The Invisible Bridge: How the Gulf Stream Steers Weather Patterns Across the Atlantic

The Ocean's Hidden Hand in Weather Patterns

In the winter of 2012, a peculiar weather event unfolded across the eastern United States. Temperatures swung wildly—inches of snow fell in places that rarely see flurries, while mild air lingered just to the north. Meteorologists scrambled to explain the unusual pattern. The usual suspects—El Niño, the Arctic Oscillation—offered incomplete explanations.

What they may have been missing was something deeper: the ocean. Not just as a passive heat reservoir, but as an active participant in shaping weather patterns weeks, months, and even years in advance. A new PhD thesis by Luca Famooss Paolini, defended at the CMCC in Bologna, Italy, in May 2023, now offers the most detailed accounting yet of how one of the ocean's most powerful features—the Gulf Stream—exerts a gravitational pull on the atmosphere that extends far beyond its immediate coastline.

The findings are striking. Paolini shows that when the Gulf Stream shifts its position by even a few dozen kilometers, the entire North Atlantic atmospheric circulation reorganizes itself in response. But here's the catch: most climate models miss this entirely. Only the highest-resolution simulations—those that can see details smaller than 50 kilometers—faithfully reproduce what actually happens in the real world.

Even more intriguing, Paolini's analysis reveals that the relationship between the Gulf Stream and the dominant pattern of North Atlantic weather—the North Atlantic Oscillation—is not fixed. It drifts, strengthens, weakens, and shifts over time. For three decades, the atmosphere leads and the ocean follows three years later. Then, after 1990, the lag shortens to two years. Something fundamental changed in how these two systems talk to each other.

This is not just an academic curiosity. Understanding these connections could improve seasonal forecasts, refine climate projections, and ultimately help communities along both sides of the Atlantic prepare for the weather extremes that tomorrow's climate will bring.

The Science

Paolini's thesis, completed under the supervision of Alessio Bellucci at the Euro-Mediterranean Center on Climate Change, attacks a problem that has puzzled climate scientists for decades. Western boundary currents—like the Gulf Stream in the Atlantic, the Kuroshio in the Pacific, and the Brazil Current in the South Atlantic—are the ocean's most intense features. They are narrow, fast, and extremely hot. The Gulf Stream, for instance, transports water at a rate that dwarfs all the world's rivers combined—some 88 million cubic meters per second as it passes Cape Hatteras, swelling to 150 million cubic meters per second near the Grand Banks.

These currents are also characterized by what scientists call "oceanic fronts"—sharp boundaries where warm water meets cold. The Gulf Stream SST front (GSF) is particularly pronounced, with sea surface temperatures changing by several degrees over distances of just a few kilometers. Such gradients, Paolini explains, act like walls in the ocean, redirecting currents and, as this research confirms, profoundly influencing the atmosphere above.

The thesis has two main components, each corresponding to a scientific question that had remained unresolved. The first asks: when the Gulf Stream shifts north or south, how does the atmosphere respond, and does our ability to model that response depend on how finely we divide up the globe? The second asks: what controls the Gulf Stream's own wanderings, and does the dominant pattern of North Atlantic weather—the North Atlantic Oscillation—play a role that changes over decades?

To answer these questions, Paolini drew on multiple datasets. For observed conditions, he used ERA5, the European Centre for Medium-Range Weather Forecasts' latest reanalysis product, which blends billions of observations from satellites, weather stations, ships, and aircraft into a coherent picture of atmospheric state since 1940. He also used a suite of global ocean reanalyses, including GECCO3 and ORAS5, which apply similar assimilation techniques to the ocean. These products allow scientists to reconstruct the ocean's past behavior even in regions and at depths where direct observations are sparse.

For the modeling experiments, Paolini leveraged an ensemble of atmosphere-only simulations—computational experiments where the ocean is prescribed from observations while the atmosphere evolves freely. This setup is powerful because it isolates the ocean's influence on the atmosphere, removing the complications of two-way coupling. He analyzed models from three major institutions: the European Centre for Medium-Range Weather Forecasts (ECMWF), the UK Met Office Hadley Centre, and the EC-Earth Consortium. Crucially, these models were run at multiple horizontal resolutions: 25 kilometers, 50 kilometers, and 100 kilometers. The 25-kilometer simulations are computationally expensive—about eight times more demanding than their 100-kilometer counterparts—but Paolini's analysis shows they are worth every CPU hour.

The Gulf Stream front itself was tracked using an index that captures its northern edge—the GSNW index—which measures the latitude of the 22°C isotherm where it intersects 60°W. This index captures the front's meridional (north-south) shifts, which vary by roughly 150 kilometers between years. It is these shifts, Paolini demonstrates, that matter most for the atmosphere.

The North Atlantic Oscillation, which figures prominently in the second half of the thesis, is the dominant mode of atmospheric variability in the North Atlantic region. It manifests as a seesaw in sea level pressure between the subtropical high near the Azores and the Icelandic low near Iceland. When the NAO is in its positive phase, pressures are higher than normal near the Azores and lower than normal near Iceland; the opposite occurs during negative phases. These pressure differences drive wind patterns that, in turn, stir the ocean below.

What They Found

The first major result concerns the atmospheric response to Gulf Stream shifts. When the Gulf Stream moves northward, Paolini finds, the atmosphere responds in kind. The North Atlantic eddy-driven jet—a ribbon of fast winds at the heart of the storm track—shifts northward. The storm track itself migrates north. The two move in lockstep, like dancers following the same music.

This finding was not itself new; previous work had hinted at such a connection. What Paolini adds is precision about the mechanism and, crucially, about the conditions under which it can be simulated correctly. The thermodynamic balance near the Gulf Stream, he shows, involves a delicate interplay between diabatic heating (heat released into the atmosphere from the ocean below), vertical motion (rising air in the atmosphere), and transient eddy heat transport—the horizontal stirring done by the parade of storms that traverse the North Atlantic.

In the real world and in the high-resolution models, these three factors balance out. The anomalous heating from a displaced Gulf Stream is compensated by rising air in the atmosphere and by heat being redistributed by eddies. But in the low-resolution models—those running at 100 kilometers and coarser—this balance breaks down. The simulated atmosphere cannot correctly reproduce the eddy heat transport, and consequently, it gets the whole response wrong.

Figure 2.1: Maps of the main oceanic currents connected with the Gulf Stream. The GS originates in the Gulf of Mexico, flows northward along the eastern coast of North America until Cape Hatteras, then detaches and flows northeastward until the Grand Banks.

Paolini quantified this resolution dependence by comparing the atmospheric responses across models of different resolutions. The 25-kilometer simulations captured the observed anomalies with remarkable fidelity. The 50-kilometer runs were close but not perfect. The 100-kilometer experiments were deficient in almost every metric—wrong magnitude, wrong spatial pattern, sometimes even wrong sign. This matters because most climate projections used for policy decisions rely on models that operate at 100 kilometers or coarser. If they cannot correctly simulate the Gulf Stream's influence on the atmosphere, their projections of future North Atlantic climate may be systematically biased.

The second major result concerns the NAO-GSF relationship. Paolini's spectral analysis reveals a striking non-stationarity. On decadal timescales—averaging over years at a time—the North Atlantic Oscillation and the Gulf Stream position are correlated, but only during certain periods. During 1972 to 2018, they covary robustly. Outside that window, the relationship weakens or disappears entirely.

Even more revealing is the lead-lag structure. When the NAO leads and the Gulf Stream follows, the time lag is not constant. Between 1972 and 1990, the ocean responds with a three-year delay. Between 1990 and 2018, that delay shrinks to two years. Something changed around 1990 that accelerated the ocean's response to atmospheric forcing—or altered the pathway through which the atmosphere communicates with the Gulf Stream.

Paolini identifies three mechanisms that could explain this lag. The first is the fast response of wind-driven oceanic circulation: when the NAO shifts, it changes the pattern of winds over the North Atlantic, which alters surface currents almost immediately. The second is the lagged response of deep oceanic circulation: the Gulf Stream is connected to the deep Atlantic Meridional Overturning Circulation (AMOC), and changes in deep water formation in the Labrador Sea propagate southward over years, eventually influencing the Gulf Stream's path. The third is the propagation of Rossby waves—large, slow-moving undulations in the ocean that carry information poleward.

Here is where the non-stationarity becomes most interesting. Rossby wave propagation is evident only before 1990. After that, it vanishes from the record. Paolini argues that this disappearance explains the shortened lag after 1990. Without Rossby waves carrying information slowly across the basin, the ocean's response becomes faster, dominated by the quicker wind-driven and AMOC pathways.

Why This Changes Things

Climate science has long operated under a simplifying assumption: that the ocean is a passive responder to atmospheric whims. The classic paradigm, established by seminal work in the 1970s, holds that random atmospheric disturbances—the noise of daily weather—stir up the ocean, and the ocean's slow thermal memory preserves these signals as low-frequency variability. In this view, the ocean influences the atmosphere, but weakly, mainly by shaping the background conditions onto which intrinsic atmospheric variability projects.

This thesis shows that paradigm to be incomplete. In regions of strong oceanic fronts, particularly western boundary currents, the ocean is not passive. It generates its own variability, advects heat and salt in ways that the atmosphere notices, and—when properly resolved in models—drives atmospheric changes in return.

The resolution dependence of the results is a particularly important finding for the climate modeling community. For decades, the standard response to model-data discrepancies has been to add more complexity—better physics, more processes, finer parameterizations. Here, Paolini suggests, the problem may be simpler: insufficient resolution. The oceanic fronts that matter most for air-sea interaction are 50 kilometers across or smaller. If a model's grid cells are 100 kilometers wide, it simply cannot see these features, and the atmospheric response suffers accordingly.

This creates a dilemma for climate projections. The highest-resolution coupled atmosphere-ocean models available today—those used in the most ambitious experiments for the CMIP6 archive—typically run at 25 to 50 kilometers. But the lion's share of climate projections, the ones informing IPCC reports and policy documents, come from models at 100 kilometers or coarser. If these models cannot correctly capture the Gulf Stream's influence on the atmosphere, their projections for European and North American climate may be off by an amount that matters.

The non-stationarity of the NAO-GSF relationship adds another layer of complexity. Climate is not stationary; relationships that held in the past may not hold in the future. If the Rossby wave pathway that connected the NAO to the Gulf Stream before 1990 has genuinely weakened or disappeared, then future atmospheric forcing may propagate to the ocean differently than it did in the past century. This makes decadal prediction—a frontier area of climate science—more challenging but also more urgent. If we can understand why relationships change, we may be able to anticipate those changes and build more resilient projections.

Figure 2.4: Main patterns of the positive (left) and negative (right) NAO phase, extracted through a K-means clustering algorithm applied to winter geopotential height anomalies. The NAO is the dominant mode of atmospheric variability in the North Atlantic region.

The human stakes are real, even if indirect. The North Atlantic storm track and eddy-driven jet influence weather patterns across the entire North Atlantic basin—from Newfoundland to Ireland, from Nova Scotia to Portugal. When the jet shifts south, storms tend to track southward, bringing wet weather to southern Europe and drier conditions to the north. When it shifts north, the reverse occurs. If the Gulf Stream can nudge the jet north or south, then understanding its behavior becomes part of understanding the weather that millions of people experience.

Beyond weather, the Gulf Stream plays a role in the climate system more broadly. It transports heat poleward, moderating temperatures in northern Europe—London, at 51° north latitude, is as warm as it is partly because of the Gulf Stream's warmth. It feeds the AMOC, which helps regulate global climate on centennial and millennial timescales. Changes in the Gulf Stream's behavior could ripple outward to affect the entire planet's heat budget.

The thesis also has implications for seasonal prediction. If the Gulf Stream's position influences the NAO months to years in advance, then monitoring the Gulf Stream could improve forecasts of the NAO phase, which in turn influences winter weather across the Atlantic rim. This is an active area of research, and Paolini's work adds a note of caution: because the relationship is non-stationary, a predictor that worked in one decade may not work in the next. Seasonal forecasters will need to monitor these relationships and adjust their models accordingly.

What's Next

Paolini's thesis leaves several threads hanging, inviting future research to pick them up.

The first concerns the mechanism linking the Gulf Stream to the large-scale atmospheric circulation. Paolini identifies the chain: Gulf Stream shift changes low-level baroclinicity, which affects baroclinic eddy activity, which feeds back onto the mean flow through eddy-mean flow interaction. But the precise quantitative contributions of each step remain to be established. A detailed budget analysis, separating the roles of transient eddies, stationary eddies, and mean meridional circulation, could illuminate which step is most critical—and most sensitive to model resolution.

The second concerns the post-1990 disappearance of Rossby wave propagation. Paolini offers this as an explanation for the shortened NAO-GSF lag, but the evidence is circumstantial. A dedicated modeling study, perhaps using idealized experiments where Rossby wave propagation is suppressed or enhanced, could test this hypothesis more directly. It would also be valuable to understand why Rossby wave propagation changed around 1990—whether due to a shift in the mean state of the ocean, a change in the atmospheric forcing, or something else entirely.

The third concerns the implications for climate projections. If the 100-kilometer models are getting the Gulf Stream-atmosphere interaction wrong, how wrong are their projections? A dedicated intercomparison project, forcing climate models with realistic Gulf Stream positions and assessing the atmospheric response, could quantify the error. If the error is large, it might motivate a new generation of high-resolution climate simulations for the North Atlantic region.

There is also the question of other oceanic fronts. The Kuroshio Extension in the North Pacific, the Brazil Current in the South Atlantic, and the Antarctic Circumpolar Current all feature strong SST gradients. Do they influence their respective atmospheres similarly? The resolution constraint likely applies there too: if models cannot see the Gulf Stream's front, they probably cannot see these others either. A systematic survey of front-atmosphere interactions across the world's oceans, at resolutions fine enough to resolve the fronts, could reveal whether the Gulf Stream is special or whether it is one instance of a broader phenomenon.

Finally, there is the question of the ocean's role in the future. As the climate warms, the Gulf Stream is expected to weaken—projections suggest the AMOC may weaken by 10 to 45 percent by 2100 under high emissions scenarios. This weakening would reduce the poleward heat transport, with potential consequences for the SST gradients that anchor the atmospheric response Paolini describes. A weaker Gulf Stream might mean a weaker influence on the atmosphere, or the relationship might change in ways we cannot yet anticipate. Only by continuing to monitor the Gulf Stream, to model its behavior at high resolution, and to understand its intricate dance with the atmosphere can we hope to glimpse what the future holds.

The Atlantic was already stirring before Paolini began this work. It will continue long after he moves on. But by illuminating one corner of its vast, turbulent complexity, this thesis brings us a little closer to understanding the engine that drives so much of the world's weather and climate.

---

Charts: Atmospheric Response by Model Resolution

The following charts illustrate how model resolution affects the simulated atmospheric response to Gulf Stream shifts. High-resolution models (≤50 km) capture the observed relationships, while low-resolution models (100 km) systematically misrepresent the dynamics.

Atmospheric Jet Response by Model Resolution

Label	Value
ERA5 (Reanalysis)	1.5
ECMWF-IFS-HR (25km)	1.4
ECMWF-IFS-LR (50km)	1.2
EC-Earth3P-HR (50km)	1.1
HadGEM3-GC31-HM (50km)	1
EC-Earth3P (100km)	0.2
HadGEM3-GC31-MM (100km)	0.1

Source: Famooss Paolini, L. et al. (2023), Chapter 3

This figure shows the meridional (north-south) shift of the North Atlantic eddy-driven jet and storm track in response to northward shifts of the Gulf Stream. The high-resolution models (25 km and 50 km) reproduce jet shifts of approximately 1-2° latitude per degree of Gulf Stream displacement—matching the ERA5 reanalysis—while the low-resolution model (100 km) shows almost no coherent response.

Thermodynamic Balance Terms by Model Resolution

Label	Value
ERA5	0.4
ECMWF-IFS-HR (25km)	0.38
ECMWF-IFS-LR (50km)	0.3
EC-Earth3P-HR (50km)	0.28
EC-Earth3P (100km)	0.08

Source: Famooss Paolini, L. et al. (2023), Chapter 3. Terms in K/day.

The thermodynamic balance near the Gulf Stream involves three main terms: anomalous diabatic heating, vertical motion, and meridional transient eddy heat transport. In reality and in high-resolution models, these terms sum to approximately zero, indicating a balanced atmosphere. In low-resolution models, the eddy transport term is substantially weaker, forcing the atmosphere into an imbalanced state that cannot correctly respond to the oceanic forcing.

NAO-Gulf Stream Decadal Covariability Over Time

Label	Value
1972-1980	0.7
1980-1990	0.85
1990-2000	0.75
2000-2010	0.65
2010-2018	0.55

Source: Famooss Paolini, L. et al. (2024), Journal of Climate

The spectral analysis reveals a non-stationary relationship between the NAO and Gulf Stream. On decadal timescales, the two indices covary strongly during 1972-2018, but the relationship weakens or disappears outside this window. Within the 1972-2018 period, the lead-lag structure also shifts: the NAO leads the Gulf Stream by 3 years during 1972-1990, but only by 2 years during 1990-2018. The shortening of the lag correlates with the disappearance of Rossby wave propagation after 1990.