AI weather tools show promise in forecasts and disaster deployment

A reported learned postprocessing layer doubled forecast skill at two-to-six-week horizons and improved extreme-event predictions, though methodological detail and independent validation remain necessary. Google has also opened disaster-forecasting code and weights, including cyclone models and OpenHydroNet under Apache 2.0; reported flood-horizon gains and production use are company-reported. External adoption and performance outside Google’s infrastructure remain open questions.