Google Announces Next-Gen AI Weather Forecasting "WeatherNext 3", Dropping Physical Simulations

Release Overview

Google has announced "WeatherNext 3," its latest weather forecasting technology. The major turning point of this technology is the abandonment of traditional physical simulations—which have been mainstream in weather forecasting—in favor of a transition to machine learning models that directly train on live data obtained from satellites.

Transforming the Approach to Weather Forecasting

WeatherNext 3 is designed to leverage live satellite imagery of the entire globe, predicting weather patterns directly from that data. The model is trained using vast amounts of historical satellite data, operating on a mechanism where it learns dynamics directly from observational data itself, rather than mathematically modeling atmospheric behavior.

Technical Background and Benefits

Traditional physics-based simulations are computationally expensive and typically require supercomputers. The data-driven approach adopted by WeatherNext 3 makes it possible to generate forecasts with equal or greater accuracy while curbing the consumption of computational resources. This can be described as a technological turning point that suggests AI is capable of inferring physical phenomena from statistical correlations.

Future Outlook

Google plans to utilize this technology as the foundation for weather forecasting. Free from the constraints of physical simulations, this method enables real-time, high-precision forecasting and is expected to contribute to addressing climate change and accelerating disaster prediction.