ENSO and QBO modulation of the relationship between Arctic sea ice loss and Eurasian winter climate

The causality between Arctic sea ice decline and cold boreal winters has been widely debated recently and remains controversial. A major source of uncertainty in the sea ice-cold winter relationship originates from that the stratosphere polar vortex (SPV) is not only affected by Arctic sea ice anomaly but can also be modulated by El Niño-Southern Oscillation (ENSO) and quasi-biennial oscillation (QBO). Using reanalysis data and hindcasts from the decadal prediction system version 4, here we show that both cold and warm winters occur over Eurasia when the Barents–Kara Sea sea ice is abnormally low. Warm winters occur during the La Niña-easterly QBO-low sea ice (LANINA-EQBO-LICE) years and cold winters during the neutral ENSO-neutral QBO (NENSO-NQBO)-LICE and El Niño (ELNINO)-NQBO-LICE years. During the LANINA-EQBO-LICE years, weakened upward-propagating planetary waves from the troposphere to the stratosphere strengthen the Arctic SPV and then weaken the Aleutian low and Siberian high, creating conditions favorable for Eurasian warming. The atmospheric responses are opposite in the NENSO-NQBO-LICE and ELNINO-NQBO-LICE years. This implies that care should be taken in using Arctic sea ice alone as the precursor to predict boreal winter climate.


Introduction
During the past few decades, several extremely cold winters occurred in Eurasia despite global warming (Coumou and Rahmstorf 2012, Liu et al 2012, Cohen et al 2020, Bailey et al 2021, Cohen et al 2021, Zheng et al 2022, which had imposed severe threats to electrical power transmission, energy supply, agriculture, air pollution, and even human lives (Kamo et al 2016, Wang et al 2016, Thornton et al 2017, Lu et al 2022. Therefore, a better understanding of the precursors and related potential mechanism for cold winters is of great importance for human life and societal development. Some observational and modeling studies attributed these cold winters to the rapid decline of Arctic sea ice in autumn or winter, especially over the Barents-Kara Sea (BKS) (Liu et al 2012, Tang et al 2013, Wu et al 2013, Mori et al 2014, Xu et al 2021. Using the atmospheric general circulation model, Zhang et al (2018b) proposed a dynamical mechanism explaining the relationship between the Arctic sea ice and midlatitude circulation based mainly on the stratospheric pathway rather than the tropospheric pathway. The enhanced upward planetary wave induced by sea ice loss could propagate into the stratosphere, and thus weaken the stratospheric polar vortex (SPV) (Jaiser et al 2013, Kim et al 2014, Peings and Magnusdottir 2014, Nakamura et al 2016, Hoshi et al 2017, Zhang et al 2018a. These stratospheric circulation anomalies forced by Arctic sea ice can persist for a couple of months and subsequently propagate downward to the midlatitude surface, resulting in cooling over most parts of northern Eurasia (Kim et al 2014, Peings and Magnusdottir 2014, Sun et al 2015, Zhang et al 2018b. Moreover, the BKS sea ice reduction could also contribute to the SPV shift, leading to a colder climate over some parts of the Eurasian continent (Zhang et al 2016). These studies imply that the SPV can act as a bridge linking the BKS sea ice and midlatitude climate.
Recently, some studies have discussed the lagged impact of the Arctic sea ice (e.g. Liu et al 2019, Ding et al 2021, Yang et al 2022. For example, Yang et al (2022) pointed out that the long-lagged impact of the Arctic sea ice in BKSs on June precipitation in eastern China can be interpreted by the long memory of the sea ice concentration, polar vortex, and the downward propagation of stratospheric anomalies. However, the response of mid-to-high latitude weather and climate to the changes in Arctic sea ice is still subject to great uncertainty (Screen et al 2014, Overland et al 2015, Perlwitz et al 2015, Semenov and Latif 2015, Wu et al 2015, Overland et al 2016, Wu 2018 Ayarzagüena and Screen (2016) demonstrated that Arctic sea ice loss is also accompanied by a reduction in the strength of cold-air outbreaks in the midlatitudes. Further, numerous modeling studies tried to separate the impact of sea ice and atmospheric internal variability on Eurasian cooling. Based on atmosphere-only and coupled atmosphere-ocean simulations, they have shown that cold Eurasian winters are attributed to atmospheric internal variability, rather than a change in sea ice forcing (McCusker et al 2016, Sun et al 2016, Collow et al 2018, indicating that external forcings only contribute to the chance of cold winters appearing with a higher probability density function. In addition, some studies indicate that the link between Arctic sea ice and midlatitude weather and climate can be modulated by tropical processes (Baxter et al 2019, Warner et al 2019. Since the stratospheric pathway appears to be an important link between the BKS sea ice and midlatitude climate, we envision that the uncertainty of stratospheric process may be responsible for the nonstationary relationship between sea ice loss and cold winters. SPV is an essential system linking the Arctic stratosphere and troposphere. Studies have shown that the SPV can be largely modulated by El Niño-Southern Oscillation (ENSO) and quasi-biennial oscillation (QBO) besides Arctic sea ice (Ren et al 2012, Calvo et al 2017, Ren et al 2017, Zhou et al 2018, Zhang et al 2020, Rao et al 2021, Kumar et al 2022. Garfinkel et al (2010) showed that QBO and Aleutian low associated with ENSO can explain ∼39.6% of polar vortex variability during winter in the reanalysis record. Specifically, during the El Niño (ELNINO) years, the associated Pacific North America pattern is accompanied by an intensified Aleutian low-pressure system, which subsequently increases the upward planetary wave and weakens the SPV before reflecting back to the surface (Ineson and Scaife 2009). As expected from dynamical considerations, Domeisen et al ( Richter et al (2020) found that the number of models that are able to simulate the QBO has increased from CMIP5 to CMIP6. This motivates us to explore the role of ENSO and QBO in the Eurasian surface air temperature (SAT) affected by BKS sea ice through stratospheric processes.
The goal of this work is to consider whether ENSO and QBO interfere with links between the BKS sea ice loss and Eurasian winter climate. Section 2 introduces the data and models used in this study. The responses of Eurasian wintertime SAT forced by sea ice (ICE), ENSO, and QBO, and the possible mechanisms are shown in section 3. Summary and conclusions are presented in section 4.

Model
The model utilized in this study is the decadal prediction system version 4 (DePreSys4) developed and deployed at the United Kingdom Met Office. This system produced the first initialized short-term climate prediction in 2007 (Smith et al 2007) but now uses the Hadley Centre Global Environmental Model version 3 as described in Dunstone et al (2016). Natural variability and human influences are simulated in this system using CMIP6 forcing datasets and it consists of four components: atmosphere, ocean, land, and cryosphere (land ice and sea ice). The horizontal resolution of atmosphere data used in this paper is 0.556 degrees of latitude by 0.833 degrees of longitude with 36 levels extending from the surface to 0.03 hPa. DePreSys4 hindcast consists of ten ensemble members each starting from different ocean analyses sample uncertainties in the initial conditions. The DePreSys4 hindcasts start each November from 1960 to 2019 and run for 10 years and 4 months. Several indices are used in this paper. Sea ice extent (SIE) index is defined as the total marine area in which the ice concentration is at least 15% over the BKS region (70 • N-82 • N, 20 • E-90 • E). The strength of the SPV is defined as the zonal wind at 60 • N and 10 hPa (Hardiman et al 2020). Strong and weak SPV winters with low SIE (the normalized SIE less than −0.5) are selected based on the threshold of ±0.5 standard deviation of the SPV strength index. According to this criterion, observed weak SPV/low SIE winters are 1965/1966, 1972/1973, 1984/1985, 2000/2001/2019, whereas strong SPV/low SIE winters are 1961/1962, 1983/1984, 1985/1986, and 1995/1996. As previous studies (Holton and Tan 1980, Son et al 2017, the QBO index is calculated by zonal mean zonal wind at 50 hPa averaged over 10 • S-10 • N. The Niño 3.4 index, which is the area-averaged SST over 5 • S-5 • N and 170 • W-120 • W, is used to represent the ENSO signal. Following Gong et al (2001), the regional averaged sea level pressure in the midlatitudes of East Asia (40 • N-60 • N, 70 E-120 • E) is used to represent the intensity of Siberian high. As presented in Wang and He (2012), the sea level pressure averaged in the region of 155 • E-130 • W and 30 • N-70 • N is defined as the strength of Aleutian low.

Observational data and index definitions
Composite analysis is the main method utilized in this paper. To obtain more samples for each composite, we superimposed the atmospheric response of opposite phases of ENSO and QBO, assuming that the atmospheric response to opposite ENSO and QBO phases can be offset. For example, the summation of the results of ENSO in positive and negative phases is considered as that for neutral ENSO (NENSO), and the summation of QBO in positive and negative phase is considered as neutral QBO (NQBO). In addition, the weights used for the summation are defined according to their polar vortex strengths, i.e. they are weighted by polar vortex strengths during 1960-2019. Taking QBO as an example, α is the ratio of the pole vortex intensity in the EQBO years to that in the WQBO years. The atmospheric response during the WQBO years is multiplied by α and then adds the atmospheric response during the EQBO years to offset the atmospheric response of QBO as much as possible, which is considered as the result of NQBO. Analogously, in this way, the LANINA-NQBO-LICE events and the ELNINO-NQBO-LICE events are included to increase the number of NENSO-NQBO-LICE events, and the NENSO-WQBO-LICE and NENSO-EQBO-LICE events are included to increase the number of NENSO-NQBO-LICE events. Figure 1 shows the subsequent zonal mean zonal wind and temperature in winter when the preceding autumn BKS sea ice is anomalously low based on the observational record. Under low BKS sea ice conditions, Eurasian winter SAT is abnormally cold when the SPV is weak and is abnormally warm when the SPV is strong. This indicates that if the link between Arctic sea ice and Eurasian SAT is through the SPV then it can be masked by other variability, which motivates us to investigate the role of ENSO and QBO due to their strong connections with SPV variability.

Results
Time series of autumn SIE and wintertime ENSO and QBO indexes and their lead-lag correlation are presented in figure S1. Consistent with previous studies, a clear downward trend is found in the autumn SIE index (e.g. Chen et al 2021, Docquier and Koenigk 2021, Yang et al 2022, while no significant trend can be seen in ENSO and QBO. All three factors exhibit significant interannual variability. Garfinkel et al (2010) pointed out that the effects of ENSO and QBO on the polar vortex were independent of each other. The lead-lag correlation between autumn SIE and wintertime ENSO and QBO presented in figure  S1(b) also indicates that the ENSO and QBO are not correlated with autumn SIE during 1960-2019.
Since only a few observed events can be found in each combination of ENSO and QBO phases under low BKS SIE conditions from the observational record, the DePreSys4 hindcasts with ten ensemble members are used to minimize the impact of internal variability in this paper. All ensemble members can well reproduce the observed Northern Hemisphere SAT climatological spatial distribution with all correlation coefficients above 0.99 and relative amplitude close to 1.0 (figure S2). Previously, Andrews et al (2019Andrews et al ( , 2020 pointed out that this model has a good performance in reproducing the QBO and Arctic sea ice. In addition, simulated QBO-Arctic Oscillation teleconnections are similar to that shown in It is necessary to investigate the relationship between the Arctic stratospheric circulation and ENSO, QBO, and ICE beforehand, as shown in figure 2. The Arctic polar vortex is negatively linearly related to ENSO, positively linearly related to QBO, and nonlinearly related to ICE. Previously, some studies pointed out the asymmetry and the nonlinearity of the influences of ENSO on the northern winter stratosphere Ren 2016a, 2016b). In this study, we discuss the role of ENSO on the premise of low Arctic sea ice, and in this situation, the nonlinearity of ENSO is relatively weaker. To explore the potential mechanism underlying the nonstationary relationship between BKS sea ice and Eurasian SAT, the wintertime zonal wind responses for all combinations are shown in figure 3. In the case of low sea ice (LICE) in the BKS, three combinations are investigated, and they show a significant downward propagation of the stratospheric circulation anomalies forced by ENSO, QBO, and ICE (figures 4 and S3) and the polar vortex anomalies of the other combinations are too weak or cannot propagate significantly to the surface (figure S3). The total occurrence of the three combinations is 25% under low BKS SIE conditions. Specifically, the occurrence is 12.5% for the first combination (LANINA-EQBO-LICE) and 6.25% for the latter two combinations. The Arctic SPV appears to be stronger during the LANINA-EQBO-LICE years ( figure 4(a)), and weaker during the NENSO-NQBO-LICE and ELNINO-NQBO-LICE years (figures 4(b) and (c)). The enhanced SPV in figure 4(a) may be mainly attributed to the LANINA since EQBO and LICE favor a weakening of the polar vortex and the weakened SPV in figure 4(b) is primarily a result of the reduction of BKS sea ice, as ENSO and QBO are in the neutral states. While the weakened SPV illustrated in figure 4(c) is ascribed to the joint effect of ELNINO and the reduction of BKS sea ice, as QBO is in the neutral state. In the case of high BKS sea ice, almost opposite responses are observed ( figure S4). Thus, only LICE conditions are discussed in this paper.
Previous studies have shown that the Arctic SPV changes forced by ENSO, QBO, and ICE are largely caused by dynamical processes (planetary waves activity) (e.g. Garfinkel et al 2010, Zhang et al 2016Zhang et al , 2018b. The responses of Eliassen-Palm (EP) fluxes for the three combinations are thus depicted in figure 5. The planetary wave activity is weakened in the LANINA-EQBO-LICE years (figures 5(a) and (d)), indicating that LANINA events strengthen the Arctic SPV by suppressing upward planetary wave activity. Consistent with previous studies, the BKS sea ice loss in autumn could excite anomalous upward planetary waves and weaken the SPV (figures 5(b) and (e)). Changes in upward planetary waves in the ELNINO-NQBO-LICE years (figures 5(c) and (f)) are also enhanced as in the NENSO-NQBO-LICE years, which eventually leads to a weakened SPV. Figure S5 shows the longitude-latitude maps of wavenumbers 1 and 2 of zonal wind composites at 10 hPa. The weakened SPV in the NENSO-NQBO-LICE and ELNINO-NQBO-LICE years (figures 4(b) and (c)) is attributable to the enhanced vertically propagating stationary waves-1 and 2. Conversely, the stronger SPV during the LANINA-EQBO-LICE (figure 4(a)) years is dominated by reduced upward    propagating stationary wave of wavenumber-2 and possibly changes in the transient waves.
Studies have shown that stratospheric circulation anomalies are closely related to the East Asian winter monsoon (EAWM) (Ma et al 2021, Lu et al 2022, which is one of the main factors in driving wintertime SAT variations over many regions of Eurasia (Chen et al 2005, Sung et al 2010. We further investigate the tropospheric responses to the stratospheric circulation anomalies. A positive North Atlantic Oscillation (NAO) signal appears during the LANINA-EQBO-LICE years (figure 6(a)), while opposite circulation responses are displayed in the latter two combinations with negative NAO signals over the North Atlantic sector (figures 6(b) and (c)). Usually, cold Eurasian winters coincide with a negative NAO and warm winters with a positive NAO. This suggests that the circulation changes shown in the first combination are favorable for Eurasian warming, while the circulation changes shown in the latter two combinations are in favor of Eurasian cooling (Hirschi andSinha 2007, Xie et al 2019). In addition, Miao and Wang (2020) pointed out that Siberian high and Aleutian low are two important components of the EAWM system. As displayed in figure 6(a), during the LANINA-EQBO-LICE years, weakened Siberian high and Aleutian low indicate that it is conducive to Eurasian warming with this anomalous circulation (figure 6(d)). Opposite responses are found in the latter two combinations (figures 6(e) and (f)).
The responses of Northern Hemisphere wintertime SAT for the three combinations are shown in figure 7 (see figure S6 for all eight combinations). The temperature anomalies are characterized  by warm Eurasia pattern during the LANINA-EQBO-LICE years (figure 7(a)) and cold Eurasia pattern during the NENSO-NQBO-LICE ( figure 7(b)) and ELNINO-NQBO-LICE years (figure 7(c)). Mori et al (2014) argued that the warm Arctic-cold Eurasia pattern is a direct atmospheric response to the decline of the BKS sea ice, which is consistent with our results shown in parts of Eurasia in figure 7(b). The cold Eurasia shown in figure 7(c) is the results of ELNINO superimposed on the LICE (in contrast to the LICE-only forcing in figure 7(b)). These Eurasian SAT anomalies are consistent with the changes in Arctic SPV and surface circulation related to the three combinations.

Conclusions
This study provides observational and modeling evidence that the BKS sea ice reduction does not always lead to a cold Eurasian winter. Using reanalysis and DePreSys4 hindcasts, we evaluated the role of ENSO and QBO in modulating the linkage between autumn BKS sea ice and winter Eurasian SAT. Results show that the stratospheric polar vortex, as a bridge linking BKS sea ice and Eurasian winter climate, can be significantly influenced by ENSO and QBO. Both cold and warm Eurasian winters can occur when BKS sea ice is abnormally low. Specifically, SAT anomalies show a warm Eurasia pattern during the LANINA-EQBO-LICE years and a cold Eurasia pattern during the NENSO-NQBO-LICE and ELNINO-NQBO-LICE years. The physical processes responsible for the modulation by ENSO and QBO on the ICE-SAT relationship can be attributable to the SPV changes associated with ENSO, QBO, and ICE. During the LANINA-EQBO-LICE years, the polar vortex is strengthened by suppressed upward planetary wave activity and couples downward with the tropospheric circulation, leading to a positive NAO, weaker Siberian high and Aleutian low in winter. These conditions are conducive to warm Eurasian winters. Opposite responses of the stratospheric and tropospheric circulation are found during the NENSO-NQBO-LICE and ELNINO-NQBO-LICE years, leading to cold Eurasian winters.

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