TechBriefAI

Google Launches WeatherNext 2 AI for Faster, Probabilistic Weather Forecasts

Executive Summary

Google DeepMind and Google Research have announced WeatherNext 2, their most advanced AI weather forecasting model. The new model generates higher-resolution global forecasts up to 15 days ahead, 8 times faster than previous methods. By creating hundreds of possible weather scenarios from a single input, it offers probabilistic predictions to better assess risks like extreme weather. WeatherNext 2 is being integrated into consumer products like Search and Maps, while its data and models are being made available to developers and enterprise customers via Google Cloud platforms.

Key Takeaways

* Product: WeatherNext 2, a new AI weather forecasting model.

* Performance: Generates forecasts 8x faster than its predecessor, with resolution down to the hour. It surpasses the previous model on 99.9% of variables for forecasts up to 15 days out.

* Core Technology: Uses a new Functional Generative Network (FGN) architecture that can predict hundreds of possible weather outcomes, enabling probabilistic forecasting for better risk assessment.

* Target Audience: Researchers, developers, businesses, and consumers.

* Availability:

* Forecast data is now available in Google Earth Engine and BigQuery.

* An early access program for custom model inference is launching on Google Cloud’s Vertex AI.

* Integrated into Google Search, Gemini, Pixel Weather, and the Google Maps Platform Weather API, with Google Maps integration coming soon.

* Pricing: No pricing details were provided in the announcement.

Strategic Importance

This announcement solidifies Google's position in applying AI to complex scientific problems, turning a research breakthrough into a scalable product that leverages its cloud and TPU infrastructure. By providing both consumer-facing forecasts and enterprise-level data access, Google aims to displace traditional supercomputer-based forecasting methods and become a foundational platform for weather prediction.

Original article