AI Could Predict Critical Climate Tipping Points
Artificial intelligence (AI) could help scientists predict when climate tipping points will occur, potentially more than two years before they happen. This capability could open up opportunities to anticipate environmental changes whose effects are difficult or impossible to reverse.
The term “tipping point” was popularised by Canadian journalist and author Malcolm Gladwell in a book published in 2000. In a social context, it describes the moment when a trend or behaviour reaches a certain threshold and spreads rapidly.
In science, a tipping point occurs when a system undergoes a fundamental change that makes it difficult or impossible to return to its original state. In the climate system, this can happen when global warming pushes an environmental component beyond a certain threshold, such as when an ice sheet begins melting in a way that is difficult to reverse.
AI has been widely used to predict daily weather and extreme events. Predicting when tipping points in the climate system will occur, however, remains a challenge because climate change can alter the Earth system’s baseline conditions.
A recent study published in the journal Proceedings of the National Academy of Sciences (PNAS) found that a new AI approach could help estimate when such tipping points will occur.
Researchers from the California Institute of Technology (Caltech), the University of Colorado Boulder (CU Boulder) and technology company NVIDIA developed an approach based on recurrent neural operators (RNO) to predict tipping points using simplified models.
Caltech professor of computer science and mathematics Anima Anandkumar explained that a tipping point is a condition in which a system loses its ability to return to its original state as instability increases.
“A tipping point is when you have no chance of going back because the system is already unstable. A change in part of the system means it cannot return to its original state. That is why a tipping point can be seen as a disaster,” Anandkumar said, as quoted by Phys.org on Saturday (17 October 2026).
Anandkumar and her colleagues’ research was also cited in a recent United Nations report on climate tipping points. The researchers said the AI approach could potentially be applied more broadly, including in climate science, ecology and epidemiology, as well as aircraft design.
“This is a stepping stone towards characterising tipping points and predicting when they will occur in the future. We demonstrated it on simplified climate models and models of air turbulence over aircraft wings—systems that could not previously be predicted,” Anandkumar said.
One scenario examined was the loss of low cloud cover over the oceans as atmospheric carbon dioxide concentrations rise. This could occur when global warming pushes the cloud system beyond a certain threshold, although the process has not yet been widely observed directly.
The loss of low cloud cover could exacerbate global warming. Low clouds generally reflect some solar radiation back into space. When they disappear, more of the Sun’s energy can be absorbed by the Earth’s surface, raising temperatures.
To test their approach, the researchers used simulation data to train and evaluate an RNO-based AI programme for estimating when the clouds would disappear. The model produced a prediction more than two years before the simulated tipping point, with an estimated timing error of about 11 days.
Despite the promising results, further testing using real-world observations is needed to establish how well the approach can be applied to actual climate systems.
Anandkumar said earlier AI approaches were generally developed on the assumption that past weather patterns could serve as a guide to predicting future conditions.