Google DeepMind open-sources WeatherNext after Nature-backed cyclone forecasting breakthrough

[Current Developments] AI20 editorial coverage based on Google DeepMind (2026-08-06). On 6 August 2026, Google DeepMind reported that its WeatherNext AI model achieved state-of-the-art accuracy in predicting tropical cyclone track, intensity and wind structure, with results published in Nature. The company states that the model gives forecasters an extra day’s worth of predictive accuracy on average — with three-day forecasts matching what prior models delivered for only the next two days — and frames that gain as roughly a decade of meteorological progress. DeepMind says the work was developed with Google Research and expert forecasters including the U.S. National Hurricane Center, CIRA, the UK Met Office and other agencies, and notes operational impact during the 2025 hurricane season, including support for Hurricane Melissa forecasting. Alongside the paper, DeepMind is open-sourcing WeatherNext 2 and WeatherNext Cyclones model assets used during the hurricane season, plus WeatherNext 2-mini for lighter experimentation, and points researchers to Weather Lab for interactive forecasts. The post emphasises ensemble forecasting at scale — up to 1,000 members — to capture rare intensification scenarios while keeping official warnings with national meteorological services. Why it matters: An extra day of cyclone lead time is a public-safety and adaptation asset, not only a research benchmark. Open model weights and shared forecast tooling can widen who can build local early-warning capacity — especially where national agencies lack proprietary systems. AI20 perspective: AI20 should treat open climate and disaster-forecasting models as alliance public goods: capability that travels with documentation, verification and human forecaster partnership rather than closed black-box alerts. (Category: Research; Region: Global; Tags: WeatherNext; Google DeepMind; cyclones; open source; climate resilience; Nature. Original source: https://deepmind.google/blog/weathernext-ai-model-achieves-breakthrough-in-forecasting-cyclones/)