Tropical origin for the impacts of the Atlantic Multidecadal Variability on the Euro-Atlantic climate

The Atlantic Multidecadal Variability (AMV), also known as Atlantic Multidecadal Oscillation (AMO), is a long-term alternation of warm and cold decades of sea surface temperatures (SSTs) in the North Atlantic (Schlesinger and Ramankutty 1994, Kerr 2000). AMV can have large impacts on rainfall, temperature and pressure in many regions of the Northern Hemisphere (Knight et al 2006), influencing the Atlantic Hurricanes intensity and frequency (Trenberth and Shea 2006), the Sahelian and the Indian/Asian Monsoons (Lu et al 2006, Zhang and Delworth 2006, Kucharski et al 2009) and even the mid-latitude climate (Häkkinen et al 2011).

The linkage between the AMV and the Euro-Atlantic climate is not totally understood: due to the large internal variability of the mid-latitude atmosphere and to the few cycles of the AMV in the observational period, it is hard to identify a robust connection between the two, especially in terms of causality. A common way to describe the mid-latitude atmospheric variability involves the North Atlantic Oscillation (NAO). The NAO is a mode of variability characterized by an oscillation of the pressure gradient between high and mid-latitudes over the North Atlantic (Hurrell et al 2003): during the positive phase the eddy-driven jet is shifted poleward and is well separated from the subtropical jet, whereas during the negative phase the eddy-driven jet is shifted equatorward and almost merged with the subtropical jet. The NAO pattern is often identified via Empirical Orthogonal Functions (EOFs) analysis (Ambaum et al 2001), but can also be characterized by the frequency of occurrence of Euro-Atlantic weather regimes (Cassou et al 2004, Ferranti et al 2014).

In the last few years the NAO phases and the jet variability have been described in terms of Rossby Wave Breaking (RWB) events, defined as the reversal of the potential temperature gradient measured at the tropopause level (McIntyre and Palmer 1983). RWB can be classified into cyclonic/anticyclonic events according to the environmental meridional shear of the zonal wind, usually known as barotropic shear (Thorncroft et al 1993, Peters and Waugh 1996, Tyrilis and Hoskins 2008). Cyclonic wave breaking is dominant on the poleward side of the jet-stream (i.e. over Greenland), whereas anticyclonic wave breaking is usually detected on the equatorward side (i.e. over Eastern Atlantic and Western Europe). Successive RWB events modulate the sign of the NAO and consequently the position of the jet stream (Benedict et al 2004, Franzke et al 2004, Strong and Magnusdottir 2008). Long-lasting RWB events are often characterized by atmospheric blocking (Pelly and Hoskins 2003), a persistent and quasi-stationary high-pressure system frequently occurring during wintertime at the exit region of the jet stream (Rex 1950, Davini et al 2012). Over the Euro-Atlantic region atmospheric blocking is intimately linked with the eddy-driven jet stream variability (Woollings et al 2010a) and with North Atlantic Oscillation phases (Woollings et al 2008). Indeed, high-latitude blocking over Greenland has been shown to be negatively correlated with the NAO, suggesting that the NAO itself could be interpreted as series of successive blocked and non-blocked conditions (Woollings et al 2010b). Since they are dynamically-based, measures of atmospheric blocking—and especially recent 2-dimensional indices which evaluate also RWB properties (Davini et al 2012, Masato et al 2012)—represent ideal diagnostics to investigate the mid-latitude climate variability in Global Climate Models (GCMs) (e.g. Davini and Cagnazzo 2014).

A few works have identified an opposite relation between the North Atlantic SSTs and the NAO phases (Peings and Magnusdottir 2014, Omrani et al 2014, 2015, Gastineau and Frankignoul 2015), with positive AMV leading to negative NAO. Recently Peings and Magnusdottir (2014) published in this journal the results of a comprehensive analysis of the relationship between the North Atlantic multidecadal fluctuations of SSTs and the insurgence of specific Euro-Atlantic weather regimes. They concluded that, in both reanalysis and GCM simulations, the positive phase of the AMV leads to an increased frequency of the NAO- and blocking regimes. Furthermore, positive AMV results in higher frequencies of cold spells over Europe and Eastern United States, in accordance with what has been recorded in the last few years (e.g. Coumou and Rahmstorf 2012).

In the present letter we start from the work of Peings and Magnusdottir (2014) to further investigate the relationships existing between the AMV and the mid-latitude climate. We make use of a group of long atmosphere-only experiments with a state-of-the-art high-resolution GCM and of a diagnostic based on the frequency of atmospheric blocking (Davini et al 2012). More specifically, we investigate the different contributions of the AMV to the Euro-Atlantic variability, distinguishing the signal generated by extratropical SSTs from that associated with tropical SSTs. This aims at providing a comprehensive assessment of the impacts of the AMV on the winter mid-latitude atmosphere and on the relevant dynamical mechanisms in action.

We use the atmospheric component of version 3.1 of the Earth-System Model EC-Earth (Hazeleger et al 2010), composed of the ECMWF Integrated Forecast System (IFS, cycle 36r4, ECWMF 2009) atmospheric component with the the H-TESSEL land surface scheme (Balsamo et al 2009)

With respect to the previous version (v3.0.1), EC-Earth 3.1 shows a reduced radiative imbalance and an improved hydrological cycle. The atmospheric configuration of IFS adopted (T255L91) is equivalent to a horizontal resolution of about 80 km. There are 91 vertical levels with the last full model level at 1 hPa.

We perform a group of atmosphere-only simulations with prescribed oceanic boundary conditions applying different anomalies representing the positive and negative phases of the AMV. Sea Surface Temperatures (SSTs) and Sea Ice Concentrations (SICs) data from the HadISST dataset (Rayner et al 2003) have been used.

We first define the NASST index as the area-averaged yearly anomalies of the North Atlantic SSTs between 75° W–5° W and 0° −70° N from 1870 up to 2012. In analogy with other works (Trenberth and Shea 2006, Peings and Magnusdottir 2014), the Atlantic Multidecadal Variability (AMV) index is then constructed by subtracting from the NASST index the 10-year running mean of the global SSTs (area-averaged between 60° S and 60° N). This is done in order to avoid the effect of the recent global warming.

Afterwards we identify years of positive/negative AMV extracting the upper/lower quartiles of the distribution of the AMV index (defined over the 1870–2012 period). SST and SIC anomalies for the chosen years are then averaged on monthly basis in order to create an ideal seasonal cycle for both positive and negative AMV phases. Even though sea-ice is set to two different climatological values before 1900 and between 1940–1952 (Rayner et al 2003), and thus its variability can be underestimated, we decided to use the full HadISST dataset in order to sample the largest possible variability for SST. We did not further detrend the anomalies in order to keep the anomalies more uniform and realistic, although this method could potentially underestimate the effect of global warming at high latitudes (Ting et al 2011).

The way in which we superimpose such ideal anomalies on the climatological seasonal cycle of the HadISST dataset for the 1990–2010 period defines 6 different experiments: FAMV+ and FAMV− (Full AMV) are two experiments where the (positive and negative, respectively) anomalies are applied to the whole North Atlantic and the Mediterranean Sea, from the equator up to the Arctic Ocean. TAMV+ and TAMV− (Tropical AMV) are two experiments in which the anomalies are applied only between 0° N and 30° N. Finally, XAMV+ and XAMV− (Extratropical AMV) are two experiments where the anomalies are applied north of 30° N. On the edges of the anomalies (e.g. at the equator for FAMV), an exponential smoothing is applied to avoid the presence of an artificial step in SST. The winter averages (December to March) of the 6 different SST and SIC anomalies are reported in figure 1. A supplementary simulation with no anomalies was run as a reference experiment: hereafter it will be named CONTROL.

Zoom In Zoom Out Reset image size

Well mixed greenhouse gases, stratospheric ozone and volcanic aerosol concentrations have been fixed at the values of year 2000 for all simulations, according to the historical scenario of the Coupled Model Intercomparison Project—Phase 5 (CMIP5) protocol (Taylor et al 2012).

The 7 simulations are run for a total of 201 years each. Given the small time needed for atmosphere and land surface to adjust to the oceanic boundary conditions, we discard the first 6 years of simulation: all the following figures are referring to simulation years 7–201 (195 years).

In order to evaluate biases of the EC-Earth simulations we use the ECMWF ERA-INTERIM Reanalysis (Simmons et al 2007) from year 1979 up to 2015. Since we want to study the AMV impact on the European climate, we focus on the season in which the mid-latitude climate variability is more pronounced, i.e. wintertime (defined as December to March, DJFM).

As mentioned in the introduction, a powerful diagnostic to evaluate the properties of the mid-latitude climate variability is provided by atmospheric blocking. To objectively recognize blocking events a 2 d index based onto the reversal of the gradient of geopotential height measured at 500hPa (Z500) has been adopted.

Firstly, both EC-Earth and ERA-INTERIM 500hPa geopotential height fields are interpolated on a common 2.5° x 2.5° regular grid with a second order conservative remapping. Then, as in Tibaldi and Molteni (1990), two meridional gradients of geopotential height are defined:

$$\begin{eqnarray}&&{GHGS}({\lambda }_{0},{\phi }_{0})=\displaystyle \frac{Z500({\lambda }_{0},{\phi }_{0})-Z500({\lambda }_{0},{\phi }_{S})}{{\phi }_{0}-{\phi }_{S}},\end{eqnarray} \tag{ 1 }$$

$$\begin{eqnarray}&&{GHGN}({\lambda }_{0},{\phi }_{0})=\displaystyle \frac{Z500({\lambda }_{0},{\phi }_{N})-Z500({\lambda }_{0},{\phi }_{0})}{{\phi }_{N}-{\phi }_{0}}\end{eqnarray} \tag{ 2 }$$

but here ${\phi }_{0}$ ranges from 30° N to 75° N and ${\lambda }_{0}$ ranges from 0° to 360°. ${\phi }_{S}$= ${\phi }_{0}$ − 15° , ${\phi }_{N}={\phi }_{0}$ + 15°. Instantaneous blocking is thus identified when:

$$\begin{eqnarray}{GHGS}({\lambda }_{0},{\phi }_{0}) & \gt & 0\qquad \\ {GHGN}({\lambda }_{0},{\phi }_{0}) & \lt & -10{\rm{m}}/^\circ \mathrm{lat}\end{eqnarray} \tag{ 3 }$$

Further constraints have been applied to instantaneous blocking. Firstly, large scale blocking is defined when an instantaneous blocking is extended for at least 15° of continuous longitude. Secondly, a large scale blocking event is defined for each grid point when a large scale blocking occurs within 5° lon (2 grid points) and 2.5° lat (1 grid point) of it. Finally, a blocking event at a certain grid point is defined when a large scale blocking event lasts for at least 5 days. Those constraints ensure that blocking events have a significant longitudinal extension, are persistent and quasi-stationary. The percentage of days per season in which blocked events occur (i.e. blocked days) defines the blocking climatology. A complete description of the blocking detection scheme may be found in Davini et al (2012).

We make use of other diagnostics to identify the properties of the signal at the mid-latitudes. We use the Eady growth rate maximum, defined following Vallis (2006):

$$\begin{eqnarray}&&{\sigma }_{{BI}}=0.3068f| \displaystyle \frac{\partial U(z)}{\partial z}| {N}^{-1}\end{eqnarray} \tag{ 4 }$$

where $U(z)$ is the vertical profile of the eastward wind component, f the Coriolis frequency and N is the Brunt–Väisälä frequency, defined as:

$$\begin{eqnarray}&&{N}^{2}=\displaystyle \frac{g}{\theta }\displaystyle \frac{\partial \theta }{\partial z}\end{eqnarray} \tag{ 5 }$$

where θ is the air potential temperature and g the standard gravity acceleration. The impact of the transient eddies is also evaluated computing the transient eddy activity as the bandpass filtered (2–6 days) standard deviation of the 500-hPa daily geopotential height. This also provides an indirect measure of the position of the North Atlantic storm track.

Finally, a 1000-trial bootstrap method is used to evaluate the level of significance of the mean differences between the simulations. Bootstrap has been chosen considering the non-Gaussianity of the blocking distribution. To test against the null hypothesis of no difference between two experiments, we artificially create a set of 1000 differences—computed subtracting the average of two random subsets of 195 years each, obtained randomly sampling the original yearly averaged data of the two experiments. The original difference between the two experiments is then compared point by point with the distribution of the differences obtained from the randomly-shuffled subsets. The level of significance is fixed at 2% if not stated differently (two-sided test).

3.1. Blocking sensitivity to AMV anomalies

The representation of atmospheric blocking is known for being a common issue in GCMs: for instance, a large systematic negative bias over Europe has been reported by both atmospheric (d'Andrea et al 1998, Neale et al 2013) and coupled climate models (Anstey et al 2013, Masato et al 2013).

Figure 2(a) shows the DJFM blocking climatology of the EC-Earth 3.1 CONTROL experiment (contours) and its bias versus the ERA-INTERIM Reanalysis (colors), both expressed as percentage of blocked days. A pattern with three relative maxima over North Pacific, Greenland and Northern Europe is evident, with values around 10–15%. The model shows a limited negative bias over North Pacific and Europe. Such bias is almost negligible over Greenland, which is a key region for the correct representation of the NAO and of the jet stream variability (Woollings et al 2008, Davini and Cagnazzo 2014). Conversely, blocking frequency is overestimated at low latitudes over both the Pacific and Atlantic oceans and over Western Russia. Overall, the representation of blocking in EC-Earth 3.1 is good. Given the linkage among blocking, NAO, eddy-driven jet and weather regimes discussed in the introduction, we can conclude that the representation of the main elements of the Euro-Atlantic mid-latitude variability can be considered as fairly realistic.

Zoom In Zoom Out Reset image size

Considered the quality of the blocking climatology in EC-Earth 3.1, we investigate the effects of the different phases of the AMV on the blocking frequencies. Figure 2(b) reports the FAMV+ minus FAMV− blocking anomaly, providing the relative difference between the positive and the negative FAMV phases. An evident dipole on the two sides of the jet stream is detected, showing an increased blocking activity over Greenland and Northern Europe and a reduction at lower latitudes over Central Atlantic. There is about a 26% relative increase of blocking over Greenland (defined as a box over 65°–15° W; 62.5°–72.5° N) and 14% relative increase over Northern Europe (15° W–25° E; 60°–65° N). A 25% relative decrease is observed over the Central Atlantic (60°–20° E; 30°–40° N). Considered the strong relationship between blocking over the North Atlantic and the NAO (figure S4 in the supplementary material, available at stacks.iop.org/erl/10/094010/mmedia), this implies the presence of an opposite relation between AMV and NAO (i.e. positive AMV resulting in a negative NAO, and viceversa), which is in agreement with works reporting an increase in blocking activity over the North Atlantic during AMV+ (Häkkinen et al 2011) and with works showing a connection between AMV+ and NAO− (Peings and Magnusdottir 2014, Gastineau and Frankignoul 2015).

An almost identical pattern with comparable intensity can be observed when anomalies are applied only to the tropical Atlantic (experiments TAMV+ minus TAMV−, figure 2(c)). Conversely, a barely noticeable signal is observed when SST and SIC anomalies are applied only to the mid-latitudes (figure 2(d), XAMV+ minus XAMV−). We measure a relative increase of about the 6% (3%) over Greenland (Central Europe) and only a 3% decrease over Central Atlantic. None of these anomalies is statistical significant even at the 10% level. This indeed suggests that the observed effects of the Atlantic Multidecadal Variability in EC-Earth 3.1 are driven by the tropics.

The seasonality of the response (DJ vs FM) is the only feature where we can find some differences between FAMV and TAMV experiments (supplementary material figures S1 and S2). FAMV experiments shows a stronger response in early winter (DJ), while TAMV experiments shows stronger response in later winter (FM).

We also investigated the linearity of the response for FAMV and TAMV, looking at the difference between each experiments and the control run (supplementary material figure S3). The response is surprisingly linear, showing the negative/positive NAO-like pattern for positive/negative AMV anomalies. A minor difference can be observed over Scandinavia: indeed here we note a weak increase of blocking in both FAMV+/TAMV+ and FAMV−/TAMV−.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

3.2. Large-scale changes

3.3. The mechanism: the Gill response and the shift of the subtropical jet

In this work we established the presence of a strong connection between winter climate variability over the Euro-Atlantic region and the phases of the AMV in a series of 200-year GCMs runs with EC-Earth 3.1. We found that the large part of the signal over Europe due to imposed AMV-like anomalies comes from the tropics, in analogy to what suggested by previous works that highlighted the influence of Tropical Atlantic SSTs on the Euro-Atlantic region (Okumura et al 2001, Terray and Cassou 2002, Peng et al 2005, Sutton and Hodson 2007). We also discussed two different mechanisms in action, one based in the tropics and one of extratropical origin.

They can be summarized as follows:

• The 'from above' mechanism: during the positive AMV phase, higher tropical SSTs enhance convection and precipitation a few degrees north of the equator, leading to a Gill-like response in the Tropical North Atlantic. This affects the subtropical jet stream structure near the Eastern North American coast, adding a negative barotropic wind shear anomaly and shifting the jet maximum to the south. The barotropic wind shear influences the propagation of Rossby waves, increasing the occurrence of cyclonic wave breaking. This ultimately leads to a negative NAO-like signal and a southward displaced eddy-driven jet stream over the North Atlantic, exactly as observed in FAMV+ and TAMV+ experiments. The opposite mechanism, driven by negative SST anomalies, is operating during the negative AMV phase.
• The 'from below' mechanism: during positive AMV phases, positive extratropical SST anomalies weaken the meridional gradient of the surface temperatures across the SST frontal zone in proximity of the Gulf Stream. This in turns affects the baroclinicity of the lower troposphere at the exit region of the eddy-driven jet stream, reducing the jet speed and shifting equatorward the jet stream. Conversely, negative SST anomalies increase the SST gradient and increase the low level baroclinicity. This mechanism appears to be extremely weak in our XAMV experiments (i.e. statistically not significant).

Peings and Magnusdottir (2014) found results similar to our FAMV experiments, for both reanalysis and GCM simulations: they concluded that the origin of the NAO-like signal was associated with changes of the North Atlantic SST gradient (i.e. the 'from below' mechanism). However, in a more recent experiment Peings et al (2015) performed another series of sensitivity integrations in which they split their forcing in a tropical and an extratropical part. In this configuration, consistently with our results, they found a larger contribution by the tropical component, although the extratropical component was a key factor to reinforce the atmospheric NAO signal. Indeed, in our simulations, the barotropic shear changes caused by the tropical precipitation anomalies are clearly leading the observed blocking pattern over the Euro-Atlantic region. However, the partial non-linearity of the signal and the different response in early and late winter (shown in the supplementary material, available at stacks.iop.org/ERL/10/094010) both suggest the presence of a non-linear interaction between the tropical and extratropical SST anomalies. In any case, some of the differences between the two studies can also be associated with the model characteristics and setup.

As can be seen by comparing figure 5(d) of Peings and Magnusdottir (2014) with our figure 4(c), EC-Earth FAMV and TAMV simulations show slightly stronger Eady growth rate anomalies, suggesting that EC-Earth might be more sensitive to SST anomalies. A first possible explanation of this discrepancy may be rooted in the larger SST anomalies that we applied at the tropical level (a methodological discussion on the reason causing these differences can be found in the supplementary material).

A second and more solid possibility can be related to the different atmospheric mean state over the North Atlantic in the model simulations. A jet stream a few degrees too poleward or too equatorward can have a dynamically different dominant mode of variability (Barnes and Polvani 2013, Davini and Cagnazzo 2014) that can translate in a different sensitivity to a diabatic heat source in the tropics. The different mean state can also explain the modest signal observed in the stratosphere (supplementary material figure S7). Such weak response suggests that the stratosphere in EC-Earth—that is a high-top model—does not play a relevant role for the propagation of the AMV signal over Europe, in contrast of what have been reported by Omrani et al (2014).

It is clear that a model response to a boundary forcing can be model dependent. On the other hand, the comparison between TAMV and XAMV simulations provides an unquestionable evidence in favor of the 'from above' mechanism. This interpretation is reinforced by the fact that the EC-Earth model—here in atmospheric-only configuration—shows minor biases in the representation of the mid-latitude climate. This also suggests that the influence of the SST anomalies along the North Atlantic frontal zone on the European climate may be overestimated.

Our findings shed further light on the importance of the representation of tropical convection in GCMs, which has been recently pointed as one of the largest open issues of state-of-the-art GCMs (Stevens and Bony 2013). Furthermore, the presence of a clear relationship between tropical convection/precipitation and mid-latitude climate variability adds further evidence to the importance of a correct simulation of equatorial dynamics, which can have wide implications for both seasonal and decadal prediction over Europe (e.g. Cassou 2008).

This work has been funded by the Project of Interest 'NextData' (http://www.nextdataproject.it) and by the PRIN 2010-11 project 20108 TZKHC, both of the Italian Ministry of Education, University and Research (MIUR). The authors would like to thank Yannick Peings and an anonymous reviewer for their constructive comments which helped to improve this manuscript.