Google DeepMind and Google Research have introduced WeatherNext 2, the company’s most advanced and efficient weather forecasting model to date, marking a major leap in AI-powered climate prediction. This next-generation system promises forecasts that are both faster and more accurate, offering a technological breakthrough that could redefine how industries and individuals plan for weather-related events.
According to Google, WeatherNext 2 can generate weather forecasts eight times faster than previous versions while achieving resolution down to one hour. The model can also produce hundreds of possible forecast scenarios within seconds, a task that would traditionally require hours of computation on a supercomputer using physics-based methods. This step effectively brings AI weather forecasting from research laboratories to practical, global-scale applications.
From experimental research to real-world use
Following years of development, Google is now integrating WeatherNext 2 into its own ecosystem. The model’s forecast data has been made available through Earth Engine and BigQuery, enabling researchers, developers, and policymakers to access high-resolution weather insights. In addition, Google has launched an early access program through Vertex AI on Google Cloud, allowing organizations to run custom inference and create specialized weather models.
The technology is also coming to Google’s consumer and enterprise products. Forecasts powered by WeatherNext 2 are now active in Search, Gemini, Pixel Weather, and the Google Maps Platform Weather API. In the weeks ahead, the upgraded forecasts will also enhance weather information in Google Maps.
Akib Uddin, WeatherNext 2 product manager at Google, said:
“Whether you're on search, Android, or Google Maps, weather affects everyone, and so by making better weather predictions, we're able to help everyone.”
Functional Generative Network drives a new forecasting approach
The improved accuracy and speed of WeatherNext 2 arise from a novel AI methodology called the Functional Generative Network (FGN). Instead of simulating atmospheric physics, FGN injects controlled "noise" into the model’s structure, allowing the system to generate a diversity of coherent, realistic scenarios.
This enables WeatherNext 2 to predict hundreds of potential weather outcomes from a single data input while maintaining physical consistency. In practice, each individual forecast can be completed in under one minute on a Tensor Processing Unit (TPU), a specialized chip optimized for AI.
“We rely on accurate weather predictions for critical decisions—from supply chains to energy grids to crop planning,” said Google DeepMind research scientist Peter Battaglia. “AI is transforming how we forecast weather.”
In testing, WeatherNext 2 outperformed its predecessor, WeatherNext Gen, in 99.9% of evaluated weather variables, including temperature, wind, humidity, and pressure, across lead times extending up to 15 days.
Battaglia added:
“It’s about eight times faster than the previous probabilistic model that we released last year, and in terms of resolution, it is six times greater. So instead of making six-hour steps, it takes one-hour steps.”
Understanding the science behind the model
What makes FGN unique is how it learns complex weather interactions. The model is trained only on “marginals”, single-variable data such as temperature or wind at a specific location, but it then learns how these separate elements interact to form “joints,” or interconnected meteorological systems.
This ability allows WeatherNext 2 to forecast wide-ranging patterns like heatwaves, cyclone tracks, or power grid fluctuations, even though it was not directly trained for those combined outcomes.
Compared with its predecessor, the model demonstrated measurable improvements in standard accuracy metrics such as the Continuous Ranked Probability Score (CRPS), with reported gains of 7.5–8.7%. It also achieved a 24-hour reduction in positional errors when forecasting tropical cyclones, providing earlier and more reliable storm trajectory predictions.
DeepMind said that it has already shared experimental cyclone-prediction tools based on this AI technology with global weather agencies for further testing.
Empowering global users through data and cloud platforms
With WeatherNext 2, Google is making its AI-driven research broadly accessible. The company aims to help both scientists and businesses make smarter, faster decisions in response to extreme weather and climate risk.
By offering open data access through Earth Engine, integration into existing Google services, and customizable deployment on Google Cloud’s Vertex AI, the tech giant is bridging the gap between cutting-edge research and real-world application.
As the world faces increasingly unpredictable climate patterns, WeatherNext 2 represents a significant advancement in using AI to help societies adapt. Faster, more detailed, and more inclusive weather forecasting could improve disaster response, logistics, agriculture, and energy distribution, all within Google’s rapidly expanding AI ecosystem.
Gunel Ismayilova
Disclaimer: All materials on this site are for informational purposes only. None of the material should be interpreted as investment advice. Please note that despite the nature of much of the material created and hosted on this website, HODL FM is not a financial reference resource, and the opinions of authors and other contributors are their own and should not be taken as financial advice. If you require advice. HODL FM strongly recommends contacting a qualified industry professional.
