AI Model WeatherNext Improves Cyclone Forecasts by a Day
Felix Pinkston Aug 07, 2026 16:00
WeatherNext by DeepMind gives forecasters an extra day of cyclone warning, a decade's leap in meteorology. Open-source release aims to save lives.
Google DeepMind's WeatherNext AI model has achieved a significant milestone in cyclone forecasting, offering an additional day of predictive accuracy for storm track, intensity, and wind structure. This advance, equivalent to a decade of meteorological progress, could save lives by providing earlier warnings for devastating tropical storms. The model is now open source, enabling the global research community to build on its capabilities.
In a study published August 6, 2026, in Nature, WeatherNext outperformed traditional models, delivering three-day forecasts as accurate as prior two-day predictions. The model's real-world success was highlighted during the 2025 hurricane season, when it helped the National Hurricane Center (NHC) anticipate Hurricane Melissa’s rapid intensification and landfall in Jamaica. This early warning allowed critical preparations on the ground, mitigating potential loss of life and infrastructure damage.
Decoding WeatherNext’s Breakthrough
WeatherNext bridges a gap faced by meteorologists for decades. Cyclone track forecasting relies on large-scale atmospheric models, while intensity predictions require localized, high-resolution data. WeatherNext combines both levels of analysis, using a Functional Generative Network (FGN) approach to train on 20 terabytes of global atmospheric data and nearly 5,000 historical cyclones from the IBTrACS database. This fusion allows the model to forecast cyclone track, intensity, and wind structure jointly and probabilistically.
The model’s efficiency is another standout feature. By operating at a resolution of 28x28 km—100 times coarser than traditional models—it can produce a 15-day forecast in under a minute. Its ensemble method generates up to 1,000 scenarios for each cyclone, enabling forecasters to evaluate a range of possible outcomes, including rare but critical events like rapid intensification.
Real-World Impacts and Open-Source Development
WeatherNext has already demonstrated real-world impact. During the 2025 hurricane season, its predictions informed NHC decisions for Hurricane Melissa, a storm that intensified to Category 5 before striking Jamaica. The model’s ability to simulate rare events proved invaluable in providing early warnings and actionable data.
By open-sourcing WeatherNext and its smaller variant, WeatherNext 2-mini, Google aims to accelerate innovation across the meteorological community. The tools are accessible via Google’s Weather Lab, which provides global weather visualizations alongside cyclone-specific forecasts.
Why It Matters
Tropical cyclones are responsible for over 700,000 deaths and $1.4 trillion in economic losses globally over the past 50 years. Improved forecasting could significantly reduce these figures by giving governments, aid organizations, and local communities more time to prepare. The additional day of accuracy provided by WeatherNext has the potential to reshape disaster response strategies and enhance resilience to extreme weather events.
Experts believe the open-source release will have ripple effects beyond cyclone prediction. By enabling researchers to refine the model, it could lead to advances in renewable energy forecasting and other extreme weather scenarios. As climate change accelerates the frequency and intensity of such events, tools like WeatherNext could become indispensable.
Looking Ahead
With hurricane seasons growing more extreme, models like WeatherNext represent a crucial step forward. Google DeepMind has invited meteorological agencies and researchers to collaborate and expand the model’s capabilities. As AI-driven forecasting continues to evolve, integrating these tools with human expertise could be key to saving lives and mitigating climate-related risks.
Image source: Shutterstock.jpg)