Anthropogenic global warming is projected to increase some weather extremes—for example, more heat waves and heavy precipitation events (
1,
2)—but not severe winter weather such as cold air outbreaks and heavy snowfalls (
3,
4). Yet, contrary to global climate model (GCM) projections, recent weather extremes have included an increase in cold air outbreaks and/or heavy snowfalls across the Northern Hemisphere (NH) since 1990 up to the recent past (
3,
5–
8). The most recent example of extreme winter weather was the anomalous cold weather of January and February 2021 in Asia (
9), Europe (
10,
11), and especially the United States (US). The US Southern Plains cold wave of February 2021 may be exceptional in the observational record for the region based on the aggregate severity of the cold intensity, cold duration, and widespread disruptive snowfall (
12,
13). The collapse of the Texas energy infrastructure could make it the state’s costliest natural disaster, even more so than previous hurricanes (
14) and at least twice as costly as the entire record-breaking North Atlantic 2020 hurricane season (
15). This event has reignited the debate whether climate change contributes to more severe winter weather (
16).
One of the more robust signatures of global warming is accelerated Arctic warming, known as Arctic amplification (AA) (
17), which has been evident since the 1990s (
5). AA is both a response to and accelerator of Arctic sea ice decline, with the greatest losses observed in the Barents–Kara and Chukchi–Bering Seas in the fall and winter (
8). AA has also coincided with increasing snow fall and snow cover at high latitudes, including across Eurasia during October through January (
18–
21), in part due to the decline in sea ice, which increases moisture availability in the Arctic (
18,
21). In figs. S1 and S2, we show that Eurasian October snow cover has increased, whereas fall Arctic sea ice has decreased over the satellite and AA periods.
A hypothesis that has received much recent attention is that AA is driving winter mid-latitude cooling (
7,
22,
23). One theory that links less sea ice and/or more Eurasian snow cover to severe winter weather in the mid-latitudes involves a pathway through the SPV (
5,
24). Less sea ice and more snow cover increase the probability of a stronger Siberian high-pressure ridge and upward atmospheric wave energy flux. Increased wave flux from the troposphere to the stratosphere can result in a sudden stratospheric warming (SSW) characterized by large rises in polar geopotential heights centered over the North Pole, often followed by an increase in NH severe winter weather (
3). However, whether AA can result in more severe winter weather, and how, is a matter of active debate (
25–
29).
Kretschmer
et al. (
30) used a machine learning (ML) technique to demonstrate that the weakest SPV state (SSWs) are increasing in frequency, whereas the strongest SPV states are decreasing in January and February over the period of AA. In a follow-up study, the same ML technique identified a less-known SPV disruption whereby the SPV is stretched (
31) (P4 in
Fig. 1) as opposed to the well-known SSW (P5 in
Fig. 1). One important difference between SSWs and SPV stretching is that the vertical component of atmospheric wave energy known as wave activity flux (WAF
z) or Eliassen Palm (EP) flux preceding SSWs converges in the polar stratosphere, resulting in rapid warming and rising of geopotential heights in the stratosphere, whereas during SPV stretching, WAF
z is reflected from the SPV back into the troposphere (and is therefore also referred to as “reflective” events) (
32–
34), where it amplifies the climatological pressure ridge and trough across North America. A second important difference is that North American cold spells tend to be more extreme following SPV stretching events (
31).
Here we expand the analysis of SPV variability and its tropospheric links to include fall and early winter, which identifies trends and links to surface snow and ice changes in the high latitudes that are consistent with surface forcing of the SPV variability. We then use modeling experiments to establish a causal link between the surface changes and the SPV variability and its tropospheric links and, therefore, between AA and extreme mid-latitude winter weather.
We extend the ML technique of Kretschmer
et al. (
30,
31) to analyze SPV variability in fall and early winter (October through December) over the reanalysis period (1980 through early 2021), shown in
Fig. 1 [in fig. S3, we update the January and February analysis of Kretschmer
et al. (
31)]. The first two clusters (P1 and P2 in
Fig. 1) show a stronger-than-normal SPV (i.e., lower geopotential heights), and the last two (P4 and P5) show a weaker-than-normal SPV (i.e., higher geopotential heights in the polar stratosphere). The stronger SPV states are experiencing a statistically significant decreasing trend in frequency, whereas the weaker SPV states are experiencing a statistically significant increasing trend, not only for January and February but also for the preceding months of October through December (fig. S4). Here we show that SPV stretching disruptions (P4) have a statistically significant increasing trend in both fall and winter, even more so than SSWs over the reanalysis period, and are increasing for the months October through February (fig. S5).
The concurrent surface temperature anomalies are presented in
Fig. 1E. The two strong SPV states exhibit a cold Arctic–warm continent pattern and the two weak SPV states exhibit a warm Arctic–cold continent pattern. Specifically, for the two weak states, SSWs (P5) are related to warming around Greenland and Baffin Bay, whereas SPV stretching (P4) is related to Arctic warming focused in the Barents–Kara and Chukchi–Bering Seas. In the mid-latitudes, both SSWs and SPV stretching are associated with relatively cold temperatures across Northern Europe, Northern and Eastern Asia, and North America; however, during SPV stretching, North American cold temperatures are more widespread and shifted eastward. It has already been shown that SSWs are contributing to an observed cooling trend across northern Eurasia for the two winter months of January and February (
30,
35), but our analysis suggests that an increasing number of SPV stretching events are a cooling influence across North America.
The tropospheric precursor pattern to SSWs has been previously identified as consisting of relatively high pressure across Northern Europe and the Urals coupled with relatively low pressure across East Asia into the northern North Pacific (
30,
36,
37). This anomalous dipole projects onto the climatological standing wave-1 of the NH (
36–
38), and through constructive interference gives rise to an enhanced WAF
z from the troposphere to the stratosphere. Although anomalous vertical wave energy flux has also been shown to precede SPV stretching disruptions (
31), other precursor features in the tropospheric circulation to these events have not yet been examined.
The precursor patterns to SPV stretching events are shown in
Fig. 2. At 100 hPa in the lower stratosphere (
Fig. 2A), there are regional ridging and positive height anomalies, which begin over the North Atlantic and then migrate to the Gulf of Alaska, Alaska, Chukchi–Barents–Kara Seas, and the Urals. The ridging amplifies shortly before and up until the time of the event (day 0). In addition, prior to SPV stretching, there is a WAF
z dipole (
Fig. 2B) with positive anomalies in Eastern Siberia and negative anomalies in northwest North America (similar to Kretschmer
et al.) (
31). Climatologically, WAF
z is upward over Siberia, reflected in the stratosphere, and then downward over Canada (
31). The low and mid-troposphere precursor patterns (
Fig. 2, C and D) project onto the climatological NH standing wave-2 (
36,
37). Finally, the observed precursors in surface temperature (
Fig. 2E) begin as positive anomalies in the Arctic focused near Greenland and to a lesser extent in the Chukchi Sea. However, shortly before and at the time of the event, two regions of positive anomalies emerge—one over the North Atlantic side of the Arctic and a second over Alaska and Chukchi–Bering Seas region of even greater amplitude and extent, whereas negative temperature anomalies emerge first in Siberia but then also develop over North America.
Could climate change have contributed to the observed increasing trends in the SPV stretching events that force cold to extreme cold in North America and East Asia? In
Fig. 3, we show the trends (1980–2021) in the late fall and winter months of the same variables used to identify the precursors to SPV stretching events. Trends in lower stratospheric WAF
z exhibit the same dipole associated with SPV stretching events, with positive trends over Siberia and negative trends over northwest North America (10 to 0 days previous;
Fig. 2B). At the surface and 500-hPa mid-to-high latitudes, increasing trends in pressure and geopotential heights are centered on the Barents–Kara Seas and Urals with a secondary maximum in the Gulf of Alaska and Alaska region, which matches what is observed in the precursors to SPV stretching (15 to 0 days previous;
Fig. 2, C and D). Surface temperatures have been rising most strongly in the Arctic with two maximum centers, one in the North Atlantic side of the Arctic and the other in the Chukchi–Bering Seas, similar to the observed Arctic warming prior to SPV stretching events (15 to 0 days previous;
Fig. 2E). There is some weak cooling in Asia. Projections of the seasonal and monthly (October through February) SPV stretching precursors (
Fig. 3A and fig. S6) onto the trends are statistically significant (figs. S7 and S8), especially relatively warm surface temperatures in the Barents–Kara and Chukchi–Bering Seas, and ridging and high pressure at 500 hPa and at the surface in the Barents–Kara Seas and Urals and to a lesser extent, the northern North Pacific.
We complete our observational analysis by correlating leading Eurasian snow cover and Arctic sea ice concentration with the lagging atmospheric fields analyzed for trends (
Fig. 3, B and C; with all time series detrended). The correlations with snow cover (
Fig. 3B) most closely resemble SPV stretching precursors (
Fig. 2), with ridging centered near Alaska and downstream troughing over eastern North America and into the North Atlantic and Europe (the correlation between snow and SPV stretching frequency is statistically significant). By contrast, correlations with Barents–Kara sea ice (
Fig. 3C) most closely resemble the observed trends, especially the pan-Arctic geopotential height rises in the lower stratosphere (100 hPa) and pressure ridging from the Urals to Greenland at 500 hPa (correlations between ice with SSW frequency and ice with SPV stretching frequency are both found to be significant), more reminiscent of the atmospheric response to SSWs (
Fig. 1C for P5). Despite the statistically significant correlations, it is a challenge to demonstrate cause and effect with observational analysis alone.
To more directly assess the physical links, we conducted numerical modeling experiments related to both increased Eurasian snow cover and reduced sea ice, using a simplified GCM. This kind of model is well-suited for isolating the atmospheric response to idealized heating perturbations (
39) (see materials and methods).
To simulate the observed trend of more extensive October Eurasian snow cover (fig. S2), the model was forced with increased surface albedo (fig. S9). About 2 months after the forcing Is imposed, the model response shows features that resemble the circulation anomalies associated with SPV stretching events, including a stretched SPV, the lower stratospheric dipole in poleward heat transport (a good proxy for WAF
z), the mid-troposphere ridging and warm anomalies in Alaska and the Bering Sea, and troughing and cool anomalies in East Asia and eastern North America (see
Fig. 4B and fig. S10B for comparison with the stretched SPV pattern from cluster analysis in
Fig. 2). The simulated atmospheric response to snow cover forcing is of comparable magnitude to the atmospheric response inferred from observational analysis (
Fig. 4A), though in many previous GCM snow sensitivity experiments, the simulated response is weaker (
40).
Because Barents–Kara sea ice shows a strong observational relationship with Ural mid-tropospheric ridging and downstream East Asian troughing (
Fig. 3C) that projects strongly onto the precursor pattern of SPV stretching events (
Fig. 2C), we further forced the GCM with anomalous heating in the Barents–Kara Seas during October through December where ice loss is observed (figs. S2 and S9). The simulated atmospheric response to both snow and ice forcing (
Fig. 4D) exhibits a pattern similar to the correlations between sea ice, with ridging and warming centered in the Barents–Kara Seas at 100, 500, and 850 hPa and troughing and cooling in East Asia in the mid-to-low troposphere (
Fig. 3C). In addition, the model forced with both snow and ice anomalies includes anomalous ridging and warming in the Chukchi–Bering seas at 500 and 850 hPa and accelerates the model response to Arctic changes by about a month. The simulated atmospheric response to the combined forcing of snow and ice is of comparable magnitude but somewhat weaker than the inferred response from observational analysis but still larger than many previous GCM studies (
3,
41).
We also examine the regression of Eurasian snow cover (
Fig. 4A) and multiple regression of snow and Barents–Kara sea ice (
Fig. 4C) with the atmospheric circulation. Correlations with snow cover exhibit anomalous ridging and warming focused on the North Pacific side of the Arctic from the surface to the lower stratosphere. Therefore, the atmospheric response to snow-cover-only forcing better matches the atmospheric anomalies associated with SPV stretching events than the atmospheric response to sea-ice-only forcing, where the anomalous ridging and warming are either focused in the Barents–Kara Seas or are pan-Arctic. However, when including both snow and ice, the atmospheric response in both observations and in the model better matches full Arctic trends, and the model atmospheric response of SPV stretching is more persistent, with anomalous troughing and relatively cold temperatures in North America persisting for more than 3 weeks (days 36 to 60; fig. S11). Both observational analysis and modeling experiments show that Chukchi–Bering sea ice loss has little impact on the SPV, consistent with previous studies (
42); however, the tropospheric response to Chukchi–Bering sea ice loss can amplify (based on observations, fig. S12) or force anomalous Alaska ridging and warming and downstream anomalous North American troughing and cooling (fig. S13). Although Chukchi–Bering sea ice loss may not force SPV disruptions, it could amplify the tropospheric response across Asia and North America (
43).
Finally, we examined the recent winter of 2020–2021 for stretching SPV events. We applied our clustering technique to observed 100-hPa geopotential heights in January and February 2021. Though our analysis shows that January 2021 was dominated by P5, it identifies P4 for more than 60% of the days from 29 January through 15 February. Also, in early February, WAFz was upward over Siberia and downward over Canada, consistent with SPV stretching events, as opposed to convergence in the stratosphere (consistent with SSW events), as observed in early January prior to an SSW on 5 January (fig. S14). Though the SSW observed in January may have also contributed to the hemispheric pattern in February, our analysis supports that the historic February (approximately 6 to 21) 2021 Texas cold wave was likely the response to the SPV stretching in February, and the atmospheric circulation can be seen transitioning from circulation anomalies associated with SSWs to those associated with SPV stretching from late January through early February (fig. S15).
In this analysis, we have demonstrated that SPV stretching events have accelerated in the era of AA. Climate change in general, but Arctic change in particular, is favorable for forcing these events. In fig. S16, we provide a generalized timeline of the principal atmospheric features beginning with Ural ridging, followed by North Pacific ridging and ending with North American and East Asian cold. It is argued that warming in the Barents–Kara and Chukchi–Bering Seas favor anomalous ridging and high pressure in these regions in the troposphere (
3,
8,
43). Autumn Siberian snowfall has also been increasing (
21), favoring anomalous troughing over East Asia. This pattern of ridging in the Urals and Barents–Kara Seas region and troughing in East Asia strongly projects onto the tropospheric pattern favorable for forcing SPV stretching that often delivers extreme cold to Canada and the United States. This interpretation is supported by a GCM forced with increased Eurasian snow cover and decreased Barents–Kara sea ice, where the atmospheric response is an increase in SPV stretching events with troughing and colder temperatures across Asia and North America 1 to 2 months after the introduction of Arctic forcing. Therefore, Arctic change is likely contributing to the increasing frequency of SPV stretching events, including one just prior to the Texas cold wave of February 2021.
These results have important societal implications. First, they highlight an important type of stratosphere-troposphere coupling, SPV stretching, that has been mostly hidden in the heretofore primary focus on SSWs, even though the impact of SPV stretching events on North American temperatures can be of greater extent and magnitude (
31). Second, the identification of the precursor pattern to stretching events can potentially extend the warning lead time of cold extremes in Asia, Canada, and the United States. Third, our analysis is informative for policy-makers. Preparing for only a decrease in severe winter weather can compound the human and economic cost when severe winter weather does occur, as exemplified during the Texas cold wave of February 2021.
Acknowledgments
We thank three anonymous reviewers whose efforts resulted in a substantially improved manuscript. J.C. thanks M. Kretschmer and S. Kazuyuki, whose creativity as graduate students made this study possible.
Funding: J.C. is supported by the National Science Foundation grant PLR-1901352. L.A. and M.B. received supported from NSF AGS-1657921 and NOAA NA20OAR4310424. C.I.G. and I.W. acknowledge the support of a European Research Council starting grant under the European Union Horizon 2020 research and innovation programme (grant agreement no. 677756).
Author contributions: Conceptualization: J.C. Methodology: J.C., M.B., L.A., C.I.G. Investigation: J.C., M.B., L.A., C.I.G., I.W. Figures: J.C., L.A., C.I.G. Supervision: J.C. Writing – original draft: J.C. Writing – review and editing: J.C., M.B., L.A., C.I.G., I.W.
Competing interests: None for all the authors.
Data and materials availability: Observational analysis was performed with MERRA2: available at
https://gmao.gsfc.nasa.gov/reanalysis/MERRA-2/. NOAA Snow cover extent is available at
http://climate.rutgers.edu/snowcover/index.php and Arctic sea ice concentration is available at
www.metoffice.gov.uk/hadobs/hadisst/. The version of MiMA used in this study can be downloaded from
https://github.com/ianpwhite/MiMA/releases/tag/MiMA-ThermalForcing-v1.0beta and Zenodo (
44). The version of MiMA used in this study follows that used in Garfinkel
et al. (
45) albeit with the albedo and ocean heat-flux modifications as listed in the materials and methods. MiMA v2.0 can be downloaded from
https://github.com/mjucker/MiMA.