Drivers of Changes to the ENSO–Europe Teleconnection Under Future Warming

The El Niño–Southern Oscillation (ENSO) teleconnection to Europe is projected to strengthen under global warming in most climate model simulations. However, given the current difference between recent observations and historical model simulations of tropical Pacific sea surface temperature trends, with models simulating an El Niño‐like warming in recent decades which is in disagreement with observations, it is important to understand the relative contributions of changes to the teleconnection forcing and background state to the overall teleconnection change. Using idealized climate model experiments, we show that both the eastward shift of El Niño precipitation and background state changes make contributions to the overall teleconnection change. These results suggest that the ENSO–Europe teleconnection can be expected to strengthen under global warming, even if ENSO precipitation anomalies do not shift eastwards as currently projected. However, the magnitude of the strengthening may depend on how much of an eastward shift does occur.


Introduction
The El Niño-Southern Oscillation (ENSO) is a major driver of worldwide climate variability, with areas of influence across the tropics and extratropics (e.g., Horel & Wallace, 1981;Ropelewski & Halpert, 1986).Although weaker than its teleconnections to regions around the Pacific basin, ENSO has been shown to have an influence in the North Atlantic/European (NAE) region (Brönnimann, 2007).In general, an El Niño event leads to surface pressure anomalies in the NAE region that project onto the negative phase of the North Atlantic Oscillation (NAO), with negative anomalies near the Azores and positive over Iceland.However, this teleconnection is dynamically distinct from the NAO in late winter, with different mechanisms responsible for establishing each pattern (Mezzina et al., 2020) and has also been shown to consist of NAO-related and non-NAO-related components (King et al., 2023).
Proposed mechanisms for ENSO's influence on the NAE winter climate can be grouped into tropospheric and stratospheric pathways.In the stratospheric pathway, El Niño causes a deepening of the Aleutian low, resulting in a weakening of the polar vortex and increasing the chance of a sudden stratospheric warming (SSW, Bell et al., 2009;Domeisen et al., 2019;Ineson & Scaife, 2009;Trascasa-Castro et al., 2019).Studies have also shown that the presence (or absence) of an SSW during the course of an El Niño event can notably alter the surface pressure anomalies in the NAE sector (Butler et al., 2014;Polvani et al., 2017).ENSO teleconnections via the troposphere also project onto the NAO, with several proposed mechanisms.Li and Lau (2012) and Jiménez-Esteve and Domeisen (2018) find that the southward shifted Pacific jet stream increases the eastward propagation of transient eddies to the NAE region, which favors the negative phase of the NAO, whereas Drouard et al. (2015) show that circulation anomalies in the North Pacific influence Rossby wave breaking in the Atlantic sector, subsequently affecting the phase of the NAO.ENSO-related anomalies in the tropical and extratropical Atlantic and the Caribbean have also been shown to exert an influence (Hardiman et al., 2019;Herceg-Bulić et al., 2023;Toniazzo & Scaife, 2006).
There is also a lack of consensus over the effect of ENSO diversity on anomalies in the NAE sector via both the tropospheric and stratospheric mechanisms (Calvo et al., 2017;Capotondi et al., 2015;Garfinkel et al., 2013;Zhang et al., 2019), in part due to the large internal atmospheric variability in this region (Deser et al., 2017).The response can also depend on the strength of event and vary between early and late winter (Ayarzagüena et al., 2018;Hardiman et al., 2019;Toniazzo & Scaife, 2006).Moreover, there is still uncertainty regarding the relative importance of the stratospheric versus tropospheric teleconnection pathways for driving NAO variability (Afargan-Gerstman & Domeisen, 2020;Albers & Newman, 2021).
The nature of ENSO events and their associated teleconnections are projected to change under global warming.There is general agreement among climate models that precipitation anomalies associated with El Niño will shift eastwards (Power et al., 2013), although such a trend is yet to materialize in observations.ENSO precipitation variability is also projected to increase under global warming (Cai et al., 2014(Cai et al., , 2015)).Such changes are likely to cause changes to ENSO teleconnections (e.g., Beverley et al., 2021;Brown et al., 2020;Meehl & Teng, 2007;Müller & Roeckner, 2008) including in the NAE sector, where the influence of ENSO is projected to increase (Drouard & Cassou, 2019;Fereday et al., 2020;Herceg Bulić et al., 2012).
In this study, we use data from CMIP6 simulations and experiments using an idealized climate model to attempt to determine the relative contributions of changes to ENSO precipitation anomalies and the background state to the overall ENSO-Europe teleconnection change.Given the disagreement between historical climate model simulations of tropical Pacific SSTs, which suggest an El Niño-like warming, and observations from recent decades, which show a La Niña-like warming (Seager et al., 2019), understanding the cause of potential changes to the ENSO-Europe teleconnection will lend important context to climate model projections of this teleconnection.The rest of the paper is arranged as follows.Section 2 gives details of the CMIP6 simulations and analysis methods used and a description of the idealized model and experiments performed.Our results, in Section 3, suggest that the projected strengthening of the ENSO-Europe teleconnection under global warming is driven by both the simulated eastward shift of El Niño precipitation anomalies in the tropical Pacific and background state (jet) changes.This gives greater confidence that this teleconnection will strengthen in a warming world, even if an eastward shift of El Niño precipitation anomalies in the tropical Pacific does not occur as simulated.We conclude in Section 4.

CMIP6 Simulations
We use simulations from the Coupled Model Intercomparison Project Phase 6 (CMIP6, Eyring et al., 2016).The control simulations against which the climate change experiments are compared are the pre-industrial control (piControl), from which we use 430 years of data for each model.These are run with forcings set to represent conditions in 1850, and have reached quasi-stable equilibrium.
For the climate change experiment, we use abrupt-4×CO2.In these experiments, the CO 2 concentrations are abruptly quadrupled from the global annual mean 1850 value.For most models, these simulations are 150 years in length.As equilibrium will not have been reached by this point, we performed Fourier harmonic analysis on the simulations to remove the long term trend and any variability with a period of longer than 10 years.Due to the abnormalities that this filtering introduces at the start and end of the time series, we do not use the first or last 20 years of the simulations.A full list of models used can be found in Table S1 in Supporting Information S1.

Indices and Analysis
Given the relatively weak nature of the ENSO teleconnection to Europe, we focus here on only the strongest El Niño events, which we define to be those which have a DJF-average Niño3.4Index anomaly of greater than 2 K.
Of the CMIP6 piControl and abrupt-4×CO2 model simulations available at the time of analysis, 17 models had El Niño events which met this criteria, and it is these models (listed in Table S1 in Supporting Information S1) that are used to calculate the multi-model means (MMMs) that we use both when examining projected changes to the teleconnection and to generate the SST forcing for our idealized experiments (see Section 2.3).Our analysis focuses on the northern hemisphere winter so we use DJF averages when exploring the teleconnection changes.Previous studies have shown that there is a difference in the pattern of El Niño mean sea level pressure (MSLP) anomalies in the NAE sector between early and late winter in observations (Hardiman et al., 2019;King et al., 2018;Moron & Plaut, 2003), however Ayarzagüena et al. (2018) showed that this change is not represented by CMIP5 models, and this is also the case for CMIP6 (not shown), therefore using DJF averages for our analysis does not affect the results.Significance for El Niño anomalies, anomaly differences and Isca anomalies is calculated by comparing to 1,000 bootstrap samples, drawn with replacement.For El Niño anomaly differences, each of the control samples contains the same number of years as the total number of El Niño years across all models in abrupt-4×CO2 (Table S1 in Supporting Information S1).A value is determined as significant if it falls above or below the upper or lower 2.5th percentile of the bootstrapped distribution, respectively-further details can be found in Supporting Information S1.

Isca Description and Experiments
We use Isca (Vallis et al., 2018), a framework for the idealized modeling of planetary atmospheres, which has previously been used to simulate atmospheric teleconnections (Jiménez-Esteve & Domeisen, 2019;Thomson & Vallis, 2018a, 2018b), including the ENSO-Europe teleconnection (Casselman et al., 2022;Jiménez-Esteve & Domeisen, 2020).It uses the Geophysical Fluid Dynamics Laboratory dynamical core along with a number of simplified parameterizations.The model used here is of intermediate complexity, with realistic radiative transfer through the rapid radiative transfer model, which allows configurable ozone and CO 2 concentrations.The model is forced using prescribed seasonally evolving SSTs.We use realistic continents and topography from the ERA-Interim reanalysis (Dee et al., 2011), and the model uses a Gaussian grid with T42 resolution and 40 vertical levels, with the model top at 0.02 hPa.Experiments are run for 120 years, and analysis is performed on the final 100 years of the simulations.Further model configuration details can be found in Supporting Information S1.
Each Isca simulation is forced with seasonally varying prescribed SSTs from CMIP6 simulations which have the same repeating annual cycle (example for El Niño experiments shown in Figure 1a).We have two climatological experiments, against which anomalies for the El Niño perturbation simulations are calculated.In these experiments, the model is forced using climatological MMM monthly SSTs from the piControl (hereafter Isca_pi-Control_clim) or abrupt-4×CO2 (Isca_4×CO2_clim) simulations, averaged over all 17 CMIP6 models and over all years available for each model (430 years for piControl and 110 years for most models for abrupt-4×CO2, Table S1 in Supporting Information S1).
The El Niño SST anomalies used in our perturbation experiments are calculated based on El Niño events as defined in Section 2.2.To calculate the prescribed seasonal cycle of El Niño SSTs, we take SSTs in the 6 months prior and 6 months post the winter peak of an El Niño event, with anomalies relative to either the piControl or abrupt-4×CO2 climatologies (piControl_clim/4×CO2_clim).To make the transition between the SSTs before and after the peak of an El Niño event smoother, we average the May, June, and July SSTs from both pre-and post-event peak when calculating the anomalies for these months.These SST anomalies (piControl_Nino_anom/4×CO2_Nino_anom) are applied on top of the relevant climatological SSTs only in the equatorial Pacific between 15°N-15°S and 150°-280°E (Figure 1b).Outside of this region, climatological SSTs from the relevant (seasonally varying) climatology are used (Table 1).Figure 1a shows the Niño3.4index anomalies in the Isca experiments forced using SSTs from the CMIP6 piControl (hereafter Isca_piControl_El_Nino) and abrupt-4×CO2 (hereafter Isca_4×CO2_El_Nino) MMMs.The anomalies peak at around 2.5 K for both piControl and abrupt-4×CO2, peaking slightly earlier in abrupt-4×CO2 (Figure 1a).The similarity of the seasonal cycle of SSTs in the two different El Niño experiments suggests that mean state SST changes are likely to be more important.
We also perform two additional El Niño perturbation experiments to attempt to separate out the influence of changes to the background state and changes to El Niño precipitation.In the first (hereafter Isca_uni-form_El_Nino), a uniform warming is applied to piControl SSTs in the tropical Pacific, equal to the difference in the climatological mean SSTs in this region between the piControl and abrupt-4×CO2 MMMs.SSTs outside the tropical Pacific are from abrupt-4×CO2 (4×CO2_clim), to keep the background state the same as that in Isca_4×CO2_El_Nino.This simulates the effect of future background state changes and some strengthening of El Niño precipitation (because of increased moisture) but without an eastward shift (i.e., all projected future changes except the eastward shift).The second additional perturbation experiment (hereafter Isca_pattern_El_Nino) simulates a future eastward shift of El Niño precipitation but on the present day (piControl) background state, without a change in its magnitude (i.e., the only change relative to Isca_piControl_El_Nino is an eastward shift of El Niño precipitation).This is achieved by adding the pattern of warming between piControl and abrupt-4×CO2 to the Isca_piControl_El_Nino SSTs in a deep tropical Pacific region (7.5°N-7.5°S,120°-280°E), while maintaining the same mean SSTs in this region as in Isca_piControl_El_Nino.In this experiment, SSTs outside the tropical Pacific are from the piControl MMM (piControl_clim) to keep the background state the same as that in Isca_piControl_El_Nino.Further details on the setup of this experiment are in Section S2.2 and Figure S1 in Supporting Information S1.Note that the sum of the El Niño SST anomalies in Isca_uniform_El_Nino and Isca_pattern_El_Nino are not the same as those from Isca_4×CO2_El_Nino, as both of these perturbation experiments contain an El Niño.

Results
The main aim of this study is to determine the relative contributions of changes to El Niño precipitation and background state changes to the overall teleconnection change under a quadrupling of CO 2 , using an idealized climate model.First, though, we examine the projected changes to the El Niño MSLP anomalies in the CMIP6 piControl and abrupt-4×CO2 simulations.In agreement with previous modeling studies (e.g., Ayarzagüena et al., 2018), the anomalies for the CMIP6 piControl MMM project onto the negative phase of the NAO, with negative values extending across the Atlantic toward Europe and positive anomalies to the north, centered over Iceland (Figure 2a).Under an abrupt quadrupling of CO 2 , this pattern is significantly strengthened, which is also in agreement with previous studies of future scenarios (Figures 2b and 2c).These changes could be driven by changes to El Niño precipitation or changes to the wave propagation characteristics due to background state changes (i.e., the jet stream).As described in Section 1, under global warming, El Niño precipitation anomalies in the tropical Pacific are projected to both significantly strengthen and shift eastwards (Figures 2d-2f) with implications for ENSO teleconnections.In addition, the upper-level jet stream in abrupt-4×CO2 extends further into the North Atlantic basin than in piControl (Figures 2g-2i).
To examine the contribution of El Niño precipitation and background state changes to the overall teleconnection change, we perform experiments using Isca, an idealized climate model, using prescribed SSTs as described in Section 2.3.To first understand the ability of Isca to reproduce the piControl and abrupt-4×CO2 ENSO-Europe teleconnection, we run Isca experiments forced by El Niño SST anomalies from the CMIP6 MMMs.Anomalies of variables from these experiments are calculated relative to their equivalent climatological run (see Table 1).
The pattern and magnitude of MSLP anomalies for the Isca experiments forced by piControl (Figure 3b, hereafter Isca_piControl_El_Nino) and abrupt-4×CO2 (Figure 3d, hereafter Isca_4×CO2_El_Nino) El Niño SSTs are very similar to those from the CMIP6 simulations (Figures 3a and 3c), with significant anomalies that project onto the negative phase of the NAO.The significant strengthening of MSLP anomalies between piControl and abrupt-4×CO2 is also captured, albeit with a greater strengthening than in the CMIP6 simulations, particularly at high latitudes (Figures 3e and 3f).This difference may in part be due to the simple representation of sea ice in Isca (see Vallis et al. (2018) and Supporting Information S1).Although the MSLP anomalies in the NAE sector are stronger in Isca_piControl_El_Nino and Isca_4×CO2_El_Nino than the CMIP6 MMM, they are still well within the spread of the CMIP6 ensemble.Figure 3g shows the MSLP anomaly difference between the Azores and Iceland for the CMIP6 piControl and abrupt-4×CO2 ensembles (box plots) and Isca_piControl_El_Nino and Isca_4×CO2_El_Nino (black cross/star-regions used for averaging are shown on Figure 3a).Although the Azores/Iceland MSLP anomaly difference is greater for the Isca experiments than the CMIP6 ensemble mean, it is only just outside the CMIP6 interquartile range.
Precipitation anomalies are also well reproduced in both Isca_piControl_El_Nino and Isca_4×CO2_El_Nino, particularly in the tropical Pacific (Figure S2 in Supporting Information S1) and while the upper level jet extends further toward Europe in both Isca_piControl_El_Nino and Isca_4×CO2_El_Nino than the CMIP6 simulations (Figures S3a-S3d in Supporting Information S1), the changes over North America and the North Atlantic are very similar (Figures S3e and S3f in Supporting Information S1).Overall, these experiments suggest that Isca is able to reproduce El Niño anomalies and teleconnections within the range of the CMIP6 simulations, giving us confidence that we can use it to further explore future changes to the ENSO-Europe teleconnection.
As described in Section 2.3, we now analyze two perturbation experiments designed to separate the contribution of changes to El Niño precipitation anomalies and the background state to the overall teleconnection change.The MSLP anomalies (relative to the equivalent climatological experiment listed in Table 1) for Isca_uniform_El_Nino and Isca_pattern_El_Nino are shown in Figure 4, along with MSLP anomalies from Isca_4×CO2_El_Nino for comparison.MSLP anomalies from Isca_uniform_El_Nino (Figure 4b) and Isca_pattern_El_Nino (Figure 4c) in the NAE region both project onto the negative phase of the NAO, as in Isca_4×CO2_El_Nino (Figure 4a).This suggests that future changes to the background state and the projected eastward shift of El Niño precipitation both make contributions to the overall ENSO-Europe teleconnection change.However, the pattern of MSLP anomalies in the NAE sector in Isca_uniform_El_Nino is much more similar to that in Isca_4×CO2_El_Nino (pattern correlation of 0.94 in the region 90°W-90°E, 0°-80°N) than Isca_pattern_El_Nino compared to Isca_4×CO2_El_Nino (pattern correlation of 0.36).This may be suggestive of greater influence of El Niño on the European circulation in Isca_uniform_El_Nino (all future changes except an eastward shift of El Niño precipitation) than in Isca_pattern_El_Nino (only an eastward shift of El Niño precipitation).This is perhaps a result of greater wave propagation from the Pacific to the Atlantic in Isca_uniform_El_Nino, due to the strengthened jet stream between the two regions, as previously hypothesized (Drouard & Cassou, 2019;Fereday et al., 2020), while the eastward shift of El Niño precipitation has a smaller influence on the pattern of MSLP anomalies over Europe.
MSLP anomaly differences between the Azores and Iceland for Isca_piControl_El_Nino and Isca_4×CO2_El_Nino are also shown on Figure 3g.The MSLP anomaly difference for both Isca_uni-form_El_Nino and Isca_pattern_El_Nino (blue pentagon/green square) have similar magnitudes, further suggesting that both background state changes and the projected eastward shift of El Niño precipitation anomalies make roughly equal contributions to the magnitude of the teleconnection change.The magnitude of both are also weaker than in Isca_4×CO2_El_Nino, which hints that a combination of both an eastward shift of El Niño precipitation and changes to the background state are needed to recover the full ENSO-Europe teleconnection change.

Summary and Conclusions
ENSO events are projected to change under global warming, with most climate models indicating an overall strengthening and eastward shift of El Niño precipitation anomalies, as a result of a projected El Niño-like sea surface temperature warming in the tropical Pacific.In addition, changes are also expected to the background state on which ENSO teleconnections depend, such as an extension of the midlatitude jet stream between the Pacific and the Atlantic basins.As a result, teleconnections associated with ENSO are expected to change, including a strengthening of the ENSO-Europe teleconnection.
However, the current disagreement between recent observations and historical climate model simulations of tropical Pacific SST changes mean that it is increasingly important to understand the relative role of changes to El Niño precipitation anomalies and background state changes such as those described above to the overall teleconnection changes.The current uncertainty regarding future changes to tropical Pacific SSTs means that teleconnection changes associated with background state changes may be more reliable than those associated with changes to ENSO precipitation anomalies.Here, we used CMIP6 simulations and the Isca idealized climate model to help address this issue and to provide insight on the robustness and likelihood of any changes.We find that Isca is able to reproduce both the CMIP6 piControl and abrupt-4×CO2 ENSO-Europe teleconnection, with the Azores/Iceland MSLP anomaly difference for the Isca experiments lying well within the CMIP6 ensemble spread.To explore the relative roles of the projected eastward shift of El Niño precipitation and changes to the background state to the overall future teleconnection change, we performed uniform and pattern warming experiments in Isca.These experiments suggest that both the eastward shift and background state changes make contributions to the overall change, although background state changes may have more of an impact on the pattern of MSLP anomalies over the European landmass.It is also worth noting that the Iceland/ Azores MSLP anomaly differences in these two individual experiments are smaller than Isca_4×CO2_El_Nino, further suggesting that both make contributions to the overall change and that both the eastward shift and background state changes are required to recover the full magnitude of the teleconnection change.
Our experiments may also provide insight on the mechanisms behind potential changes to the ENSO-NAE teleconnection.It is perhaps unlikely that the changes are occurring via the Aleutian low and the stratosphere as, while these anomalies are slightly weakened and shifted eastwards in the CMIP6 abrupt-4×CO2 experiments, they are actually strengthened in the Isca experiments.It therefore seems more likely that the changes are being communicated via changes to the midlatitude jet stream, with a strengthened jet across southern parts of North America and the North Atlantic enabling greater penetration of Rossby waves from the Pacific to the Atlantic basin.The potential roles of Atlantic or Caribbean precipitation anomalies are difficult to ascertain from our experiments.
Overall, our results suggest that the ENSO-Europe teleconnection can be expected to strengthen under global warming, whether or not ENSO precipitation anomalies shift eastwards as is suggested by most climate model simulations.However, the magnitude of the strengthening and the exact nature of the change may well depend on how much of an eastward shift does occur.Therefore, it is increasingly important to understand the cause of the current disagreement between observations and climate models relating to tropical Pacific SST trends to put these projected changes in context.This work was funded by the Natural Environment Research Council (NERC) through the Emergence of Climate Hazards project (NE/S004645/1).MC and FHL were also supported by the NERC CIRCULATES project (NE/T006285/1), and RC was also supported by NERC grant NE/W005239/1.The authors are grateful to Adam Scaife and John Albers for very helpful discussions and suggestions, and the two anonymous reviewers whose comments helped to improve the manuscript.

•
The El Niño-Southern Oscillation teleconnection to Europe is projected to strengthen under global warming • This change is driven by an eastward shift of El Niño precipitation anomalies in the tropical Pacific and background state changes globally • Background state changes have a greater impact on European circulation than precipitation changes, particularly over central-eastern Europe Supporting Information: Supporting Information may be found in the online version of this article.

Figure 1 .
Figure 1.(a) Seasonal evolution of the prescribed SSTs in the Nino3.4region used in Isca_piControl_El_Nino (blue) and Isca_4×CO2_El_Nino (orange).Anomalies are relative to the climatology of each simulation (b) DJF El Niño SST anomaly used in Isca_piControl_El_Nino.

Figure 2 .
Figure 2. DJF mean (a) mean sea level pressure anomalies (d) precipitation anomalies, and (g) mean 200 hPa zonal wind in El Niño years for the piControl multi-model mean.(b, e, and h) Same as (a, d, and g) except for abrupt-4×CO2.(c, f and i) Difference between piControl and abrupt-4×CO2.Anomalies on (a-f) that are not significant at the 5% level are hatched.

Figure 3 .
Figure 3. Mean sea level pressure (MSLP) anomalies in El Niño years from the full CMIP6 experiments for (a) piControl, (c) abrupt-4×CO2, and (e) the difference between them (abrupt-4×CO2 minus piControl).MSLP anomalies from the Isca El Niño experiments: (b) Isca_piControl_El_Nino, (d) Isca_4×CO2_El_Nino, and (f) the difference between them (Isca_4×CO2_El_Nino minus Isca_piControl_El_Nino). Anomalies that are not significant at the 5% level are hatched.(g) Azores minus Iceland MSLP anomaly in the full CMIP6 experiments and the Isca experiments-regions used for averaging are shown on panel (a) as boxes.The box and whisker plots represent the distribution among the CMIP6 models, with a red diamond for the ensemble mean and an orange line for the median.The black cross, black star, blue pentagon, and green square represent the values from Isca_piControl_El_Nino, Isca_4×CO2_El_Nino, Isca_uniform_El_Nino, and Isca_pattern_El_Nino, respectively.

Table 1
List of Isca Perturbation Experiments Performed, the Prescribed SSTs Used in Each in the Tropical Pacific Region (15°N-15°S, 150°E-280°E) and Elsewhere, and the Climatological Experiment Against Which the Perturbation Experiment Is Compared