Recurrent neural operator flags a simulated ocean-cloud tipping point more than two years ahead
Caltech-led researchers report that a neural operator predicted a simulated cloud tipping point more than two years ahead, with an error of about 11 days. The PNAS paper uses only simplified models, so the real climate remains untested.
The Scientist · Science desk
What happened
- The work comes from Anima Anandkumar's group at Caltech with colleagues at the University of Colorado Boulder and NVIDIA.
- The cloud test used a scenario in which low ocean clouds vanish as carbon dioxide rises, adding to global warming.
- The neural operator differs from other machine learning programs by carrying a built-in memory of how the physics change over time.
- A second test used simulated airflow over a tilting wing, where lift drops sharply, and the model predicted that tipping point too.
Why it matters
- What's now possible capability Earlier AI weather work covered day-to-day variation. This paper reaches the timing of an irreversible shift, but only inside simulations.
- What it limits constraint The cloud collapse has never been observed, so no real record exists to score the forecast against. The only ground truth is the simulator's own.
- Who has to decide decision The method flags departures from previously defined physics. We'd expect a team in ecology or epidemiology, both named as possible uses, to need a model of its system's normal behavior first.
The design starts from a limit in earlier weather AI. "In the past, we used AI to learn mostly stationary systems that assume the weather from before will be the same as weather in the future," said Anima Anandkumar, the study's principal author at Caltech [8]. Add rising carbon dioxide and that assumption fails. In her words, "the system is no longer stationary, and you can no longer use standard machine learning principles" [11].
The recurrent neural operator's warning is a consistency check. It detects when its predictions about a system's physics no longer make sense, which indicates a tipping point is imminent [12]. Anandkumar said: "The system is drifting away over time, and the laws from physics are no longer holding at the tipping point" [9].
The headline figure needs its denominator. An error of about 11 days against a lead of more than two years is roughly 1.5 percent of the lead time or less [14]. That is one scenario in one simulator, and the expected tipping point comes from simulated data [4]. Phys.org's account does not report an error for the wing, a false-alarm rate or a comparison with another method.
Anandkumar said the team used "a simplified climate model and a model of air turbulence over a wing" and called them "systems that weren't tractable with earlier methods" [10]. We would read the 11-day figure as a result about the method on a simulator. The thing this doesn't tell you is whether a physics-breakdown signal survives the noise of real measurements.
The paper has already been cited in a recent United Nations report on the importance of tipping points in climate science [15].
What to watch
- Whether the full PNAS paper reports a wing error, a false-alarm rate and comparisons with other methods.
- Any test of the neural operator on measured records from a real system instead of simulated data.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence50
- Adoption
- Not scored
- Hype gap+25
- Incentives40
- Confidence55
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
A Caltech-led paper published in the Proceedings of the National Academy of Sciences shows a recurrent neural operator (RNO) can predict when tipping points occur in simple models, including a climate-change system model and simulations of turbulent airflow over a plane's wing.
- [2]
The research was performed by Anima Anandkumar and colleagues at Caltech, the University of Colorado Boulder and the tech company NVIDIA.
- [3]
In one part of the study the researchers examined a climate scenario in which low clouds over the oceans rapidly disappear due to rising carbon dioxide levels, a point-of-no-return loss of low cloud cover that would add to global warming.
- [4]
Tested on simulated data, the RNO was able to look more than two years ahead of the expected tipping point and predict its timing with an error of about 11 days.
- [5]
The RNO differs from other machine learning programs because it incorporates a built-in memory that tracks how the physics change over time.
- [6]
The PNAS paper also tested the RNO on simulations of airflow over a tilting plane wing, where a tipping point occurs when the wing's lift drops precipitously, and the RNO succeeded in predicting it.
- [7]
Scientists have previously used AI models to predict smaller weather changes such as day-to-day variations; forecasting tipping points comes with more uncertainty and has proven more challenging.
- [8]
"In the past, we used AI to learn mostly stationary systems that assume the weather from before will be the same as weather in the future,"
- [9]
"The system is drifting away over time, and the laws from physics are no longer holding at the tipping point,"
- [10]
Anandkumar said the team demonstrated the method on "a simplified climate model and a model of air turbulence over a wing", which she called "systems that weren't tractable with earlier methods".
- [11]
"if you want to now model how that climate is changing and account for changes such as the carbon dioxide content, the system is no longer stationary, and you can no longer use standard machine learning principles."
- [12]
The RNO detects when the AI's predictions about the physics of a system no longer make sense, indicating a tipping point is imminent; when predictions no longer follow the rules of previously defined physics, a significant shift in the model is about to occur.
- [13]
The low-cloud loss scenario has been predicted to occur in the future but has not yet been observed.
- [14]
An error of about 11 days against a lead time of more than two years (more than 730 days) is about 1.5 percent of the lead time or less.
- [15]
The research has been cited in a recent United Nations report on the importance of tipping points in climate science.
- [16]
Because tipping points are so universal, the AI tool has potential applications in climate science, ecology, epidemiology, aeronautical design and more.
Get this ledger inside ChatGPT. Ask about any of our stories and the answer brings its figure, each claim’s verdict and the sources. Free, no account.
Sources
1 independent publisher whose own reporting we read for this story.
- phys.orgForecasting 'tipping points' with AI climate simulation
1 article · October 16, 2026
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.
Topics
- AI for ScienceFollow
- Climate Tipping PointsFollow
- Neural operatorsFollow
Entities
- California Institute of TechnologyFollow
- Jane BaeFollow
- Anima AnandkumarFollow
- NvidiaFollow
- University of Colorado BoulderFollow
- Proceedings of the National Academy of SciencesFollow
- Recurrent neural operatorFollow