Google DeepMind WeatherNext: Improving Hurricane Intensity and Track Prediction

Google DeepMind WeatherNext: Improving Hurricane Intensity and Track Prediction

Google DeepMind's WeatherNext AI model enabled the National Hurricane Center (NHC) to predict that Hurricane Melissa would reach Category 5 intensity and make landfall in Jamaica five days in advance. This represents a historic milestone, as it was the first time a storm was predicted to reach Category 5 strength starting from Category 1 wind speeds.

Solving the Challenge of Rapid Intensification

Predicting "rapid intensification"—defined as a hurricane's winds increasing by at least 35 mph within 24 hours—is historically difficult due to the scale of the weather systems involved. Traditional meteorological models typically faced a trade-off between track and intensity:

  • Global Models: Effective at predicting a storm's path (track) but lacked the resolution to see small-scale thunderstorms that drive intensity.
  • Local Models: High-resolution and better at predicting intensity, but lacked the global context necessary for accurate track forecasting.

WeatherNext bridges this gap by excelling at both track and intensity prediction. Developed by Google DeepMind and Google Research as part of the Earth AI initiative, the model is trained on decades of global weather patterns and specialized datasets of extreme tropical cyclones.

Performance and Capabilities of WeatherNext

WeatherNext provides probabilistic forecasting rather than a single estimate. It can run ensembles of 50 different "what-if" scenarios to provide experts with a broader range of possibilities for decision-making.

During the tracking of Hurricane Melissa, WeatherNext demonstrated superior predictive power compared to traditional models:

  • Early Prediction: While traditional models were uncertain whether the storm would hit Haiti as a weak system or intensify toward Jamaica, WeatherNext predicted a Category 5 landfall in Jamaica five days in advance with 80% confidence.
  • Confidence Scaling: This confidence level increased to nearly 100% three days before landfall.
  • Seasonal Performance: According to the NHC's annual verification report, WeatherNext was the top-performing individual model for both track and intensity across the 2025 hurricane season.

Real-World Impact and Global Scaling

The integration of WeatherNext into the NHC's guidance suite, alongside physics-based models like HAFS and real-time satellite and hurricane hunter data, provided the Meteorological Service Jamaica with unprecedented lead time for evacuations and resource mobilization.

"With early evacuation and better preparation, that reduction in harm really does make a difference to our people. [...] It does actually save their lives, and it saves the livelihoods that they want to secure."

— Evan Thompson, Principal Director, Meteorological Service Jamaica

Google is currently expanding these research capabilities to other regions prone to extreme weather, including:

  • The Philippines (PAGASA)
  • Taiwan (CWA)
  • Indonesia (BMKG)
  • Vietnam (VNMHA)

Future collaborations are planned for agencies in Japan, Australia, and India. Google has also integrated this technology into Google Search forecasts for geographic areas covered by NOAA.

Sources