Google DeepMind has announced "WeatherNext 3," its most advanced and accurate global weather AI model to date, which performs predictions using real-time satellite observation data. The company stated that by leveraging fresh and abundant observation datasets instead of traditional physical simulations, the model is capable of providing clearer predictions than previous versions.

Regarding resolution, while the previous WeatherNext 2 utilized a 25km grid with 6-hour intervals, WeatherNext 3 provides hourly predictions at a 5km resolution, making it approximately five times clearer than the previous model. Specifically, the resolution breakdown is 5km for surface variables such as temperature and humidity, 10km for other surface variables, and 25km for atmospheric variables such as wind speed. The company also stated that precise predictions for precipitation have become possible.

Samier Merchant, a research engineer at Google Research, said that a key development of WeatherNext 3 is its ability to utilize newer datasets that go beyond the data used for training by many other global AI models.

This model has been integrated into Google Search, Gemini, Maps, Google Maps Platform, and Google Cloud. The technology is expected to be utilized in agriculture, renewable energy, supply chains, and daily planning.

Sources: