Identification of Shortcomings in Simulating the Subseasonal Reversal of the Warm Arctic–Cold Eurasia Pattern

Subseasonal reversal of warm Arctic–cold Eurasia (SR‐WACE) pattern has significant impacts on transitions of weather and climate extremes in Eurasia. This study explored the performances of climate models to simulate the main features of SR‐WACE. For real‐time predictions, most of the state‐of‐the‐art climate models showed limited ability to accurately forecast SR‐WACE in advance. Furthermore, most of the historical simulations from Phase 6 of the Coupled Model Intercomparison Project (CMIP6) had also difficulties in well simulating the SR‐WACE. Further exploration showed that the simultaneous reversal of the Ural blocking high (UB) and Siberian high (SH) is the key atmospheric driver of the SR‐WACE occurrences, which were verified by both of the real‐time predictions and historical simulations. Our results implied that the simulation of SR‐WACE was a huge challenge and the critical solutions included improving simulation of subseasonal reversals of UB and SH in the atmosphere.


Introduction
A dipole pattern of surface air temperature (SAT) exists across the Arctic-Eurasian region at mid-to-high latitudes known as the warm Arctic-cold Eurasia (WACE) pattern (Jin et al., 2020;Kug et al., 2015), which has received widespread recognition as the dominant pattern of interannual variation in the winter season (Sung et al., 2018).However, with the weakening trend of WACE in the past decade (Blackport & Screen, 2020), subseasonal reversals of WACE (SR-WACE) pattern in early and late winter have become increasingly commonplace (Yin et al., 2023).Meanwhile, it also maintains a strong intensity on the subseasonal timescale, which leads to extreme climate events.For instance, the record-breaking cold and warm transition in eastern China in the winter of 2020/ 2021 (Zhang et al., 2021) and a super-sandstorm in North China in the spring of 2021 (Yin et al., 2022) were both heavily influenced by the subseasonal transition from the WACE to the cold Arctic-warm Eurasia (CAWE) pattern.These seesaw temperature extremes exerted pronounced impacts on human activities and the global economy (WMO, 2022).Additionally, the subseasonal variability lies between weather forecasting and longrange climate prediction (Mariotti et al., 2020), improving the subseasonal predictability remains a huge challenge (Li et al., 2020), especially in the mid-to-high latitudes (Wang et al., 2022).
Different from the seasonal WACE pattern, which is closely related to North Atlantic Oscillation (NAO) (Zhuo et al., 2023), Arctic sea-ice (Mori et al., 2014), North Atlantic sea surface temperature (Luo, Wu, et al., 2019), and the Atlantic Multidecadal Oscillation and Pacific Decadal Oscillation (Luo et al., 2017;Tokinaga et al., 2017), previous studies have linked the SR-WACE to subseasonal reversals of anomalies in the midlatitude westerlies and stratospheric polar vortex (X.Xu, He, Zhou, & Wang, 2022).Meanwhile, the subseasonal variability of wintertime SAT is obvious under the frequent influence of large-scale atmospheric circulation (Li et al., 2022;Lin, 2018).Two critical atmospheric circulation systems play key roles in linking SAT variations in the Arctic-Eurasian region: the Ural blocking high (UB) and Siberian high (SH) (X.Xu, He, Zhou, & Wang, 2022;Yang & Fan, 2022).The persistent strengthening of the UB and SH, as well as the northerly winds to the east, are significant causes of enhanced cold advection (Liu & Zhu, 2019), which can lead to cold anomalies across Eurasia (Cheung et al., 2013).Moreover, it can also tend to transport warm air from the mid-latitudes to the Arctic along its western flank (Xie et al., 2022), which is favorable for Arctic warming.In contrast, when the UB and SH exhibit subseasonal out-of-phase shifts throughout the winter, a corresponding reversal of SAT occurs (Yan et al., 2022;Yu et al., 2022).Therefore, accurate simulation of the SH and UB is likely to be a key factor in the ability to simulate the SR-WACE.
Climate models are widely applied to comprehend and enhance forecasts of climate systems over the globe.Meanwhile, the ability to simulate the WACE pattern in climate models is a critical prerequisite for revealing the causes of winter climate change and occurrences of extreme weather in the Northern Hemisphere (Zhao et al., 2023).Some previous studies have been carried out to evaluate the capacity of climate models in simulating the WACE pattern.Under the ongoing Arctic sea ice reduction (Simmonds & Li, 2021), there's a close relationship between sea ice and WACE in observations, but the causal link needs further model validation.The coupled Model Intercomparison Project Phase 5 (CMIP5) simulated a coherent relationship between the sea-ice variability and the WACE pattern (Wang et al., 2020).However, the results of ensemble simulations in other studies indicated that the response of the WACE pattern to Arctic sea-ice change was small (Ogawa et al., 2018), possibly because of underestimation of the atmospheric response to sea-ice change (Mori et al., 2019).Meanwhile, SR-WACE might be responsible for the subseasonal transition of Arctic sea-ice (Zhang et al., 2023).The potential influence of sea-ice variation on the WACE pattern shown in model simulations remains a hot topic of debate (Li et al., 2021;Luo, Chen, et al., 2019;Screen & Blackport, 2019).Furthermore, the simulation of UB strongly affects the performance of models in simulating the WACE pattern in the sixth phase of CMIP (CMIP6) (Zhao et al., 2023).The deep Arctic warming and weakening of the midlatitude westerly jet have also been proposed as a key feature generating cold anomalies in Eurasia (He et al., 2020;X. Xu, He, Zhou, Wang, & Outten, 2022).In the prediction models, the WACE pattern is strongly dependent on the Barents-Kara Sea temperature and the atmospheric circulation in winter (Komatsu et al., 2022).In short, these latest studies demonstrate that the WACE pattern is essentially caused by internal atmospheric variability.
It is worth noting that previous studies have tended to focus on interannual, interdecadal, and long-term trend changes in the WACE pattern, with comparatively fewer studies having examined the subseasonal timescale.It is currently uncertain if existing climate models are sufficiently capable of reproducing the SR-WACE.Therefore, we explored the performance of both real-time prediction models and CMIP6 models in their simulations of the SR-WACE and analyzed the key atmospheric circulation processes involved.

Data
Daily meteorological data for winter (December-February) over the period 1959-2021 were obtained from ERA5 (Hersbach et al., 2020), including SAT, sea level pressure (SLP) and geopotential height at 500 hPa (Z500); and the daily outputs of the historical simulations over the period 1900-2013 in winter of 22 models from CMIP6 (variant label: r1i1p1f1; Table S1 in Supporting Information S1) were employed (Eyring et al., 2016).Note the different periods and lengths of observation (OBS) records used to compare with the model results, OBS was not intended to be a strict reference but to serve as a guide to facilitate model intercomparison (McKenna et al., 2020).We used daily forecast data of SAT, SLP and Z500 in seven seasonal prediction systems: the ECMWF System 5 seasonal forecast model (SEAS5) (EC; Johnson et al., 2019), CMCC Operational Seasonal Prediction System SPS3.5 (CMCC; Sanna et al., 2016), DWD German Climate Forecasting System Version 2.1 (DWD; Fröhlich et al., 2021), UK Met Office's GloSea5 (UKMO; MacLachlan et al., 2015), Meteo-France System 7 (MF; Batté & Déqué, 2016), JMA Coupled Prediction System version 3 (CPS2) (JMA; Takaya et al., 2018) and ECCC's GEM5-NEMO (ECCC; Lin et al., 2020).The forecast data of DWD and EC cover the period 1993-2021, while other models cover 1993-2016.The prediction results are released on 1 November (JMA is released on 12th November).All the data have had their daily linear trends removed and been remapped onto the same (2.5°×2.5°) grid.
The SAT difference between the Barents-Kara Sea (SAT A ; 65°-85°N, 30°-90°E) and Eurasia (SAT E ; 40°-60°N, 60°-120°E) (former minus latter) is defined as the SAT AE , which can adequately depict the variability characteristics of the WACE pattern (Kim et al., 2021;T. B. Xu et al., 2023).The SH index (SHI) is defined as the areaaveraged SLP over central Siberia (40°-60°N, 80°-120°E) and the UB index (UBI) is defined as the regional average of Z500 within the region of (55°-70°N, 50°-100°E) (Zhang et al., 2021).The winter season is defined as December in the current year and January to February in the next year.The reversal point of SR-WACE generally falls in the mid-January, around 14 January (Yin er al., 2023).Therefore, from 1 December to 14 January and 15 January to 28 February are defined as the early and late winter, respectively.The difference in the indexes defined above between early and late winter are SAT A -S, SAT E -S, SAT AE -S, SHI-S and UBI-S, which can represent the subseasonal variations between early and late winter (T.B. Xu et al., 2023).

Methods
In this study, the multi-model ensemble (MME) is calculated by averaging the variables over all the models with equal weighting.The Taylor diagram is used to analyze the performance of the CMIP6 models in terms of the spatial correlation coefficient (R), root mean square error, and ratio of their standard deviation (STD) (Taylor, 2001).The skill score is used to quantitatively evaluate the simulation ability of the models (Hirota et al., 2011): where R represents the spatial correlation between the model and OBS, and SDR is the ratio of the spatial STD of the model against that of OBS.The skill score lies between 0 and 1; the closer to 1, the better the simulation effect.
Season-reliant empirical orthogonal function (S-EOF) analysis is used to examine the SR-WACE.The derived spatial patterns for each S-EOF mode contained two sequential patterns illustrating the subseasonal evolution of the SAT.They share the consistent annual value in their respective normalized time series of the standardized principal component (PC) (Wang & An, 2005).The criteria for identifying SR-WACE in S-EOF modes within CMIP6 models involve detecting a reversal sign of the SAT AE indices in regressed early and late winter SAT on each normalized PCs, and the sum of their absolute values is the largest among the first four modes.

Observed and Model-Predicted Subseasonal Reversal of WACE Pattern
To identify the inherent pattern of SR-WACE, the S-EOF analysis is used over the Arctic-Eurasian region (40°-90°N, 20°-130°E).The spatial distribution of the first four leading modes of S-EOF are shown in Figure S1 in Supporting Information S1 and the first three are independent of each other (North et al., 1982).The third S-EOF mode (S-EOF3) is characterized by an out-of-phase change in the WACE pattern, indicating the WACE and CAWE pattern in early and late winter, respectively.Regression maps of SAT on the normalized PC3 in early and late winter are shown in Figures 1a and 1b.Both the Arctic and Eurasian centers show substantial signals of subseasonal phase reversal.To assess whether the results are sensitive to the selection of partition for early and late winter, we perturbed the division dates and found that the results were consistent (Figure not shown).
Meanwhile, the time series of PC3 implies that the SR-WACE also displays interannual variability (Figure 1c).PC3 shows a positive value in 2020 and negative value in 2021, which means that S-EOF3 can capture the transition of the WACE/CAWE pattern in early and late winter of 2020 (Figures S2c and S2d in Supporting Information S1) and the CAWE/WACE pattern in 2021 (Figures S2e and S2f in Supporting Information S1) (Yin et al., 2023).In addition, we selected positive (WACE-CAWE) and negative (CAWE-WACE) cases by taking ±0.8 STD of the PC3 as the threshold.The mean SAT AE index (Figure 1d) and SAT (Figures S2a and S2b in Supporting Information S1) were composited for positive and negative cases in early and late winter.In both the positive and negative composite, it is evident that the SAT experiences a subseasonal reversal.Furthermore, the R

Geophysical Research Letters
10.1029/2023GL105430 between the PC3 and SAT AE -S index is 0.77 (exceeding the 99% confidence level), indicating that the time series of SAT AE -S appropriately reflects the variation in PC3 (Figure 1c).
Realizing accurate prediction of the SR-WACE has great practical significance.However, it is difficult for existing state-of-the-art real-time prediction models to provide skillful prediction results (Figure 2a).The seven prediction models are unable to capture the S-EOF mode of the SR-WACE (Figure not shown).Meanwhile, the prediction skill for the SAT A -S, SAT E -S and SAT AE -S, measured by the R with OBS, are all below 0.2 during 1993-2016, except for JMA (Figure 2a).For the MME, the Rs are below 0.1 for all indices.DWD and EC still have poor forecasting skill in recent years.In short, existing real-time prediction models have limited prediction ability with respect to the SR-WACE, which possibly restricted the subseasonal to seasonal forecasts skill of climate anomalies in mid-low latitudes (Zhang et al., 2021).
Furthermore, there is a significant positive correlation between the SAT AE -S in OBS and prediction bias of winter SAT AE , with R being 0.43 (exceeding the 95% confidence level) for the MME (Figure 2b).It is interesting that when there is a negative prediction bias (third quadrant in Figure 2b), its corresponding SAT AE -S is almost negative.This shows the frequent SR-WACE is an important source for prediction bias of seasonal WACE (T.B. Xu et al., 2023).To improve the prediction skill, further exploration is needed regarding the key physical processes associated with the SR-WACE in climate models.We therefore explored CMIP6 models in their simulation of the SR-WACE.

Subseasonal Reversal of WACE Pattern in CMIP6
To explored the performance of CMIP6 models in simulating SR-WACE, we performed S-EOF analysis on the SAT of each model.Compared with the SR-WACE appearing in S-EOF3 in OBS, only 4 models appearing in S-EOF3 (Figure 3a).However, we extended the mode separated by S-EOF to the fourth mode (S-EOF4) and found that all other CMIP6 models can capture the SR-WACE in S-EOF4 (Figure 3b).This shows that CMIP6 models have generally limited simulation ability for the inherent mode of SR-WACE.The mean SAT AE indices in regressed early and late winter SAT on the normalized selected PC (PC3 in 4 models and PC4 in the other 18 models) in models were compared (Figures 3a and 3b) and the difference between them (SAT AE -S) was used to evaluate the ability of CMIP6 to simulate the reversal strength (Figure 3c).Large inter-model spread existed among the 22 CMIP6 models (Figure 3c).CESM2-FV2, IITM-ESM and CESM2-WACCM show considerable strength with OBS, while NorESM2-LM, MPI-ESM-1-2-HAM, IPSL-CM6A-LR and NorESM2-MM overestimate the reversal strength.In contrast, the other 15 models all underestimate the reversal strength.From the perspective of the variance contribution rate, except for GFDL-CM4, all models underestimate the explained variance (Figure 3d).
In order to further explore the performance of CMIP6 models in representing the SAT anomalies associated with the SR-WACE, a Taylor diagram is shown in Figure S3 in Supporting Information S1.In early winter, the spatial Rs range from 0.3 to 0.89, and roughly half the models have a normalized STD greater than 1, and the other half on that is less than 1 (Figure S3a in Supporting Information S1).In late winter, the spatial Rs range from 0.1 to 0.9 and most models underestimate the spatial STD (all except CESM2-WACCM, EC-Earth3, GFDL-ESM4, INM-CM5-0 and NorESM2-MM; Figure S3b in Supporting Information S1).Quantifying the differences between models through their skill scores, we find scores ranging from 0.42 (0.21) to 0.86 (0.89) in early (late) winter (Figure 3e).Interestingly, the orderings of models in terms of early and late winter skills are similar, demonstrating consistency.High-and low-skill model groups are selected in terms of the early plus late winter skill score.CMCC-CM2-HR4, CMCC-CM2-SR5, NorESM2-MM and CESM2 are selected as the high-skill models, and EC-Earth3-AerChem, GFDL-CM4, INM-CM5-0 and CESM2-FV2 are the low-skill models.Figure S4 in Supporting Information S1 compares the MME SAT associated with the SR-WACE between high-skill and lowskill groups.With a spatial R of 0.92 (0.94) between OBS and the MME in early (late) winter in the high-skill In general, large inter-model spread exists in the simulation ability of the SR-WACE among models and the skill scores can clearly distinguish between high-and low-skill models.However, it is worth noting that in CMIP6, although some models can capture the appearance of SR-WACE in S-EOF3 with a higher variance contribution, their ability to simulate the spatial distribution of the WACE/CAWE pattern centers is poor.In contrast, other models can better simulate the spatial distribution, but there is a significant underestimation of this inherent mode by the low variance contribution (Figures 3d and 3e).Therefore, there is serious uncertainty in terms of the CMIP6 models' ability to simulate the SR-WACE, which is a huge challenge faced by climate models.
It is noteworthy that all CMIP6 models can simulate the seasonal WACE pattern in the second EOF mode, which is the same as OBS (Zhao et al., 2023).However, most CMIP6 models can capture SR-WACE in the fourth mode of S-EOF, rather than the observed third mode (Figure 3).It indicates that the CMIP6 models differ greatly in their simulation capabilities with respect to the seasonal and subseasonal WACE pattern.The relationship between the simulation ability of the seasonal WACE pattern and SR-WACE in CMIP6 was analyzed (Figures S3c and S3d in Supporting Information S1).The R between the skill scores of the seasonal WACE pattern and that in early plus late winter of SR-WACE in each model is 0.5 (exceeding the 95% confidence level), which indicates that the ability of CMIP6 to simulate the seasonal WACE pattern is closely related to the ability to simulate the SR-WACE.Therefore, improving the ability of climate models to simulate the SR-WACE is critical.Next, we further analyze the key atmospheric circulation processes associated with the SR-WACE based on the grouping of high-and low-skill CMIP6 models mentioned above.

Atmospheric Circulation Associated With the Reversal of WACE Pattern
Regression maps of Z500 and SLP on the normalized PC3 for OBS are shown in Figures S5a, S5b in Supporting Information S1.In the Z500 and SLP fields, the NAO enters a positive phase and an anticyclonic anomaly stretches from the Arctic to Eurasia in early winter, which is replaced by a cyclonic anomaly over Eurasia in late winter (X.Xu, He, Zhou, & Wang, 2022).This indicates that the UB and SH show subseasonal out-of-phase changes, concurrent with reverse changes in the East Asian trough, which leads to a corresponding reversal of SAT occurring in winter.For the MME of the high-skill CMIP6 models, only the local atmospheric circulation anomalies can be simulated, which resembles the OBS spatial feature (Figures S5c and S5d in Supporting Information S1).However, there is almost no simulation ability for the circulation field over the North Atlantic.In the MME of the low-skill models, the range of anticyclonic anomalies over Eurasia in early winter is larger than that in OBS and the MME of the high-skill models, and the center of the cyclonic anomalies in late winter is more to the north (Figures S5e, S5f in Supporting Information S1).
In order to further quantify the relationship between them, scatter diagrams of the skill scores against the spatial Rs of Z500 and SLP anomalies between models and OBS in the region bounded by 40°-90°N and 20°-130°E are shown in Figure S6 in Supporting Information S1.The Rs between the skill scores and spatial Rs of Z500 UB and SH are essential components of the atmospheric circulation in the Eurasian mid-high latitudes, which play a crucial role in influencing SAT variations at multiple time scales (X.Xu, He, Zhou, & Wang, 2022;Yang & Fan, 2022).A scatter diagram of the skill scores against the Rs between the selected PC and UBI-S and SHI-S in CMIP6 is shown in Figure S7 in Supporting Information S1.The Rs between them are 0.71 and 0.56, respectively, which indicates that the reversal of the UB and SH is closely related to the SR-WACE and are perhaps the key atmospheric circulation processes.In order to further explore the impact of out-of-phase changes in the UB and SH, we classified the strength of UB and SH reversals in CMIP6.A strong UB (SH) reversal is identified when the UBI-S (SHI-S) index is above/below 0.8/ 0.8 STD; and a weak UB (SH) reversal is identified when the UBI-S (SHI-S) index is between 0.8 and 0.8 STD.This can divide the reversal relationship between the UB and SH into three categories: "strong UB and SH reversal," "strong UB and weak SH reversal," and "weak UB and strong SH reversal" (Figure 4).Notably, only when strong reversals of the UB and SH are simultaneously simulated does SR-WACE occur (Figure 4).This indicates that for CMIP6 models to simulate the key physical process of the SR-WACE, a strong reversal of both the UB and SH is key.The above physical processes were also validated in prediction models (Figure S8 in Supporting Information S1).Only by accurately predicting the simultaneous reversal of the UB and SH in the prediction models can the SR-WACE be accurately predicted.Meanwhile, the stronger the predicted reversal intensity of the UB (SH) between early and late winter, the stronger the SR-WACE, which demonstrates the critical role of blocking (Kim et al., 2021).The R between them reached 0.63 (0.61), which is significant at the 99% confidence level (Figures 2c and 2d).Therefore, the key to improving the prediction of the SR-WACE is to improve the ability of prediction models to capture UB and SH reversals.

Conclusions and Discussions
In this study, different to most previous studies that focused on the interannual, interdecadal or long-term trend changes of the WACE pattern, we explored the SR-WACE in climate models.We found that existing real-time prediction models have limited prediction ability with respect to the SR-WACE, which possibly restricted the subseasonal to seasonal forecasts skill of climate anomalies in mid-low latitudes.Furthermore, the historical simulations from CMIP6 had also difficulties in well simulating the SR-WACE.The most of CMIP6 models only reproduce the SR-WACE phenomenon with higher order modes and lower explained variances than OBS.Meanwhile, the capacity of models to simulate the seasonal WACE pattern is highly correlated with their capacity to replicate the SR-WACE.Therefore, more concerns are needed to better simulate the SR-WACE in climate models.We further explored the atmospheric circulation associated with the SR-WACE.The ability of CMIP6 models to reproduce the SR-WACE was found to be closely related to their performance in capturing the location and extent of the atmospheric circulation pattern.Specifically, the reversal of UB and SH are the key atmospheric circulation processes linked with the SR-WACE.Only when a strong reversal of the UB and SH are simultaneously simulated can the SR-WACE occur.The above physical processes were also validated in realtime prediction models.Our results implied that the simulation of SR-WACE was a huge challenge and the critical solutions included improving simulation of subseasonal reversals of UB and SH in the atmosphere.
The reasons for the deficiencies in simulating SR-WACE in the models may also stem from other factors, such as the models underestimating blocking frequency (Davini et al., 2020) and insufficient modeling of stratospheretroposphere interactions (Ding et al., 2023;M. Xu et al., 2023), which requires further exploration.Meanwhile, the impact and role of transient weather systems on SR-WACE also requires further attention (Kim et al., 2021).Furthermore, previous research has shown that the WACE/CAWE pattern in early and late winter, is significantly impacted by the preceding sea surface temperature anomalies in the tropical Atlantic and Indian Ocean, respectively (Yin et al., 2023).The central Pacific ENSO has been found to contribute more to the reversal of SATs over China in recent decades (Li et al., 2021;Luo et al., 2021), and the southern center of the NAO shifting westward is closely related to the subseasonal inversion of the Eurasian SATs (Song et al., 2023).The ability of climate models to simulate the comprehensive effects of internal atmospheric variability and background forcing on SR-WACE is unclear and needs to be investigated further.

Figure 1 .
Figure 1.Regressions of surface air temperature (shading; unit: °C) on the normalized PC3 in panels (a) early and (b) late winter in OBS during 1959-2021.The black slashes indicate regression coefficients exceeding the 95% confidence level.(c) Temporal variations of the normalized PC3 (solid) and SAT AE -S (dashed) in OBS.(d) Boxplot of the mean SAT AE for the transition from WACE to cold Arctic-warm Eurasia (CAWE) (red boxes) and from CAWE to WACE (blue boxes) in early and late winter.

Figure 2 .
Figure 2. (a) Temporal variation of SAT A -S, SAT E -S and SAT AE -S in OBS (black), prediction models (gray) and the multi-model ensemble (blue) during 1993-2016.The forecast results for 2017-2021 only include DWD and EC.Blue shading shows the ±2 standard deviation ensemble spread.The Rs between OBS and models are shown in the right-hand bars.(b) Relationship between the model-predicted winter-mean SAT AE bias and the SAT AE -S in OBS during 1993-2016.The Rs between them are also shown in the right-hand bars.(c, d) Scatterplot of SAT AE -S versus UBI-S (c) and SHI-S (d) in prediction models.

Figure 3 .
Figure 3. (a) Subseasonal reversal of WACE pattern appearing in S-EOF3 mode of the CMIP6 models.The vertical and horizontal axes represent the mean SAT AE in regressed early and late winter surface air temperature on the normalized PC3, respectively.(b) As in panel (a) but for S-EOF4 and (c) of subseasonal reversal in individual CMIP6 models.The dashed line represents the intensity in OBS.(d) Explained variance the S-EOF mode in OBS (dashed and CMIP6 models.(e) scores for each CMIP6 model, with red bars representing skill scores for early winter and blue for late winter.

Figure 4 .
Figure 4. (a) Differences of early and late winter-mean Z500 (shadings; unit: gpm), sea level pressure (contours; unit: gpm) and surface air temperature (shadings; unit: °C) anomalies composited with strong UB and Siberian high (SH) reversal from positive to negative phase and negative to positive phase in the multi-model ensemble of CMIP6 models.(b, c) As in panel (a), but for strong UB/weak SH reversal and weak UB/strong SH reversal, respectively.The black slashes and red contours indicate regions significant at the 95% level.
(SLP)anomalies are as high as 0.88 (0.7), 0.68 (0.82) and 0.82 (0.72) in early winter (FiguresS6a and S6din Supporting Information S1), late winter (FiguresS6b and S6ein Supporting Information S1) and early plus late winter (FiguresS6c and S6fin Supporting Information S1), respectively.This confirms that the ability of the CMIP6 models in reproducing the SR-WACE is closely related to the performance in capturing the atmospheric circulation pattern.