Google’s latest AI weather model gives you no excuse to forget your umbrella
techcrunch ·
Google DeepMind and Google Research have released WeatherNext 3, a new AI weather forecasting model that produces more detailed and frequent predictions than previous versions. It can forecast at a resolution of 5 kilometers, generate hourly updates instead of every six hours, and shows a 60% improvement in rainfall evaluation over its predecessor. The model outperformed other leading AI weather systems and traditional government forecasts on the Operational WeatherBench benchmark. Google plans to integrate the model into products like Search, Maps, and Gemini, and make it available on cloud platforms. The designers trained it to forecast variables at specific weather stations and to incorporate raw satellite data in real time. These advances aim to address previous AI forecasting weaknesses such as coarse resolution, difficulty with rain, and dependence on formatted datasets. The article notes that WindBorne's WeatherMesh 6 has also been using raw observations, and both models still rely on national weather datasets. Researchers believe such improvements can benefit agriculture, renewable energy, and regions where accurate forecasting has been too costly.
techcrunch