
What happened
Google DeepMind released WeatherNext, an AI model that predicts cyclone track, intensity, and wind structure with state-of-the-art accuracy. The model gives forecasters an extra day's worth of predictive accuracy—three-day forecasts match what prior models could only deliver for two days. The code and weights are now open source. During the 2025 hurricane season, WeatherNext helped the National Hurricane Center predict Hurricane Melissa's rapid intensification and landfall in Jamaica, enabling an advance warning.
Why it matters
Tropical cyclones have caused more than 700,000 deaths and $1.4 trillion in economic losses globally over the past 50 years. An extra day of forecast lead time is equivalent to roughly a decade's worth of meteorological progress. By open sourcing the model, Google aims to help meteorological agencies, researchers, and nonprofits worldwide better predict weather events and protect lives and infrastructure.
What to watch
WeatherNext now predicts 1,000 possible scenarios for each cyclone (up from 50 last year) to help forecasters assess rare but severe outcomes like rapid intensification. A smaller version, WeatherNext 2-mini, runs on a single TPU in a free public Colab notebook. Forecasts and visualizations are available on Weather Lab, which Google recently expanded to show global weather predictions alongside cyclone tracks.
Summaries like this, in your inbox every morning.
WeatherNext represents a fundamental shift in how AI tackles the dual challenge of cyclone forecasting. Traditionally, forecasters faced a trade-off: global atmospheric models could predict a cyclone's track (steered by massive, coarse weather patterns) but struggled with intensity (driven by fine-scale thermodynamic processes around the storm's core). WeatherNext bridges this gap with a single AI model trained end-to-end on nearly 20 terabytes of global atmospheric data and the historical IBTrACS database spanning nearly 5,000 storms. By learning complex atmospheric patterns and extreme-weather dynamics, the model gains more than 24 hours of lead time—a level of improvement unseen in decades of incremental progress.
The real-world proof came during the 2025 hurricane season, when WeatherNext enabled the National Hurricane Center to predict Hurricane Melissa's rapid intensification and landfall in Jamaica, translating into advance warnings that gave communities critical preparation time. This operational success validates the model's practical value beyond benchmarks. The decision to open source WeatherNext 2 and WeatherNext Cyclones further amplifies impact: meteorological agencies, researchers, and nonprofits worldwide now have access to the same breakthrough technology, rather than waiting for indirect adoption through official channels. The scaled ensemble approach—now generating 1,000 possible scenarios per cyclone instead of 50—better captures rare but catastrophic outcomes like rapid intensification, giving forecasters a fuller picture of tail risks they must prepare for.
For example, today's edition would include:
AI-summarized, only the topics you pick: one digest a day via Email, LINE, or Slack.
Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.