and discussion

In this work, after investigating the changes in polar amplifications in response to different feedback mechanisms (Planck, lapse rate, cloud,water vapor, albedo feedback, CO 2 radiative forcing, OHU, and AHT)under three warming scenarios in CMIP6 (SSP1-2.6, SSP2-4.5, and SSP5-8.5) compared with the historical run, we find that the main contributors to the AA and ANA are: (1) the lapse rate feedback, (2) Planck feedback, (3) albedo feedback, and (4) AHT ( Fig. 1 (a, b)). The AA is much stronger than the ANA in almost all the models, caused by the lapse rate, albedo, and Planck feedbacks ( Fig. 1 (d)). With the seasonal effect on polar amplification taken into consideration, winter warming and summer warming are compared, which shows that warming is stronger in winter than in summer, and mainly derives from the OHU due to the release of heat stored in the ocean, followed by the lapse rate feedback,in both the Arctic and Antarctic ( Fig. 2 (a—c)). During winter, the AA is also larger than the ANA, and the lapse rate feedback is the largest contributor ( Fig. 2 (d)) —the same as in Pithan and Mauritsen (2014) .

With the increasing warming, both the AA and ANA decrease in terms of the multi-model mean, with more than half of the CMIP6 models showing the same downward trend ( Fig. 3 (a, b)). Through attribution analysis, in the Arctic, the Planck, lapse rate, and albedo feedback are dominant in the change of the AA with increasing warming ( Fig. 3 (c)).

The results shed light on polar amplification and feedback mechanisms. In future work, observations and model simulations need to be combined and a more accurate investigation of each individual process’s effects on polar amplification needs to be carried out. Besides, the internal changes in the contributions of different feedback mechanisms to different models and the changes of the AA with increasing warming need further discussion.

Funding

This work was supported by the National Natural Science Foundation of China [grant number 41420104002 ] and by the Natural Science Foundation of Jiangsu Province [grant numbers BK20150907 and 14KJA170002 ].

Supplementary materials

Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.aosl.2021.100043 .