Wednesday, November 18, 2026Wed, Nov 18

Moon: waxing gibbous, 58% litAurora: active

Español

Space, Earth and the natural world, explained.

Google DeepMind AI Gives Hurricane Forecasters an Extra Day

A DeepMind model described in Nature gives hurricane forecasters about an extra day of accuracy. The NHC uses it, but specialists make the final call.

A powerful tornado over the ocean during a dramatic sunset.
Photo: Johannes Plenio / Pexels

Google DeepMind’s weather model gives forecasters about an extra day of predictive accuracy for tropical cyclones, according to a paper the company published in Nature in August. Three-day forecasts from the model were as good as what earlier models could deliver for only the next two days.

The US National Hurricane Center (NHC) began using AI models in operations in late June of 2025, its operations division chief Dan Brown told AFP. Human specialists still decide which forecasts to trust.

What the DeepMind model does

Ferran Alet, a DeepMind research scientist who leads the company’s cyclone work, told AFP that Google has no hurricane-specific model. The same system that handles everyday weather is pushed to its limits on extreme storms.

It predicts a cyclone’s track (the path it follows), its intensity (how strong it gets) and its rainfall. For each storm it can generate 1,000 possible scenarios.

Hurricane Isaias, which struck the south-eastern United States in October as a major storm, was an example. The model recognized a full week before the storm formed that it could become a tropical cyclone. Brown said the AI models performed quite well on Isaias’s track, and that they appear good at quickly picking up shifts in complex steering patterns, the large-scale winds that guide a storm.

How AI differs from older forecasting tools

Before AI, the NHC relied on three kinds of models. Dynamical models run on supercomputers and solve the equations of fluid motion and thermodynamics that govern the atmosphere. Statistical models use historical relationships between storm behavior and variables such as location and date. Hybrids combine the physics and the statistics.

AI models also learn from data, like statistical models, but from far larger datasets and with the ability to find more complicated patterns. Brown noted that physics-based models built the baseline on which the AI models were trained.

Where human forecasters still step in

The NHC added the DeepMind model to some of its track and intensity consensus aids in 2026. It also uses AI models from the European Centre for Medium-Range Weather Forecasts and the National Weather Service, and it has an agreement with Google to develop the model’s successor. Brown named DeepMind and the NHC’s own model, called HCCA, as two of the best-performing aids for track and intensity.

Individual models have a weakness. Brown said they can “wildly flip” their track or intensity projections from one forecast cycle to the next, so hurricane specialists use experience to judge which models fit a given storm and to blend them. An algorithm can show a track probability, he said, but it does not look at live aircraft and satellite data to determine the current structure of the storm. In his words, “the specialist becomes more important than ever.”

Why it matters

AFP reported that 2026 brought an intense El Nino pattern and a record number of hurricanes in the Pacific, which put the new tools to work in a busy year. Climate change is making hurricanes more powerful, wetter and slower-moving, which worsens their impacts, according to AFP. An extra day of accuracy gives more lead time for the same kind of storm.

The technology has a cost of its own. AFP cited an analysis by Cleanview finding that planned newer US data centers are set to be largely powered by natural gas, which adds to the country’s carbon footprint. Readers in hurricane-prone areas should follow official NHC warnings and local authorities.

Source: The Straits Times