← All research storiesWeather
Can AI weather forecasters grow up to predict climate? Top scientists lay out what it would take
AI already rivals traditional weather forecasts. A new review asks whether it can also simulate decades of future climate, and sets two minimum tests it must pass.
What the scientists did
AI weather models learn from decades of past weather data and can now compete with traditional physics-based forecasts for time spans from the next few hours up to a season ahead, at much lower computing cost. A team of 13 climate scientists, including Tom Beucler, J. David Neelin, Chris Bretherton and Tapio Schneider, reviewed this progress and asked whether the same approach can work for climate.
Climate is a harder problem. Instead of predicting tomorrow from today, a climate model must show how the whole system responds to changes, like more greenhouse gases or different land use, that push the planet into conditions not seen in the historical record the AI learned from.
Words to know
Weather vs. climate
Weather is what the atmosphere does over hours to weeks; climate is the long-term pattern over decades.
- Weather: rain tomorrow afternoon
- Climate: how much summers warm by 2050
External forcings
Outside influences that push the climate to change.
- Greenhouse gases like carbon dioxide
- Changes in land use, such as cutting forests
- Volcanic eruptions
- Changes in sunlight
Out-of-distribution
Conditions that are different from anything in the data an AI was trained on.
- A heat wave hotter than any on record
- A future world with much more carbon dioxide
What they found
The authors propose two minimum requirements for AI climate models. First, forcings like greenhouse gases must be built in clearly enough that scientists can change them one at a time in experiments. Second, models must be tested in situations outside their training data, including extreme events and “what if” scenarios.
They also suggest AI can save time by predicting only the variables people need at the needed detail, and by taking longer time steps, and that AI “downscaling” could provide cheap local estimates of hazards.
Why it matters to you
Climate projections help decide where to build flood defenses, which crops to plant and how cities prepare for heat. If AI can make those projections faster and cheaper, communities could get more detailed local information about future risks. This review is a roadmap, not a new forecast.
Where this could lead
Faster local climate risk maps
AI downscaling could give towns low-cost estimates of future floods or heat.
3–10 yearsTrustworthy AI climate models
Clear tests could help scientists decide when AI climate models can be relied on.
10+ yearsThese are possibilities the research points toward, not promises. Most early findings take years of testing before they reach everyday life, and some never do.
Why scientists care
AI has transformed weather forecasting in just a few years. Whether it can do the same for climate is one of the biggest open questions in Earth science, because climate change itself pushes the planet beyond the past data AI learns from.
Keep in mind
This is a review and perspective, not a new experiment, and it is a preprint that has not yet been peer-reviewed. The requirements it proposes are the authors’ suggestions for the field.
How far along is the science?
- Peer-reviewedReviewed by independent experts for a journal before publication. Peer review is a quality check, not a grade of evidence strength: confidence still depends on study design, limitations, replication and the wider evidence.
- PreprintA manuscript shared publicly, often before or alongside journal review. Check its separate journal and review status (each story says what we know). Results may change.
- Conference talkPresented to other scientists at a meeting. Usually short and early; a full paper may come later.
- Agency reportPublished by a science agency such as NASA, NOAA or NIH. Reviewed internally, but not by an outside journal.
- Press releaseAn announcement from a university or company. Useful, but always check the research it describes.
How soon could it reach you?
- Could reach you soonCould show up in products, services or advice soon. Our rough editorial guess is about 1 to 3 years, not a forecast from the research.
- A few years outPromising, but needs more testing or engineering first. Our rough editorial guess is several years, with a lot of uncertainty.
- Long-term scienceBasic research that builds knowledge. Real-world uses, if any, are far off.
- Small stepA small update to earlier work. Useful to researchers, but not a change you would notice yet.
Every story links to its original source: the paper itself, or the conference program when no paper is out yet. Explanations are written with AI, usually from the paper’s abstract (each story says what it was written from), then checked against the source. We only use journals, recognized preprint servers, science agencies, scientific conferences and established science news outlets.