A surface temperature dipole pattern between Eurasia and North America triggered by the Barents–Kara sea-ice retreat in boreal winter

The Arctic has experienced dramatic climate changes, characterized by rapid surface warming and sea-ice loss over the past four decades, with broad implications for climate variability over remote regions. Some studies report that Arctic warming may simultaneously induce a widespread cooling over Eurasia and frequent cold events over North America, especially during boreal winter. In contrast, other studies suggest a seesaw pattern of extreme temperature events with cold weather over East Asia accompanied by warm weather in North America on sub-seasonal time scales. It is unclear whether a systematic linkage in surface air temperature (SAT) exists between the two continents, let alone their interaction with Arctic sea ice. Here, we reveal a dipole pattern of SAT in boreal winter featuring a cooling (warming) in the Eurasian continent accompanied by a warming (cooling) in the North American continent, which is induced by an anomalous Barents–Kara sea-ice decline (increase). The dipole operates on interannual and multidecadal time scales. We find that an anomalous sea-ice loss over the Barents–Kara Seas triggers a wavenumber one atmospheric circulation pattern over the high-latitude Northern Hemisphere, with an anomalous high-pressure center over Siberia and an anomalous low-pressure center over high-latitude North America. The circulation adjustment generates the dipole temperature pattern through thermal advection. Our finding has important implications for Northern Hemisphere climate variability, extreme weather events, and their prediction and projection.


Introduction
Since the 1970s, climate change over the Northern Hemisphere high latitudes has intensified compared to those of the global mean, as a result of a process known as Arctic amplification [1,2]. The most significant changes occurred in boreal autumn and winter [3], characterized by a rapid surface air temperature (SAT) increase, more than twice as fast as the global warming rate [2], accompanied by a continuous sea-ice retreat [4][5][6][7]. These rapid Arctic changes have been associated with a series of physical processes [8,9], including the local anthropogenic greenhouse gas forcing [10,11], the lapse rate and Planck feedbacks [12,13] in the atmosphere, as well as the cloud feedbacks [14,15], the ice albedo-feedback [16], and tropical-polar teleconnections [17]. In particular, the ocean-atmosphere heat exchange induced by the recent Arctic sea-ice loss [18] and the heat and moisture transport between the Arctic and lower latitudes [19,20] play a crucial role in heating the North Pole region, especially during boreal winter when the shortwave radiative forcing is absent over the Arctic region.
Arctic Amplification has a broad implication on climate changes and variability over the mid-and high-latitude Northern Hemisphere, especially over the continental areas [21][22][23]. The warm Arctic, associated with the sea ice melt, and the rapid increase of Eurasian snow cover [24][25][26] in boreal autumn and winter, are often followed by a decrease in SAT over the mid-latitude Eurasian continent [27,28], and sometimes over the mid-latitude North American continent [29] in boreal winter. This pattern has been termed the 'warm Arctic-cold continent' pattern which includes cooling in eastern North America [24,26]. Hypotheses proposed to explain mechanisms associated with this pattern include: stratospheric-tropospheric coupling induced by an enhanced upward propagation of planetary-scale waves with wavenumbers of one and two [30,31], a weakening of the Arctic Oscillation [32], a persistent shift of Arctic polar vortex towards the Eurasian continent [33], and more frequent Eurasian blocking events associated with an intensified Siberian High [34,35]. These mechanisms are in part triggered by the Arctic sea-ice retreat, especially over the Barents-Kara Seas (BKS) region.
Several recent studies questioned the significance level of the impacts of the Arctic sea-ice loss on the Northern Hemisphere mid-latitude climate in boreal winter [7,[36][37][38]. Large discrepancies exist between the observations and numerical simulations, as well as among the experiments results using different numerical models [7]. Some modeling studies based on large ensembles [39][40][41][42][43] indicated that the observed cold anomaly over the Eurasian continent may be more attributed to the atmospheric internal variability and the forcing from the tropics, rather than driven by the Arctic sea-ice loss. Other studies indicated that the atmospheric response to the Arctic sea-ice loss may be underestimated in climate models [44,45] in comparison to that in the observations. In addition, the linkage between the Arctic and the mid-latitudes is weakening according to a recent study [46].
The sea-ice loss and the circulation changes over different sectors of the Arctic Ocean may have distinct impacts on mid-latitude North America [47]. The surface warming and the sea-ice loss over the Chukchi-Bering Seas usually drive an anomalous cooling over North America [29]. In addition, the weakening of stratospheric polar vortex [48][49][50] associated with the Arctic sea-ice loss can cause the increase of the extreme cold events over North America [51,52], other studies [53,54] revealed a seesaw pattern of extreme temperature events on subseasonal time scales, with a cold event over East Asia accompanied by a warm event in North America, usually lasting for several weeks [54]. Despite the weak cooling temperature trend in February, there has been a warming trend in wintertime monthly-mean SAT over North America during the past decades [55,56], which contravenes the hypothesis that a warm Arctic is associated with a cold North America.
This study reveals a 'cold Eurasia-warm North America' SAT dipole pattern between the midlatitude Eurasian continent and much of North America in boreal winter (December-January-February, DJF), active on interannual and multidecadal time scales triggered by BKS sea-ice variability. In particular, the BKS sea-ice loss, while cooling the Eurasian continent, instead heats the North American continent through atmospheric circulation adjustment and its associated thermal advection.

Data analysis
The UK Met Office Hadley Centre's sea surface temperature (SST) and sea-ice datasets [57] have been used in this study to estimate variability and trends of the Arctic sea ice concentration (SIC) and SST. Sea level pressure (SLP) and SAT from the three reanalysis datasets are also used. To estimate the continental dipole pattern and its relationship with the anomalous Arctic sea ice and atmospheric circulation over Northern Hemisphere in boreal winter (DJF), we use (a) the Modern Era Retrospective-Analysis for Research and Applications, version 2 (MERRA2) [58]; (b) the European Centre for Medium-Range Weather Forecasts Reanalysis version 5 (ERA5) [59]; and (c) the Japanese 55 year Reanalysis (JRA55) [60]. To estimate the multidecadal trend of the SAT, we use a two-meter air temperature (T2m) of 10 252 landbased weather stations from the National Center for Environmental Information, National Oceanic and Atmospheric Administration [61].

Statistical methods
Sen's slope method [62] is used to calculate the observed trends in SAT and the Arctic SIC, with the confidence intervals estimated using the Mann-Kendall test [62].
We use the linear regression coefficients to evaluate the relationship between the temperature, the atmospheric circulation, and the sea ice. We use the Student's t-test to calculate the confidence intervals of these coefficients.
We use empirical orthogonal function (EOF) [63] to obtain the leading modes of SAT over the Northern Hemispheric mid-latitude continents, focusing on boreal winter (DJF). The EOF decomposition is applied based on both trend-retained and detrended reanalysis datasets. The results of these two types of decomposition are identical in this study.
A maximized covariance analysis (MCA) method [64] is used to retrieve the most important coherent modes between Arctic sea ice and SAT over Northern Hemispheric continents to determine the linkage between these two variables. The leading modes of MCA maximize the covariance between two highdimensional datasets through a singular value decomposition of the covariance matrix between these two variables. This method gives the most important modes that dominate the variability of each time series, and shows a strong correlation with each other.
We use a dynamical adjustment (DA) method [65] to isolate anomalous atmospheric circulation modes associated with the evolution of the areaaveraged SAT over mid-latitude Eurasia and North America. This method, based on partial least-squares regression and spatial pattern correlation, separates the SAT component associated with atmospheric circulation. Based on the above two procedures, SLP modes that significantly contribute to the area-averaged SAT time series are obtained. Crossvalidation with a bootstrap method is used to evaluate whether the contribution of these modes is statistically significant.

Model simulation
The NCAR climate model, the Community Atmosphere Model version 5 (CAM5), is used to investigate the teleconnection between the Arctic sea-ice and the continental SAT and SLP. We employ the finitevolume dynamical core with a global horizontal resolution of about 2 • (F19). The Community Land Model and the Community Sea-Ice Model thermodynamic module provide the heat and moisture fluxes on the lower boundary. A transient experiment with CAM5 forced by observed time-varying (with trend) SST and sea ice (i.e. AMIP-like) is performed for the 1979-2019 period, with 20 ensemble members accompanied by different perturbed initial conditions. The experiment is forced by the SST and sea ice variability over the BKS region (0 • -80 • E, 65 • N-80 • N). The SST and SIC over other areas are set to the climatological mean states (1981-2010) with the annual cycle, while other forcings, including the concentration of greenhouse gases, the aerosols, and the solar radiation, are all set to fixed values during the entire integration period. The linear trend of the ensemble mean state from 1980 to 2019 is calculated, which represents the impact of the SST and sea ice over the BKS region on the changes of the Northern Hemispheric SAT and SLP.  c) and (e)) shows a coherent warming pattern over almost the entire Eurasian and North American continents, except the areas around Hudson Bay and Greenland. This mid-latitude warming pattern is tightly associated with the anthropogenic global warming trend. Mode two (see supplementary figures 1(b), (d) and (f)) shows a dipole pattern between Eurasia and North America, with anomalous cooling over the Eurasian continent and a broad warming pattern over North America. The Eurasian cooling-North American warming pattern should be reversed when the principle component (PC) is negative. Mode one and two explain about 24% and 15% of the total variability, respectively. We also perform an EOF analysis with a linear trend removed before the decomposition, and the resulting patterns are identical (see supplementary figure 2).

Dipole pattern on interannual time scales
To further investigate the potential linkage between the Arctic sea ice and the land-area SAT over the mid-latitude Northern Hemisphere, we conduct a Maximum Covariance Analysis (MCA) (see section 2.2) between these two variables in DJF. The spatial pattern of the SIC ( figure 1(a)) of the leading MCA mode is characterized by sea-ice retreat over the BKS and the Greenland Sea. The spatial pattern of the SAT ( figure 1(b)) resembles the second EOF mode of the SAT (see supplementary figure  1(b)), with a broad cooling pattern over mid-latitude Eurasia and continental-wide warming over entire North America, despite a cooling signal around the west coast of the United States. This SAT dipole pattern between the Eurasian and the North American continents may be reversed when the sea ice increases over the BKS. To confirm the relationship revealed by the decomposition, we also perform an MCA using detrended SIC and SAT (see supplementary figure 4). The second mode (see supplementary figures 4(c) and (d)) resembles the linkage between the BKS sea-ice retreat and the continental SAT dipole pattern. Using other reanalysis datasets produces nearly identical MCA results (see supplementary figure 6).
The EOF decomposition reveals a SAT dipole pattern between the Eurasian and North American continents, while the MCA links it to the variability of the Arctic SIC, especially that over the BKS and Greenland Sea. We then regress the SAT ( figure 2(a)) onto the area-averaged BKS (20 • E-80 • E, 70 • N-80 • N) SIC time series (black curve in supplementary figure 3). The results are similar to the dipole-like SAT pattern ( figure 1(b)) in the first MCA mode, confirming that this SAT dipole pattern is related to the BKS sea-ice variability.
Considering that the adjustment of the atmospheric circulation may play an important role in mediating these kinds of remote effects [66][67][68], we further regress the Northern Hemisphere SLP and the 500 hPa geopotential height (figure 2(b)) onto the area-weighted mean BKS SIC time series (black curve in supplementary figure 3). The result shows a wavenumber one pattern over the high-latitude Northern Hemisphere, with a high-pressure center over north Siberia, and a low-pressure center over the high-latitude North America. Further analysis (see supplementary figure 9) indicates that the largescale circulation anomalies related to the BKS sea-ice retreat may intensify the cold advection over central Eurasia, meanwhile driving an anomalous warm advection to mid-latitude North America, contributing to the SAT dipole pattern between the two continents in boreal winter.
With a combination of statistical analyses, including EOF, MCA, and linear regression, we reveal that the Arctic sea-ice variability, especially over the BKS, may contribute to this SAT dipole pattern through modulating the atmospheric circulation over the mid-and high-latitude Northern Hemisphere. This phenomenon operates on interannual time scales. Below, we show that such a diploe pattern operates on multidecadal time scales.

Dipole pattern in multidecadal trend
We calculate the DJF SIC and SAT trends from 1980 to 2019. As revealed by many previous studies [7,[69][70][71], the wintertime Arctic sea ice experienced a rapid retreat over the BKS, and part of the Nordic-Greenland Seas (figure 1(c)), mainly due to the warm surface water intrusion associated with the shift of the Gulf stream extension [72,73]. Central Eurasia experienced a cooling trend [28,45,74,75]. In contrast, North America experienced a broad warming trend in boreal winter [55,56,76] (figure 1(d), based on the MERRA2 reanalysis), despite a spot of mild cooling over the Rocky Mountain area. We also calculate the mid-latitude Northern Hemisphere SAT trend using other reanalysis datasets (ERA5 and JRA55) and in-situ observations (see supplementary figure 8). Results show a similar dipole pattern, with a cooling signal over central Eurasia and a broad warming pattern over mid-latitude North America. However, the intensity of the temperature trends among these datasets varies.
To evaluate the potential trigger of these opposite SAT trends between the two continents, we perform a DA analysis (see section 2.2) to the area-weighted mean DJF SAT time series over the central Eurasian continent (40 • N-65 • N, 50 • E-110 • E, blue box in figure 1(d)) and the North American continent (40 • N-65 • N, 60 • W-130 • W, red box in figure 1(d)), respectively. The DA technique isolates the effects of dynamical processes on the SAT trends. The dynamical component is determined through the identification of atmospheric circulation patterns (e.g. anomalous high-/low-SLP centers) that impact the SAT through thermal advection [65]. Figures 3(a) and (b) show the anomalous SLP patterns, which significantly contribute to the recent observed SAT trends over central Eurasia ( figure 3(a)) and North America ( figure 3(b)), respectively through the DA. Our results indicate that both the negative SAT trend over central Eurasia and the positive SAT trend over North America can be at least partially attributed to a similar atmospheric circulation pattern (figures 3(a) and (b)). This pattern is characterized by a strong high-pressure center over north Siberia, and a relatively weaker low-pressure center over northern Canada and Hudson Bay. It constitutes a wavenumber one pattern over the northern high latitudes along with a high-pressure center over the North Pacific. The thermal advection associated with the high-and low-pressure centers helps advect cold and warm air to central Eurasia and North America, respectively, forming a SAT dipole pattern between these two continents in boreal winter.
According to the DA, the dipole-like SAT trend between the Eurasian and North American continents may be associated with a wavenumber one circulation pattern over the high-latitude Northern Hemisphere ( figure 3). This pattern resembles the anomalous circulation pattern related to the BKS sea-ice retreat ( figure 2(b)). This similarity implies that the BKS seaice loss may impact the Northern Hemispheric landarea SAT trend through an adjustment of the atmospheric circulation. Additional evidence is needed to validate the causality and to clarify the mechanisms of these linkages, which we provide below with numerical model experiment.

Causality and mechanism
We conduct an ensemble simulation experiment using the NCAR climate model, the Community Atmosphere Model version 5 (CAM5), driven by the observed evolution of the SST and SIC over the BKS region from 1980 to 2019 (see section 2.3). The simulation results provide the atmospheric temperature and circulation responses to the forcings from the Arctic ( figure 4). The simulated SAT trend shows a similar dipole-like pattern, with cooling signals over central Eurasia and a broad, significant warming pattern over North America ( figure 4(a)). However, the cooling signal is not strong in all regions. The highlatitude Eurasian continent is dominated by a strong warming, which agrees well with our MCA results. The impact of the Arctic sea ice on the Eurasian cooling trend has been intensely investigated [27][28][29]. The robustness of this effect has been questioned [39-41, 45, 77, 78], as debate surrounding the uncertainties of the simulation results continues. Previous studies [28,79,80] indicated that simulating the impact of Arctic sea ice on the Eurasia cooling may need a large sample size due to the low signal-to-noise ratio. Remarkably, the relationship between the BKS sea-ice loss and the heating over North America, as revealed in this study, is significant in both the statistical analysis (figure 2(a)) and numerical simulation (figure 4), presenting a robust linkage between these processes.
The DA and linear regression analysis imply that the linkage between the Arctic sea-ice retreat and the SAT dipole pattern is likely mediated by atmospheric dynamics, mainly through a zonal wavenumber one mode over the northern high latitudes. A high-pressure center characterizes the circulation pattern over Siberia and a low-pressure center over North America, along with a high-pressure center over the mid-latitude North Pacific (figures 2(b) and 3), despite a slight deviation of the intensity among different high-and low-pressure centers. We further analyze the simulation results ( figure 4(b)) to elucidate the circulation response to the BKS sea-ice retreat. The SLP response to the BKS SST/SIC ( figure 4(b)) forcings reproduce a high-pressure center over Siberia and a low-pressure center over North America. However, the former is stronger in the simulation results than in the statistical analyses (figures 3 and 4). It extends to the east coast of the Eurasian continent and merges with a high-pressure center over the North Pacific. The low-pressure center in the simulation results is relatively weaker than that in the statistical analyses. . Schematic diagram of the mechanism for the Cold Eurasia-Warm North America pattern triggered by the sea-ice retreat over the Barents-Kara Seas in boreal winter. The sea-ice retreat over the Barents-Kara Seas heats the surface atmosphere and triggers an anomalous high-pressure center over northern Siberia, propagating to the east through stationary Rossby wave dynamics (with wavenumber one), forming an anomalous low-pressure center over high-latitude North America. The anomalous high-pressure center drives an anomalous cold atmospheric advection from the Arctic to central Eurasia. In contrast, the anomalous low-pressure center drives an anomalous warm advection to the central-east North American continent.
These circulation patterns in both statistical analyses and the numerical experiment are significant almost everywhere, which reemphasizes the role of the circulation patterns in mediating the Arctic sea-ice variability and the continental SAT dipole pattern. Recent studies indicated that the BKS sea-ice retreat might shift the stratospheric polar vortex to the Eurasian continents [33]. The BKS sea-ice retreat may also intensify the Siberian high or Ural blocking by heating the atmosphere, expanding the geopotential heights, and weakening the westerly wind [28,35,81]. This signal may propagate to the east through a stationary Rossby wave train, contributing to the formation of the low-pressure center over North America.
To examine the Rossby wave dynamics, we estimate the horizontal component of Plumb flux [82] at 500 hPa (vectors in figure 2(b)), which represents the direction and the intensity of the propagation of a Rossby wave train. Results show a clear wave flux from the Siberian High to the anomalous low-pressure center over North America, forming a wave number one pattern of the mid-to high-latitude Northern Hemisphere ( figure 2(b)), agreeing well with our simulation results ( figure 4(b)).

Conclusion and discussion
We reveal a continental dipole pattern of the SAT in boreal winter, with opposite temperature signals over central Eurasia and North America, operating in both the interannual variability and multidecadal trend of wintertime land-area SAT over the mid-latitude Northern Hemisphere. This pattern is seen in the EOF decomposition of both the trend-retained and detrended land-area SAT time series over the midlatitude Northern Hemisphere. This dipole pattern in MCA and linear regression is linked to an Arctic seaice retreat, especially over the BKS region. We show that the BKS sea-ice retreat drives a wavenumber one atmospheric circulation pattern over the highlatitude Northern Hemisphere (figure 5). The circulation anomalies further advect cold and warm air to the Eurasian and North American continents (see supplementary figure 9) respectively, forming the SAT dipole pattern in boreal winter.
Our results indicate that the BKS sea-ice loss may contribute to the cooling over Eurasia during winter, agreeing well with previous studies [27,28,74]. On the other hand, we clarify that the BKS sea-ice loss also intensifies the warming trend over North America, although the impact of Arctic sea ice in different regions is different [47]. Recent studies show robust evidence [50,52,83] that the Arctic sea-ice variability may drive extreme cold events over both Asia and North America through stratospheric polar vortex disruption. Additional investigation is needed to determine to what extent Arctic sea-ice retreatinduced mean warming over North America is offset by the more frequent cold extreme events.
Eurasia and North America are two heavily populated continents. The dipole pattern we find may also influence the precipitation and air pollution with broad societal implications, including economic and public health consequences in these regions. The pattern potentially contributes to climate predictability on interannual and multidecadal time scales.

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