Didier Swingedouw (Credits: https://tipesm.eu/about/people/didier-swingedouw/)
The Atlantic Meridional Overturning Circulation — the AMOC — has become one of the most discussed elements of the climate system. It is often portrayed as a looming tipping point, a potential disruptor of European climate, and a symbol of the uncertainties that still surround climate projections. We spoke with Didier Swingedouw, a leading researcher in ocean–climate interactions from the CNRS and based at University of Bordeaux, who coauthored a recent work aiming to refine our understanding of how the AMOC may evolve in the coming decades. The paper, appeared on Science Advances, is authored by Valentin Portmann, and is available for reading at this link.
From the very beginning of our conversation, Swingedouw stressed that several factors influence AMOC changes, but different parts of the AMOC respond differently to climate forcing, and not all of them are equally relevant when discussing long‑term stability. When it comes to timing, the message is both reassuring and sobering. According to Swingedouw, no climate model currently predicts a full collapse of the AMOC before 2100. “According to CMIP6 models there won’t be any full collapse before the end of the century,” he said, summarizing the consensus across modelling centres. But the picture becomes more complex beyond 2100. Some studies, such as work by Drijfhout and colleagues, show that shortly after the turn of the century, many models reach a very low level of AMOC strength, below 5-6 Sv at 26°N, a threshold value that has been associated with irreversible AMOC changes in a recent study of Jackson et al. based on idealized NAhosMIP experiments. The difficulty is that most modelling centres still stop their simulations at 2100, a legacy of early IPCC practice that no longer matches the lifespan of today’s newborns. As Swingedouw put it, “People born today will likely know 2100.”
Why does all this matter? Because the AMOC is not just an abstract oceanic conveyor belt — it is a major regulator of climate. Its influence extends from European and North American climate and seasonality, from the position of the Intertropical Convergence Zone to the strength of monsoon systems in Africa and Asia. It affects fisheries, oceanic carbon uptake, and even regional sea level, which could rise by several tens of centimetres along parts of the Atlantic coast if the AMOC weakens strongly. And yet, despite its importance, the uncertainty surrounding its future evolution remains enormous. In his study, Swingedouw found that the projected weakening by 2100 ranges from almost zero to around 70 percent. “That’s huge,” he emphasized — and it has direct consequences for climate adaptation strategies, especially in Europe.
This is where their recent work comes in. The study applies a statistical framework known as observational constraint, which attempts to reduce uncertainty by linking present‑day model biases to future projections. The idea is simple in principle: if a model misrepresents key aspects of the current climate, that misrepresentation may also affect its projections. By quantifying this relationship, one can weight models according to how well they match observations.
Their approach stands out for several reasons. Instead of relying on a single observational dataset — such as the RAPID array at 26°N — the study incorporates multiple variables, including sea‑surface temperature and salinity patterns that are closely tied to the density gradients driving the AMOC. It also tests several statistical methods, from traditional linear regression to ridge regression and random forests, and validates them using a leave‑one‑out approach to estimate overfitting. Therefore, machine‑learning techniques, he explained, were also tested, but they were simply too unstable given the small sample of available models. “We have so few data,” he said. “Neural networks were even more unstable.” The method with the lowest LOO is found to be muti-variate Ridge Regression, a method hardly used in climate studies.
The result is striking: the method yields an estimated 51 percent weakening of the AMOC by 2100, with an uncertainty of only ±8 percent (at 90% probability). This is a much narrower range than raw model projections, and it points toward a stronger weakening than the ensemble mean would suggest. While the exact number should not be taken as definitive, the reduction in uncertainty is itself a major step forward.
Of course, no method is without assumptions. Swingedouw acknowledged that the choice of regions used for the constraints is partly subjective, and that observational uncertainties may be underestimated. His team has already begun addressing these issues by incorporating more variables, using clustering techniques to define regions more objectively, and refining the statistical formulation of the method.
Our conversation eventually turned to the broader climatic implications of an AMOC that weakens by half. The strongest effects would likely be felt in winter in the Northern high latitudes, particularly in Europe, where the AMOC plays a key role in shaping temperature patterns and storm tracks. A weaker AMOC could mean colder winters than standard projections suggest, greater seasonal contrasts, and shifts in atmospheric circulation that can affect ecosystems and agriculture. Summer, by contrast, is less sensitive: heatwaves are driven primarily by atmospheric blocking, and the ocean plays a more modest role. Globally, the AMOC can influence teleconnections through Rossby waves and other mechanisms, but quantifying these effects requires targeted experiments — something the upcoming TIPMIP-OCEAN initiative aims to provide. He therefore invites as many modeling center as possible to contribute to this MIP (see protocol here).
Looking ahead, Swingedouw sees several priorities for future research. Understanding the physical processes behind AMOC weakening remains essential, as does studying the possibility of recovery after a collapse or after large‑scale CO₂ removal. Another key question is how different degrees of weakening — say 30 percent versus 50 percent — translate into concrete climatic impacts. And, of course, extending projections beyond 2100 will be crucial for capturing the full trajectory of the AMOC under continued warming.
In the end, the message is nuanced. The AMOC is not on the verge of collapsing within this century, but it is very likely to weaken substantially — and that weakening will matter. Thanks to studies like Swingedouw’s, we now have a clearer, more constrained picture of what to expect. But the AMOC remains a complex, dynamic system, and understanding its future will require both scientific innovation and sustained international collaborations and observations.
