North American cold events following sudden stratospheric warming in the presence of low Barents-Kara Sea sea ice

The ERA-Interim reanalysis on a resolution of 1.5° × 1.5° grid in 1979–2019 is used [32]. The sea ice concentration (SIC) is 25 km × 25 km passive microwave satellite observation, Nimbus-7 SMMR and DMSP SSM/I-SSMIS, with the NASA team algorithm Version 1 from NSIDC [33]. We define the late autumn BKS SIC index using the sea ice area in the region of 15°–100°E and 70°–82°N during November (figure S1) (available online at https://stacks.iop.org/ERL/15/124017/mmedia. The reason that we focus on the BKS region is that the sea ice variability and associated warming over this area is more effective in influencing the stratosphere-troposphere coupling than other regions over the Arctic [3, 34]. SSW in this study is simply defined as the reversal of 60°–90°N zonal mean zonal wind at 10 hPa during November to March [35]. A discussion of the choice of the SSW definition can be seen in Supplementary Material (SM). The 1st day with zonal wind reversal is identified as the central day (day 0) listed in table S1. 34 SSWs are identified in 1980–2019 for composite analysis. Since the SSW sample size is limited in observations, all SSW cases are simply divided into two groups conditioned on the BKS SIC index lower and greater than 0 (17vs17), referred to as LoSIC and HiSIC respectively. The departure from the corresponding (HiSIC or LoSIC) daily climatology is extracted for analysis, but the results are insensitive to the choice of climatology and similar results are found with 1980–2019 climatology (not shown). The correlation coefficient between BKS SIC index and Nino3.4 index is small and not significant, indicating the results in this study are insensitive to ENSO. In addition, since this study focuses on the current climate, the grouping of low and high SIC winters is on the basis of current climatology defined by WMO, i.e. mean of 1981–2010. Similar conclusions are found with 2–9 year bandpass filtered observations (with emphasis on the interannual variability) and with the first 30 years only (figures S1 and S2).

A stratosphere-resolving atmospheric circulation model, SC-WACCM4, the atmosphere component of CESM 1.2, with specified chemistry rather than the full interactive chemistry version WACCM4, is used [36]. The removal of interactive chemistry much reduces the computational cost without changing the climatology and variability of the atmospheric circulation in the stratosphere and troposphere, which makes SC-WACCM4 suitable for studies of stratosphere-troposphere dynamical coupling [36]. The simulated frequency of SSWs is about 0.66 per winter and is, therefore close to the observed value of 0.84 (figure S9). The model has 1.9° × 2.5° horizontal resolution and 66 vertical levels with the model lid at 140 km.

In experiments, SST and SIC from the coupled simulations with CESM1-WACCM4 for CMIP5 are prescribed as the model's surface boundary condition. In the control run (hereafter referred to as Ctrl), the repeating climatological seasonal cycle of SST and SIC, from the historical simulations (average of 7 ensembles) and averaged during 1980–1999, is prescribed. In the perturbation run, referred to as BKSIC, the settings are identical to that of the Ctrl but the SIC in the BKS region is replaced by that in the CMIP5 RCP8.5 outputs (average of four ensembles) averaged during 2080–2099. The SST in the BKS region is also replaced with RCP8.5 SST in the open water areas that used to be covered by sea ice in Ctrl run. Although the projected sea ice changes were used in this study, the autumn sea ice difference between the Ctrl and the BKSIC run (−20%) is comparable to the forcing magnitude (−18%) obtained from HadISST in [37] (see supplementary discussion in [30]) due to the underestimation of sea ice changes in climate models [38, 39]. The Ctrl run is regarded as the higher sea ice condition compared to the BKSIC run. There are two sets of 274 winters and 181/217 SSWs in Ctrl and BKSIC runs for analysis, respectively. Details of experiments and supplementary nudging experiments to support the conclusions can be seen in SM.

Statistical significance is assessed by using 10 000 times bootstrap resampling with replacement. The same method is employed to calculate the confidence interval in figure S9. The two-tailed two-sample Kolmogorov–Smirnov (KS) test is used to compare the probability density function (PDF) of surface air temperature anomaly (SATa) following SSW to that of the whole winters in figure 3. The null hypothesis, that the distributions of two samples are the same, is rejected when P ≤ 0.01. Two-dimensional kernel density estimate is employed to explore the distribution changes of the bivariate density of North American cold status and stratospheric ridge anomaly in figure 5. The 2-D density estimate is based on 10 000 times bootstrap resampling with replacement for the mean of 17/17 samples (the sample sizes in LoSIC/HiSIC) from the simulated 217/181 events in BKSIC/Ctrl runs. We also compare the distributions in each dimension using KS test. Wave activity flux (WAF) is employed to diagnose the wave propagation anomalies in section 4.2 (see details in SM).

We begin by showing the importance of the SIC in the BKS region for the influence of SSWs on the cold events over North America. Figures 1(a)–(b) compares the SAT and near-surface wind anomalies averaged over days 20–40 following the onset of SSWs, for the winters in which the preceding late autumn SIC over the BKS region is lower and higher than normal, respectively, in observations. The reason we focus on days 20–40 is that the SAT anomaly difference between these two cases is distinct during this period (will be discussed in figures 3 and 4). In contrast to very weak SAT anomalies in the winters with higher BKS SIC (HiSIC; figure 1(b)), a large decrease in SAT is found over Canada and central and midwestern US, 20–40 d after the onset of SSWs in the winters with lower BKS SIC (LoSIC; figure 1(a)). These SAT anomalies are significantly different between the two sea ice conditions (figure 1(c)). These surface cold anomalies in lower BKS SIC winters are consistent with cold air advection by anomalous northeasterlies blowing from the Arctic region to North America. The results remain robust if the lower and higher SIC winters are defined with 2–9 year bandpass filtered BKS SIC time series that focus on the interannual variability and with the removal of the last 10 years in which the sea ice rapidly declines (figures S1–S2).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The identification of causality among the BKS sea ice, SSWs and North American cold events in observations is challenging, since midlatitude circulation anomalies could also possibly contribute to changes in BKS sea ice [11]. Thus, we have used an AGCM with a well-resolved stratosphere, and conducted numerical model experiments with and without prescribed BKS sea ice reduction to demonstrate the critical role of BKS sea ice conditions. The model results support the observed connection between the SSWs and North American SATa (figures 1(d)–(f)). Comparing the model results with and without lower BKS sea ice (referred to as BKSIC run and Ctrl run), an enhancement of cold anomalies is found over North America in the former.

During the latest 2019 cold event, the record-breaking cold temperature over the midwestern U.S. was observed on Jan. 30–31, 2019 [40], lagging the SSW that occurred on 30 December 2018 by about one month. The preceding early winter BKS SIC anomaly was −14%, the fifth lowest value in the past 40 years (figure S1(a)). As shown in figure 1(g), the North American SAT anomaly pattern during days 20–40 in the 2019 case highly resembles that of the LoSIC composite as shown in figure 1(a) except that the 2019 event exhibits a larger amplitude than the composite likely from the internal variability of the atmosphere.

In addition, the impact of BKS sea ice reduction on cold events is also reflected in the number of extreme cold days after SSWs (figure 2). Here we define an extreme cold day as a day during which the SAT is at least 1 standard deviation (σ) below its climatology. Compared to 2–3 extreme cold days in the higher sea ice cases (figure 2(b)), more than five extreme cold days occur across the most part of Canada and Midwest U.S. in days 20–40 following the SSWs in the lower BKS sea ice cases (figure 2(a)). The increase of North American cold days is statistically significant (figure 2(c)) and also holds for thresholds of 1.5σ and 2σ (not shown). A similar increase of extreme cold days can be found in model simulations (figures 2(d)–(f)) with consistent spatial patterns, although the increase is smaller.

Figures 3(a)–(b) displays the PDF of area averaged daily SAT anomaly and its temporal evolution after the onset of SSWs under different sea ice conditions. In the winters with lower BKS sea ice, there is a shift (P < 0.01) in the PDF toward colder SAT during days 10–50 after SSWs (solid blue) as compared to that for all winter days (dashed blue) and to that for higher BKS sea ice SSWs (solid red). As the largest surface cooling over North America takes place 20–40 d after the onset of SSWs and its differences to that in higher sea ice conditions are significant (figure 3(b)), it is promising that SSWs combined with BKS sea ice conditions can be used as a predictor for North American cold events at the sub-seasonal timescale. It is worthy to note that the aim of this study is to explore the potential subseasonal predictability of BKS sea ice on SSW related weather extremes and thus only the SATa during the weeks following SSW are discussed here, and the changes in these SATa may have little impact on the whole winter SATa distribution over North America (contrast the dashed red and blue curves in figure 3(a)). Although the modeled PDF shapes are slightly different from the observations due to the distinct sample size, a shift of PDF toward the colder flank of the SATa after SSWs in model experiments is in agreement with observations (figures 3(c) and (d)). While the magnitude of surface cooling is slightly weaker in the model experiments than that in observations, the broad consistency in the spatial patterns and timing of the SAT anomalies between model simulations and observations confirms the critical role of the lower BKS sea ice on the SSW-induced cold events over North America.

Moreover, the relative contribution of stratospheric versus tropospheric processes to anomalous surface cooling is assessed by supplementary experiments, in which the stratospheric and tropospheric pathways are separated by isolating the circulation response in the stratosphere and troposphere, respectively (see details in SM). The stratospheric pathway experiment, in which no sea ice reduction is prescribed but the stratospheric circulation is nudged to that of the lower sea ice experiment (BKSIC run), also shows significant anomalous surface cooling over North America following the onset of SSWs (figure S3(a)). In contrast, the SATa following SSWs is nearly absent and even a warm anomaly over the western Canada is found in the tropospheric pathway experiment, in which the same sea ice reduction is imposed in the model but the stratospheric processes are suppressed by nudging the stratospheric circulation to that of the control simulation (figure S3(b)). This further confirms that the cold anomalies identified in figure 1 are largely the downward influence of SSWs in the presence of lower BKS sea ice rather than the direct tropospheric circulation response to sea ice anomaly.

How can SSWs influence the SAT over North America with a lead time of about one month? The Northern Annular Mode (NAM) index, commonly used to explain the downward influence of SSWs, shows that the downward propagation from the stratosphere to the surface is enhanced under the lower BKS sea ice condition (figure S4). However, it is hard to employ the downward propagation of anomalous zonal mean zonal flow to explain the regional temperature signals over North America. Thus, an alternative explanation is provided here from the perspective of the regional effect of planetary waves.

We first present the mid-tropospheric circulation pattern following SSW events associated with lower BKS SIC (figure 4). In the mid-troposphere, there is a ridge over the Gulf of Alaska in observations in winter climatology (blue contours in figure 4). Under lower SIC conditions in observations, an extensive positive 500 hPa geopotential height (Z) anomaly is found over the Bering Sea to Gulf of Alaska (BSGA) (150°E–150°W), manifesting a reinforced ridge in this case (figure 4(a)). The resulting enhanced northwesterly wind to the east of the ridge brings more cold air from the Arctic to North America. With the intensified southward sinking of cold air mass, near-surface northeasterlies and associated cold air advection inland are developed downstream of the intensified mid-level BSGA ridge (figure 1). On the contrary, SSWs in higher BKS sea ice conditions are accompanied by mostly negative Z anomalies over the BSGA (figure 4(b)). The most pronounced feature in the differences between the two composites is a high-pressure anomaly over the BSGA, favoring a downstream northerly anomaly and cold air outbreaks (figure 4(c)). Similar results in the mid-tropospheric circulation are simulated in the AGCM experiments with and without BKS sea ice loss, especially for the BSGA ridge anomaly, despite that the Z anomalies are only found over the BSGA while the observed Z anomalies occupy the whole northern North Pacific (figures 4(d)–(f)). Thus, the deepened mid-tropospheric BSGA ridge is the key to the anomalous surface cooling in response to SSWs under lower BKS SIC.

The mid-tropospheric BSGA ridge anomaly is associated with the planetary WAF that first displays an enhanced upward propagation into the stratosphere around the SSW onset and subsequently turns eastward (i.e. suppressed upward propagating waves) in the troposphere over the North Pacific (120°E–120°W) and converges over 180°–150°W favoring the ridging over the BSGA (figure S5), which is consistent with the planetary wave characteristics following stratospheric warming events in previous studies [41, 42] (see more discussion in SM). Accordingly, a coherent vertical wave structure from the mid-troposphere to the stratosphere can be found in Z differences between low and high sea ice conditions over the North Pacific to North America (120°E–60°W) (figures S6(a) and (b)). Moreover, the daily evolution of the Z anomaly averaged over the BSGA region, coinciding with a planetary wave-2 ridge anomaly that starts descending from the lower stratosphere (50 hPa) around day 16 to the mid-troposphere around day 24 (figures S6(c) and (d)), shows a persistent ridge anomaly in the troposphere starting from approximately day 24 after the onset of SSW events, confirming the downward coupling timescale revealed in figure S4. The downward descent of the wave-2 ridge anomaly from the lower stratosphere demonstrates the key role of the planetary wave-2 component in the establishment of the mid-tropospheric BSGA ridge and associated North American surface cooling to SSWs.

These lower stratospheric anomalies arise from the slow recovery of the stratospheric polar vortex after SSW and associated upward propagating planetary wave activity. Figure S7 depicts the evolution of the stratospheric polar vortex anomalies and associated wave-2 component in the lower stratosphere. Figure 5 is the same but for the average of days 20 to 40. It can be seen that, under low BKS SIC conditions, the stratospheric polar vortex is persistently distorted with two ridges of wave-2 anchored over the Eastern Siberia and Greenland and two troughs over Alaska and Ural Mountain for about 40 d after the onset of SSWs, indicating a persistent stretching of the polar vortex, and this is true in both observations and experiments (figures 5(a), (d) and S7(a), (d)). However, this is not the case under high BKS SIC conditions (figures 5(b), (e) and S7(b), (e)), instead the wave-2 anomaly pattern shifts in longitude and decays with time. Therefore, following the SSW in low BKS SIC winters, the persistent ridge of wave-2 anomaly over eastern Siberia extends from the stratosphere to the troposphere, enhancing the mid-tropospheric ridge over BSGA that favors downstream cold air outbreaks. While the maximal wave-2 differences between the low and high BKS SIC conditions in AGCM are slightly shifted to the east by about 30 degrees compared to observations (figures 5(c) and (f)), the basic features of the wave-2 vertical structures are similar (figure S6).

Given the small sample size of SSWs in observations (17/17), we further demonstrate the robustness of observational results in the view of probability distribution based on a stochastic resampling of 17 events from the abundant samples in model simulations (217/181). Figure 6 shows the 2-D kernel density estimate for North American SAT anomaly and stratospheric wave-2 anomaly over the northern North Pacific (green box in figures 5(c) and (f)) following the SSWs. The wide range of probability density indicates that the mean of 17 random samples could have a large spread due to the internal variability of the atmosphere. However, it is clear that there are significant changes in the probability distribution between lower and higher BKS SIC. The density distribution shifts to an enhancement of the stratospheric wave-2 ridge and colder North American SAT under the lower BKS sea ice conditions (figure 6(a)). This shift between the two resampling groups is similar to that of the means of two observational groups. Similar conclusions are also found in the extreme cold days, with more cold days following SSWs due to the enhancement of the persistent wave-2 ridge under lower BKS sea ice condition (figure 6(b)).

Zoom In Zoom Out Reset image size

A remaining question is how the BKS sea ice reduction changes the stratospheric polar vortex geometry during SSWs and thereafter its downward influence. It has been recognized that the sea ice decline over the BKS region can effectively enhance the upward planetary-scale wave propagation into the stratosphere via linear constructive interference and thereby modify the stratospheric polar vortex [43, 44]. Heat flux ($v^{^{\prime}}T^{^{\prime}}$) activity is generally employed to measure the wave activity during SSWs. Figure S8 shows the anomalies of 100hPa heat flux by wave-1 and wave-2 averaged during days −8 to −1 prior to SSW onset, the embryonic stage of the SSW during which upward coupling is prevailing. For the wave-1 heat flux component prior to SSW onset, a similar amplified heat flux is found in both observations and AGCM experiments regardless of BKS sea ice conditions (figures S8(a)–(f)).

The differences are, however, notable in the wave-2 heat flux anomalies. Under lower BKS sea ice conditions, an enhancement of wave-2 heat flux can be seen at high latitudes (figures S8(g), (j)). In comparison, in the absence of BKS sea ice reduction, the wave-2 heat flux anomalies are weaker (figures S8(h), (i), (k) and (l)). The significantly intensified wave-2 heat flux anomalies with lower BKS sea ice highlight the important role of wave-2 component, favoring polar vortex stretching, in the evolution of the polar vortex during SSW, which is also consistent with previous section and previous finding that the preceding enhanced wave-2 activity from the troposphere favors the split SSWs (the extreme case of polar vortex stretching) [45–47].