← All research storiesWeather

PreprintWeather

An AI given some cause-and-effect rules beat other systems at warning when typhoons will suddenly power up

Researchers built a deep-learning system that combines satellite images with a map of what causes what, and it gave more skillful warnings of storms rapidly strengthening.

How far alongPreprintA 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.
How soon could it reach you?A few years outPromising, but needs more testing or engineering first. Our rough editorial guess is several years, with a lot of uncertainty.
Where it comes fromarXiv (preprint server)Oct 29, 2026Read the original →

What the scientists did

Sometimes a tropical storm suddenly gets much stronger in just a day. This is called rapid intensification, and it is one of the hardest things for forecasters to predict. It is dangerous because people may have less time to prepare than they expected.

The team built a deep-learning system called CausalVisionRI for typhoons in the western North Pacific Ocean. It looks at infrared satellite images plus information about the atmosphere and ocean around each storm. First, the researchers used data from 522 storms between 2000 and 2017 to work out a “causal graph,” a map of which conditions seem to drive which. That map was then used to guide how the AI learned, instead of letting it pick up any pattern it liked.

Words to know

Rapid intensification

When a tropical storm’s winds get much stronger within 24 hours.

  • A common definition used by forecasters is a jump of at least 30 knots (about 55 km/h) in a storm’s top winds in 24 hours (example; the paper’s summary does not state its exact threshold)

Causal graph

A diagram showing which things are thought to cause or influence which other things.

  • Warm ocean water → more energy for the storm
  • Strong winds that change with height (wind shear) → can tear a storm apart (example factors; the summary does not list the paper’s exact variables)

Brier skill score

A score for yes-or-no forecasts that compares them with a simple baseline guess: 0 means no better than the baseline, 1 means perfect.

  • Here the baseline was the usual long-term chance of rapid intensification (the “climatology”)

What they found

The researchers tested the system on storms from 2020 to 2024 that it had not seen during training. A team of five versions of the model working together reached a Brier skill score of 0.259 compared with a fixed climatology baseline. That beat the study’s baseline models and four existing reference systems.

When the researchers removed the causal graph to test its role, performance dropped, which suggests the cause-and-effect map added a measurable share of the skill.

522storms (2000–2017) used to build the cause-and-effect map
0.259Brier skill score on 2020–2024 test storms
2020–2024storm seasons held back for testing

Why it matters to you

If you live near a coast, the scariest storms are the ones that suddenly get stronger right before landfall. Better warnings of rapid intensification could give families and emergency managers extra hours to evacuate, board up windows or move boats and supplies.

Where this could lead

Extra tool for forecasters

Warning centers could test systems like this alongside their current tools to flag storms likely to strengthen quickly.

3–10 years

More trustworthy weather AI

Building cause-and-effect knowledge into AI could make its forecasts easier for scientists to check and explain.

3–10 years

These 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

Many AI models learn patterns without understanding why they happen, which can make them fail in surprising ways. Guiding AI with scientific cause-and-effect knowledge is one way researchers are trying to make weather AI both more accurate and easier to trust.

Keep in mind

This is a preprint and has not yet been peer-reviewed. It covers only the western North Pacific, and a skill score of 0.259 means the warnings are better than the baseline but still far from perfect. We read only the summary, not the full paper.

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.