Scientists warn that the Gulf Stream system could shut down after 2100, driving extreme winters, drying summers, and chaotic rainfall shifts.
In scenarios where greenhouse gas emissions remain high, the Atlantic Meridional Overturning Circulation (AMOC) could completely collapse sometime after the year 2100. The AMOC, which includes the Gulf Stream, is one of the most important systems of ocean currents on Earth.
A new study involving the Potsdam Institute for Climate Impact Research (PIK) warns that its shutdown would sharply reduce the ocean’s ability to transport heat northward. The result would be much drier summers and far harsher winters in northwestern Europe, along with major changes to rainfall patterns in the tropics.
“Most climate projections stop at 2100. But some of the standard models of the IPCC – the Intergovernmental Panel on Climate Change – have now run centuries into the future and show very worrying results,” says Sybren Drijfhout from the Royal Netherlands Meteorological Institute, the lead author of the study published in Environmental Research Letters.
“The deep overturning in the northern Atlantic slows drastically by 2100 and completely shuts off thereafter in all high-emission scenarios, and even in some intermediate and low-emission scenarios. That shows the shutdown risk is more serious than many people realise.”
Ocean Conveyor Belt and Its Tipping Point
The AMOC functions like a massive conveyor belt. Warm tropical waters travel northward near the surface, while colder, denser water sinks and returns south at depth. This circulation keeps Europe’s climate relatively mild and shapes weather patterns across the globe.
In the new simulations, the critical tipping point comes when deep winter convection collapses in the Labrador, Irminger, and Nordic Seas. Rising global temperatures reduce how much heat escapes from the ocean in winter because the air is not cool enough. This weakens the vertical mixing of ocean waters. As a result, the surface remains warmer and lighter, which makes it less able to sink and mix with the colder waters below. That loss of mixing disrupts the AMOC and reduces the flow of warm, salty water moving north.
Self-Reinforcing Feedback Loops Intensify
In northern regions, then, surface waters become cooler and less saline, and this reduced salinity makes the surface water even lighter and less likely to sink. This creates a self-reinforcing feedback loop, triggered by atmospheric warming but perpetuated by weakened currents and water desalination.
“In the simulations, the tipping point in key North Atlantic seas typically occurs in the next few decades, which is very concerning,” says Stefan Rahmstorf, Head of PIK’s Earth System Analysis research department and co-author of the study. After the tipping point, the shutdown of the AMOC becomes inevitable due to a self-amplifying feedback. The heat released by the far North Atlantic then drops to less than 20 percent of the present amount, in some models almost to zero, according to the study.
Lead author Drijfhout adds that “recent observations in these deep convection regions already show a downward trend over the past five to ten years. It could be variability, but it is consistent with the models’ projections.”
Emission Cuts Could Still Reduce the Risk
To arrive at these results, the research team analyzed CMIP6 (Coupled Model Intercomparison Project) simulations, which were used in the latest IPCC Assessment Report, with extended time horizons from 2300 to 2500. In all nine high-emission simulations, the models evolve into a weak, shallow circulation state with the deep overturning shutting down; this result is produced in some intermediate and low-emission simulations as well. In every case, this change follows a mid-century collapse of the deep convection in the North Atlantic seas.
Global Consequences and the Urgent Call
“A drastic weakening and shutdown of this ocean current system would have severe consequences worldwide,” PIK researcher Rahmstorf points out. “In the models, the currents fully wind down 50 to 100 years after the tipping point is breached. But this may well underestimate the risk: these standard models do not include the extra fresh water from ice loss in Greenland, which would likely push the system even further. This is why it is crucial to cut emissions fast. It would greatly reduce the risk of an AMOC shutdown, even though it is too late to eliminate it completely.”
Reference: “Shutdown of northern Atlantic overturning after 2100 following deep mixing collapse in CMIP6 projections” by Sybren Drijfhout, Joran R Angevaare, Jennifer Mecking, René M van Westen and Stefan Rahmstorf, 28 August 2025, Environmental Research Letters.
DOI: 10.1088/1748-9326/adfa3b
Never miss a breakthrough: Join the SciTechDaily newsletter.
Follow us on Google and Google News.
16 Comments
Sure thing, buddy.
Scientists have been predicting this since at least the 1970s. It was supposed to be imminent then.
The only difference between now and the 1970s is the amount of data that’s been collected that _confirms_ the hypothesis that the earth is warming due to human action. On a geological timescale, this is happening more or less instantly. Natural climate change takes thousands of years and we’re doing it in mere decades.
It’s true, that we’ve been talk about the “imminent” problem of climate change for 50 years at this point. But there is an important shift that’s occurred in how we’ve discussed it.
In the 70s and 80s, it was all prevention. If we act, we can stop it from happening in the first place.
As the 90s and 2000s slid by, we stopped talking about prevention and started talking about mitigation. If we act now, we can avoid the worst effects.
In the 2010s and 2020s, we’re increasingly discussing adaptation. If we act now, we might _survive_ the worst effects. No one talks about prevention these days because it’s too late for all that.
Climate change is here and now, and even if we snapped our fingers and stopped all human emissions today, things would still get worse for at least a decade or two. But we haven’t stopped emissions. 2024 saw the most emissions ever recorded. Things are going to get worse _faster_ from here on out.
The study doesn’t say we are ok until 2100. It gives the computer generated likely percentage of a complete shutdown under 3 different RCP scenarios. Their results depend on the accuracy of the computer model. There are physical processes yet to be added to the model. The addition of fresh water to the North Atlantic from Greenland runoff and sea ice meltwater in the Boufort gyre are not included. The imminent collapse of the SMOC in the southern ocean is not included. The imminent collapse of Greenland glacier 79 N is not included. The authors of the study say that very thing yet the media keep misrepresenting study after study. You are NOT safe until 2100.
You are certain of your claim despite not being able to provide readers with an objective, numerical assessment of the veracity of the models and no information on what the glacial ice melting will do? In other words, you are willing to go with what you want to believe, rather than facts.
The last time I did a ‘back of envelope’ calculation for the “imminent collapse” of the Greenland Ice Sheet, I came up with a number of about 15,000 years for it to go back to what it probably was during the Eemian interglacial. To put that into context, it is generally estimated that the agricultural revolution and civilization began about 10,000 years ago. That doesn’t seem very “imminent” to me.
Except you and other media are reporting it wrong. The AMOC is severely weakening and could shut down any day now. Hardly any scientists believe the AMOC will stop after 2100, which is a date the politicians and billionaires made up. Truth is, the AMOC could collapse anytime now. Climate is now accelerating exponentially and multiple tipping points have been breached. That’s the truth about climate change.
Your view of ‘reality’ is not supported by historical, empirical evidence:
https://scitechdaily.com/climate-puzzle-scientists-uncover-mysterious-halt-in-the-atlantic-conveyor-belt-slowdown/
Do you have anything to offer to counter this article other than your unsupported opinion?
It’s always been amazing to me that people think drilling a well extracting gooey black slime, transporting it hundreds of miles, then refining it in gigantic complex refineries, then burning it, then hacking a lung out because of smog is preferable to putting a solar cell in the sun and charging your batteries.
I think it was the SciFi writer Robert Heinlein that pointed out that calling a rare steak smothered in fried onions and mushrooms was more appealing that calling it a burned segment of bloody muscle tissue from a castrated bull. You have learned well, padawan.
What you have overlooked is what goes into mining all the materials for manufacturing solar cells and batteries, the energy needed fabricate and install them, and the conversion of land that can produce food and timber into solar cell ‘farms’ and manufacturing plants.
Once again, the specter of the Four Horsemen of the Climate Apocalypse — Could, May, Might, and Possibly — is raised to try to scare the public into turning the world energy economy on its head. This is despite a study by Curry et al. that calls the SSP585 high-emissions scenario into serious question because of insufficient fossil fuel resources to support it long-term. Furthermore, the empirical evidence does not support the predicted high-emissions scenario warming.
All the lawyer words — Could, May, Might, and Possibly — say is that the high-emissions scenarios are not impossible. No mention is made in the published peer-reviewed article of the probability and uncertainty envelope (margin of error) for the individual or ensemble projections. Bear in mind that none of the models have been verified, or can be for more than 75 years!
I’m not a fan of the ensemble approach because, logically, there can only be one best model. When it is averaged with all the inferior models, the prediction of the best model is corrupted. The unstated assumption of the ensemble approach is that the individual model runs produce results that regress around the mean of the true value, for which no evidence is presented. However, even the ‘best’ model may not be accurate enough to be truly useful.
As usual, there are a lot of colorful graphs and qualitative arm waving, but an unwillingness to assign numbers to the probability and uncertainty of the event happening. As close as they come is, “As far as current models suggest, we conclude that the risk of a northern AMOC shutdown is greater than previously thought [10%], at least when concerning the recent CMIP6 model ensemble. … In the CMIP6 ensemble a northern AMOC shutdown by 2300 occurs in 67% of all model-runs in an SSP585 scenario; 30% of all model-runs in an SSP245 and 21%, in an SSP126 scenario.” There is an assumption that the model runs provide a probability, but no proof is provided of the veracity or ability to replicate the number of runs, or, thus, the uncertainty.
Note also that the authors have moved the implied goal posts. The title states their predicted “Tipping Point” will happen after 2100 CE, while the later text claims that the impact of the (improbable) high emissions will take place between 2300 and 2500 CE. They aren’t actually wrong, but I consider it disingenuous to suggest that the problem will hit us in 75 years, and then put a stake in the ground out 275 to 375 years. It is unlikely that we will still be burning significant quantities of fossil fuels in 300 years, regardless of what we do. Think of how much the world has changed since 1725 CE.
This is the language of actual science. Could, probably, may, and my all time favorite “it depends”. Because real science is humble about what is not currently known (which is a lot!). It includes uncertainty and room for errors in its calculations. It admits the possibility of being wrong and changes when there is evidence to indicate it should.
The current state of a scientific field (any scientific field) is nothing more or less than our collective best guess based on _all_ the available evidence.
Certainty is the arena of charlatans and people with an agenda.
“This is the language of actual science.”
Not so. “Could, might, may, possible” are ‘lawyer’ words intended to create doubt in the minds of jurors. It is commonly accepted that mathematics is the language of science. Anything less is tantamount to an admission that the things that characterize a phenomenon are not measurable, or the researcher didn’t attempt to measure them. The best that one can say about such claims as “could,” is that it is a subjective observation about something that appears to the observer to not be impossible. However, it tells us nothing about the magnitude, probability of occurrence, or timing/frequency of occurrence.
There is a difference between “uncertainty,” which is commonly expressed numerically as the standard deviation of a number of measured samples that regress around the mean, and “ambiguity” that isn’t defined, leaving the interpretation of the meaning to the biases of the reader.
Inductive reasoning about observations is a requirement of science, preferably measurements that can be summarized as quantitative statistical descriptors, which can result in a formal hypothesis that can be stated as a function with an independent and dependent variable(s). Would Einstein’s work have been as valuable as E = mc^2 if all he had said was “Energy could have an equivalence to mass”?
You’re conflating two different things. The research itself, and how researchers talk about their work.
No scientist worth their degree is going to say something like “this is definitely the cause”. They will use what you term “lawyer words”. “Given our findings, we believe this may be the cause (and here is our suggestions for future research)”. To do otherwise is irresponsible since we know we don’t know everything.
Their research on the other hand _may_ include things like measurements of standard deviation or other relevant statistics for errors and outliers.
In this case, it appears they took CMIP6 models and ran them for longer and noted the implications of running them longer on different SSPs. You don’t need standard deviation to observe that 15 of 25 runs (or whatever the actual number is) lead to the shutdown of AMOC. Which is why they just report a ratio and none of the other uncertainty numbers you deem necessary. There is no uncertainty in those particular numbers to report (in the run, AMOC either failed or it did not). They then make the rather unsurprising claim that the risk is higher than previous though under the specific assumption that the given models hold true. Notice their wording: “As far as current models suggest, we conclude…”
What you appear to want is for the authors to justify CMIP6 and its relevance under various SSPs to the real world. This would be akin to asking Einstein to first prove all of algebra and calculus before using them as tools to prove relativity. Their work simply builds on that other work, and if you have issues with that other work (I.e. the veracity of CMIP6) you need to take it up with people that did that research.
The value to society of E=mc squared is questionable. Some idiots understood it could inspire the creation of nuclear weapons and ultimately nuclear-powered submarines to launch MIRVs at assorted cities. From that point of view, the statement, had Einstein said it, “Energy could have an equivalence to mass” might have led us to dismissing it as a cranky idea. of patents clerk Although no doubt some other bright spark might have got a PhD grant to refine the notion.
“Could/might” is a very useful word for expressing possibilities that may be refined subsequently according to your demands for numerical calculations of uncertainties. Scientific thought and consequent ideas start with non-quantified, and indeed non-quantifiable, uncertainties
“Certainty is the arena of charlatans and people with an agenda.”
Actually, I agree with that statement. Which is why I feel compelled to challenge someone who states, “After the tipping point, the shutdown of the AMOC becomes inevitable due to a self-amplifying feedback.” There is obviously an assumption that tipping points exist, without offering proof of climate tipping points, and the author uses the word “inevitable,” which does not allow for ANY uncertainty. It is a declarative statement that allows little wiggle room for alternative interpretations or new facts.
Again, the issue here seems to be that you’re taking the research as a statement about what _will_ happen in the real world rather than a statement about what _did_ happen in the model runs.
The models exhibited what is colloquially known as tipping point behavior. In their observations of the model runs, once that happens self-reinforcing feedback loops appear that make a drastic decline in AMOC inevitable. That’s the whole point of models. They allow us to understand the interplay between complex mechanisms in dynamic systems. Their use for predicting the real world is only as good as the fidelity of the model. Since we know all models make simplifying assumptions (they have to or they would be impossible to run), we know that they are not a perfect predictor of what will happen.
Which incidentally is _why_ we use ensembles. Different models make different simplifying assumptions. Taking the combined output of many models (along with human expertise) is how we have such a great ability to predict the paths of hurricanes for example.