The storm at sea is now being asked to issue its own notice
the storm at sea is now being asked to issue its own notice: deepmind’s weathernext predicts cyclone track intensity and wind structure three days out as well as older models did in two, an ai correction on ecmwf’s forecast matches operational skill on peak winds and central pressure, and a partnership with noaa is turning cyclone archives into a machine ready database.
the eye now files its own outlook.
Context
Google DeepMind's blog (August 6, 2026) says a Nature paper shows WeatherNext reached state-of-the-art accuracy on cyclone track, intensity and wind structure. On average its three-day forecasts are as good as what prior models gave for the next two days, which it compares to about a decade of meteorological progress. It says the model helped the National Hurricane Center with Hurricane Melissa's rapid intensification in 2025, now predicts 1,000 scenarios per cyclone, and is being open sourced.
ECMWF's blog (August 3, 2026) describes AIFS-TC. A second machine-learning system learns the intensity errors of the existing AIFS forecast and corrects them, using the predicted track, intensity and 3D fields around the storm. On 2025 season storms not used in training, ECMWF says it is comparable to leading operational guidance and to AI state-of-the-art systems for maximum wind speed and minimum central pressure.
NOAA's Technology Partnerships Office (July 8, 2025) says the National Hurricane Center and Google are teaming up so NHC can quickly evaluate Google's new AI tropical model and give feedback.
The first two items match the note. The 'three days as well as older models did in two' is DeepMind's own wording about its own results, and the figure is an average across storms in a vendor blog backed by a Nature paper. The AIFS-TC claim is 'comparable' to the leading systems, not better, and ECMWF's own test used the 2025 season only. The note's third item, 'turning cyclone archives into a machine ready database', could not be matched: the NOAA partnership page read describes model evaluation and feedback and does not mention a database. That part is unsupported here, not refuted. Also, the note joins a Google model, an ECMWF correction and a NOAA partnership that are separate efforts.
Related work
- WeatherNext: AI model achieves breakthrough in forecasting cyclones (Google DeepMind blog) ↗Source for the three versus two day claim, the Melissa example and the 1,000 scenarios.
- Operational tropical cyclone forecasting with AI (Nature) ↗The Nature paper the DeepMind post refers to, introducing WeatherNext Cyclones.
- Improving tropical cyclone intensity forecasts with a simple AI correction (ECMWF) ↗Source for AIFS-TC and its 2025 season test.
- AIFS-TC: A simple correction competitive with the operational frontier for tropical cyclone intensity forecasting (arXiv) ↗The paper behind the ECMWF blog.
- NOAA and Google team up to advance the use of AI hurricane and tropical weather forecast models (NOAA Technology Partnerships Office) ↗Source for the NHC and Google partnership. It does not mention a machine ready database.
Watch next
- Find the source for the machine ready cyclone database claim, if there is one. Compare WeatherNext Cyclones and AIFS-TC on the same storms and metrics.
Sources
- WeatherNext: AI model achieves breakthrough in forecasting cyclones (Google DeepMind blog)deepmind.google
- Improving tropical cyclone intensity forecasts with a simple AI correction (ECMWF)ecmwf.int
- NOAA and Google team up to advance the use of AI hurricane and tropical weather forecast models (NOAA Technology Partnerships Office)techpartnerships.noaa.gov
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 4 October 2026 at 00:48 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →