Google DeepMind WeatherNext: Breakthrough in Cyclone Forecasting

Google DeepMind has introduced WeatherNext, an AI model designed to predict the track, intensity, and wind structure of tropical cyclones with state-of-the-art accuracy. The model provides forecasters with an average of one additional day of predictive accuracy, meaning three-day forecasts are now as accurate as previous models' two-day forecasts. According to DeepMind, this improvement is equivalent to roughly a decade of meteorological progress.

High-Accuracy Cyclone Prediction at Low Resolution

WeatherNext bridges the gap between global atmospheric currents (which steer a cyclone's track) and localized thermodynamic processes (which drive intensity). Traditionally, these required two different modeling techniques: coarse global models for track and high-resolution local models for intensity. WeatherNext is a single AI model that handles both.

A surprising finding in the research is the model's ability to maintain high accuracy despite using significantly lower resolution data. WeatherNext Cyclones operates at a resolution of 28x28km—100 times coarser than traditional models. A smaller version, WeatherNext 2-mini, operates at an even coarser 111x111km resolution and still demonstrates strong performance. The exact mechanism by which the model produces accurate predictions at this resolution remains an open research question.

Training Data and Model Architecture

WeatherNext was co-trained on two distinct data modalities to learn both general atmospheric patterns and extreme weather events:

• Global Weather Dynamics: Nearly 20 terabytes of global atmospheric data.
• Historical Cyclone Observations: The IBTrACS database, which includes data from nearly 5,000 historical storms.

To handle the inherent uncertainty of weather, the model utilizes Functional Generative Networks (FGNs) to produce ensembles of predictions. This allows the system to generate a 15-day forecast in less than a minute on a TPU. The ensemble size has been scaled to 1,000 members, enabling the capture of rare but consequential "tail-risks," such as the rapid intensification events seen during Hurricane Melissa in 2025.

Real-World Impact and Open Source Release

WeatherNext has already been applied operationally. During the 2025 hurricane season, it helped the National Hurricane Center (NHC) predict the rapid intensification and landfall of Hurricane Melissa in Jamaica, allowing for earlier advance warnings.

DeepMind is open-sourcing the following models and code:

• WeatherNext Cyclones: The model used during the hurricane season.
• WeatherNext 2: An operational update released in October.
• WeatherNext 2-mini: A compact version capable of running on a single TPU via a public Colab notebook.

Community Insights and Technical Perspectives

While the researchers highlight the breakthrough, the technical community has noted several critical dependencies and considerations:

"The impression that industry can predict weather better than the government agencies totally misses that the industrial models utterly rely on government data for inputs."

Observers emphasize that AI models like WeatherNext rely entirely on the critical ground-truth data provided by government infrastructure (such as weather balloons and satellites) managed by agencies like NOAA and the UK Met Office.

Other discussions centered on the efficiency of AI-driven forecasting compared to Numerical Weather Prediction (NWP) models, with some users noting that problem-specific models are more impactful than general-purpose LLMs. There are also ongoing questions regarding the model's explainability and how uncertainty estimates are handled for high-stakes government evacuation orders.