HTML
--> --> -->Thermodynamically, global warming is expected to reduce the chances of cold winters or extreme cold days while increasing the likelihood of warm winters and heat waves (IPCC, 2012; Wallace et al., 2014; Shepherd, 2016). Its direct warming effect determines the statistics of cold and warm events during the cold season at mid–high latitudes in the Northern Hemisphere (van Oldenborgh et al., 2015). AA causes cold days with northerly air mass to warm more rapidly than warm days with southerly winds, and thus leads to a decline in daily SAT variance (Screen, 2014). However, cold extremes in boreal winter, which have occurred more frequently at northern midlatitudes in recent decades, might well have their origin in the rapidly warming Arctic (Honda et al., 2009; Mori et al., 2014; Screen et al., 2015; Ma et al., 2018; Ma and Zhu, 2019). Thus, whether global warming is increasing the likelihood of record-breaking cold events is an active topic of debate (Palm, 2013, 2014; Wallace et al., 2014; Shepherd, 2016).
Actually, the climate system is not a thermodynamic, but a fluid dynamic system, and its dynamic effect can often be counterintuitive (Palmer, 2013). Due to the positive feedback between Arctic warming and sea-ice loss, enhanced Arctic warming has been accompanied by extensive loss of sea ice, and Arctic sea-ice loss acts as both a response to and a driver of AA (Cohen et al., 2014; Francis et al., 2017). A number of works have shown that the rapid warming in the Arctic accompanied by sea-ice decline augments some natural patterns, favors more persistent weather patterns by modulating wave interference or the jet stream, and ultimately leads to more extreme weather events, such as the increased “weather whiplash” events at midlatitudes (Honda et al., 2009; Francis and Vavrus, 2012; Liu et al., 2012; Mori et al., 2014; Palm, 2014; Gao et al., 2015; Horton et al., 2015; Ma et al., 2018; Ma and Zhu, 2019). Therefore, both dynamic and thermodynamic effects of global warming have contributed to the observed temperature extremes in recent decades (Horton et al., 2015; Screen et al., 2015; Ma et al., 2018).
The response of SAT variability to AA has been identified to vary spatially. In North America, because the dynamic response is overwhelmed by the thermodynamic effect, the daily SAT variability has shown a significant reduction in recent decades, and cold extremes are expected to become less frequent owing to the continuing Arctic ice loss (Screen, 2014; Screen et al., 2015; Rhines et al., 2017). In East Asia, the thermodynamic effect of AA causes more warm extremes, and meanwhile its dynamic effect has resulted in frequent cold extremes owing to the enhanced Urals high-pressure ridge and Siberian high, leading to a significant increase in daily SAT variability in boreal winter since 1988 (Ma et al., 2018). Evidence shows enhanced northerly flow has contributed substantially to the increased cold extremes over central Asia during boreal winter (Horton et al., 2015). Climate model projections also depict an increased response of winter SAT variability to AA in this region (Screen et al., 2015). Intensi?ed Eurasian anticyclones and more frequent Eurasian blockings, resulting from sea-ice decline, have increased extremely cold SATs and severe winters at midlatitudes over the Eurasian continent (Liu et al., 2012; Zhang et al., 2012; Mori et al., 2014).
In fact, extreme cold surges at midlatitudes in the Northern Hemisphere are not only affected by the climate anomaly in the Arctic, but also by Pacific/Atlantic SST anomalies (Ding et al., 2014; Palmer, 2014; Trenberth et al., 2014; Lee et al., 2015; Kennedy et al., 2016; Sigmond and Fyfe, 2016; Tokinaga et al., 2017; Li et al. 2018; Zyulyaeva et al., 2019). For example, the severe winter in North America during 2013/14 has been identified as having been forced by the compounding effect of the Pacific SST and Arctic sea-ice decline (Lee et al., 2015). The anomalous extratropical circulation forced by the tropical Pacific SST is believed to have contributed greatly to the recent Arctic warming in northern Canada and Greenland (Ding et al., 2014), winter cooling in northwestern North America (Sigmond and Fyfe, 2016), and increased probability of cold winters in Europe (Trenberth et al., 2014).
Several recent studies have focused on changes in wintertime sub-seasonal SAT variability in the northern mid–high latitudes with respect to the recent rapid Arctic warming background. However, they focused either on the zonal mean signal alone (Screen, 2014; Cohen, 2016), within a specific region (Zhang et al., 2012; Mori et al., 2014; Rhines et al., 2017), or solely on the linkage with Arctic warming or extratropical SST anomalies (Liu et al., 2012; Mori et al., 2014; Trenberth et al., 2014; Sigmond and Fyfe, 2016; Francis et al., 2017; Cohen et al., 2018). This study investigates the regional differences of changes in wintertime sub-seasonal SAT variability and how they are orchestrated by the compounding effects of Arctic warming and extratropical SST anomalies, with a particular focus on regional differences between eastern Eurasia and North America and their possible driving mechanisms.
Following previous studies (Ito et al., 2013; Screen, 2014; Cohen, 2016), the daily SAT variability was defined as the variance of the daily SAT anomalies over 90 days in each winter. A larger (smaller) variance of daily SAT anomalies implies more active (stable) weather, which is referred to as stronger (weaker) “weather whiplash” (Cohen, 2016; Swain et al., 2018). We applied standard least-squares linear regression to examine the trends. The linear trends were calculated based on the winter mean during 1988/89–2015/16, which has been used in previous studies (Ma et al, 2018; Ma and Zhu, 2019), matches the modern period of AA (Cohen et al., 2020), and shares coinciding data with the model simulations we used. The correlation coefficients between variables and regressed circulation anomalies were calculated using detrended data during 1979–2016. The statistical significances for the trends, correlation coefficients and regression coefficients were estimated by a two-tailed Student’s t-test. Following Liu et al. (2012), an extreme cold day was defined as when the daily SAT anomaly was below 1.5 standard deviations (σ), and a cold spell was defined as an uninterrupted sequence of extreme cold days. An extreme cold day over eastern Eurasia and North America was defined as when the regional average SAT anomaly was more than 1.5σ colder than the climatology in the domain (40°–60°N, 50°–120°E) and (40°–60°N, 70°–120°W), respectively. Accordingly, we identi?ed 283 and 262 extreme cold days over eastern Eurasia and North America during the period 1979–2016, accounting for approximately 8.5% and 7.9% of the total days, respectively.
To confirm the results obtained from ERA-Interim, we employed the daily SAT from the outputs of the atmosphere-only model simulations, which were forced by observed SST, sea-ice concentrations and historical anthropogenic and natural external forcing agents and provided by the International CLIVAR Climate of the 20th Century Plus Detection and Attribution Project. The utilized experiments consisted of a 10-member ensemble simulation conducted using the Model for Interdisciplinary Research on Climate, version 5 (MIROC5), with a horizontal resolution of 1.4° × 1.4° and 40 vertical levels (Shiogama et al., 2014), and a 50-member ensemble simulation carried out with the Community Atmosphere Model, version 5.1 (CAM5.1), with a horizontal resolution of 0.94°(long.) × 1.25°(lat.) and 30 vertical levels (Stone et al., 2018). These experiments are equivalent to the historical Atmospheric Model Intercomparison Project. The multi-size of the ensemble helped to distinguish boundary-forced signals of change from internal noise of the atmospheric circulation itself.
3.1. Trends of daily SAT variability
Figure 1a shows the linear trends of winter mean SATs over the extratropical Northern Hemisphere during 1988/89–2015/16. Disproportionate Arctic warming relative to the midlatitudes appears in all seasons, with the strongest warming trend in winter (Cohen et al., 2014). The SAT trend exhibits a warm Arctic and cold continents, and the significant warming is especially strong over the Barents–Kara seas, Hudson Bay, Baffin Bay, as well as the East Siberian–Laptev seas. The rapid Arctic warming with an unprecedented sea-ice decline represents the AA signature of global warming (Liu et al., 2012; Mori et al., 2014; Screen et al., 2015). In contrast, apparent cooling trends are displayed in a large patch over eastern Eurasia, but weakly over North America. The “warm Arctic, cold continents” pattern began in the 1970s and became more prominent around 1990 owing to an acceleration in the rate of Arctic warming (Cohen et al., 2014; Kug et al., 2015; Sun et al., 2016; Mori et al., 2019). The decreasing daily SAT variance encircles the high latitudes (Fig. 1b), consistent with the atmospheric response to the ongoing sea-ice loss under Arctic warming in model simulations (Scree et al., 2015; Sun et al., 2016). Corresponding to the synchronous cooling of winter mean SAT, despite the difference in the cooling rates, the daily SAT variance shows an opposite trend over the midlatitudes of eastern Eurasia and North America (Fig. 1b). Overall, for the trends of the daily SAT variance at midlatitudes, the Eurasian continent is dominated by increasing trends with prominent trends over eastern Eurasia, while North America is dominated by decreasing trends with prominence over southern Canada and the northern United States. Hence, the positive daily SAT variance trends averaged over the core region bounded by 40°–60°N and 50°–120°E were employed to characterize the SAT variability in association with the long-term change over eastern Eurasia, while the negative daily SAT variance trends averaged over the core region bounded by 40°–60°N and 50°–120°E were employed to characterize the SAT variability in association with the long-term change over North America. The area-weighted averaged variance over the eastern Eurasian continent and North American continent shows an increasing and decreasing trend of 4.31°C2 (10 yr)?1 and ?4.24°C2 (10 yr)?1, respectively, and both passed the significance test at the 95% confidence level (Figs. 1e and f).
Figure1. Observed changes in winter SATs from 1988/89 to 2015/16: (a) linear trend of DJF-mean SAT [units: ℃ (10 yr)?1]; (b–d) linear trends of DJF SAT variance [units: ℃2 (10 yr)?1] derived from the (b) total (original field), (c) high-frequency (≤ 10 days) and (d) low-frequency (> 10 days) components; (e, f) time series of DJF SAT variance anomalies averaged over (e) eastern Eurasia (40°–60°N, 50°–120°E) and (f) North America (40°–60°N, 70°–120°W), marked by the green boxes in (b). Stippling indicates significance at the 95% confidence level. Data are taken from ERA-Interim.The total sub-seasonal SAT variance includes the high-frequency synoptic-scale variation and low-frequency intraseasonal-scale components. To identify the dominant contributor of sub-seasonal variance change, the trends of both high-frequency (≤ 10 days) and low-frequency (> 10 days) variation were examined, where high-frequency and low-frequency variation were extracted by applying a 10-day Lanczos high-pass and low-pass filter to daily data with 41 weights, respectively. The high-frequency variation shows a significant deceasing trend only over western Europe and the high latitudes of North America (Fig. 1c), with little change over lower latitudes. Over eastern Eurasia, however, the increasing counterpart is largely attributable to the low-frequency counterpart. The decreased daily SAT variability is largely contributed by its low-frequency component over North America (Fig. 1d). The trend of low-frequency variance is 4.65°C2 (10 yr)?1 over eastern Eurasia, which is larger than the trend of the total variance. The high-frequency synoptic-scale counterpart, meanwhile, shows a slightly weak and non-significant decreasing rate of ?0.32°C2 (10 yr)?1 (Fig. 1e). In contrast, the low-frequency variance over North America displays a significant decreasing trend of ?3.39°C2 (10 yr)?1, accounting for approximately 80% of the total trend variance. The high-frequency variance also shows a decreasing trend, of ?0.77°C2 (10 yr)?1, but it is not significant at the 95% confidence level (Fig. 1f).
The total variance of daily SAT reflects the spread of day-to-day SAT fluctuation. A high (small) variance usually indicates that the daily SAT is very distant from (close to) the mean and from each other. It is possibly caused by the shift in the mean, the shape of the distribution, or both (IPCC, 2012; Ito et al., 2013). We calculated the standardized deviation of daily SAT anomalies and compared the frequency distributions before and after 2000/01 corresponding to the colder and warmer Arctic phase, respectively. According to the Kolmogorov–Smirnov test, the distributions of daily SAT anomalies are significantly different (p < 0.001) between the two Arctic phases in eastern Eurasia and North America (Figs. 2a and b). During the warmer Arctic period, the extreme cold days become more frequent over the eastern Eurasian continent, but less frequent over the North American continent. Days within ±1σ of daily SAT anomalies decrease over eastern Eurasia, while days with ?1σ < daily SAT anomalies < 0 increase over North America. In contrast, the extreme warm days exhibit almost the same occurrence frequency in both regions during the colder and warmer epochs (Figs. 2a and b). Moreover, in terms of the time series of the frequencies of the extreme cold days (Fig. 3), the 1988/89–2015/16 period saw a steady increase of 2.40 d (10 yr)?1 in extreme cold days over eastern Eurasia. This increase is statistically significant at the 0.05 level. In contrast, the extreme cold days over North America display a decline of ?1.56 d (10 yr)?1, which is statistically significant at the 0.1 level. These statistical results imply that the increasing trend of daily SAT variance in eastern Eurasia is mainly contributed by the increase in extreme cold days, while the decreasing trend in North America can be attributed to the decrease in cold extreme days.
Figure2. (a) Histogram distribution of the standardized daily SAT anomalies over eastern Eurasia (left-hand green box in Fig. 1b) during the warmer Arctic epoch of 2000/01–2015/16 (red curve) and colder Arctic epoch of 1988/89–1999/2000 (blue curve). (c, e) Quantile–quantile plot for (c) cold-spell durations and (e) cold-spell minimum SATs during the warmer Arctic epoch of 2000/01–2015/16 and colder Arctic epoch of 1988/89–1999/2000. The horizontal (vertical) dashed lines indicate the P10, P25, P75 and P90 quantiles of 1988/89–1999/2000 (2000/01–2015/16), and the solid lines indicate P50. (b, d, f) As in (a, c, e), respectively, but over North America (right-hand green box in Fig. 1b). The inset plots in (c, d) are for the mean cold-spell SAT as a function of duration. Data are taken from ERA-Interim.
Figure3. Time series of extreme cold day anomalies averaged over eastern Eurasia (solid line with open circles) and North America (dashed line with ?lled circles) and their linear trend during 1988/89–2015/16. Data are taken from ERA-Interim.Extreme cold days are often associated with strong cold spells, regulated by the sub-seasonal activity of the East Asian winter monsoon (Park et al., 2011; Ito et al., 2013). The significant upward (downward) trend of sub-seasonal SAT variation suggests an opposite change in cold spells over eastern Eurasia (North America) under recent Arctic warming. To verify this assertion, we applied a quantile–quantile plot to compare the duration and intensity distributions of cold spells during the two Arctic epochs. Over eastern Eurasia, the durations of cold spells are basically above the y = x line, suggesting that the cold spell durations during the warmer Arctic epoch are longer than those of the colder Arctic epoch (Fig. 2c). Over North America, cold spells, especially the 10% longest-lasting cold spells, during the warmer Arctic epoch, are shorter-lasting than those of the colder Arctic epoch (Fig. 2d). The mean cold-spell SAT drops with increasing duration (inset plots in Figs. 2c and d). As expected, over eastern Eurasia, the minimum SAT of cold spells during the warmer Arctic epoch become colder than those of the colder Arctic epoch, while the North American cold-spell minimum SAT shows fewer cold anomalies during the warmer Arctic epoch than those of the colder Arctic epoch (Figs. 2e and f).
2
3.2. Distinct circulation regimes over eastern Eurasia and the North American continent
The circulation regime related to extreme cold days is different over eastern Eurasia and North America. Figure 4 shows the composites of Z500, SLP and SAT anomalies corresponding to extreme cold days over eastern Eurasia and North America. In eastern Eurasia, the positive center of Z500 anomalies is situated over the Ural Mountains and extends southeastward and westward to the midlatitude Pacific and North Atlantic, while a negative center exists over the Asian continent and western Europe. At 500 hPa, the circulation exhibits a deepened trough–ridge–trough structure and displays a southeastward propagation of the Rossby wave train over the Eurasian continent according to the wave activity ?ux, which is independent of wave phase and parallel to the local group velocity on a zonally varying basic ?ow (Takaya and Nakamura, 2001). The wave train originates from the warmer Barents–Kara seas (Figs. 4a and e). At the surface, the Eurasian continent is covered almost entirely by positive SLP anomalies (Fig. 4c). This suggests that the circulation regime of a strengthened high-pressure ridge over the Urals and surface Siberian high is closely related to extreme cold days over eastern Eurasia.
Figure4. Composite anomalies of (a) geopotential height at 500 hPa (Z500; shading; units: 10 gpm) and attendant wave activity flux (arrows), (c) SLP (units: hPa), and (e) SATs (units: ℃) during extreme cold days in eastern Eurasia (total of 283 days). (b, d, f) As in (a, c, e), respectively, but for extreme cold days in North America (total of 262 days). Signi?cant values at the 95% con?dence level are represented by white dots. The green boxes in (a, b) mark the Urals blocking region (55°–85°N, 20°–80°E) and the Alaska blocking region (50°–80°N, 160°E–140°W), respectively. Data are taken from ERA-Interim.For extreme cold days in North America, the positive center of Z500 is located over the high-latitude North Pacific, which extends northward into the Arctic and displays a notable ridge anomaly over the mid-latitude North Atlantic. In contrast, significantly negative height anomalies are mainly located over most areas of North America and the high-latitude North Atlantic. The Z500 anomalies over the North Pacific and North America resemble the negative phase of the North Pacific Oscillation (Wallace and Gutzler, 1981), corresponding to an eastward propagation of the Rossby wave train originating from the Bering Sea (Fig. 4b). At the surface, positive SLP anomalies are observed over Northwest America, which develop from the North Pacific to the mid-latitude North Atlantic and extend northward into the Arctic (Fig. 4d). Meanwhile, a strong warm SAT anomaly appears over the East Siberian–Chukchi seas (Fig. 4f). Therefore, the cold extreme–related circulation regime is associated northerly winds and characterized by a strengthened high-pressure ridge at upper levels over Alaska and an amplified Canadian high near the surface. This result is consistent with the previous study of Cohen et al. (2018).
4.1. Data analyses
It has been suggested that the cold extreme–related circulation regime resembles the similar long-term changes of its seasonal mean associated with Arctic warming (Mori et al., 2014, 2019; Kug et al., 2015; Sun et al., 2016). After removing the linear trends of the frequency of extreme cold days, we found that extreme cold days over both eastern Eurasia and North America are negatively correlated with their seasonal means, but positively correlated with their counterparts in the upstream regions of the Arctic (Figs. 5a and b). This suggests that the increased frequencies of extreme cold days over eastern Eurasia and North America are closely related to the dipole SAT anomalies, i.e., cold continents and a warm Arctic. The seasonal anomalies of Z500 and SLP (Figs. 5c–f), corresponding to winters with a more frequent occurrence of extreme cold days, resemble the similar circulation patterns of cold extremes in both regions (Figs. 4a–d). During winters with a more frequent occurrence of extreme cold days over eastern Eurasia, stronger positive and negative anomalies of DJF-mean Z500 prevail over the Urals and East Asia and western Europe, respectively. Near the surface, the northern Eurasian continent is occupied by positive winter-mean SLP anomalies. This circulation regime is synchronously associated with the upper-level enhanced Urals high-pressure ridge and East Asian trough with an amplification of the Siberian high at the surface. Similarly, cold winters over North America are dominated by a stronger North Pacific high-pressure ridge and North American trough with an enhanced Canadian high near the surface.
Figure5. Regression coefficients of DJF-mean anomalies of (a) SAT (units: ℃), (c) Z500 (units: gpm) and (e) SLP (units: hPa) with respect to the detrended standardized area-weighted average of the frequency of extreme cold days over eastern Eurasia (left-hand green box in Fig. 1b). (b, d, f) As in (a, c, e), respectively, but with respect to the detrended standardized area-weighted average of the frequency of extreme cold days over North America (right-hand green box in Fig. 1b). Dots represent significance at the 95% level. Data are taken from ERA-Interim.The recently frequent cold winters at midlatitudes in the Northern Hemisphere have been attributed to the impacts of amplified Arctic warming and extratropical SST forcing (e.g., Honda et al., 2009; Francis et al., 2012; Liu et al., 2012; Ding et al., 2014; Mori et al., 2014; Palmer, 2014; Trenberth et al., 2014; Kug et al., 2015; Lee et al., 2015; Cohen, 2016; Sigmond and Fyfe, 2016; Mori et al., 2019). Melting Arctic sea ice contributes and results from AA (Screen and Simmonds, 2010; Cohen et al., 2014; Gao et al., 2015; Francis et al., 2017), and is usually used as one important forced factor of the impacts of AA on midlatitude weather and climate (Honda et al., 2009; Liu et al., 2012; Mori et al., 2014, 2019; Lee et al., 2015; Screen et al., 2018). Here, following Kug et al. (2015) and Tokinaga et al. (2017), to reveal the possible linkage between AA and the atmospheric circulation regime associated with extreme cold days over eastern Eurasian and North America, we examined the linear correlation between the Arctic SAT and the key circulation factors. Figure 6 shows the correlations of DJF-mean SAT anomalies with the Z500 index anomaly averaged over the Urals and Alaska region, respectively (depicted by the green boxes in Figs. 4a and b). The significantly positive correlations of the Urals Z500 anomaly extend from the Greenland Sea eastward to the Laptev Sea, but no significant correlations are found over the tropical and extratropical oceans (Fig. 6a). Therefore, the enhanced Urals high-pressure ridge at Z500 is closely linked to the warmer SAT anomalies over the Barents–Kara seas. In contrast, the Alaskan height anomaly is positively correlated with the warmer SAT centered over the Chukchi Sea. Meanwhile, it is also closely correlated with the Pacific decadal oscillation (PDO)–like SST anomalies, with significant negative correlations over the western subtropical North Pacific and positive correlations over the extratropical North Pacific and eastern tropical North Pacific (Fig. 6b). Therefore, the enhanced Alaskan high-pressure ridge is possibly forced by the warm SAT anomalies over the East Siberian–Chukchi seas and the PDO-like SST anomalies in the North Pacific. The warmer winter SAT anomalies over the Barents–Kara seas is closely associated with the cold extreme–related circulation regime over eastern Eurasia, characterized by the enhanced Urals high-pressure ridge at Z500, the Siberian high near the surface, and the increased frequency of extreme cold days over eastern Eurasia (Figs. 7a, d and g). The warmer SATs over the East Siberian–Chukchi seas, however, are significantly linked to the cold extreme–related circulation regime in North America (Figs. 7b, e and h).
Figure6. Correlation coefficients of DJF-mean surface temperature (SATs over land and ice, SSTs over the oceans) with detrended DJF-mean Z500 anomalies averaged over the (a) Urals blocking region (55°–85°N, 20°–80°E; green box in Fig. 4a) and (b) Alaska blocking region (50°–80°N, 160°E–140°W; green box in Fig. 4b). Dots represent significance at the 95% level. The yellow box in (a) marks the Barents–Kara seas region (70°–90°N, 0°–105°E), and in (b) it marks the East Siberian–Chukchi seas region (65°–85°N, 155°E–130°W). The green boxes in (b) mark the western subtropical North Pacific (WSubNP; 20°–40°N, 120°E–160°W), the extratropical North Pacific (ExtraNP; 40°–60°N, 140°E–135°W), and the eastern tropical North Pacific (ETNP; 10°–25°N, 160°–110°W); the PDO-like SST index is defined as ExtraNP + ETNP ? WSubNP. The SST data are taken from the monthly HadISST dataset, and other variables are from ERA-Interim.
Figure7. Linear regression of DJF-mean Z500 (units: 10 gpm) with respect to the standardized detrended DJF-mean SATs averaged over the (a) Barents–Kara seas (yellow box in Fig. 5a), (b) East Siberian–Chukchi seas (yellow box in Fig. 5b), and (c) the PDO-like SST index (averaged over the green boxes in Fig. 5b). (d–f) As in (a–c), respectively, but for SLP (units: hPa). (g–i) As in (a–c), respectively, but for the frequency of DJF extreme cold days (units: days). Stippling indicates significance at the 95% level. Arrows in (a–c) denote the horizontal wave activity flux associated with the Z500 anomalies. The SST data are taken from the monthly HadISST dataset, and other variables are from ERA-Interim.To reveal the impact of the PDO-like SST anomalies, we defined the regional SST difference of the extratropical North Pacific (ExtraNP) and eastern tropical North Pacific (ETNP) from the western subtropical North Pacific (WSubNP), i.e., ExtraNP + ETNP ? WSubNP. The central PDO-like SST anomalies shift southward approximately 10° of latitude relative to those SST anomalies for the commonly used PDO. The correlation coef?cient (r) between the PDO-like SST index and PDO index is 0.42, which is statistically signi?cant at the 0.05 level. In contrast, the atmospheric circulation anomalies associated with the positive PDO-like SST anomalies are almost the same as those associated with the warm anomalies of the DJF-mean SATs over the East Siberian–Chukchi seas (Figs. 7c, f and i), albeit with a slight southward shifting of the strong center of anomalous circulation.
Corresponding to the significant warming over the Barents–Kara Sea region (Fig. 1a), the seasonal Z500 anomalies show strong increasing trends over the Urals region and slightly decreasing trends over the west coast of Europe and Lake Baikal (Fig. 8b). It looks like a southeastward-propagating Rossby wave train over the Eurasian continent, originating from the Barents–Kara seas. The SLP anomalies show significant increasing trends across almost the entire northern Eurasian continent (Fig. 8c). The enhanced circulation regime related to cold extremes provides a favorable background and increase in extreme cold days over eastern Eurasia. Consequently, the winter-mean SATs show cooling trends (Fig. 1a), cold extremes occur more frequently (Figs. 2 and 3), and the daily SAT variance increases over eastern Eurasia (Figs. 1b and e).
Figure8. Linear trends of DJF-mean (a) SST [shading; °C (10 yr)?1], (b) Z500 [gpm (10 yr)?1] and (c) SLP [hPa (10 yr)?1] during the period 1988/89–2015/16. Stippling indicates significance at the 95% confidence level. Arrows in (b) are the horizontal wave activity flux associated with the Z500 trends. (d) Time series of the standardized anomalies of the PDO-like SST index. The SST data are taken from the monthly HadISST dataset, and other variables are from ERA-Interim.Increasing trends of DJF-mean Z500 anomalies are observed over the north of the Bering Sea (Fig. 8b), associated with prominent warming over the East Siberian–Chukchi seas (Fig. 1a). They exhibit a wave-like distribution with negative–positive–negative values over the Aleutian, northwestern American and eastern American regions, respectively (Fig. 8b). The response of Z500 to the strong warming trend over the western subtropical North Pacific and the weak cooling trend over the extratropical North Pacific and eastern tropical North Pacific, i.e., the negative PDO-like SST change (Fig. 8a), partially cancels out the response to Arctic warming. Consequently, the Z500 trends are not significant from the Aleutian low region to the east coast of North America. The SLP shows an increasing and decreasing trend over western and eastern North America, respectively, but they are weak and not significant (Fig. 8c). The offsetting effects of the Arctic warming and the North Pacific SST suppress the intensity of the Alaskan high-pressure ridge and surface Canadian high region, consequently resulting in weak cooling and even a warming trend over North America (Fig. 1a). It is suspected that the remote response of cold extremes to Arctic warming is overwhelmed by the effect of the North Pacific SST change, which reduces the frequency of extreme cold days in the North American continent (Figs. 2b and 3), and therefore the corresponding daily SAT variance shows a significant decreasing trend (Figs. 1b and f). It is found that the decreasing rate of daily SAT variability in North America has slowed down since the 2010s (Fig. 1f), when the PDO-like SST anomaly changed from a negative to positive phase (Fig. 8d).
2
4.2. Model simulations using MIROC5 and CAM5.1
Numerical model simulations using MIROC5 and CAM5.1 were employed to further test the physical mechanism responsible for the opposite trends of sub-seasonal temperature variability over eastern Eurasia and North America derived from the analysis using ERA-Interim data. MIROC5 can simulate the circulation regime of extreme cold days very well over both eastern Eurasia and North America (Figs. 9a–d). The pattern correlations between the model and observation over eastern Eurasia (North America) is 0.90 (0.94) for Z500, 0.94 (0.91) for SLP, and 0.92 (0.95) for SAT. Its historical simulation can reproduce the temporal evolution of winter-mean SAT very well over the Barents–Kara seas and East Siberian–Chukchi seas, as well as the upstream Arctic region of eastern Eurasia and North America (Fig. 9e). Similarly, CAM5.1 simulates a realistic circulation regime of extreme cold days over both eastern Eurasia and North America (Figs. 10a–d). The pattern correlations between the model and observation over eastern Eurasia (North America) is 0.92 (0.86) for Z500, 0.96 (0.88) for SLP, and 0.95 (0.95) for SAT. The observed temporal evolution of winter-mean SAT over both the Barents–Kara seas and East Siberian–Chukchi seas are also reasonably captured by CAM5.1 (Fig. 10e).
Figure9. Composite anomalies of (a) Z500 (shading; units: 10 gpm; dotted regions exceed the 95% confidence level) and SLP [contours drawn for ±1, ±3, …, ±11 hPa; solid (dashed) contours denote positive (negative) anomalies] and (c) SATs during extreme cold days in eastern Eurasia derived from the 10-member ensemble of MIROC5 simulations. (b, d) As in (a, c), respectively, but for North American extreme cold days. (e) Time series of DJF-mean SAT anomalies averaged over the Barents-Kara seas (BK; yellow box in Fig. 6a) and East Siberian–Chukchi seas (SC; yellow box in Fig. 6b) during 1988/89–2015/16. The sub-plot in (e) is a box-and-whisker plot for the BK and SC SAT trends; the minimum, lower quartile, median, upper quartile, and maximum values of MIROC5-simulated trends are shown; solid circles correspond to the observed trends derived from ERA-Interim.
Figure10. As in Fig. 9 but for CAM5.1 simulations.In addition, the warming trends over both the Barents–Kara seas and East Siberian–Chukchi seas are reproduced in both the MIROC5 and CAM5.1 simulations, albeit the observed warming trends are underestimated in both models (Figs. 9e and 10e). During the winters of 1988/89–2015/16, the area-weighted average SAT over the Barents–Kara seas exhibit a strong warming trend of 2.13°C (10 yr)?1 in the observation, but 1.02° (10 yr)?1 and 1.11°C (10 yr)?1 in the ensemble mean of the MIROC5 and CAM5.1 simulations, respectively, with ranges of 0.58–1.44°C (10 yr)?1 and 0.18–1.87°C (10 yr)?1, respectively. The SAT averaged in the East Siberian–Chukchi seas shows an observed warming trend at a rate of 1.47°C (10 yr)?1, and a rate of 0.89°C (10 yr)?1 and 1.02°C (10 yr)?1 in the ensemble mean of the MIROC5 and CAM5.1 simulations, respectively. Both trends are significant at the 0.01 level. The observed warming trend over the Barents–Kara seas falls outside the ranges of both the MIROC5 and CAM5.1 ensembles, while the observed warming trend over the East Siberian–Chukchi seas lies within the simulated trend ranges of both the MIROC5 and CAM5.1 ensembles.
Having established that MIROC5 and CAM5.1 can reproduce the observed atmospheric circulation regimes for extreme cold days in both eastern Eurasia and North America, as well as the warming trends over both the Barents–Kara seas and East Siberian–Chukchi seas, we then investigated the simulated trends of wintertime sub-seasonal SAT variance during 1988/89–2015/16. The winter daily SAT variance in the ensemble mean of both the MIROC5 and CAM5.1 simulations display significant increasing trends over large patches of eastern Eurasia and decreasing trends over most of North America (Figs. 11a and b), similar to the observations (Fig. 1b). However, both the increasing trend over eastern Eurasia and the decreasing trend over North America are underestimated. This underestimation may be closely related to the performance of the model itself, as well as the tendency for the current generation of climate models to respond too weakly to sea-ice change (Screen et al., 2018). In addition, the natural atmospheric variability, remote response to climate fluctuations in the tropical SSTs, and the forced response to climate change act together to affect midlatitude weather and climate (Shepherd, 2016; Sigmond and Fyfe, 2016; Ma and Zhu, 2019; Mori et al., 2019). Thus, it is possible that the stronger natural atmospheric circulation leads to the underestimation of the model ensemble mean compared to the observed change of sub-seasonal SAT variance.
Figure11. Linear trends of DJF SAT variance [units: °C2 (10 yr)?1] in the ensemble mean of (a) MIROC5 and (b) CAM5.1 simulations during 1988/89?2015/16. Stippling indicates significance at the 95% confidence level. Green boxes denote the observed increasing trend core regions in eastern Eurasia and the observed decreasing trend core regions in North America. (c) Scatterplot of the DJF-mean SAT trend averaged over the Barents–Kara seas [BK; units °C (10 yr)?1] and SAT variance trend averaged over eastern Eurasia. (d) Scatterplot of the DJF-mean SAT trend averaged over the East Siberian–Chukchi seas [SC; units °C (10 yr)?1] and SAT variance trend averaged over North America. The red, blue and green dots denote the results based on ERA-Interim, the 10-member ensemble mean of MIROC5 simulations, and the 50-member ensemble mean of CAM5.1 simulations, respectively.To further explore the realism of the physical linkages between the warming over the Barents–Kara seas and the changes in daily SAT variance over eastern Eurasia, as well as the linkages between the warming over the East Siberian–Chukchi seas and the changes in daily SAT variance over North America, we show the scatter relationship between the warming trends over the Barents–Kara seas (East Siberian–Chukchi seas) and the daily SAT variance trend over eastern Eurasia (North America) in the 60-member ensembles of the MIROC5 and CAM5.1 simulations. An obvious linear relationship exists, with a correlation of 0.60 between the warming trends over the Barents–Kara seas and the daily SAT variance trends over eastern Eurasia, and a correlation of 0.44 between the warming trends over the East Siberian–Chukchi seas and the daily SAT variance trends over North America. Both correlations are significant at the 0.01 level. By comparison with the model ensembles, both the observed warming trend over the Barents–Kara seas and the increasing SAT variance trend over eastern Eurasia stand out as an extreme. However, the linearity between the warming trend over the Barents–Kara seas and the increasing trend over North America describes the observed results well. Similarly, the observed SAT variance trend over North America is also consistent with a simple physical process of East Siberian–Chukchi seas warming driving as indicated by the agreement with the linear relationship among model ensembles. Meanwhile, the preponderance of decreasing trends of daily SAT variance over North America among the model ensembles indicates the overwhelming effect of SST forcing.
Because the historical simulations of MIROC5 and CAM5.1 are mainly driven by the observed SST and sea ice under global warming, their successful capture of the cold extreme–related circulation regime and opposite trends of daily SAT variance over eastern Eurasia and North America confirms the different impacts of the upstream Arctic warming and PDO-like SST anomalies on the trends of SAT over eastern Eurasia and North America.
5.1. Summary
Frequent cold extremes have occurred recently at the midlatitudes of the Northern Hemisphere, and concurrently with the unprecedented Arctic warming. However, we found the response of cold extreme events are opposite over the eastern Eurasia and North American continents. Accordingly, we analyzed the daily SAT variability over the Northern Hemisphere midlatitudes during the winters of 1988/89–2015/16 and discussed the possible underlying mechanism based on statistical analyses and model simulations. The main findings can be summarized as follows:The variations of cold extremes at midlatitudes of the Northern Hemisphere can be represented by the daily SAT variances, which have different regional variation. The trends of daily SAT variance are mainly contributed by their sub-seasonal components, showing an increasing and decreasing trend over eastern Eurasia and North America, respectively. The daily SAT variance shows a significant upward and downward trend of 10.15% (10 yr)?1 and ?8.16% (10 yr)?1 over eastern Eurasia and North America, respectively. The trends of daily SAT variability can be attributed to the frequency of extreme cold days, regulated by the sub-seasonal component of SAT variance over eastern Eurasia and North America.
The cold extreme–related circulation regime is characterized by an enhanced Urals high-pressure ridge at Z500 and Siberian high near the surface over eastern Eurasia. The recent significant warming over the Barents–Kara seas has enhanced the cold extreme circulation regime over eastern Eurasia and led to more extreme cold days, and increased the daily SAT variability of the sub-seasonal component. This is consistent with previous studies (Honda et al., 2009; Liu et al., 2012; Mori et al., 2014, 2019; Ma et al., 2018; Ma and Zhu, 2019). Over the North American region, the circulation regime of extreme cold days is characterized by a strengthened Alaskan high-pressure ridge at Z500 and amplified Canadian high near the surface. The warming over the East Siberian–Chukchi seas in the Arctic region may strengthen the Alaskan high-pressure ridge, but a PDO-like SST anomaly in the North Pacific may weaken the Alaskan high-pressure ridge. When the effect of warming over the upstream Arctic is overwhelmed by the effect of PDO-like SST anomalies, it reduces the sub-seasonal SAT variability and extreme cold days over North America, resulting in the increasing and decreasing trends of daily SAT variability over eastern Eurasia and North America, respectively. The regional Arctic warming driving the changes in SAT variance over both eastern Eurasia and North America, and the overwhelming contribution of SST forcing to the decreasing SAT variance over North America, were confirmed by the multi-ensemble simulations of both MIORC5 and CAM5.1.
2
5.2. Discussion
It is suggested that the increased frequency of extreme cold days over eastern Eurasia, with an enhanced Urals high-pressure ridge and Siberian high, is closely associated with the warming over the Barents–Kara seas in the Arctic region. Besides the impact of Arctic warming, it is also likely that atmospheric circulation anomalies at midlatitudes (Feng and Wu, 2015; Dobricic et al., 2016; Wu, 2017), extratropical ocean warming (Li et al., 2015; Wu, 2017), and atmospheric transport of moisture and energy from lower latitudes (Kim and Kim, 2017; Hao et al., 2019) may also affect the daily SAT variability over eastern Eurasia. However, the increasing trend of cold extremes in this region is mainly attributable to Arctic warming, particularly during the era of AA. The PDO-like SST anomalies show decadal variability, which may enhance the Alaskan high-pressure ridge and the cold extremes over North America when it turns to a positive phase in the future. In the present study, we compared the daily SAT variability between 40° and 60°N with the strong and opposite SAT variance trends over eastern Eurasia (50°–120°E) and the North American continent (70°–120°W). We also compared the trends of extreme cold days derived from the selected core domain to small domains by shrinking the lateral boundaries slightly (within 8°), as well as domains defined by a slight shift northward or southward with respect to the core domain (within 8°). We found that the slightly shrunken domain and slightly northward- or southward-shifted domain exhibit similar trends in extreme cold days as the core domain, albeit the magnitude and significance of the corresponding trends differ. The selected core region in North America corresponds only to the northernmost part, and the SAT trends are weaker than those in the core region selected over Eurasia. The results would be very different if other areas of North America had been selected, such as the southern United States, where the SAT variance trends are positive.Acknowledgements. This study was jointly supported by the National Key R&D Program (Grant No. 2018YFC1505904), the National Natural Science Foundation of China (Grant Nos. 41830969 and 41705052), and the Basic Scientific Research and Operation Foundation of CAMS (Grant No. 2018Z006).
