DeepMind's WeatherNext Model Revolutionizes Cyclone Forecasting
AI technology provides an extra day of early warning for cyclone predictions.
1Revolution in Cyclone Forecasting
DeepMind's WeatherNext model achieved the highest accuracy in predicting the path, intensity, and wind structure of cyclones. This model provides forecasters with an extra day of predictive accuracy. Three-day forecasts are equivalent to the accuracy previous models provided for only two days. This advancement is equivalent to roughly a decade of progress in meteorology.
2Real-World Impact
During the 2025 hurricane season, the WeatherNext model enabled a historic forecast for Hurricane Melissa by predicting its rapid intensification and landfall in Jamaica. This gave local teams critical preparation time. The model predicts 1,000 possible scenarios for each cyclone, supporting forecasters' decision-making processes.
3Open-Source Model
DeepMind is open-sourcing the WeatherNext 2 and WeatherNext Cyclones models. By making this technology openly accessible, they aim to empower the research community and amplify AI's impact in building more resilient communities. This ranges from providing local forecasters with tools to prepare for natural disasters to supporting the growth of renewable energy.
4Technological Advancement
WeatherNext Cyclones predicts global weather patterns and fine-scale cyclone tracks up to 15 days in advance, starting from the global atmospheric conditions during Hurricane Milton (October 2024). Running a 1,000-member ensemble generates localized probability maps of winds ranging from tropical storms to hurricane force.
5Model's Unique Approach
The WeatherNext model predicts global weather and cyclones with a single AI model, marking a first in this field. The model achieved this success through a unique combination of training, architecture, and approach to low-resolution inputs. This sets a new standard in cyclone forecasting.
6Evaluation with Historical Data
WeatherNext Cyclones was evaluated on historical cyclones from 2023-2024, measuring its deterministic and probabilistic performance against other top weather models. On average, it provided more than a full day (24 hours) of lead time advantage for predicting cyclone tracks, intensity, and wind structure.
7Data and Training Process
The model was trained on two distinct data modalities: global weather dynamics and expert-curated historical cyclone observations. It underwent end-to-end training on nearly 20 terabytes of global atmospheric data and the IBTrACS database spanning 5,000 historical storms, learning to model complex atmospheric patterns and extreme weather conditions.
8Future and Community Contribution
DeepMind is open-sourcing the model's code and weights alongside their Nature paper, allowing anyone to build on this technology. They hope to work with the research community to understand how the models produce such accurate predictions at this resolution. This could pave the way for even greater advancements in future weather forecasting.
Original article: Hacker News — DeepMind'in WeatherNext Modeli Kasırga Tahminlerinde Çığır Açtı