Hastrup Review of Science

← All episodes

Deep divefrom Episode 14Climate & weather21 min

Do Climate Models Give Up on the Atlantic Too Easily?

A deep dive into one paper: what was known before, what the researchers did, what the numbers say and what they don't.

The paper: Is AMOC collapse too unstable in climate models? (arXiv preprint)

MP3

Chapters

Also linked

Transcript

The script the two AI voices read from. The times are worked out from the script, not heard, so a line can start a few seconds early or late.

Read the whole episode

Do climate models tip too easily?

Host 1If you run today's climate models all the way to the year twenty-three hundred under the highest emissions, how many of them lose the deep part of the Atlantic's great ocean circulation?

Host 2Every single one that was run that far. This new paper, written by two of the oceanographers behind that result, asks an uncomfortable question about their own finding: what if the models are now giving up the circulation too easily?

Host 1Welcome to the Hastrup Review of Science, where artificial intelligence reads real science, so you don't have to pretend you understood the paper.

Host 1Before we get to the models, what exactly is this circulation that everyone is worried about losing?

Host 2It is called the Atlantic Meridional Overturning Circulation, AMOC for short, and it is a slow loop of seawater. Warm, salty water flows north near the surface, gives its heat to the air around Greenland and Iceland, becomes cold and heavy, sinks, and creeps back south at depth, roughly one to two kilometres down.

Host 1So it works like a conveyor belt that carries heat north on top and brings cold water back underneath?

Host 2That is the classic picture, yes, and the heat it delivers is a big part of why northwest Europe is milder than places at the same latitude in Canada. The paper is mostly about the bottom half of that loop, the deep southward return, which they call the lower limb.

Why oceanographers worry about one current

Host 1Why did scientists start worrying that a current this big could simply stop?

Host 2As the wider history of the field goes, it started with a very simple model in the early nineteen sixties by the oceanographer Henry Stommel, which showed that a circulation driven by heat and salt can have two stable states, one on and one off. Later, ice cores and ocean sediments suggested that the real Atlantic circulation did weaken or switch off abruptly during past cold spells, for example near the end of the last ice age.

Host 1Why would salt matter for whether the loop keeps running or stops?

Host 2Because the loop only works if surface water near Greenland gets dense enough to sink, and both cold and salt make water dense. If the surface gets warmer or fresher, from rain, rivers or melting ice, it stays light and floats, and the sinking that drives the whole thing weakens.

Host 1And this is where the phrase tipping point comes in, I assume?

Host 2Yes, the paper opens with that idea: a change that may be abrupt and impossible to undo once it starts. The authors note how much money now goes into it, including a British research programme on forecasting tipping points that has funded twenty-seven international projects, and both authors are funded by several of these efforts.

What the models said before

Host 1What did the climate models used to say about all this, before the new result?

Host 2For years the common view, which the paper calls a long-held paradigm, was that the big coupled climate models were too stable. In other words, people suspected the models made it too hard for the circulation to collapse, so they underestimated the risk.

Host 1What do they mean by a coupled climate model, in plain words?

Host 2It is a computer model of the whole planet where the ocean, the atmosphere, sea ice and land all talk to each other as the simulation runs. Climate centres around the world run them in a coordinated exercise called the Coupled Model Intercomparison Project, and the sixth round of it, called CMIP six, is the one this paper is about.

Host 1So what changed in that sixth round?

Host 2A study led by the same first author, Sybren Drijfhout, and published in twenty twenty-five, took the CMIP six runs that had been extended to the year twenty-three hundred. In many of them, the deep circulation shut down, leaving a residual loop, at least one model did it in every emission scenario, including the low ones, and every model given the highest emissions did it.

Host 1When you say the deep circulation shut down, does the whole Atlantic loop disappear?

Host 2Not entirely, and that detail matters. A weaker, shallow loop survives in the subtropics, but it no longer carries heat north of the boundary between the subtropical and subpolar parts of the ocean, so the authors say that for Europe the effect is the same as if the circulation had vanished completely.

Host 1Is that leftover state permanent, or can it come back?

Host 2They say plainly that nobody can yet tell whether it is truly stable, or whether it is a kind of halfway state from which recovery is still possible. What they do know is that in the models it lasts for many centuries, which, as they put it, makes the question of its exact stability somewhat academic.

The question this paper asks

Host 1So the old worry was that the models were too stable, and now they collapse a lot, so what exactly is the open question?

Host 2The question is whether this round of models has swung too far the other way. The authors give two reasons to take that seriously: real measurements so far show no clear downward trend in the circulation, and some CMIP six models are known to warm too much for a given rise in carbon dioxide and to react too strongly when air pollution is removed.

Host 1Why would air pollution have anything to do with an ocean current?

Host 2Tiny pollution particles in the air, called aerosols, reflect sunlight and cool the surface, so cleaning them up adds warming on top of the greenhouse effect. If a model overreacts to both, it gets pushed harder than the real world would be, and the authors say that would shift the timing of any collapse.

Host 1So the authors are trying to poke holes in a result that they themselves helped to publish?

Host 2In a sense, yes, and they are careful about it. They write that they want to list which missing or crudely represented processes could make the models too unstable, without suggesting that they probably are, because if the models are right, the implications are, in their words, very worrying if not threatening.

How the collapse unfolds in the models

Host 1What kind of paper is this, then, a new experiment or something else?

Host 2It is a perspective, essentially an argued essay backed by existing model output, and they state that no new data were generated. The figures come from the CMIP six archive, and the core of the work is laying out, step by step, how the collapse happens inside the models and where each step could be wrong.

Host 1What is the first thing that breaks down in the models as the collapse begins?

Host 2Deep convection, which is the winter mixing south of Greenland and in the Nordic Seas where heavy surface water churns down into the deep. In the collapsing models, the depth of that mixing in March, when it is normally deepest, drops from about a thousand metres to less than two hundred, often to just one hundred.

Host 1Is that a fast change or a slow one, in ocean terms?

Host 2Fast: the mixing collapses within twenty to thirty years. The shutdown of the deep circulation itself then takes fifty to a hundred years, so the convection goes first and the circulation follows decades later.

Host 1Why would there be a delay of several decades between the two?

Host 2Because the deep southward flow is pushed by a pressure difference between the west and east sides of the deep Atlantic, at roughly one to two kilometres depth. That pressure comes from a pile of heavy water along the western edge, which convection keeps topping up, and when the supply stops, it takes about fifty years for that heavy water to drain away and the push to fade.

Host 1So if I have this right, the convection is like a tap filling a bathtub, and when you turn the tap off the bath keeps draining for decades before it is empty?

Host 2That is a fair picture of the lag. One refinement: the tap has to be off everywhere, because if convection survives anywhere in the North Atlantic or the Nordic Seas, the deep pressure difference can still be held up, just weaker.

Host 1Where does the sinking actually happen, then, if not at the convection sites?

Host 2Not in the open ocean where the mixing happens, according to a line of earlier work. If water sank there, the swirling currents of the region would have to be about ten times stronger than observed, so the sinking has to happen in boundary currents along the edges, and in the models net sinking in the region stops several decades after convection does.

Host 1What tips off the convection in the first place, is it mostly the melting ice from Greenland?

Host 2No, and that is one of the less intuitive points. In the analyses they cite, the change in heat flow at the surface over the convection areas is at least ten times larger than the change in fresh water, even counting Greenland meltwater, so warming from the atmosphere is the trigger and Greenland melt makes it worse.

Host 1So the first link is warming, and then what follows from it?

Host 2They describe a chain of six links. First, warming over the subpolar Atlantic and Nordic Seas reduces deep mixing, so less dense water forms; second, the lighter water that does sink weakens that west-to-east pressure difference at depth, and the deep return flow slows.

Host 1And once the deep return slows, does the problem start feeding on itself?

Host 2Yes, through two feedbacks that make it worse. Third, a weaker circulation brings less warm, salty water north, so the region gets fresher; fourth, less mixing means less salty water is stirred up from below, so the surface freshens further, and one model, in a run to twenty-three hundred, shows exactly that pattern, the top two hundred metres freshening south of Greenland and in the Nordic Seas while the rest of the Atlantic gets saltier.

Host 1Does anything in the system push back against the collapse?

Host 2Some things do, but in the models none of them wins. The air above changes too: the atmosphere tends to carry more heat north to make up for the ocean carrying less, which is called the Bjerknes feedback, and that further reduces the ocean's heat loss where convection happens, so it actually adds to the problem as link five.

Host 1What about sea ice, since it is so cold up there?

Host 2That is link six, and it does push back. As the ocean brings less heat, sea ice grows again, and growing ice squeezes salt into the water below, which makes it denser, but in the models this only starts after the convection has already collapsed, and on its own it is too weak to stop it.

Host 1Is there a figure in the paper where you can actually watch the surface water getting lighter over time?

Host 2Yes, in the Labrador Sea of a Japanese model called MRI-ESM two point zero. Its surface water first gets warmer and only later gets fresher, so the water grows lighter step by step, under both a low and a high emission scenario.

A word from our sponsor

Host 2This episode is brought to you by the Mathmos Astro lava lamp. Deep convection for the living room, without the inconvenience of an ocean.

Host 1I bought mine the year the first monitoring array went into the Atlantic, and I have logged its mixed layer depth every March since.

Host 2Mathmos has not paid for this, was never asked, and does not know this programme exists.

Host 1In twenty twenty-one my blobs stopped sinking, but the sofa kept its deep return flow for another fifty years, ending next spring.

Host 2Each lamp ships with a box model of convective instability, tuned to its owner.

Host 1The wax at the top is now lighter than everything below it even when cooled to minus one point nine degrees, and it has been formally listed as a new intermediate water mass.

Where the models might be wrong

Host 1If each of those six links is a place where the models could go wrong, which ones do the authors suspect most?

Host 2They go through all six and propose a check for each. Their central suspicion is resolution: as is typical across the field, standard climate models divide the ocean into boxes of roughly a hundred kilometres, too coarse to capture eddies, the swirling ocean storms tens of kilometres across, so those are replaced with simplified formulas.

Host 1What could eddies change about the very first link, the warming that weakens the mixing?

Host 2Two things. If a model already has the water column too layered, too stratified, in its starting state, the trigger is too easy to pull, which can be tested by comparing that bias against how much the circulation declines; and in some places resolved eddies behave differently from the formulas, for example western boundary currents warm about three times faster than the ocean interior because eddies keep the heat in the current.

Host 1So the idea is that real eddies might hold heat near the surface and keep the water mixing longer than the coarse models allow?

Host 2That is the hypothesis they want tested, comparing high-resolution models with the standard ones. For the second link, they suggest tracing the path of water parcels backwards in time in the models, to see whether eddies make the route from convection to the deep western edge so wobbly that the direct link weakens.

Host 1Has anyone already checked whether the salt feedback is too strong in these models?

Host 2Partly. For the average across all the models, earlier work says it is not too strong, but nobody knows yet whether the models that collapse have a stronger salt feedback than the ones that do not, or how either group compares with the real ocean, and the authors say that should be checked.

Host 1And the atmospheric part, the Bjerknes feedback, why might that be overdone?

Host 2Because it was demonstrated for natural ups and downs in a climate in balance. When the ocean is soaking up extra heat, as it is now, they argue the feedback should be less effective, and clouds may also play a role, so they want collapsing and non-collapsing models compared against a reconstruction of the real ocean called ECCO.

Host 1Is there a single thing they would fix first, if they could?

Host 2Yes: how well the models' layering of the water in the early twenty-first century matches observations in the convection regions, and whether the salt feedbacks there are trustworthy. Second, they point to dense bottom water south of Denmark Strait, which they say is missing in almost every climate model, leaving the deep water in the mixing regions too light.

Host 1Why would deep water that is too light make the models more fragile?

Host 2Convection happens when cold surface water becomes heavier than the water below it. The paper describes a critical salinity: if the surface, even cooled to its freezing point of about minus one point nine degrees, still stays lighter than the water at depth, convection stops, so the density of the deep water sets how close the system is to that line.

Host 1Could that critical salinity work as a warning sign before a collapse?

Host 2Unfortunately not, according to the paper. Crossing the threshold generally happens at the same moment as the collapse, so it cannot come before it, but it shows that the layering of the water matters far more than the weather once the threshold is reached.

What the real ocean shows

Host 1You said real measurements show no clear decline, how do we actually measure a current this big?

Host 2With arrays of moored instruments stretched across the ocean. The RAPID array at twenty-six degrees north has measured the circulation continuously since two thousand and four, a newer array called OSNAP further north has run since twenty fourteen, and the Florida Current has been measured through a submarine cable since nineteen eighty-two and has stayed broadly stable for about four decades.

Host 1And what does that record of about twenty years show?

Host 2A weak downward trend that is only marginally statistically significant. Several other observational studies also find no significant long-term trend, but the paper cites work estimating that you need continuous records of thirty to seventy years to detect a climate-scale trend, and RAPID has just over twenty.

Host 1So the absence of a trend does not mean the circulation is fine, it means the record is still too short to tell?

Host 2Exactly, and the authors say so directly: current observations can neither confirm nor refute a weakening. Yet they also list strong evidence of unprecedented change in the North Atlantic, including a decline of the deep western boundary current at twenty-six north over the past two decades.

Host 1What else in that evidence stands out?

Host 2A patch south of Greenland that has cooled while the planet warms, a freshening between twenty twelve and twenty sixteen that was the largest in over a century, temporary shutdowns of deep convection in the Labrador Sea in twenty twenty-one to twenty twenty-three, and a seventy-year record from the Greenland Sea showing lighter water and a new water mass forming at middle depths. Whether any of it is linked to a weakening or collapse, they say, remains open.

Host 1Haven't some studies claimed they can already see warning signs of a coming collapse?

Host 2Yes, and the paper is sceptical of them. Some studies used a single long record of sea temperatures south of Greenland as a stand-in for the circulation and looked for statistical warning signals, and a follow-up study argued that such estimates are too uncertain to be robust.

Host 1What kind of statistical warning signal were they looking for?

Host 2Critical slowing down: as a system nears a tipping point, it recovers more slowly from small knocks, which leaves a trace in the data. The authors doubt it will work here, because the circulation may be tipped by random noise, the climate is changing fast, and the CMIP six models do not show this slowing down in their historical runs.

Why it matters

Host 1So what does all this actually change for someone living in Europe today?

Host 2Not their weather this year: the collapse in these runs unfolds over many decades and is mostly a story after twenty-one hundred. What it changes is how much weight to put on one of the most serious climate risks, because these model results feed into how governments and researchers think about tipping points.

Host 1If the models are too unstable, is that good news for us?

Host 2It would mean the risk is lower or later than the models suggest, but the authors stress they are not claiming that. Their principle is that a frightening projection deserves the same scrutiny as a reassuring one, and if the checks pass, trust in the projections goes up, while if they fail, the checks show modellers what to fix.

Host 1Which people would actually do something differently because of this paper?

Host 2Mostly modellers and the people funding observation programmes. It gives them a concrete to-do list: compare collapsing and non-collapsing models link by link, run high-resolution models alongside, and measure salinity and density profiles in the convection regions properly.

What this paper cannot tell us

Host 1What should we be careful not to conclude from this paper?

Host 2That the models are wrong. It contains no test showing that they are too unstable; it is a list of questions with suggested checks, and the authors deliberately avoid saying which way the answer will go.

Host 1Are there other limits to how far we can lean on it?

Host 2Several. It is a preprint, so not yet peer reviewed; its detailed illustrations lean on just a couple of models; and much of it rests on the authors' own earlier study, so it is a well-informed view from inside one line of research rather than an independent review.

Host 1And on the measurement side, is there a limit as well?

Host 2Yes, the observations share the models' blind spot. Gridded ocean data are typically about one degree across, too coarse to see the eddies that matter for convection or the quick swings in mixing depth around its March peak.

What to watch next

Host 1What would tell us, in the next few years, which way this is going?

Host 2The checks they propose: studies comparing stratification, salt feedbacks and the Bjerknes effect between models that collapse and models that do not, plus high-resolution runs that resolve eddies. On the ocean side, every extra year of the RAPID and OSNAP records, and more profiles of salinity and density where the deep mixing happens.

Host 1Is there one result that would really settle it?

Host 2Not one, but if the models that collapse turn out to have water columns that were already too layered in their historical runs, compared with the real ocean, that would be a strong sign they tip too easily. If they match the real ocean well, the warning in the twenty twenty-five study gets heavier.

Sign-off

Host 1So, in one sentence, are the models crying wolf about the Atlantic?

Host 2Nobody knows yet, and that is the honest point of the paper: the models collapse the deep circulation readily, the measurements are too short to confirm or refute it, and the authors have written down exactly which links to test. That is a modest contribution, and a useful one.

Host 1That was the Hastrup Review of Science, and every weekend this same feed carries the roundup, the week's new papers in one short episode, which is roughly how long it takes the Labrador Sea to change its mind.

More from this week

Get the next one

A roundup of new papers every week and a deep dive into one paper most nights. Free, no account needed: paste this address into any podcast app that can add a show by its address, or find “Hastrup Review of Science” by name in an app that searches Podcast Index, like Fountain or Podverse.

https://reviewofscience.org/podcast.xml

Only want the weekly roundup, or only the deep dives? There are two smaller feeds: see Listen and subscribe.