Winter Eurasian cooling linked with the Atlantic Multidecadal Oscillation

In this paper, we analyze observational and reanalysis data to demonstrate that the Atlantic Multidecadal Oscillation (AMO) significantly modulates winter Eurasian surface air temperature through its impact on the shape, frequency and persistence of Ural blocking (UB) events that last for 10–20 d. This impact results from changes in mid-high latitude westerly winds over Eurasia associated with the warming in the Barents–Kara Seas (BKS) through the AMO-driven high sea surface temperature and sea-ice decline and resultant weakening in meridional temperature gradients. The BKS warming has a strongest positive correlation with the AMO at a time lag of about 14 years. During the recent positive AMO phase, more persistent northwest–southeast (NW–SE) oriented UB events are favored by weakened westerly winds in Eurasian mid-high latitudes. Through cold atmospheric advection and radiative cooling, such UB events produce a strong, persistent and widespread cooling over Eurasia and enhance BKS warming during 1999–2015. However, the positive AMO phase cannot directly produce the Eurasian cooling if the UB is absent. Thus, we conclude that the recent AMO phase change is a major cause of the recent winter cooling over Eurasia through its impact on BKS temperature and sea ice, which in turn affect the meridional temperature gradient, the westerly winds and the UB events.


Introduction
Rapid winter warming over the Arctic and large winter cooling over Eurasia are the most important manifestations of the recent climate change over northern latitudes (Cohen et al 2014). While some studies have attributed the Arctic warming to anthropogenic forcing (Gillett et al 2008), Arctic surface air temperature (SAT) and sea-ice cover (SIC) were found to exhibit clear multidecadal variations associated with the Atlantic Multidecadal Oscillation (AMO) (Levitus et al 2009, Wyatt et al 2012, Mahajan et al 2011, Miles et al 2014, which reflects 60-80 year quasi-periodic oscillations in the sea surface temperature (SST) over the North Atlantic basin (Mann et al 1995. Many studies have indicated that the winter Eurasian temperature shows a clear multi-decadal variability linked to the AMO , Wyatt et al 2012, Hao et al 2016, while it is also influenced by the North Atlantic Oscillation (NAO) (Li et al 2013, Hanna et al 2015, tropical forcing (Trenberth et al 2014), stratospheric processes (Scaife et al 2005, Omrani et al 2014, the Inter-decadal Pacific Oscillation (IPO) , Steinman et al 2015 and decadal modulated oscillation (Huang et al 2017).
In particular, the winter Eurasian continent has exhibited a large cooling since the 1990s (Cohen et al 2014) that concurs with the recent global warming hiatus , Huang et al 2017. Some studies have attributed the recent Eurasian cooling to recent winter sea-ice loss (Honda et al 2009, Outten and Esau 2012, Cohen et al 2014, Mori et al 2014, Kug et al 2015 because it is coincident with the Arctic sea-ice loss in the recent decades. However, a recent modeling study suggests that the winter Eurasian cooling is not related to the winter sea-ice loss in the Barents-Kara Seas (BKS), but may be due to other internally-generated atmospheric circulation anomaly patterns (McCusker et al 2016), while the atmospheric responses to the winter (Luo et al 2016a) and autumn (Wang et al 2017) sea ice anomalies are different. On the other hand, our recent studies (Luo et al 2016aLuo et al 2016b suggest that individual Eurasian cold events are often linked to the presence of Ural blocking (UB) and their increased frequency can form a warm Arctic-cold Eurasia pattern to lead to a pause of global warming during 2000-2013 (Yao et al 2017, Deser et al 2017. The exact physical cause of the winter Eurasian cooling trend is still under debate (Semenov and Latif 2015, Kug et al 2015, Overland et al 2016, Huang et al 2017. In particular, it is unclear how the AMO contributes to the recent winter Eurasian cooling.
Because the SST and surface air temperature (SAT) over the BKS and sea-ice loss in winter are related to the AMO through the incursion of warm Atlantic water (Levitus et al 2009, Day et al 2012, Miles et al 2014, it is suggested that the AMO may affect the winter Eurasian cooling through its influence on the BKS warming related to the SIC decline in that the cold Eurasian anomaly occurs together with the BKS warming (Outten and Esau 2012, Cohen et al 2014, Luo et al 2016a. In this paper, we will investigate how the AMO influences the winter Eurasian temperature by analyzing observational and reanalysis datasets. We found that the AMO can contribute to the winter Eurasian cooling through changes in the frequency, persistence and shape of the UB patterns due to changes in westerly winds over Eurasia that are related to the BKS warming.

Data and method
The winter (December to February, DJF) daily data with 2.5 • × 2.5 • , which includes daily 500 hPa geopotential height, 500 hPa zonal wind and air temperature at 995 hPa defined as the SAT, is taken from the National Centers for Environmental Prediction-National Center for Atmospheric Research (NCEP-NCAR) reanalysis from December 1950to February 2016(1950-2015 (www.esrl.noaa. gov/psd/data/gridded/data.ncep.reanalysis.html). The data has been detrended for a linear trend during 1950-2015 and non-seasonalized.
The monthly mean SST and SIC data with 1 • × 1 • grid points from 1950-2015 is available from the Hadley Centre Global Sea Ice and Sea Surface Temperature (HadISST) datase (www.metoffice.gov. uk/hadobs/hadisst/data/download.html). The winter AMO index during 1950-2015 is calculated by averaging the monthly SST anomalies over the North Atlantic (0 • -70 • N, 80 • W-0 • ) in a winter. The anomaly at each grid point during 1950-2015 was calculated as the deviation from its long-term  mean for each day of the winter.
The method used to identify individual atmospheric blocking events with 10-20 d timescales occurring in the Ural Mountains during 1950-2015 is based on the one-dimensional blocking index of Tibaldi and Molteni (1990) or TM index in terms of meridional gradients of 500 hPa geopotential height. A UB event is defined if the TM index is satisfied in the region (30 • -90 • E) and persists for at least three consecutive days. We also use a regression-based method to estimate the contributions of the AMO and UB events to the SAT variations in the BKS and central Asia. Mathematically, the linear regression model can be written in the form of T A = ΔΣ+ , where and are the regression coefficients, respectively. If the AMO index or UB days is denoted by ΔΣ, then T A represents the AMO-(UB-) related SAT variation.
To reflect the contribution of UB events to the DJFmean SAT anomaly, we can express the DJF-mean SAT in BKS (central Asia) and 500 hPa zonal wind anomaly fields as T BKS = ΔT BKS + Δ̃B KS (T C = ΔT C + Δ̃C) and U E = ΔU E + Δ̃E, where T BKS and T C are the DJF-mean SAT anomalies in BKS and central Eurasia, respectively; and U E is the DJF-mean 500 hPa zonal wind anomaly over mid-high latitude Eurasia. Moreover, ΔT BKS (ΔT C ) and ΔU E represent the DJF-mean SAT and 500 hPa zonal wind anomalies without the UB days respectively. Here, the UB days are defined to be the occurrence days of the UB identified by the TM index. Moreover, Δ̃B KS (Δ̃C) and Δ̃E denote the DJF-mean values of the SAT and 500 hPa zonal wind anomalies with the UB days, and thus ΔT BKS (ΔT C ) and ΔU E can be considered as the background or prior conditions of UB events because they do not include the effect of UB events.

Results
To understand the possible role of the AMO in the Eurasian air temperature change, it is useful to show the normalized time series of the winter-mean AMO index, SST and SIC anomalies averaged in the BKS in figure  1  Year 1950 1960 1970 1980 1990 2000 2010 Year  and this cooling is associated with increasing SAT and SST and decreasing SIC in the BKS, as well as the recent phase change of the AMO (figure 1(a) and (b)). A decadal cooling from 1960-1973 was also seen over CA when the AMO was on a declining phase, similar to that happened from about 1999-2009 although the AMO index decreased less during this later period (figure 1(a) and (b)). The detrended DJF-mean SAT in the BKS region (T BKS ) exhibits a significant correlation of 0.76 (−0.72) with the SST (SIC) there during 1950-2015. It is also not surprising that the detrended DJF SST and SIC are strongly anti-correlated (r = −0.95) in the BKS. The positive SST anomaly in the BKS has been partly attributed to the intrusion of warm Atlantic water into the BKS during AMO positive phases (Levitus et al 2009). Luo et al (2016b) also found that the SST in the North Atlantic basin is warmer during 2000-2013 than during 1979-1999, which corresponds to a case that the AMO index tends to be positive. We further see that the SAT in BKS has a maximum positive correlation with the AMO index as the later leads the former by about 14 years (figure S1(a) available at stacks.iop.org/ERL/12/125002/mmedia). The fact that the detrended SST and SAT (SIC) anomalies on interannual timescales have peak positive (negative) correlations with the AMO index when the later leads the former by about 3 years (figure S1(a)) suggests that this intrusion through advection from the northern North Atlantic Ocean into the BKS may take about 3 years. In fact, some studies indicated that the advection time of the Atlantic SST anomaly from the Norwegian Sea to the BSK is of about 2∼3 years (Vinje 2001, Årthun et al 2012Årthun and Eldevik 2016, while the warm Atlantic water has a decadal timescale with a dominant period of 14 years (Eldevik et al 2009). To understand the relationship between the SIC, SST and SAT anomalies and the AMO index on a timescale longer than 10 years, their lead-lag correlations are calculated and shown in figure S1(b) using a 9 year smoothing. It is obvious that for 9 year smoothed time series, the AMO leads the SIC (SST) in BKS by about 10 (10−12) years and the SAT in BKS by about 10∼16 years (figure S1(b)). While we cannot infer the cause and effect relationship between the SST in the BKS and AMO from their correlation coefficient (figure 1(a)), it is believed that the high SST over the BKS is partly due to the inflow or advection of the warm Atlantic water into the BKS during the positive AMO (AMO + ) phase (Levitus et al 2009, Eldevik et al 2009, Årthun et al 2012, Årthun and Eldevik 2016 although sea ice loss can enhance the local warming. This notion is consistent with the fact that AMO leads the BKS SST by about 10−12 years (figure S1(b)), since the oceanic advection takes a number of years. Gulev et al (2013) recently found that on decadal or multidecadal timescales the oceanic process plays a major role in the SST change and associated atmospheric variability over the North Atlantic. Thus, an AMO + would enhance surface warming over the BKS through the intrusion of warm surface waters from the North Atlantic and the subsequent positive feedback from sea-ice loss ( figure S2). This process can amplify the AMO's impact and generate a strong negative correlation (r = −0.72) between BKS SIC and T BKS (figure S2(d)).
The SAT over CA (T C ) and AMO index show peak correlations when the AMO leads T C by about 15 (13) years for the unsmoothed (9 year averaged) time series (figure S1(c)), although they have no in-phase correlation (r = 0.03). This suggests that the AMO could be a cause of the T C variations. Here we explore the physical links behind this possible cause. As explained above and in other studies, the positive AMO is likely a cause of the recent BKS warming (Chylek et al 2009) and seaice loss (Day et al 2012, Miles et al 2014. In fact, the SAT over the BKS and CA is modulated by not only the AMO (Chylek et al 2009, Wyatt et al 2012, but also by mid-latitude large-scale atmospheric circulations over the Eurasian continent (Luo et al 2016a) and North Atlantic (Luo et al 2016b). But, how the AMO modulates the Eurasian SAT via the sea-ice loss over the BKS is still unclear, because the linkage between the BKS seaice loss and Eurasian cooling is not better established and still controversial (Cohen et al 2014, Mori et al 2014, Kug et al 2015, Outten and Esau 2012, McCusker et al 2016. Below we reveal that the AMO-driven BKS warming can cause large Eurasian cooling by changing the frequency, shape and persistence of atmospheric blocking over the Ural Mountains (around 60 • E).
One way for the AMO-driven warming over BKS to affect SAT in central Asia is through reducing meridional temperature gradients, which in turn would weaken westerly winds and alter atmospheric circulation over Eurasia (Yao et al 2017). Most of Eurasia has undergone pronounced cooling since the early 1990s (Cohen et al 2014), and this cooling is associated with a positive height trend in the Ural region (figure 1(c)) that resembles the recent change pattern of the UB (Luo et al 2016a). The composite SAT and height anomaly patterns associated with UB events (figure 1(d)) also match those of the recent SAT and height trend patterns over Eurasia ( figure 1(c)). These results suggest a possible role of UB in recent Eurasian cooling.
To examine the further impact of UB, we classified the UB events into three types of orientation (figure 2): northwest-southeast (NW-SE)-(figure 2(a)), northeast-southwest (NE-SW)-(figure 2(b)) and north-south (NS)-oriented (figure 2(c)) blocking patterns, similar to Luo et al (2015). The total number of the detrended UB events (figure 3(a)) is correlated with detrended T BKS (T C ) with a correlation coefficient of 0.35 (−0.41) during 1950-2015 that is statistically significant. It reflects the contribution of UB to the BKS warming and Eurasian cooling (Luo et al 2016a). The total number of UB events (figure 3(a)) has a peak positive correlation with the AMO index as the AMO leads the UB by about 14 years ( figure S3(a)). This result is also held for the NW-SE-oriented UB events (red line in figure S3(b)). The NS-oriented UB exhibits peak positive correlations with the AMO index as the AMO leads the NS-oriented UB by about 21 years (green line in figure S3(b)), while the NE-SW-oriented UB has a peak 6 year leading negative correlation with the AMO (blue line in figure S3(b)). For a 9 year moving average the NW-SE-(NS-) oriented UB lags the AMO by about 11 (20) years ( figure S3(c)). Thus, on decadal or multidecadal timescales the shape and frequency of UB events are related to the phase of the AMO. To examine the impact of the AMO phase on the blocking frequency, during the AMO + (1951−59 and 1999−2015) and AMO − (1968−1992) periods based on ±0.8 standard deviations of the AMO index ( figure 1(a)) the number of the total UB events and the corresponding ratios of the three blocking types relative to the total UB events can be calculated and shown in figure 3(f). We see that there are 21 NW-SE-and 4 NE-SW-oriented UB events during the AMO + phase, but there are 12 NW-SE-and 12 NE-SW-oriented UB events during the AMO − phase. While the change of the total UB events from the AMO − to AMO + phases is not significant because it corresponds to a small change from 36 to 32 cases (figure 3(f)), the variation of the NW-SE-oriented UB events between the AMO − and AMO + phases is significant. A further calculation shows that the mean duration of the NW-SE-(NE-SW-) oriented UB events is 9.5 (6.4) days, though the NS-oriented UB events have a mean duration of 7 d (figure 3(e)). Thus, while the BKS sea-ice loss associated with the AMO + phase does not significantly increase the frequency of total UB events, it can correspond to an increased frequency of NW-SE-oriented UB events and the lengthening of their mean life period due to intensified BKS warming. As we show below, the NW-SE-oriented UB pattern associated with AMO + can produce a strong winter Eurasian cooling. The time-mean fields of composite daily geopotential height and SAT anomalies, averaged over a mature period of the UB from lag −5 to 5 day (lag 0 denotes the peak day of the UB), show that the anticyclonic (cyclonic) anomaly of the NW-SE-oriented UB events is stronger ( figure 4(a)) than the other two types (figures 4(b) and (c)). The corresponding cold anomaly is also more intense, widespread (blue shading in figure  4(a)) and persistent (red line in figure 4(e)) than those of NE-SW-and NS-oriented UB events. This is easily explained because the NW-SE-oriented UB is more long-lived than the other two types (figure 3(e)) and because the persistent and strong cooling occurs easily over central-eastern Asia through cold advection and radiative cooling (Yao et al 2017). While the magnitude of the maximum warming in the BKS is almost the same for NW-SE-, NE-SW-and NS-oriented UB events (figure 4(d) for day 0), it seems that the intraseasonal BKS warming of NW-SE-oriented UB events is most persistent amongst the three blocking types (figure 4(d)). Thus, the presence of NW-SE-oriented UB events is able to generate a strong, persistent and widespread Eurasian cooling (figure 4(a)) and to further amplify the BKS warming prior to the block onset (blue line in figure 4(d)).
Here, a key question we need to answer is why the BKS warming is favorable for NW-SE-oriented UB events. The regressed DJF-mean 500 hPa zonal wind against the T BKS time series shows that the Eurasian westerly wind is weakened in mid-high latitudes, but intensified in lower latitudes (figure 4(f)). Such a zonal wind distribution will lead to a NW-SE orientation of the UB dipole because the anticyclonic (cyclonic) anomaly of the UB dipole in the reduced (intensified) westerly wind region is displaced westward (eastward). Thus, the BKS ice loss inevitably brings such an effect because the sea ice loss in the BKS corresponds to an intensified warming ( figure S2(d) figure 1(b) (red line). Lag 0 denotes the peak day of the UB. In (a)-(c), the contour and shading represents the regions above the 95% confidence level, whereas in (f) the dot represents the region above the 95% confidence level.
warming includes the effect of the UB. This result is also held for the prior BKS warming without UB days because the prior BKS warming ΔT BKS is highly proportional to the BKS warming (T BKS ) with UB days ( figure S4(a)). The calculation seems to indicate that the Eurasian SAT (ΔT C ) without UB days is unrelated to the prior BKS warming (ΔT BKS ) ( figure S4(b)). This suggests that on an interannual timescale the Eurasian cooling is not directly linked to the BKS warming as the UB is absent, though the Eurasian mid-high latitude westerly wind (EWW) is closely related to the BKS warming (figures S4(c) and (d)). The DJF-mean warming in the BKS tends to weaken the EWW even for the case without UB days ( figure S4(b)) through reducing the meridional temperature gradient (not shown). Because the reduced EWW corresponds to an increased persistence of UB (Luo et al 2016a, Yao et al 2017), the BKS warming lengthens the lifetime of the UB events or increases the blocking days (figures S4(e) and (f)). Thus, the above results clearly indicate that the BKS ice loss provides an environment that promotes NW-SEoriented UB events and lengthens their lifetime due to the reduction of the EWW related to BKS warming. On interannual timescale the BKS warming and Eurasian cooling have no in-phase correlations with the AMO (figures 5(a) and (b)). However, on a decadal timescale the BKS warming and Eurasian cooling show maximum correlations with the AMO as they lag the later by about 14 years (figure S1). Thus, the AMO may have a delay effect on the Eurasian cooling through the BKS warming. It is expected that the winter Eurasian cooling is strongly modulated by the AMO because the BKS warming corresponds to more UB days ( figure 5(c)) and because the persistent UB pattern The value of R in the map denotes the correlation coefficient. In (a) and (b), the slope between T BKS (T C ) and the AMO index is not statistically significant, but in (c) and (d) the slope between T BKS (T C ) and blocking days is statistically significant at the 95% confidence level. In (e)-(f), the blue line represents the SAT variation obtained from the reanalysis data and the green (black) line denotes the time series of the regressed AMO-(UB-) driven SAT anomalies over the BKS and Eurasia. The dashed line represents a second 9 year running average. The unit of S is K day −1 in (c) and (d).
leads to a stronger Eurasian cooling (figure 5(d)) through cold advection and radiative cooling (Yao et al 2017). Thus, it is concluded that the AMO has an important contribution to the Eurasian cooling in winter through the BKS warming and its influence on the UB.
Figures 5(e) and (f) show the time variations of the SAT anomalies in the BKS and Eurasian continent obtained from reanalysis data (simply defined as an observed one) and the AMO-and UB-induced components using a simple linear regression model. It is found that the AMO-induced SAT anomalies have consistent variations over the BKS and CA regions (green line in figures 5(e) and (f), though their values are small, thus indicating that the AMO + cannot directly produce a cooling in the CA during 2000-2015 if the UB is absent (green line in figure 5(f)). The correlation calculation shows that the lagged correlation coefficient of the AMO-induced SAT in BKS with the AMO index is the same as that in CA ( figure S5(a)). It is also found that the UB-induced BKS SAT variation (black line in figure 5(e)) has a significant positive correlation of 0.43 (0.5) with observed T BKS on an interannual timescale (a 9 year moving average). This means that the UB can amplify the AMO-induced BKS warming, although the AMO-induced BKS SAT change is relatively small. Thus, on a decadal timescale the AMO + and UB also contribute to the BKS warming. On interannual and decadal (9 year running average) timescales (figure 5(f)) in the CA region, while the AMO-induced CA SAT is hardly correlated with observed T C , the UB-induced CA SAT (black line in figure 5(f)) can have a strong positive correlation of 0.59 (0.81) with observed T C (blue line in figure 5(f)). Moreover, it is found that the peak values of the UB-induced BKS warming ( figure S5(b)) and Eurasian cooling (figure S5(c)) lag the AMO by about 14 years. This result is also held for 9 year smoothed time series. Therefrom, it inferred that in the recent decades the UB, especially the NW-SE-oriented UB, has a large contribution to the central Asian cooling through the modulation of the BKS warming triggered by the AMO + and enhanced by the sea-ice loss. We further examined the sensitivity of the results in figure 5 to other regression methods such as the Mann-Kendall and robustfit methods. The results were found to be insensitive to the method used (not shown), although based on the reanalysis data. However, it must be pointed out that many other factors such as tropical forcing (Trenberth et al 2014), stratospheric processes (Scaife et al 2005) and IPO  and so on can also influence the winter Eurasian cooling. Thus, how much the AMO contributes to the winter Eurasian cooling compared to these factors needs to be further quantitatively evaluated from different reanalysis and model datasets.

Conclusion and discussions
In this paper, we have examined the modulation of the AMO on the winter Eurasian SAT and its contribution to the recent winter Eurasian cooling in the  midlatitudes through analyses of observational and reanalysis data. It is found that the recent positive phase of AMO (AMO + ) causes BKS warming during 1999-2015, likely through the advection of North Atlantic warm surface waters and the positive feedback associated with sea-ice loss. The BKS warming weakens the Eurasian mid-high latitude westerly wind (EWW), which in turn leads to changes in the shape, frequency and persistence of the UB events that contribute to the recent Eurasian cooling. We found that the AMO + is able to produce frequent long-lived NW-SE-oriented UB events due to the weakened EWW resulting from the BKS warming. The NW-SE-oriented UB is an optimal circulation pattern that produces a strong, persistent and widespread cold anomaly over Eurasia through cold atmospheric advection and radiative cooling. We also find that the AMO + cannot directly generate a cooling over Eurasia if the UB is not present. Thus, the AMO can affect the Eurasian SAT likely only by means of its influence on BKS SST and sea ice, and subsequently the EWW and the frequency, persistence and shape of the UB through the BKS warming. The logic chain of how the AMO + leads to the BKS changes and the winter Eurasian cooling since the 1990s is summarized in figure 6 as a schematic diagram. Our results help understand the formation of the recent winter Eurasian cooling (Cohen et al 2014, McCusker et al 2016 as well as the changes over BKS. In this paper we did not examine NAO's impact on Eurasian temperatures (Scaife et al 2005, Li et al 2013, even though the UB is modulated by the phase of the NAO (Luo et al 2016b). In fact, the NAO has been found to have an important impact on the Northern Hemisphere air temperature and sea ice near Eurasia (Karpechko et al 2015). Eden and Willebrand (2001) indicated that the NAO can drive the North Atlantic SST variability mainly on an interannual to sub-decadal timescale, while Gulev et al (2013) found that the surface turbulent heat fluxes driven by the ocean can force the atmosphere on timescales longer than 10 years. To some extent, the AMO may be considered as a driving source for the multi-decadal variability of the SST, SAT and SIC in BKS (Levitus et al 2009, Chylek et al 2009, Day et al 2012, Miles et al 2014 and UB. Our results suggest that the winter Eurasian cooling is partly caused by the recent phase change of the AMO due to the AMO-induced decadal or multi-decadal changes in the frequency, persistence and shape of the UB events, in addition to other influences on the Eurasian air temperature such as stratospheric processes (Scaife et al 2005). However, further investigations are needed to differentiate the different contributions of the decadal changes in the NAO and UB to the decadal trend of the winter Eurasian cooling, although they are related.