Impact of internal variability on recent opposite trends in wintertime temperature over the Barents–Kara Seas and central Eurasia

The large ensembles of the IPSL-CM6A-LR model output for the historical forcing experiment were employed to investigate the role of internal variability in the formation of the recent “warm Arctic–cold Eurasia” trend pattern in winter surface air temperature (SAT). The ensemble-mean SAT shows a positive trend over Arctic during 1990–2014, indicating a positive contribution of anthropogenic forcing to the warming Arctic. Over the region of central Eurasia, the ensemble-mean SAT trend is opposite to the observed trend. The winter SAT trends display remarkable inter-member diversity over the Barents–Kara Seas (BKS) region and central Eurasia, suggesting an important role played by internal variability. In addition to anthropogenic forcing, the results suggest that the barotropic anticyclone over northern Eurasia arising from internal variability can also contribute positively to the warming anomalies over the BKS region. On the other hand, through a fingerprint pattern matching method, it is found that the observed cooling trend over central Eurasia tends to be predominantly due to the internal variability. Finally, the results estimate that the internal variability can contribute to about 50–60% of the observed warming trend over the BKS region.


Introduction
Since the 1990s, rapid surface warming has occurred over the Arctic region (Polyakov et al. 2002;Johannessen et al. 2004;Serreze et al. 2008;Screen and Simmonds 2010a, b;Serreze and Barry 2011). During boreal winter, the strongest surface warming signal can be observed over the Barents-Kara Seas (BKS) region (Kug et al. 2015;Park et al. 2015;Wang et al. 2020a). Over the same period, cooling trends have been observed in mid-latitudes, especially over central Eurasia (Wu et al. 2011;Cohen et al. 2012aCohen et al. , 2012bLiu et al. 2012). The pattern of this Northern Hemisphere temperature signal has been referred to as the "warm Arctic-cold continents" pattern (Overland et al. 2010;Cohen et al. 2013;Cohen et al. 2014).
The occurrence of the "warm Arctic-cold continents" trend pattern is accompanied by recovery of the Siberian high (Jeong et al. 2011;Wu et al. 2011) and enhancement of stationary Rossby wave activity over the northern part of Eurasia (Wang et al. 2020b). It is well recognized that the decrease in winter surface air temperature (SAT) over mid-latitudes has arisen from anomalous atmospheric circulation over Eurasian land (Wang and Chen 2013;Mori et al. 2014;Sun et al. 2016). However, the causality between the anomalous atmospheric circulation and the Arctic warming is disputed. Some studies have reported that the colder continental temperatures and associated change in atmospheric circulation may be caused by Arctic sea-ice loss and Arctic warming (Tang et al. 2013;Mori et al. 2014;Wang et al. 2020c). For example, Liu et al. (2012) reported that Arctic sea-ice decline can excite negative Arctic Oscillation (AO)-like structural circulation anomalies, which in turn result in more frequent episodes of blocking patterns and cold surges over large parts of northern continents. But, other studies indicate that the observed co-variability between the Arctic sea ice and the winter extratropical atmospheric 1 3 circulation does not necessarily imply causality (Wang et al. 2021;Warner et al. 2020). In addition, several modeling studies have also reported that sea-ice reduction over the BKS region may decrease winter SAT over central Eurasia (Honda et al. 2009;Mori et al. 2014). However, the trends towards cooling in the midlatitudes have reversed recently (Blackport and Screen 2020), which suggests that the cooling in the midlatitudes and associated atmospheric circulation may arise mainly from internal variability (Blackport and Screen 2020;Sun et al. 2016). Furthermore, the warmer Arctic may be a consequence of the increased waviness in the atmospheric flow (Trenberth et al. 2014). For example, some studies have indicated that recent recovery of the Siberian high can contribute positively to the warming trend over the BKS region (Feng and Wu 2015;Wang et al. 2020a). This implies that the internal atmospheric variability can also drive Arctic warming in addition to anthropogenic forcing (Serreze and Barry 2011;Stroeve et al. 2012;Gillett et al. 2008).
The internally generated variability can be extracted using a given climate model with large ensemble simulations forced by the same external forcing (Deser et al. 2012). This methodology has been adopted to study the impact of internally generated variability on regional climate changes over periods of several decades (Deser et al. 2016;Wang et al. 2018;Ding et al. 2019;Hu et al. 2019). For example, previous studies have investigated the impacts of internal atmospheric variability on the SAT trends over North American and East Asian regions during boreal winter and summer (Deser et al. 2016;Hu et al. 2019). Ding et al. (2019) presented an important internal mechanism arising from lowfrequency Arctic atmospheric variability in models that can cause substantial summer sea-ice loss in recent decades.
Adopting the methodology formulated by Deser et al. (2012), we examine the influence of internal variability on the winter SAT trends over the BKS region and central Eurasia, separately. Furthermore, the relative contribution of internal variability to the observed trends in wintertime temperature remains an open question. Thus, this study will try to reveal the relative contribution of internal variability to the recent warming trend over the BKS region and cooling trend over central Eurasia, respectively.
The rest of the paper is organized as follows: Sect. 2 describes the observational data, model outputs and methodology employed in this study. Section 3 describes the results. Section 4 gives a summary and a discussion.

Data and methods
The present study employs monthly mean atmospheric reanalysis data from ERA-Interim, which span from 1 January 1979 to 31 August 2019 and are on horizontal grids of 2.5° × 2.5° (Dee et al. 2011). The analysis in this paper uses 31-member ensemble simulations to evaluate the role of internal variability in observed winter temperature trends. The simulations were conducted by the IPSL-CM6A-LR model, set up and developed by the Institute Pierre-Simon Laplace (IPSL) Climate Modelling Centre for Phase 6 of the Coupled Model Intercomparison Project (Eyring et al. 2016). Each of the ensemble members were forced by the same historical radiative forcing but differed in their initial atmospheric conditions. More information about the IPSL-CM6A-LR model is provided in Boucher et al. (2020). All the modeled atmospheric variables in the free atmosphere have been bilinearly interpolated to a horizontal resolution of 2.5° × 2.5°.
According to previous studies, the biases among the different members of the IPSL-CM6A-LR model can be attributed to the internal climate variability (Deser et al. 2012(Deser et al. , 2014(Deser et al. , 2016. Thus, the model's internal climate variability can be extracted as the inter-member spread from the multi-member mean. Following previous studies (Deser et al. 2014;Hu et al. 2019), empirical orthogonal function (EOF) analysis was applied to extract the leading patterns of the internal climate variability. To explore the relationships between the inter-member spread of the trends in two variables, the linear regressions were used in the present paper. And the two-tailed Student's t-test was used to evaluate the statistical significance of regression and composite analysis.
In the present study, the variance fraction of the intermember spread explained by the leading patterns of the internal climate variability was calculated as following: First, the inter-member spread of winter SAT trends was regressed against the principal components (PCs) using a simple linear regression method: where Y i denotes the winter SAT trend from the ensemble member i , and X i is the values of PCs in the ensemble member i . The and were calculated so as to minimize the RMS of the residuals i . Then, the fraction of variance explained by the PCs (FVE) can be defined as: To investigate the relative contribution of anthropogenic forcing and internal variability to the observed winter SAT trends, it is assumed that the trend in any variable can be replicated by a combination of forced and internal trend patterns derived from the IPSL-CM6A-LR 31 ensemble members. Following Ding et al. (2019), the fingerprint pattern matching method was used to find the linear combination of the fingerprint patterns that best matched the observed trend pattern ( OBS(x) ) derived from the ERA-Interim: (1) where F(x) is the pattern for forced trend, I 1 (x) and I 2 (x) are internal trends, respectively. Coefficients a, b and c are scaling coefficients that determine the linear contribution of forced response and internal trend to the observed trend pattern. In this study, the ensemble-mean SAT trend over the Arctic-Eurasia region (30°-90°N, 0°-180°E) was used as the the fingerprint of anthropogenic forcing ( F(x) ). The composite SAT trends over the Arctic-Eurasia region with respect to the PC1 and the PC2 were used as the fingerprint of internal variability ( I 1 (x) and I 2 (x) ). The composite trends of any varible with respect to the PCs were calculated as the difference in the trends between the high and the low values of PCs, with a criterion of ± 0.75σ. The best match was achieved when the root mean square difference between the observed trend pattern and the linear combination of forced and internal fingerprints is minimized and their spatial correlation is maximized. Using the resulting coefficients and input fingerprint patterns of any variable (such as SAT and air temperature), the observed trend patterns of the variable can be reconstructed. Figure 1a displays the observed winter SAT trend during 1990-2014. The period 1990-2014 was chosen because the winter SAT trend pattern during this time exhibits a strongly pronounced "warm Arctic-cold Eurasia" pattern. As shown in Fig. 1a, significant warming anomalies can be observed over the Arctic, especially over the BKS region (72.5°-85°N, 15°-90E°). Meanwhile, an apparent cooling trend can be seen over central Eurasia (40°-60°N, 50°-130°E) in boreal winter (Fig. 1a), which is consistent with previous studies (Wu et al. 2011;Cohen et al. 2013;Wang and Chen 2013;Wang et al. 2020b). The winter SAT trend shown in Fig. 1a was determined by a combination of external forcing and internal climate variability. Following previous studies (Deser et al. 2012(Deser et al. , 2014Kang et al. 2013), the trends due to external forcing can be calculated as the average of trends in all the ensemble members. Figure 1b displays the winter SAT trends during 1990-2014 generated by external forcing, which are defined as the average of trends in the IPSL-CM6A-LR 31 ensemble members. The ensemble-mean winter SAT trend (Fig. 1b) shows robust warming trend over the Arctic Ocean and American and Eurasian land areas. The forced response is not uniform, with higher warming rates over the Arctic, especially over the BKS region (Fig. 1b). However, the magnitude of the ensemble-mean SAT trend over the BKS region is smaller

Trend analysis
than that in the observational data. Thus, it is speculated that the internally driven SAT trends may exacerbate the forced response over the BKS region and mask the forced response over central Eurasia.
The relative importance of external forcing and internal climate variability to the winter SAT trend can be estimated quantitatively by calculating the signal-to-noise ratio (SNR). The SNR is calculated as the ratio of the ensemble-mean winter SAT trends to the standard deviation of the deviation from the ensemble mean trend among the 31 ensemble members (Deser et al. 2012). Figure 1c, d display the spatial distributions of the inter-member standard deviations and SNR for the winter SAT trends during 1990-2014, respectively. Large inter-member standard deviations of the winter SAT trends (more than 1.0) can be observed over the BKS region and over the mid and high latitudes of the Eurasian continent (Fig. 1c). The SNRs of winter SAT trends are larger than those over the Pacific sector of the Arctic and the Russian Far East (Fig. 1d). In contrast, low SNRs (values lower than one) of winter SAT trends appear over the Atlantic sector of the Arctic, especially over the BKS region and over the mid and high latitudes of the Eurasian continent (Fig. 1d). Thus, the winter SAT trends over the BKS region and central Eurasia are both strongly influenced by internal climate variability. Figure 2 display the winter SAT trends during 1990-2014 averaged over the BKS region and central Eurasia in the IPSL-CM6A-LR 31 ensemble members, respectively. As expected, there are considerable inter-member spreads in the winter SAT trends over the BKS region and central Eurasia across the 31 ensemble members. This implies that internal climate variability can exert a larger impact on the winter SAT trends over the BKS region and central Eurasia. The trend values over the BKS region are positive in the majority of the 31 ensemble members (Fig. 2a), suggesting that the warming trend induced by the external forcing is hard to be offset by the internal climate variability. Even over the region of central Eurasia, negative values of the SAT trend can be only seen in 5 ensemble members (Fig. 2b), suggesting that the recent cold trend over central Eurasia may be an extreme event of internal variability (Sun et al. 2016).
To further examine vertical structure of the temperature change in the Arctic-Eurasia region during 1990-2014, we display the vertical cross sections of the observed trend of winter temperature averaged between 30°-150°E in Fig. 3a. Over the high latitudes (north of 70°N), the air temperature exhibits warming trends throughout large parts of the troposphere. The Arctic warming shows maxima at the surface and in the upper troposphere. In midlatitudes (40°-65°N), the cooling anomalies can be observed in nearly entire troposphere below 300 hPa. The ensemble-mean trend of winter air temperature averaged between 30°-150°E (Fig. 3b) shows warming trend between 40°-90°N at the heights below 400 hPa. The maxima of the warming anomalies are located in the lowermost part of the troposphere over the Arctic. However, the external forcing cannot produce the Arctic warming trends in the upper troposphere and cooling trends in the midlatitudes. According to the inter-member standard deviations and the SNR for the trends of winter air temperature averaged between 30° and 150°E (Fig. 3c, d), the trends in the temperature at the surface and in the upper troposphere of Arctic are both strongly influenced by internal climate variability. Except in the lowermost part of the troposphere over the midlatitudes, the influence of external forcing usually exceeds that of internal variability at the heights below 300 hPa.

Physical processes for internally driven Arctic SAT trends
The above analysis has indicated that the observed winter SAT trends over the BKS region during 1990-2014 were strongly influenced by internal climate variability. In this part, we examine the physical processes responsible for the internal variability of the winter SAT trends over the BKS region.
Previous studies have reported that the warming anomalies over the BKS region can be explained by an increase in wintertime water vapor and accompanying enhancement of downward longwave radiation (DLR) (Wang et al. 2020a;Park et al. 2015;Gong et al. 2017). The inter-member spread of winter SAT trends over the Barents-Kara Seas may be driven by the same mechanism. As shown in Fig. 4a, a strong warming trend over the BKS region corresponds to positive winter SAT trends over most parts of the Arctic and across the Arctic coastline. Inspection of the winter DLR trends indicates that the inter-member spread of the winter SAT trends over the Barents-Kara Seas is closely associated with the variation there of the winter DLR trends (Fig. 4b). Although the warming trend in Arctic can be driven by the enhancement of DLR, the close relationship between the trends of SAT and DLR does not necessarily imply causality. Actually, the DLR is closely coupled to the SAT anomalies. For example, Vargas Zeppetello et al. (2019) indicated that response of DLR can be determined by the change in surface temperature, regardless of its cause.
Several studies also indicated that the change in largescale atmospheric circulation can influence the SAT anomalies over the BKS region through affecting the moisture flux (Wang et al. 2020a;Gong et al. 2017;Woods and Caballero 2016;Woods et al. 2013). Here, the regressed trends of wintertime atmospheric circulation with respect to the domainaveraged SAT trends over the BKS region (Fig. 4c, d) are further examined. Similar to the observed pattern of anomalous atmospheric circulation which drives the BKS warming (cf. Wang et al. 2020a), a barotropic anticyclone can be simulated over northern Eurasia accompanied with the warming trend over the BKS region (Fig. 4c, d). Under the influence of this barotropic high-pressure circulation, more moisture tends to be transported to the BKS region by the southerly wind anomalies (Fig. 5a). As a result, an increase in moisture flux convergence can be observed over the BKS region (Fig. 5b). The enhancement of DLR caused by strong moisture convergence tends to melt sea ice, which in turn leads to increased evaporation (Fig. 5c, d). The increased evaporation may also contribute positively to the wetting and warming anomalies over the BKS region.
It should be noted that the trends of winter SAT and SIC cannot simply be considered as the response to the anticyclonic circulation over northern Eurasia. Actually, a positive feedback exists between the Arctic SAT/SIC anomalies and the circulation anomalies. Previous studies have reported the importance of variations in sea ice over the BKS region in influencing atmospheric circulation (Inoue et al. 2012;Mori et al. 2014). According to Inoue et al. (2012) and Wang et al. (2020b), the weak thermal contrast over northern Eurasia can lead to weak synoptic-scale eddy activity there, which drives an anomalous anticyclone around the Ural Mountains. According to some studies, the warming trend over the BKS region can provide a favorable condition for enhancement of Urals blocking activity (Mori et al. 2014;Luo et al. 2017;Yao et al. 2017), which is also suggested to be the responsible internal atmospheric process for the anomalous ridge around the Ural Mountains (Cheung et al. 2013).

Dominant modes of internal climate variability
To extract the dominant modes of internal climate variability governing winter SAT trends, EOF analysis was performed on the ensemble SAT trends in the domain of Arctic-Eurasia (30°-90°N, 0°-180°E) among the 31 ensemble members. The first and second EOF modes can explain 40.1% and 22.7% of total variance in the winter SAT trends across the ensemble members, respectively. These two modes can be well separated from the others according to the criterion of North et al. (1982). The first EOF (EOF1) mode of winter SAT trends among ensemble members represents a pattern of uniform cooling over the Arctic-Eurasia region (Fig. 6a). The second EOF (EOF2) mode exhibits a dipole structure, with positive trend anomalies over the BKS region and negative trend anomalies over central Eurasia (Fig. 6b). The SLP trends associated with the EOF1 mode (Fig. 6c) resemble those associated with the negative phase of the AO (Thompson and Wallace 2001;Wang and Chen 2013), with significant positive trends over the high-latitudes and negative trends over the mid-latitudes. The anomalies of SLP trends in association with the EOF2 (Fig. 6d) are characterized by positive SLP trends over the northern Eurasian continent, which implies an amplification of the Siberian high (Chen et al. 2005).
The above results indicate that both the EOF1 and the EOF2 can influence the winter SAT trends over the BKS region and central Eurasia. The correlation coefficient between the PC1 and the domain-average SAT trend over the BKS region is as high as −0.73 (Fig. 7a). The PC1 also has a significant correlation with the SAT trend averaged over the central Eurasia with a correlation of −0.54 (Fig. 7c). The strong linear relationships also exist between the PC2 and the SAT trend averaged over the BKS region and central Eurasia, with correlation coefficient being 0.64 and −0.78, respectively (Fig. 7b, d). These two  Fig. 1, but for the latitude-height winter temperature trends (K/10 year) averaged over 30°-150° E modes are complementary in describing local percentage variances over the BKS region and central Eurasia. The EOF1 mode explains the majority (53.28%) of the local percentage variance over the BKS region, whereas the EOF2 accounts for 41.27% of the local percentage variance there (Fig. 7a, b). Over the region of central Eurasia, the two leading modes together can explain majority of the total fractional variance. In terms of its contribution to the local variability, the EOF2 is more important than the EOF1. The local percentage variance over central Eurasia explained by the EOF2 exceeds 60% (Fig. 7d). As for the EOF1, it explains 28.79% of the local percentage variance over central Eurasia (Fig. 7c). Therefore, the two leading EOF modes dominate the inter-member variations of winter SAT trend over the BKS region and central Eurasia. And the trend component of winter SAT arising from and horizontal winds (vectors; K/10 year) at 1000 hPa on the normalized winter SAT trend averaged over the Barents-Kara Seas. The dotted areas denote the anomalies that are significant at the 95% confidence level according to the Student's t-test. In d, only the winds anomalies that exceed 95% significance level are drawn internal variability can be determined by a linear combination of the EOF1 and EOF2 modes. Figure 8 displays composite differences in the trend of winter SAT and temperature averaged over 30°-150°E between the high and the low values of PC1 (left) and PC2 (right). As displayed in Fig. 8a, b, the fingerprints of the SAT trend due to internal variability have the comparable magnitudes to those of observed SAT trends and the forced fingerprint (cf. Fig. 1a, b). The footprints of the EOF1 and EOF2 mode are not limited in the lowermost part of the troposphere. As shown in Fig. 8c, the negative values of composite temperature trend with respect to the PC1 can extend to 300 hPa between 50°N and 65°N. The warming anomalies over Arctic and the cold anomalies over central Eurasia associated with the EOF2 mode can be observed throughout large parts of the troposphere below 300 hPa (Fig. 8d). However, the composite temperature trends with respect to the PCs weaken gradullay in their magnitudes with the increase of height in the troposphere (Fig. 8c, d). 1 3

The relative contribution of internal variability
One issue that needs to be addressed is what the relative contribution of the internal variability in the formation of the observed winter SAT trend over the Arctic-Eurasia region is. To answer this question, a fingerprint analysis has been performed to reconstruct the winter SAT trends over the Arctic-Eurasia region based on the fingerprints of the SAT trend due to internal variability and anthropogenic forcing as described in Sect. 2. Considering the comparable magnitudes of the observed SAT trends and the fingerprints of the SAT trend due to internal variability and anthropogenic forcing, coefficients a, b and c were chosen to be varied from 0.1 to 3 independently (at 0.1 increments) to form as few combinations as possible. The lowest root mean square error and highest spatial correlation is obtained when a = 0.8 , b = 0.2 and c = 0.9 . As shown in Fig. 9a, b, the reconstructed trend patterns of winter SAT well resemble Fig. 6 Regressions of winter SAT trends (K/10 year) upon the standardized a PC1 and b PC2 of the winter SAT trends over 30°-90° N and 0°-180° E across the 31 IPSL-CM6A-LR ensemble members. c, d as in a, b, but for winter SLP trends (hPa/10 year). Dotted areas denote anomalies that are significant at the 95% confidence level according to the Student's t-test the observed trend patterns. Moreover, the magnitudes of the reconstructed trends are very similar to those observed. These suggest that the linear combination of forced and internal variability accounts fully for the observed trends of winter SAT during 1990-2014. Using the coefficients of a = 0.8 , b = 0.2 and c = 0.9 and the fingerprint patterns of the tempertature trend averaged over 30°-150°E, we can also reconstruct the vertical structure of the temperature change in the Arctic-Eurasia region. The comparison of Fig. 9c with Fig. 9d indicates the that the linear combination of forced and internal variability can reproduce well the temperature change at the heights below 500 hPa. However, there are some discrepancies between the observed and reconstructed temperature trend at the heights above 500 hPa. This implies that other physical processes may account for the change in the air temperature in the upper troposphere. Figure 10a, b display the winter SAT trends due to anthropogenic forcing and internal variability, respectively. The anthropogenic forcing can drive warming trend over the whole Arctic Ocean (Fig. 10a). The maximum warming values can be observed over the BKS region (Fig. 10a). In the Arctic, the warming anomalies driving by the internal variability can be only observed over the BKS region (Fig. 10b). A comparison of Fig. 10b with Fig. 7 Scatterplots of the standardized winter SAT trends averaged over BKS against the standardized a PC1 and b PC2 of the winter SAT trends over 30°-90° N and 0°-180°E across 31 IPSL-CM6A-LR ensemble members. c, d as in a, b, but for winter SAT trends averaged over central Eurasia against the standardized c PC1 and d PC2 Fig. 10a suggests that the magnitude of internally generated warming trends over the BKS is comparable to that of forced warming trends. And the internal variability is estimated to contribute to about 50-60% of observed warming trend over the BKS region during 1990-2014 (Fig. 10d). As for other parts of Arctic, the warming trends are dominated by the anthropogenic forcing (Fig. 10c). Over the Eurasian land, the internal variability and the anthropogenic forcing play different roles in influencing the observed winter SAT trend during 1990-2014. The internal variability contributes dominantly to formation of the cold anomalies over central Eurasia, whereas the anthropogenic forcing acts to impede the cold anomalies weakly (Fig. 10). In short, the cold Eurasia tends to arise predominantly from the internal variability, whereas the warming BKS region is a combined result of anthropogenic forcing and internal climate variability. The composite trends of winter SAT (K/10 year) constructed for a PC1 and b PC2 of the winter SAT trends over 30°-90° N and 0°-180° E. c, d as in a, b, but for the cross section of the composite trends of winter temperature (K/10 year) averaged over 30°-150° E. Dotted areas denote anomalies that are significant at the 95% confidence level according to the Student's t-test

Summary and discussion
In this paper we have investigated the role of internal variability in the formation of the opposite trends in winter SAT over the BKS region and central Eurasia during 1990-2014 through analysis of large ensembles of fully coupled climate model simulations with historical radiative forcing. Measured by the standard deviation of the winter SAT trends, the trends in winter SAT over the BKS and central Eurasia display large inter-member diversity across all the ensemble members. This suggests that internal climate variability is vital to the recent observed trends in wintertime SAT over the BKS region and central Eurasia. Our study presents that the change in atmospheric circulation arising from internal variability in models can cause the warming anomalies over the BKS region. The barotropic anticyclone over northern Eurasia can transport more moisture to the BKS region. The increase in the moisture and the resulted DLR anomalies tend to contribute positively to the warming trend and the sea ice loss over the BKS region. Furthermore, the sea ice reduction is conductive to the increase in the evaporation, which Fig. 9 The a reconstructed and b observed trends of winter SAT (K/10 year). c,d as in a, b, but for the cross section of trends of winter temperature (K/10 year) averaged over 30°-150° E also contributed positively to the wetting and warming anomalies over the BKS region.
The EOF1 mode of winter SAT trends over the Arctic-Eurasia region represents a pattern of uniform trend anomalies, whereas the EOF2 mode exhibits opposite anomalies over the BKS region and central Eurasia. The results indicate that majority of the inter-member variations of the winter SAT trends over the BKS region and central Eurasia can be explained by the two leading modes of winter SAT trends over the Arctic-Eurasia region. Thus, the trend component of winter SAT arising from internal variability can be determined by the linear combination of the EOF1 Fig. 10 The winter SAT trends (K/10 year) as contributed to by a external forcing and b internal variability. The relative contribution (%) of c external forcing and d internal variability to the reconstructed winter SAT trends and EOF2 modes. In addition, using the fingerprint pattern matching method, it is found that the observed cooling trend over central Eurasia is predominantly due to the internal variability. Further analysis indicates that the internal variability contributes to about 50-60% of the warming trend observed over the BKS region. Cohen et al. (2012b) have reported the important role of decrease in the winter AO index in favoring anomalously cold temperatures over northern Eurasia. However, as estimated in Fig. 11a, the EOF1 mode, which is reminiscent of the negative phase of the AO, made minor contribution to the recent decrease in the temperature over northern Eurasia. The recent Eurasian cooling was predominantly contributed by the EOF2 mode (Fig. 11b). The EOF2 mode was referred as the warm Arctic-cold Eurasian (WACE) mode in Mori et al. (2014). Mori et al. (2019) argued that nearly 44% of the WACE variance was driven by sea ice loss over the BKS region (Mori et al. 2019). However, this conclusion is disputed by Screen and Blackport (2019) and Zappa et al. (2021).
It should be noted that the used fingerprint methods rely on the assumption that the forced response in the real world does not strongly project onto either the two leading modes of variability. Furthermore, the obtained results rely on the assumption that the models can reproduce all the relevant processes. However, the models cannot capture all the relevant processes correctly. For example, some previous studies have emphasized the role of Arctic sea-ice loss in exciting the atmospheric circulation anomalies and cold continents in the winter (Honda et al. 2009;Nakamura 2015;Zhang et al. 2018). However, several studies argue that models do not capture the response to sea ice loss (e.g., Cohen et al. 2020;Wang et al. 2020c). Furthermore, due to possible inter-model diversity, the estimated contribution of the internal variability may depend on the model used. Anyway, in our study, the successful reconstruction of winter SAT trend using the combination of simulated forced and internal patterns suggests that the internal variability plays an important role in driving the opposite trends in wintertime temperature over the BKS region and central Eurasia.