The team’s focus was the Atlantic Multidecadal Variability, or AMV, a slow oscillation in North Atlantic sea surface temperatures that swings between warm and cool phases over decades. When the AMV is in its positive phase, a characteristic ‘horseshoe’ pattern emerges at the ocean surface: the subpolar gyre south of Greenland warms strongly, and a band of warm water spreads across the tropical North Atlantic, lagging slightly behind. Decades of research have linked this pattern to European summer climate, yet the observational record of such multidecadal variability is barely 45 to 70 years long — far too short to cleanly separate the tropical and extratropical contributions to the phenomenon using observations alone.

To overcome that fundamental limitation, the researchers turned to one of the largest decadal prediction experiments ever assembled: an 80-member hindcast ensemble built with the low-resolution Max Planck Institute Earth System Model, MPI-ESM-LR. The members were initialised every November from 1960 to 2019, with the ocean nudged toward the EN4 ocean reanalysis through an ensemble Kalman filter and the atmosphere constrained toward observed conditions. Each member was run for at least ten years beyond its start date, and a subset of sixteen members was extended to twenty. By comparing this initialised ensemble with fifty uninitialised historical runs from the MPI Grand Ensemble, the team could isolate exactly what initialisation — the act of telling the model what the ocean is doing right now — buys in terms of predictive skill.

The first result concerns the ocean itself. In the uninitialised historical simulations, the model reproduced the horseshoe pattern of AMV sea surface temperatures only weakly, with the subpolar gyre signal considerably underdone and displaced toward the south and east. In the initialised hindcasts, by contrast, the subpolar gyre anomalies were dramatically stronger and closely matched the observations from the HadISST dataset and the ERA5 reanalysis. When the researchers defined AMV phases using ERA5 rather than the model’s own index, the strong subpolar signal survived intact across forecast lead years one to seven — meaning those temperatures are highly predictable — while the tropical signal essentially vanished, revealing that AMV-driven tropical SST anomalies are poorly predicted at these lead times.

That contrast matters enormously because the two halves of the AMV drive European summer weather through completely different atmospheric pathways. The extratropical pathway is grounded in classic theory: warm subpolar sea surface temperatures act as a diabatic heat source, and the atmosphere responds with a shallow cyclonic circulation a quarter-wavelength downstream, producing a low pressure anomaly west of Great Britain and Ireland. This feature, closely related to the East Atlantic Pattern of variability, advects warm air polewards and has been linked to hotter summers in central Europe. The tropical pathway is more exotic: warm tropical Atlantic waters generate upper-level divergence, launching Rossby waves — vast undulations in the jet stream — that arc from the Caribbean toward the extratropical North Atlantic and Europe.

When the team examined sea level pressure composites, a striking divergence appeared. Observations from HadSLP and ERA5 showed the expected negative pressure anomaly west of Britain and Ireland during positive AMV summers, but the uninitialised historical ensemble produced almost nothing in the extratropics, even though its tropical response was comparable to reality. The initialised hindcasts, however, did produce the cyclonic anomaly — in the right place, but with an amplitude notably weaker than observed. Crucially, when the hindcasts were evaluated against ERA5-defined AMV phases, this extratropical signal persisted, indicating genuine predictability. Correlation skill maps confirmed it: the highest skill for multi-year mean sea level pressure sat precisely off the west coast of Europe, where the AMV response is strongest and statistically significant.

Yet the weakness of the modelled response opened a door to one of climate science’s most perplexing puzzles: the signal-to-noise paradox. In a well-behaved forecast system, the correlation between the ensemble mean and observations should be no higher than the average correlation between the ensemble mean and its individual members. In the North Atlantic-Europe sector, this is routinely violated — the real atmosphere appears more predictable than the models themselves suggest. By sampling 100,000 single-member timeseries of East Atlantic pressure against the ERA5 AMV index, the researchers found that the observed regression slope was stronger than 99 percent of the hindcast samples, and the ratio of predictable components exceeded two with high significance. Most remarkably, the joint distribution revealed a correlation of minus 0.50 between the weak AMV response and the paradox’s diagnostic errors, directly linking the underestimated ocean-driven signal to the reliability failure.

The story grew richer still when the researchers climbed into the upper troposphere at 200 hectopascals. In reanalyses, the surface cyclonic response extends upward as a deep, near-equivalent-barotropic column over the eastern Atlantic. In the model simulations, however, the dominant upper-level response instead resembles a Rossby wave train emanating from the Caribbean — the tropical pathway. By varying the length of the rolling averaging window from one to fifteen years, the team showed why: tropical Atlantic SSTs carry relatively more interannual variability, while extratropical SSTs dominate on longer timescales, and the correlation between the two basins strengthens with window length. On interannual timescales, both models and reanalyses agree on the Caribbean wave train; on decadal timescales, reanalyses shift toward the extratropical response while the model, whose surface response is too weak, remains dominated by the tropical mechanism.

The implications are profound. The study demonstrates that dynamically driven, multi-year predictability for European summers genuinely exists — rooted in ocean temperatures that models can observe and initialise — and that large ensembles are essential to realise it. But it also shows that the extratropical diabatic heating response, though simulated, is severely underestimated, and that this deficiency cannot simply be fixed by adding more ensemble members or post-processing the output, because the two competing mechanisms interfere destructively in the upper troposphere. The authors point toward higher atmospheric and oceanic resolution, better representation of ocean-atmosphere coupling and eddy feedback, and targeted bias reduction as the most promising routes forward. Until the signal-to-noise paradox is solved, they conclude, large-ensemble forecasts with appropriate correction will be necessary to deliver the summer predictions that a continent increasingly vulnerable to extreme heat urgently needs.

Subject of Research: Decadal predictability of European summer climate through Atlantic Multidecadal Variability mechanisms in large-ensemble climate model hindcasts

Article Title: Predictability of European summer climate: the influence of competing mechanisms related to Atlantic Multidecadal Variability

Article References: Williams, N. C., Müller, W. A., & Pinto, J. G. (2026). Predictability of European summer climate: the influence of competing mechanisms related to Atlantic Multidecadal Variability. Earth System Dynamics, 17(4), 1007-1023. https://doi.org/10.5194/esd-17-1007-2026

Image Credits: AI Generated

DOI: 10.5194/esd-17-1007-2026

Keywords: Atlantic Multidecadal Variability, decadal prediction, European summer climate, sea surface temperature, subpolar gyre, Rossby waves, signal-to-noise paradox, MPI-ESM-LR, climate models, North Atlantic, predictability, diabatic heating