As the climate crisis intensifies, with extreme weather events becoming more frequent and severe, researchers are searching for faster and more reliable forecasting tools. Aurora, a new AI model developed by Microsoft Research AI for Science in Amsterdam and Cambridge, and involving experts from the University of Amsterdam, offers a transformative approach.

Aurora leverages a 1.3 billion parameter foundation model trained on over a million hours of Earth system data. According to UvA researcher Max Welling, the model sets a new benchmark in environmental forecasting by significantly outperforming traditional methods in both accuracy and efficiency.

Aurora achieves:

• 74% improved performance in air quality forecasting
  
• 86% superiority over numerical ocean wave models
  
• 100% success rate in outperforming existing cyclone prediction centres
  
• 92% accuracy in high-resolution weather forecasting during extreme events

One of Aurora’s major advantages is accessibility. Traditional forecasting models require expensive infrastructure and supercomputing power. Aurora, however, offers fast, high-quality predictions using far fewer resources—making it a viable tool for smaller weather services and countries in the Global South.

“This kind of acceleration allows engineering teams to build local solutions in weeks instead of years,” says Ana Lucic, AI researcher at the UvA.