HTML
--> --> -->In recent decades, the sea-ice cover has dramatically decreased and this directly affects local heat fluxes and atmospheric circulation (Deser et al., 2010; Screen and Simmonds, 2010). Meanwhile, the Arctic warms faster than other regions and this is called Arctic Amplification (AA; e.g., Graversen et al., 2008). In the late 21st century, when the radiative forcing becomes much stronger than the present climate, the Arctic is expected to become ice-free in summer (Wang and Overland, 2012) and the AA will be stronger. In response to the sea-ice decline and AA, some numerical studies have shown a negative AO-like/NAO-like circulation (e.g., Magnusdottir et al., 2004; Sokolova et al., 2007; Peings and Magnusdottir, 2014; Blackport and Kushner, 2017), and a higher SLP over northern Eurasia (Deser et al., 2016). These circulation changes potentially advect more cold polar air equatorward. However, due to the warmer cold polar air, the AA-related cold-air outbreaks would be weaker than the present climate (Ayarzagüena and Screen, 2016). The thermodynamic effect due to increasing sea surface temperature (SST) would also outweigh the dynamic cooling caused by AA (Deser et al., 2016). The temperature over extratropical Eurasia would generally increase (Deser et al., 2010, 2016) and the probability of cold winters would decrease (Yang and Christensen, 2012), unlike the increasing tendency for the warm-Arctic-cold-midlatitude temperature pattern during the recent AA period (e.g., Cohen et al., 2014).
Although there is high confidence that sea-ice cover will rapidly diminish under future global warming, its rate of retreat has a large intermodel spread across the CMIP5 models (Wang and Overland, 2012). Meanwhile, the midlatitude circulation changes could result from the competing effect of AA and other drivers (Vihma, 2014; Chen et al., 2016; Deser et al., 2016). Thus, the intermodel spread in sea-ice projection could be related to uncertainties in midlatitude atmospheric circulation change. For instance, over North America and the North Atlantic, models disagree on the sign of change of the wintertime midlatitude westerly wind and speed, but the intermodel spread of these quantities is significantly correlated to that of AA (Barnes and Polvani, 2015). Through analyzing model outputs from the CMIP5 archive, we address the following questions that have not been analyzed thoroughly by previous studies: (1) What are the projected uncertainties of Arctic sea-ice cover and the Eurasian climate? (2) What is the relationship between these projected uncertainties? (3) Do the models showing a stronger sea-ice decline correspond to cooling or less pronounced warming in any parts of Eurasia?
To answer these questions, we focus on the intermodel spread of the forced response to the Arctic sea-ice decline. To minimize the internal atmospheric variability, we only analyze the 11 CMIP5 models (https://esgf-node.llnl.gov/projects/esgf-llnl/) that include at least three ensemble members for both the historical and RCP8.5 simulations (Table 1). The forced response is regarded as the climatological difference between the period 2069-98 in the RCP8.5 run and 1971-2000 in the historical run; below, we refer to this difference simply as "the response". In each model, the climatology is the unweighted average of all ensemble members listed in Table 1. The multimodel ensemble mean (MME) response is defined by averaging the individual model means. The reason for choosing 2069-98 instead of 2071-2100 is due to some missing outputs in 2099 and 2100. All atmospheric and oceanic variables are interpolated to a horizontal resolution of 2.5°× 2.5° and 1.0°× 1.0°, respectively.
Figure1. (a, c) MME mean global warming response (2069-98 relative to 1971-2000) in DJFM and (b, d) the corresponding intermodel standard deviation of the response. (a, b) SIC (%), and (c, d) mean SLP (hPa), where the green box indicates the domain of the SVD analysis in Fig. 2. In (a) and (c), white and black dotted regions indicate at least 7 (～ 65%) and 10 (～ 90%) out of 11 models agreeing on the sign of change. In (b) and (d), contours indicate the MME mean response in (a) and (c).Associated with a robust decline in SIC, the SLP consistently decreases over the Arctic in the MME (Fig. 1c). In other polar regions, the strongest SLP decline is over the North Pacific and this is associated with a stronger Aleutian low (Fig. 1c). In contrast, the SLP over the North Atlantic and Greenland increases, but there is not a large agreement among models on the sign of response (<90%; Fig. 1c). These regions also have a large intermodel standard deviation, suggesting a large projected uncertainty of the Icelandic low and the NAO.
There is also a robust response in SLP outside the polar region. This includes an increase in SLP near the Mediterranean Sea and a decrease in SLP over western Africa (Fig. 1c), suggesting a northeastward extension of the Azores high. In addition, the SLP increases robustly over Southeast Asia and south of Japan (Fig. 1c), reflecting an expansion of the subtropical high over the western North Pacific and a northward shift of the East Asian trough. In other parts of Eurasia, the MME response of the SLP is comparable and even smaller than the intermodel standard deviation, and most models do not agree on the sign of the response (Fig. 1d). As cold air originating from the polar region strongly influences the Eurasian climate, it is important to assess the potential links between the projected uncertainties in the Arctic SIC and the Eurasian climate.
During the late 21st century, most models agree in simulating an ice-free Arctic in boreal autumn, whereas this agreement has a large spread in boreal winter. As the polar air strongly affects the lower-latitude regions, we hypothesize a simultaneous linkage between the intermodel spread of the forced response of the SIC (?SIC) and the Eurasian circulation in boreal winter. To test our hypothesis, we use singular value decomposition (SVD) to identify the spatial pattern accounting for the largest fraction of their covariability.
In the SVD analysis, the left-hand vector is the DJFM-mean response in (?SIC) over the entire Arctic, whereas the right-hand vector is the DJFM-mean SLP response (?SLP) over (0°-90°N, 60°W-180°), which is a domain able to capture the large-scale circulation features in Eurasia, of the 11 models. Note that the series of our SVD analysis is a time-invariant model-dependent parameter (i.e., the forced response of different models) instead of a time-varying parameter in a conventional SVD analysis, as mentioned in (Bretherton et al., 1992) and (Wallace et al., 1992). Therefore, the expansion coefficient of our SVD analysis is a series of 11 models instead of time steps. This approach was adopted in (Wang et al., 2014), who analyzed the covariability between the biases of the global SST and the meridional overturning circulation across CMIP5 models. Note that the power of the SVD analysis here is to identify the spatial patterns accounting for the largest covariability between the model uncertainties of ?SIC and ?SLP, and to quantify their covariability. The coupling between these uncertainties can be due to different physical processes. This is not trivial using simpler composite or correlation analysis, which requires an index (say, the sea-ice change in the Arctic) to be first defined. Such a composite analysis cannot measure how strongly the uncertainties of ?SIC are coupled to ?SLP.
The first three SVD modes (SVD1-3) explain 70.5%, 16.4% and 6.63% of the total squared covariance, and hence we focus only on the first mode (SVD1). The correlation between the expansion coefficients of ?SIC from the SVD1 and ?SIC (the homogeneous correlation map) is shown in Fig. 2a. The spatial pattern of ?SIC represents a strong decline in SIC over most of the Arctic, except the Barents Sea openings. Consistently, the expansion coefficient of ?SIC is almost perfectly correlated (across the 11 models) to the DJFM-mean Arctic total sea-ice extent (the total area of grid points with SIC >0.15) for the period 2069-98, and the difference between 2069-98 and 1971-2000 (r=-0.97 in both cases).
Figure2. The dominant relation between uncertainties in the global warming response in winter of sea ice and SLP over Eurasia, explaining 70% of intermodel covariability. Intermodel correlation between the expansion coefficients of the DJFM SIC response from the SVD1 and the response in DJFM-mean (a) SIC, (b) SLP, and (c) SLP (shading), 250-hPa wind (magenta arrows) and 850-hPa vector wind (black arrows); the left-hand vector and the right-hand vector of the SVD1 is the DJFM-mean change of sea-ice cover and the SLP, respectively. Thick gray lines denote p=0.1; white dotted regions and vectors indicate p<0.1.The correlation between the expansion coefficients of ?SIC of the SVD1 and ?SLP in Eurasia (the heterogeneous correlation map) is shown in Fig. 2b. The spatial pattern of ?SLP consists of a north-south-oriented dipole over the Euro-Atlantic region and a strong anticyclone over the Eurasian continent (Fig. 2b). First, the positive ?SLP over the North Atlantic suggests a weaker Icelandic low; here, models with stronger sea-ice loss reinforce the MME response (Fig. 1b). Second, the negative ?SLP over the Mediterranean region suggests a weaker northeastward extension of the Azores high; here, models with increased sea-ice loss suppress the MME response (Fig. 1b). Such a dipole ?SLP pattern suggests a linkage between the SVD1 and the projected difference of the NAO. Third, the signal over Asia represents a positive ?SLP anomaly at the northwestern flank of the Siberian high, where the climatological center is located at (40°-65°N, 80°-120°E) (Panagiotopoulos et al., 2005; Fig. 2b). This is opposite in sign to the MME response.
The linkage between ?SIC and the large-scale circulation response (Fig. 2b) could be associated with thermal interaction or a large-scale response to an eddy-mean flow interaction. To assess the relative importance of these two effects, we correlate the ?SIC of the SVD1 to the wind response at 250 hPa (the upper troposphere) and at 850 hPa (the lower troposphere). When the linkage is related to an eddy-mean flow interaction, the atmospheric response has an equivalent barotropic structure. In contrast, the atmospheric response due to a thermal forcing has a baroclinic structure (Hoskins and Karoly, 1981; Overland and Wang, 2010; Jaiser et al., 2012). As shown in Fig. 2c, models with a stronger SIC decline exhibit lower SLP over the entire Arctic. In the lower troposphere, the associated wind response is an anticyclonic flow over the Asian side, suggesting a baroclinic response. On the other hand, the associated wind response is a cyclonic flow over the Euro-Atlantic side, suggesting an equivalent barotropic response associated with an eddy-mean flow interaction. Next, in section 3, we investigate if the atmospheric response related to the ?SIC of the SVD1 is accompanied by forcing originating from outside of the Arctic.
Note that the objective of this study is to analyze the significant link between the projected uncertainties of the Arctic sea-ice decline and the Eurasian circulation. However, one may be interested to know if the intermodel response to the ?SIC of the SVD1 has the same or opposite sign to the MME response. One may also be interested to know if the sign of these responses is in large agreement across the models. Accordingly, we also show the MME response of the large-scale parameters and highlight the regions that the models agree on the sign of the response by at least 65% and 90%, i.e., the same as Figs. 1a and b.
2
3.1. SST and turbulent heat fluxes
During the recent AA period, one of the potential causes of the SIC decline is the remote signals of SST originating from the tropical Pacific (Ding et al., 2014; Trenberth et al., 2014). In particular, model studies suggest the Pacific Decadal Oscillation can contribute to AA (Svendsen et al., 2017; Tokinaga et al., 2017), and could modulate the response to sea-ice loss (Screen and Francis, 2016). Because part of the projected uncertainties of ?SIC is probably linked to the forcing outside the Arctic, it is interesting to see if the ?SIC of the SVD1 shows a strong linkage with the simultaneous response of the SST (?SST) and the associated turbulent heat fluxes anywhere.As shown in Fig. 3, only the Barents-Kara Sea and the midlatitude North Atlantic have pronounced differences in DJFM-mean ?SST associated with a stronger SIC decline. In the former region, the models robustly simulate an increase in SST and turbulent heat fluxes (Figs. 3c and d), which is related to the SIC decline. Associated with a stronger SIC decline of the SVD1, both the SST and turbulent heat fluxes have a stronger increase (Figs. 3a and b). For the second region, the majority of models simulate a weakened Atlantic meridional overturning circulation in the 21st century, although with large uncertainties in strength (Cheng et al., 2013; Collins et al., 2013; Reintges et al., 2017). Whereas the models robustly simulate a reduction of turbulent heat fluxes (Fig. 3d), they have a small agreement for the SST projection in this region (Fig. 3c). Because a stronger SIC decline of the SVD1 accompanies an anomalous SST warming in this region (Fig. 3b), the projected uncertainty of ?SIC may be related to the Atlantic meridional overturning circulation, either through an oceanic pathway (?rthun et al., 2012) or an atmospheric connection (Sato et al., 2014). Specifically, models with stronger SST warming coincide with stronger turbulent heat fluxes locally (Fig. 3b). This is associated with a decrease in the low-level baroclinicity (figure not shown) and weaker westerly winds in the lower and upper troposphere (Fig. 2c). Therefore, the midlatitude circulation response uncertainties associated with the ?SIC of the SVD1 could be due to both the projected uncertainties of the SIC decline and the SST warming in the North Atlantic (Woollings et al., 2012). The tropical SSTs seem to play an insignificant role in the dominant linkage between the uncertainties of sea-ice-Northern Hemisphere atmospheric responses in winter.
Figure3. (a, b) Intermodel regression of the forced response against the standardized expansion coefficient of SVD1 in boreal winter: (a) SST (K); (b) turbulent heat fluxes (shading; W m-2; positive upwards) and SLP (contours; hPa). Thick gray lines denote p=0.1 and dotted regions indicate p<0.1 for the shaded variable. (c, d) As in (a, b) but for the MME response of the shaded terms in (a, b), where white and black dotted regions indicate at least 7 (～ 65%) and 10 (～ 90%) out of 11 models agreeing on the sign of change.2
3.2. Zonal-mean circulation
The spatial pattern of both the ?SIC of the SVD1 and its associated turbulent heat fluxes in the polar region exhibit strong zonal wave number-0 components (Fig. 2a and Fig. 3b). Thus, we explore the linkage between the ?SIC of the SVD1 and the DJFM zonal-mean circulation changes at different altitudes. Among the 11 models, only three are high-top models with a model top above the stratopause (Table 1). Assuming that the low-top models do not resolve the stratospheric dynamics well, we only show the composite differences up to the 100-hPa level (the lower stratosphere).A stronger SIC decline associated with SVD1 is linked to an increased zonal-mean Arctic warming confined to the lower troposphere (Fig. 4a). Compared to the MME response, models with a stronger SIC decline (Fig. 4a) do not contribute significantly to the intermodel spread in the pronounced upper-tropospheric warming aloft in the Arctic and outside of the Arctic (Fig. 4d). This is consistent with other studies (e.g., Screen and Simmonds, 2010; Manzini et al., 2014; Blackport and Kushner, 2017; Ogawa et al., 2017 a). Models with more pronounced lower-tropospheric warming in the Arctic than in the low-latitude region exhibit weakening of the equator-to-pole temperature gradient and midlatitude westerlies (Fig. 4b). These tropospheric circulation features are the first-order response of AA (Cohen et al., 2014; Vihma, 2014). The SVD analysis suggests the uncertainties in the MME response seen in the midlatitude westerlies (Fig. 4e) are related to pan-Arctic sea-ice decline.
The dynamical response corresponding to a stronger SIC decline of the SVD1 can be approximated by weaker tropospheric Polar and Ferrell cells, where the mass stream function response is opposite in sign to the climatology, and the boundary between these two cells shifts southward (i.e., the zero-line shifts southward; Figs. 4b and c). When less cold polar air sinks near the surface, the SLP becomes lower across the polar region (Fig. 2b). This is associated with an anomalous upward motion in the poleward branch of the Polar cell, and an anomalous downward motion in the equatorward branch of the Polar cell and the poleward branch of the Ferrell cell (Figs. 4b and c). Due to the linkage between the vertical velocity and the surface divergence, there is a stronger increase in SLP around 60°N (Fig. 2b), where the anomalous zonal-mean downward motion is strongest (Fig. 4b). At the southern flank of the positive SLP response linked to a stronger Arctic warming response (Fig. 2b and Fig. 4a), the deceleration of westerly winds is strongest (～50°N; Fig. 4b). This anomalous zonal-mean zonal wind response has a barotropic structure, with pronounced easterly anomalies in the upper troposphere and the lower stratosphere (Fig. 4b).
Figure4. (a-c) Latitude-height cross sections showing the intermodel regression of the forced response of the zonal-mean fields against the standardized expansion coefficient of SVD1: (a) air temperature (K); (b) zonal-mean zonal wind (shading; m s-1) and meridional wind together with the vertical velocity (vectors; m s-1 in the meridional direction and 0.01 Pa s-1 in the vertical direction); (c) mass stream function (109 kg s-1), where the black contours represent the 2069-98 climatology (109 kg s-1). Thick green lines denote p=0.1 and dotted regions have p<0.1. (d-f) As in (a-c), but for the MME response of the shaded terms in (a-c), where white and black dotted regions indicate at least 7 (～65%) and 10 (～90%) out of 11 models agreeing on the sign of change. In (e), the lines represent the intermodel standard deviation (interval: 0.25 m s-1) of the zonal-mean zonal wind change.It should be noted that the models do not robustly simulate a weaker Polar cell in the lower troposphere by the end of the century (Fig. 4f), although the SIC decline is a robust signal (Fig. 1a). This suggests that the MME response (not its uncertainties) of surface circulation changes in the Arctic are also influenced by the forcing other than the sea ice, such as tropical SST forcing (e.g., Ding et al., 2014). Moreover, the models tend to simulate a strong Polar cell in the upper troposphere (Fig. 4f). Similarly, whereas the models robustly simulate weakening of the upper-tropospheric zonal wind aloft in the Arctic (Fig. 4e), the zonal-mean zonal wind here is slightly weakened by a stronger SIC decline of the SVD1 (Fig. 4b). Although the regressed anomalies are statistically significant, the magnitude is small compared to the MME response (Figs. 4b and e). These again suggest that the strong sea-ice decline of the SVD1 is not associated with strong upper-tropospheric circulation changes aloft in the Arctic.
In the midlatitudes, the zonal-mean zonal wind generally strengthens and this MME response is most robust near the tropopause and in the lower stratosphere (Fig. 4e). This is due to an intensification and a northward shift of the subtropical jet in response to global warming (Seidel et al., 2008). Because the zonal-mean zonal wind response that is linked to a stronger SIC decline of the SVD1 is opposite in sign to the MME response (Figs 4b and e), the SIC-related forcing appears to weaken the global warming response. This contrast can also be seen in the mass stream function of the Ferrell cell, where the anomalous response to a stronger SIC decline of the SVD1 is positive in sign (Fig. 4c) and the MME response is negative in sign (Fig. 4f). The positive anomalous response suggests an anomalously weaker Ferrell cell (Fig. 4c), which accompanies less poleward transport of eddy momentum and heat fluxes. In addition, the intermodel spread of the zonal-mean zonal wind is largest in the stratosphere (above 100 hPa; not shown) and it extends downward into the lower troposphere (Fig. 4e). The strengthening of the midlatitude zonal-mean zonal wind in the MME response appears to be linked to the stratospheric signals, whereas the weakening of the zonal-mean zonal wind in the SVD1 is due to SIC-related signals. The former is consistent with (Manzini et al., 2014), who highlighted the importance of stratospheric forcing in future surface circulation changes.
In addition to the linkage with anomalously weaker Polar and Ferrell cells, the stronger SIC decline of the SVD1 is linked to an overall weaker Hadley cell (Fig. 4c). Similar to the MME response, the anomalous response to a stronger SIC decline of the SVD1 suggests a stronger Hadley cell at its northern edge and in the upper troposphere (Fig. 4f). This represents a northward shift and a deeper Hadley cell. In short, the ?SIC of the SVD1 is linked to the hemispheric-scale circulation in boreal winter, where the classical three-cell meridional circulations are weakened, consistent with weaker poleward heat transport (Kang et al., 2008).
2
3.3. Eurasian circulation
Whereas the ?SIC of the SVD1 has a strong linkage with the projected difference of the zonal-mean circulation, it also has a zonal asymmetric component (Fig. 2a). But how strongly does it affect the intermodel agreement of the large-scale circulation features in Eurasia, including the heterogeneous SLP pattern as shown in Fig. 2b? To demonstrate these linkages, we show the intermodel regression of different large-scale atmospheric variables against the expansion coefficient of the SVD1 for the DJFM period in Fig. 5. Because the ?SIC of the SVD1 is almost perfectly correlated to the response of the total sea-ice extent, we also define several large-scale circulation indices (Table 2) and show their scatterplot against the response of the total Arctic sea-ice extent for the DJFM period in Fig. 6.3.3.1. Central and East Asia
Recall that the MME of ?SIC shows the largest decrease in SIC around the sea-ice edge, where the primary center is located at the Barents-Kara Sea (>60%) and the secondary center is located at the Bering Strait (>40%; Fig. 1a). The intermodel regression shows that the largest decrease north of the Kara Sea (>50%) and the difference over the Barents Sea opening is insignificantly small (<10%; figure not shown). The local response to stronger pan-Arctic sea-ice decline exhibits the largest increase in surface air temperature near the Kara Sea (Fig. 5a). Meanwhile, the stronger sea-ice decline leads to an increase in the water vapor content in the air column (Bintanja and Selten, 2014). This also enhances the precipitation (Fig. 5b) and decreases the vertical stability (Fig. 5d) locally. These changes reinforce the MME response (Figs. 5e-g). As the Arctic warming extends upward in the lower troposphere, the 1000-500 hPa thickness height increases and attains a maximum over the Barents-Kara Sea (～75°N, 50°E; Fig. 5a). This is associated with a stronger surface anticyclone over the Urals-Siberia region (～60°-110°E) and stronger southerly winds near the Barents Sea (Fig. 5b).
Figure5. Intermodel regression against the standardized expansion coefficient of SVD1 in DJFM: (a) surface air temperature (shading; K) and thickness height between 1000 and 500 hPa; (b) precipitation (shading; mm month-1) and 850-hPa wind (black vectors; m s-1), (c) meridional surface air temperature gradient (10-5 K m-1); (d) vertical stability at 925 hPa (K hPa-1). Thick white lines denote p=0.1 and dotted regions and vectors have p<0.1. (e-h) As in (a-d) but for the MME response of the shaded terms in (a-d), where white and black dotted regions indicate at least 7 (～ 65%) and 10 (～ 90%) out of 11 models agreeing on the sign of change.
Figure6. Scatterplots of the forced response of large-scale circulation indices against the decrease in sea-ice extent in DJFM: (a) Urals-Siberia SLP; (b) Icelandic low index; (c) Mediterranean SLP; (d) NAO index. In each plot, the number denotes the response of individual models listed in Table 1, whereas the open circle represents the MME response. The correlation of the intermodel regression line (thick solid line) and the corresponding level of significance are shown at the top.The intermodel correlation between the SLP response over the Urals-Siberia region and the pan-Arctic sea-ice decline is -0.752 (～ 57% of the total variance; Fig. 6a). The SIC signals of the SVD1 appear to modulate instead of dominate the SLP response, as most models (9 out of 11) simulate a negative SLP response over this region (Fig. 6a). The anticyclone related to the increased SIC decline extends across the whole of northern Asia. Whereas the stronger anticyclone likely strengthens the northerly cold-air advection, the meridional temperature gradient over the high-latitude region sharply decreases and this weakens the northerly cold-air advection (Fig. 5c). Hence, it is unclear if the seasonal-mean cold-air advection is strengthened by a larger sea-ice decline of SVD1. Note that correlation analysis does not imply any causality of the linkage (i.e., increased Arctic SIC decline could instead be driven by the Eurasian SLP changes, or both the sea ice and SLP might be independently affected by a third factor).
The anomalous surface air temperature response of the SVD1 shows a more pronounced warming spread across the high-latitude region of Asia (Fig. 5e), whereas part of the Siberian-Mongolian region (～ 40°-55°N, 90°-120°E) has a slight and insignificant "cooling" associated with the SVD1 [note that this "cooling" means the warming is less pronounced, as the magnitude of the intermodel regression is much smaller than the MME response (Figs. 5a and e). The stronger increase in temperature over northern Asia (Fig. 5a) is mainly due to the stronger reduction in the meridional temperature gradient (Fig. 5c). Part of the stronger warming over Northeast Asia (～ 100°-140°E) is related to the increase in vertical stability (Fig. 5d). The change in the downwelling shortwave radiation and the turbulent heat fluxes play an insignificant role (figure not shown).
3.3.2. Euro-Atlantic region
Over the Euro-Atlantic region, the intermodel regression against the SVD1 projects on to a negative NAO-like dipole pattern, with an anomalous high near Iceland, weak anomalies over the subtropical Atlantic, and an anomalous low near the Mediterranean Sea (Fig. 2b). On the one hand, the majority of models (9 out of 11) simulate a weaker Icelandic low that is intensified in models simulating a stronger sea-ice decline of the SVD1 (Fig. 6b). A stronger sea-ice decline is associated with a weaker Polar cell and anomalous downward motion near 60°N (Fig. 4b), which is close to the center of action of the Icelandic low. Moreover, a stronger sea-ice decline of the SVD1 is accompanied by a stronger Arctic warming and a smaller equator-to-pole temperature gradient. According to (Harvey et al., 2015), this is related to the lower tropospheric baroclinicity and is hence crucial for reducing the storm tracks in the northern North Atlantic (see their Fig. 5c). As the sea-ice decline is a robust feature in the future climate, the increase in SLP near the Icelandic low region appears to be linked to the storm track changes. Under a stronger SIC decline of the SVD1, the meridional surface temperature gradient becomes weaker along the Gulf Stream (Fig. 5c). As can also be seen in Fig. 5b, this accompanies an anomalous anticyclonic flow and negative precipitation anomalies extending northeastward from Iceland toward Scandinavia. All the aforementioned features suggest a further reduction in the Northeastern Atlantic storm tracks (Rogers, 1997), which needs to be investigated in future studies.
On the other hand, all but one of the models simulate an increase in SLP in Mediterranean Europe (Fig. 6c), and this response is strongly suppressed by a stronger SIC decline of the SVD1 (Fig. 2b). As can be inferred from Fig. 5b, a stronger sea-ice decline is associated with an anomalous cyclonic flow over the tropical and subtropical North Atlantic. The Azores high might have a smaller northeastward extension toward Mediterranean Europe, where the SLP robustly increases (Fig. 1c). This anomalous response can be regarded as a weaker Hadley cell (Fig. 4c), where the intermodel correlation between the zonal-mean mass stream function averaged over 10°-20°N in the 850-500 hPa levels and the SLP over Mediterranean Europe is +0.872. The anomalous low over Mediterranean Europe is associated with a stronger southerly advection of the warm subtropical air toward southeastern Europe. This accompanies an anomalous increase in surface air temperature and precipitation over part of Central Europe, Mediterranean Europe and the Middle East (Figs. 5a and b).
Because a stronger pan-Arctic sea-ice decline is linked to weakening of the Icelandic low but little change to the intensity of the Azores high, it has a significant negative correlation with the NAO response (Fig. 6d). However, it is noticeable that the NAO response does not robustly show a negative tendency. The spread is consistent with the inconsistency of the NAO response among previous studies (Vihma, 2014), suggesting other factors also affecting the NAO change. Moreover, the stronger negative NAO response does not correspond to a colder and even a less warm climate over Europe (Fig. 5a). Altogether, a stronger pan-Arctic sea-ice decline in boreal winter might significantly modulate the key circulation features over Eurasia, where the anomalous SLP and precipitation responses (Fig. 2b and Fig. 5b) are often opposite in sign to the MME response (Fig. 1c and Fig. 5f). However, an anomalous high does not correspond to anomalous cooling, unlike the warm-Arctic-cold-Eurasia temperature pattern during the recent AA period (e.g., Cohen et al., 2014).
Our results suggest that a stronger Arctic sea-ice decline of the SVD1 is associated with an anomalous increase in SLP over high-latitude Eurasia, including the Urals-Siberia region and the Icelandic low region. However, we did not find significantly stronger northerly winds over Eurasia. Nor did we find any cooling or even less pronounced warming in any part of Eurasia. This is different from the present climate, where a stronger high pressure and persistent snow cover over Eurasia might enhance the upward propagation of the planetary waves from the troposphere to the stratosphere (e.g., Allen and Zender, 2011; Cohen et al., 2012; Kim et al., 2014). One possible reason is that only 3 out of the 11 models have a well-resolved stratosphere, which is crucial for simulating the midlatitude atmospheric response associated with the troposphere-stratosphere interaction (Omrani et al., 2014, 2016; Nakamura et al., 2015; Zhang et al., 2017).
Other possible reasons are that the impact of the pan-Arctic sea-ice loss is different from the regional sea-ice loss (Screen, 2017), or the recent changes are dominated by internal atmospheric variability (McCusker et al., 2016; Ogawa et al., 2017 a). Regarding the former factor, a stronger anticyclone over high-latitude Eurasia driven by a regional sea-ice decline over the Barents-Kara Sea could enhance the downstream cold-air advection, as the sea ice is still present downstream. The resultant dynamic effect of regional sea-ice loss could cause cooling in the mid and high latitudes (Mori et al., 2014; Kug et al., 2015; Overland et al., 2015).
Conversely, in response to the pan-Arctic sea-ice loss, the cold-air intensity over the entire Arctic becomes weaker. The meridional temperature gradient in the high latitudes decreases and the midlatitude westerlies weaken. Unless the northerly winds become much stronger (such as a higher amplitude flow), the northerly cold-air advection would be weaker. (Screen, 2014) also suggested that the northerly wind makes a larger contribution to the warming trend in the high-latitude region than the southerly wind during boreal winter. Recently, (Meleshko et al., 2016) showed that the ocean heat transport is more important for the higher amplitude planetary wave in the midlatitudes. Based on these findings, it is unlikely that the pan-Arctic sea-ice loss causes any cooling effect in the extratropical region on seasonal timescales, which agrees with the results of (Deser et al., 2016) and (Screen, 2017).
A stronger Arctic sea-ice decline of the SVD1 is also linked to weakening of the three-cell circulations and a warmer SST in the midlatitude North Atlantic. In the midlatitudes, a weaker Ferrell cell is characterized by a higher SLP in the midlatitudes and weaker zonal-mean zonal winds. However, we cannot assess the causality of these linkages. Indeed, the stronger SST warming over the North Atlantic could enhance the poleward ocean heat transport and could then melt more sea ice (Mahlstein and Knutti, 2011; Jung et al., 2017; Nummelin et al., 2017). The basin-wide Atlantic warming is also crucial for the negative tendency of the NAO, via an atmospheric wave train (Sato et al., 2014) and troposphere-stratosphere interaction (Omrani et al., 2016). The atmospheric response to the sea-ice decline might also be highly nonlinear (Petoukhov and Semenov, 2010). In a future study, we intend to design several sensitivity experiments based on the intermodel spread of SST and SIC, in order to assess the relative importance of these projected uncertainties in the future climate change of the Northern Hemisphere.
