Robust Arctic warming caused by projected Antarctic sea ice loss

Over the coming century, both Arctic and Antarctic sea ice cover are projected to substantially decline. While many studies have documented the potential impacts of projected Arctic sea ice loss on the climate of the mid-latitudes and the tropics, little attention has been paid to the impacts of Antarctic sea ice loss. Here, using comprehensive climate model simulations, we show that the effects of end-of-the-century projected Antarctic sea ice loss extend much further than the tropics, and are able to produce considerable impacts on Arctic climate. Specifically, our model indicates that the Arctic surface will warm by 1 °C and Arctic sea ice extent will decline by 0.5 × 106 km2 in response to future Antarctic sea ice loss. Furthermore, with the aid of additional atmosphere-only simulations, we show that this pole-to-pole effect is mediated by the response of the tropical SSTs to Antarctic sea ice loss: these simulations reveal that Rossby waves originating in the tropical Pacific cause the Aleutian Low to deepen in the boreal winter, bringing warm air into the Arctic, and leading to sea ice loss in the Bering Sea. This pole-to-pole signal highlights the importance of understanding the climate impacts of the projected sea ice loss in the Antarctic, which could be as important as those associated with projected sea ice loss in the Arctic.


Introduction
The Arctic has lost over 40% of its summer sea ice extent over the past forty years (see, e.g. the NSIDC Sea Ice Index, Fetterer et al 2017). Meanwhile, Antarctic sea ice extent has fallen to record lows over the past four years after a 35-year period of small but significant sea ice growth (Parkinson 2019). More importantly, by the end of this century, climate models project that both Arctic and the Antarctic sea ice covers will shrink considerably (Collins et al 2006, Notz et al 2020, and a welter of studies have focused on determining if and how the projected sea ice loss at the poles could impact the climate system at lower latitudes (Shepherd 2016, Screen 2017, Screen et al 2018, Cohen et al 2020. Observational and modeling evidence has shown that sea ice loss causes a robust warming and moistening of the atmosphere at the high-latitudes, especially in the lower troposphere (Deser et al 2010, Screen and Simmonds 2010, Screen et al 2013, England et al 2018. Sea ice loss also has an important impact on the mid-latitude tropospheric jet, with Arctic sea ice loss causing an equatorward shift of the Northern Hemisphere mid-latitude jet (Peings andMagnusdottir 2014, Screen et al 2018), and Antarctic sea ice loss causing a weakening of the Southern Hemisphere mid-latitude jet (England et al 2018, Ayres and. In fact, several studies with coupled oceanatmosphere models have suggested that the response to sea ice loss can be global in nature , Screen et al 2018. Specifically, the effects of sea ice loss have been shown to extend to the tropics (Wang et al 2018, Kennel and Yulaeva 2020, with enhanced warming and precipitation in the equatorial regions, and even reaching deep into the opposite hemisphere (Deser et al 2015, Liu andFedorov 2018). A detailed examination of the pole-to-pole effects of projected sea ice loss, however, is still lacking.
And yet, from a paleoclimatic perspective the idea that the polar regions may be connected is not new. Evidence from ice cores from the last glacial and deglacial period indicates that past periods of warming in the northern high-latitudes coincided with periods of cooling in the southern highlatitudes, and vice versa (Blunier et al 1998, Blunier and Brook 2001, Barbante et al 2006, Pedro et al 2011: this phenomenon, whereby temperature at the poles are at opposite phases on millennial timescales, is known as the 'bipolar seesaw' hypothesis (Broecker 1998, Marino et al 2015, Pedro et al 2018. Most studies have pointed to the deep ocean circulation as the likely mediator of this anti-correlated behavior of the climate at the two poles (Crowley 1992, Stocker 1998, Stocker and Johnsen 2003, Knutti et al 2004. Recently it has been suggested that the 'bipolar seesaw' mechanism might operate on much shortermulti-decadal-timescales, and that this may be seen in the observed temperature record from the last century (Chylek et al 2010, Wang et al 2015. It seems, however, that this phenomenon is likely an artifact of the limited Antarctic data coverage (Schneider and Noone 2012). In any case, most of the literature on pole-to-pole linkages is focused on the two polar regions behaving asynchronously.
More recently, a handful of modeling studies have suggested that future Arctic sea ice loss can potentially impact the climate of Antarctica. Notably, Liu and Fedorov (2018) have reported that in the first fifteen years following a large, abrupt loss of Arctic sea ice, the southern high-latitudes cool and Antarctic sea ice cover expands (via an atmospheric connection), in a manner reminiscent of the 'bipolar seesaw'; however, unlike the initial transient phase, the equilibrium response to Arctic sea ice loss in the same model simulations features a clear warming of the southern high-latitudes. Such Antarctic warming is consistent with the results of Deser et al (2015), who show that projected Arctic sea ice loss leads to upper-tropospheric warming in the tropics and lower-tropospheric warming at both poles (termed a 'mini global warming' , due to its resemblance to the atmospheric warming pattern caused by increased green-house gases). The potential influence in the other direction, however, remains unexplored.
Hence the goal of our paper: to investigate climate change in the Arctic caused by projected endof-the-century sea ice loss in the Antarctic. Analyzing model integrations specifically designed for this purpose, we here demonstrate that Antarctic sea ice loss also causes a 'mini global warming' signal, with enhanced warming over the Arctic and a significant reduction in Arctic sea ice cover. To understand the underlying mechanism, we perform atmosphereonly runs, and show that the tropical SST anomalies caused by projected Antarctic sea ice loss drive a substantial portion of the enhanced Arctic warming, via a Rossby wave trains and a deeper Aleutian Low. We start by detailing the model we use and the simulations we analyze, then present the results, and conclude with a brief discussion.

Model
In this study we analyze climate model simulation performed with the Community Earth System Model (CESM1) Whole Atmosphere Coupled Chemistry Model (WACCM4). CESM1-WACCM4 is fully documented in Marsh et al (2013), to which the reader is referred to for details. The atmospheric component, WACCM4, has a horizontal resolution of 1.9 • latitude by 2.5 • longitude, with 66 vertical levels and a model top in the lower thermosphere. The representation of the stratospheric chemistry and dymamics in WACCM4 is much superior to the one in typical low-top models, owing to improved vertical resolution, gravity wave parameterisation for the upper atmosphere, and interactive stratospheric chemistry. This is important because previous studies have identified the stratosphere as a potential pathway for polar sea ice loss to influence the lower latitudes , De and Wu 2019. The atmospheric component is coupled to land, ocean and sea ice components, making CESM1-WACCM4 a CMIPclass fully-coupled climate system model.

Fully-coupled runs
To understand the climate response to projected Antarctic sea ice loss we analyze two simulations with perturbed sea ice cover, described detail in England et al (2020). Both are 350-year long, time-slice integrations of the fully-coupled CESM1-WACCM4 model, with all anthropogenic forcings fixed at year 1955 values. These include CO 2 , methane, nitrous oxide and, most importantly, ozone depleting substances (which may have contributed to the recent warming in the Arctic, as reported in Polvani et al 2020). The midtwentieth century was chosen as the control period so as to avoid the impacts of stratospheric ozone depletion; stratospheric ozone concentrations are severely perturbed at present (WMO 2018), but are expected to return to pre-1960 values in the second half of this century. We discard the first 100-years of these integrations, and focus on the average of the remaining 250-years.
The only difference between these two integrations is their Antarctic sea ice cover. In the 'control' run Antarctic sea ice conditions are nudged to match the mean of a sixmember ensemble CESM1-WACCM4 historical runs, averaged over the period 1955-69 (figure S1a(https://stacks.iop.org/ERL/15/104005/mmedia)). In the 'future' run Antarctic sea ice conditions are nudged to match the mean of a three-member ensemble of CESM1-WACCM4 RCP8.5 scenario simulations, averaged over the period 2085-2099 (figure S1b). In both cases, Antarctic sea ice conditions are constrained following the methodology of Deser et al (2015), which consists of adding an additional 'ghost flux' to the sea ice component of CESM1-WACCM4 so as to maintain the desired sea ice concentrations. This approach does not conserve energy but it does conserve the fresh water budget, and has been found to be more effective than the commonly-used albedo-reduction method . A detailed explanation can be found in England et al (2020).
The difference in sea ice concentrations between the control and future runs is shown in figure S1c, and corresponds to reduction in Antarctic sea ice extent of 6.6 × 10 6 km 2 . In the remainder of the paper, we will refer to the difference between these two runs, averaged over the last 250 years, as 'the response' to Antarctic sea ice loss.

Atmosphere-only runs with prescribed tropical SSTs
To investigate the role of tropical SST anomalies in driving Arctic warming, we carry out two additional model integrations. The first is a 251-year-long 'control' run with WACCM4 in atmosphere-only configuration, i.e. with sea ice and SSTs prescribed from the climatology (with a monthly-mean repeating seasonal cycle) of the six-member mean of the CESM1-WACCM4 historical runs, averaged over the period 1955-69, and with radiatively active gases fixed at year 1955 levels. The second run is nearly identical, except for the tropical SSTs, where the response to Antarctic sea ice loss is added onto the SSTs used in the control run. Specifically, the SST response to Antarctic sea ice loss-computed with the fully-coupled CESM1-WACCM4 as described above-is added to the control SST equatorward of 25 • , and linearly tapered so as to vanish poleward of 30 • , as shown in figure 1. By taking the difference between these two atmosphereonly runs, we can isolate the Arctic response to the tropical SST changes caused by Antarctic sea ice loss. We discard the first year of each simulation, and then take the average of the remaining 250 years.

Response of the Arctic to Antarctic sea ice loss
Let us start by examining the global impacts of Antarctic sea ice loss in the fully-coupled runs: the responses of sea ice and temperature-indicated by the letter ∆ -are shown in figure 2 (for context, we show the imposed annual mean Antarctic sea ice loss in the black box in panel 2a). First, we note that the response involves an overall surface warming across the planet (figure 2b), with the largest increase in the southern high-latitudes. The enhanced warming in the tropical Pacific was documented in England et al (2020). Here, we focus on the pole-to-pole impacts, notably the amplified surface warming in the Arctic. In our simulations, the Arctic polar cap (60-90 • N) surface warms by approximately 1 • C in response to Antarctic sea ice loss. This is a substantial effect, as it accounts for 10-15% of the projected endof-century Arctic warming of 7.5 • C under RCP8.5. Viewed another way, although this signal has traveled all the way from the southern high-latitudes to the northern high-latitudes, it is still 20% as large as the 5 • C Antarctic polar cap (60-90 • S) surface warming, and twice as large as the 0.
This results in an Arctic amplification factor, which we define here as the ratio of the Arctic (60-90 • N) warming to tropical (25 • S-25 • N) warming, of 2.2. We note that this is not statistically different from the Arctic amplification factor of 2.1 under projected changes under RCP8.5 for this model, as determined from the difference between the period 2085-2099 for the RCP8.5 simulations and the period 1955-69 for the historical transient simulations. We note that the warming under RCP8.5 would include the effects of projected Antarctic sea ice loss. This could suggest that this Arctic warming is part of the 'mini global warming' response, where local feedbacks are the dominant processes in Arctic amplification (Stuecker et al 2018). However, in section 3.2, we show that a sizable fraction of the Arctic warming response to projected Antarctic sea ice loss is actually driven remotely from the lower latitudes.
Zooming into the Arctic, one sees that the amplified atmospheric warming at low-levels in response to Antarctic sea ice loss (figure 3a), which extends up to the tropopause (figure 4a), is associated with a deepened Aleutian Low and high pressure over the central Arctic ( figure 3(b), Svendsen et al 2018). This is consistent with the atmosphere-only experiments of Tomas et al (2016). The low pressure response in the Pacific sector brings warmer air from the south into the Arctic and carries colder air into Northern Eurasia (Trenberth and Hurrell 1994). We note that a deepened Aleutian Low is also a robust feature of the modeled response to Arctic sea ice loss (Screen et al 2018). In addition to the warming and sea level pressure response, Antarctic sea ice loss also causes a reduction in Arctic sea ice cover, with an annual mean loss of 0.5 × 10 6 km 2 of Arctic sea ice extent, largely concentrated in the Bering Sea (figure 3c), and thinning of sea ice across the central Arctic ( figure 3d).
This Arctic response to Antarctic sea ice loss has an important seasonal dependence. Since the Aleutian Low occurs primarily in boreal winter (Trenberth and Hurrell 1994, Bograd et al 2002, Gan et al 2017, the deepening of this low pressure circulation is found to be strongest in that season ( figure S3(b)). By contrast, in boreal summer the North Pacific high extends further westward, limiting the extent of the Aleutian Low (Bograd et al 2002); the summertime mean sea level pressure response involves high pressures across much of the northern high-latitudes with a swath of  low pressure further south across the North Pacific ( figure S4(b)). Thus the warming response is largest in wintertime and weakest in summertime (compare figure S3(a) and figure S4(a)).

Connecting Antarctic sea ice loss to the Arctic
Having shown that Antarctic sea ice loss can have important impacts on Arctic climate, we now ask: how does the signal reach all the way to the other pole? We propose that the tropics play a key role in enabling these substantial pole-to-pole effects. As documented in England et al (2020), Antarctic sea ice loss in these model simulations causes enhanced surface warming and increased precipitation in the Equatorial Pacific, as well as a warming of the tropical upper troposphere (see figure 4a): ocean dynamics was shown to be key for connecting the loss of Antarctic sea ice to the tropics. Now, we suggest that the tropical response signal is quickly propagated into the Arctic by atmospheric teleconnections. Our suggested pathway is in line with the modeling study of Tomas et al (2016), which showed that many of the impacts of Arctic sea ice loss on the northern mid-and highlatitudes are first mediated through the tropical SST response to sea ice loss. It is also consistent with the We investigate this proposed mechanism by performing and analyzing two additional, atmosphereonly model simulations, to isolate the Arctic impacts of the tropical SST response to Antarctic sea ice loss. These runs are detailed in section 2.3. In essence, one is a control simulation, the other is forced by the SSTs in the control simulation plus the tropical SSTs anomalies resulting from Antarctic sea ice loss in the fullycoupled model simulations (see figure 1). The difference between these two runs illustrates the impact of such SST anomalies onto the Arctic, as communicated by the atmosphere alone.
These prescribed-SST runs reveal five important points. (i) As expected, the tropical upper tropospheric warming response to Antarctic sea ice loss (figure 4a) is driven from below by the tropical SST anomalies (figure 4b). (ii) The tropical SST anomalies cause amplified warming throughout the lower troposphere in the Arctic (figure 4b), albeit somewhat smaller than in the fully-coupled Antarctic sea ice loss runs (figure 4a). This is clearly seen in the warming response at 850 hPa (compare figure  3a and figure 3e). (iii) The enhanced Arctic warming in response to tropical SST anomalies, as in the fullycoupled runs, is related to a deepening of the Aleutian Low (compare figure 3b and figure 3f). This suggests that the tropical SSTs anomalies, via a Rossby wave train, are also responsible for the deepened Aleutian Low-and accompanying Arctic warming-in response to Antarctic sea ice loss in the fully-coupled runs. This is in agreement with the findings of Svendsen et al (2018), who identify the same mechanism as contributing to recent Arctic warming in the observed record, as well as the modeling study of Screen and Deser (2019). (iv) We are, of course, unable to diagnose the response of Arctic sea ice cover to the tropical SSTs anomalies because the surface conditions are prescribed in the atmosphere-only runs; however, both the near surface circulation response and the near surface warming in these runs are consistent with a loss of Arctic sea ice in the Pacific sector. Finally, (v) the wintertime response to prescribed tropical SSTs anomalies well captures the response to Antarctic sea ice loss in the fully-coupled simulations (compare the top and bottom rows in figure  S3), whereas that response is much weaker, and less similar, in the summertime (compare top and bottom rows in figure S4). Also, note that in that season the sea ice loss occurs most prominently over the central Arctic (figure S4(c)), rather than over the Bering Sea ( figure S3(c)). A different mechanism from the one examined here, in which the ocean and ice feedbacks are likely involved and persist throughout the year, may be needed to fully explain the summertime response.
We confirm that, in the coupled simulations, a Rossby wave train initiates in the tropical Pacific and connects to the North Pacific, by showing the eddy geopotential height response at 200 hPa and the associated wave activity flux ( figure 5(a)). It is clear that this wave train is driven by Antarctic sea ice induced changes in tropical SSTs because the same mechanism occurs in the atmosphere-only simulations in response to prescribed tropical SSTs ( figure 5(b)). This wave train mechanism is largely consistent with the one reported in the modelling and observational studies of Wettstein andDeser 2014, Tokinaga et al (2017), Svendsen et al (2018), Screen and Deser (2019), but opposite to the tropical-polar teleconnections reported in Ding et al (2014), Baxter et al (2019), andDing et al (2019). This discrepancy is likely explained by the differing spatial patterns of anomalous SSTs imposed in these studies, especially in the West Pacific. In addition, the fact that Baxter et al (2019) and Ding et al (2019) focus on the relationship between tropical SSTs and Arctic conditions in summer rather than the winter could play a role; however it is also possible that climate models have limitations in their representation of tropical-polar linkages (Topal et al 2020).
We conclude, therefore, that fast atmospheric teleconnections from anomalous tropical SSTs offer a plausible pathway allowing the signal caused by Antarctic sea ice loss to reach into the Arctic, with the amplitude of the Arctic response largest in the boreal winter. This suggests that once the tropics begin to respond to Antarctic sea ice loss , which could take multiple decades owing to the long timescale of the ocean response (Wang et al 2018), the effects on the Arctic would then appear relatively quickly (on a timescale of years, rather than decades). To be clear, the ocean plays a pivotal role in this process, because there is no Arctic response to Antarctic sea ice loss in atmosphere-only model runs, as shown in England et al (2018) (which use exactly the same model as the one employed here).

Summary and Discussion
In this study, we have demonstrated the existence of a substantial Arctic impact from projected twenty-first century Antarctic sea ice loss. In our fully-coupled climate model runs, in response to imposed Antarctic sea ice loss, the Aleutian Low deepens causing approximately 1 • C warming in Arctic near-surface air temperature ( 0.7 • C at 850 hPa), with a larger warming over the Bering Sea, East Siberia Sea, Chukchi Sea, and Alaska regions than in the Atlantic sector of the Arctic Ocean. The loss of Antarctic sea ice also leads to an annual mean loss of 0.5 × 10 6 km 2 of sea ice extent in the Arctic, primarily in the Bering Sea. With the aid of additional atmosphere-only model runs, we have shown that a fast atmospheric response to the Antarctic-sea ice-loss-induced tropical SST anomalies is responsible for at least half of this pole-to-pole signal.
We acknowledge that the pole-to-pole effects documented here are relatively small compared to the internal variability of the climate system in the highlatitudes. However, the polar cap warming and loss of Arctic sea ice in our model are statistically significant at a 95% confidence level for every month of the year, not just in the annual mean. Furthermore, the sheer fact that as much as 10-15% of the end-of the century Arctic warming projected under RCP8.5 could be induced from climate change at the opposite pole offers a striking example of the huge geographical extent of the couplings at play among various components in the Earth's climate system.
We also acknowledge that the magnitude of the Arctic warming in our atmosphere-only runs with prescribed tropical SST anomalies is, approximately, only half as large as the one in the fully-coupled runs (compare figure 4a and 4e). It is important to appreciate, however, that our aim was not to fully replicate the exact Arctic response from the fully-coupled simulations (which, in fact, may no be feasible with an atmosphere-only model). Instead, our goal has been to demonstrate a plausible pathway which could explain the pole-to-pole connection. In fact, since prescribing SSTs and sea ice cover does not allow them to freely evolve with the atmospheric conditions, the Arctic warming response is likely underestimated. For example, one would expect Arctic warming to be amplified if sea ice cover is allowed to change, via the sea ice albedo feedback. There may also be other pathways through which Antarctic sea ice loss could influence the Arctic, the main candidates being atmosphere-ocean coupling and ocean circulation changes which could alter the heat transport into the northern high-latitudes. However, the results presented above suggest that tropics-to-pole mechanism we have proposed is likely a dominant one in facilitating the pole-to-pole response, especially in the boreal winter.
Taken together, previous studies (e.g. Ding et al 2014, Dong et al 2019, McCrystall et al 2020 suggest that the Arctic response to tropical warming is sensitive to the exact tropical forcing pattern and is likely model-dependent. This is an important caveat for our results, which are only based on one climate model. Consistent with our study, however, most mechanisms that have been proposed to explain a connection between the tropics and the Arctic have been based on tropospheric Rossby waves initiating in the tropical Pacific (Yuan et al 2018). In our fully-coupled model simulations, we find warming throughout the tropics, but the strongest warming is located in the Central and Eastern Equatorial Pacific (figure 1). However, results from Dong et al (2019), in agreement with earlier studies (Yoo et al 2012, Ding et al 2019, suggest that the Arctic is responding primarily to the warming in the Western Equatorial Pacific, the region of tropical ascent. Dong et al (2019) show that in abrupt 4 × CO 2 experiments, despite the Eastern Pacific warming more, it is the warming in the Western Pacific which is responsible for the temperature increase over the Arctic. Additional experiments with our atmosphere-only model could be carried out to test the relative importance of the Eastern vs Western Tropical Pacific for Arctic climate warming. However, such work is beyond the scope of this short letter, whose primary goal is to highlight the pole-to-pole impact of future Antarctic sea ice loss.

Acknowledgment
We appreciate the constructive feedback from two anonymous reviewers which have led to considerable improvements in the manuscript. The work of MRE is funded by grants OPP-1643445 and OPP-1744835 from the US National Science Foundation. The work of LMP is funded, in part, by a grant from the US National Science Foundation to Columbia University.

Data Availability
The data that support the findings of this study are available upon reasonable request from the authors.