Topic — Artificial Intelligence

Google DeepMind’s WeatherNext 2 Brings High-Resolution Forecasting

Published November 18, 2025

Written by

We may earn from vendors via affiliate links or sponsorships. This might affect product placement on our site, but not the content of our reviews. See our Terms of Use for details.

Will it ‘rain’ supreme? The system can produce a wide range of physically coherent outcomes, critical for anticipating worst-case weather scenarios.

Image: Envato

Move over fortune-tellers, Google DeepMind and Google Research have launched yet another impressive leap in AI prediction.

The new system, WeatherNext 2, is the company’s latest upgrade in AI-powered meteorology, promising faster, more accurate, and higher-resolution global forecasts.

Announced in a Google DeepMind blog post, the new system delivers decent capabilities, including generating hourly-resolution predictions up to 15 days ahead, creating hundreds of scenario forecasts, and running eight times faster than its predecessor, while using far less computational power than traditional physics-based models.

How WeatherNext 2 works

At the core of WeatherNext 2 is a Functional Generative Network (FGN). This architecture injects noise directly into the model’s internal functions, allowing the system to produce a wide range of physically coherent outcomes critical for anticipating worst-case weather scenarios.

Although the model is trained only on isolated weather variables (“marginals”), it has learned to accurately capture large-scale atmospheric interactions (“joints”), enabling it to map out complex systems, including multi-state heat domes, and predict wind-farm-scale power generation with greater precision.

WeatherNext 2 is now embedded directly into platforms, including Google Search, Gemini, Pixel Weather, and the Google Maps Platform’s Weather API. Forecast data is now accessible to researchers and developers through Earth Engine, BigQuery, and a Vertex AI early-access program, as Google strives to make climatology tools available to the broader community.

Real-world use cases and validation

The release happens to coincide with increasing interest in AI’s role in forecasting dangerous storms. Earlier this year, a related Google DeepMind hurricane model outperformed traditional government systems during a highly active Atlantic season.

National Hurricane Center forecasters relied heavily on Google’s AI during Hurricane Melissa, using ensemble predictions to support an unusually early forecast of rapid intensification, which is the critical information that gave Jamaica additional time to prepare before the Category 5 landfall.

This evolution in meteorology marks a turning point, as fast, inexpensive AI models can now detect atmospheric patterns that once required hours of supercomputer time. WeatherNext 2 demonstrates how high-precision forecasting can become faster and more accessible while strengthening climate resilience and empowering scientific research.

The future impact on AI meteorology

Still, AI forecasting is not yet without its limitations. Weather prediction relies on neural networking AI, which bases its predictions on the training of identifying patterns in historically recorded weather data.

However, a University of Chicago-led study published research earlier this year suggesting that neural networks are unable to predict weather events that could occur beyond the scope of the existing training data. Therefore, AI weather models may fail to forecast significant events, such as prolonged floods, droughts, and unprecedented heatwaves, for which there is a limited amount of recorded data for training.

WeatherNext 2’s speed and accuracy could very likely nudge the weather industry toward wider adoption of AI-driven forecasting, especially as the technology continues to improve.

Private forecasting firms may rely less on expensive supercomputing, while government agencies could blend AI outputs with traditional models to improve early warnings.

Overall, the technology may propel the industry toward faster and more cost-efficient predictions, hopefully without replacing human expertise just yet.

Sunny days seem to be on the way. Google is supercharging NotebookLM with a powerful upgrade that turns it into a serious research engine.

Madeline Clarke

Madeline is a content writer specializing in copywriting and content creation. After studying Art and earning her BFA in Creative Writing at Salisbury University she applied her knowledge of writing and design to develop creative and influential copy. She has since formed her business, Clarke Content, LLC, through which she produces entertaining, informational content and represents companies with professionalism and taste.