Google Says WeatherNext Buys Forecasters a Crucial Extra Day

Google Says WeatherNext Buys Forecasters a Crucial Extra Day
For communities in a cyclone’s path, the difference between a forecast and a useful warning can be measured in hours. Google says its latest weather AI may add a full day to that shrinking window.

Google DeepMind said WeatherNext can forecast a tropical cyclone’s track, intensity and wind structure as far as 15 days ahead, with the practical gain coming in earlier accuracy rather than simply a longer outlook. A summary of the announcement described the advance as giving forecasters “an extra day’s worth of predictive accuracy.”

The company and Google Research detailed the claim in a Nature paper, saying WeatherNext’s three-day forecasts match what earlier models could offer at two days. “Every hour counts,” the researchers wrote, framing the work around storms that have caused more than 700,000 deaths and $1.4 trillion in global economic losses over the past half-century.

The technical challenge is unusually stubborn: a cyclone’s path is governed by broad atmospheric currents, while its strength is shaped by fine-scale processes around its core. Google says WeatherNext bridges that divide in one model, trained on global atmospheric data and nearly 5,000 historical storms. In testing on 2023-24 cyclones, it reported more than 24 hours of lead-time advantage across track, intensity and wind structure.

Google also points to operational use during the 2025 hurricane season, when it says the National Hurricane Center used the model’s prediction of Hurricane Melissa’s rapid intensification and Jamaica landfall to issue an advance warning. The company is now open-sourcing WeatherNext 2 and WeatherNext Cyclones, while expanding its forecast ensembles to 1,000 scenarios to better expose low-probability but dangerous outcomes.

That is a powerful claim, but not a declaration that machines should issue warnings alone. Google explicitly directs the public to local meteorological agencies for official forecasts, presenting WeatherNext as decision support for human forecasters—not a substitute for them.

Continue reading https://foxvector.com/stories/019fd8ad-59f0-29ef-71fa-20a0efffc691