An artificial intelligence model developed by Microsoft, led by Greek professor Paris Perdikaris from the University of Pennsylvania, can successfully predict various aspects of Earth’s behavior—from weather patterns and atmospheric pollution to ocean waves. The model, called Aurora, is featured in a publication by the journal Nature.
Forecasting Earth systems is an essential tool for providing timely warnings about extreme events. Such predictions come from complex models that rely on decades of data and often require supercomputers and specialized teams to maintain—making them inaccessible to many communities around the world.
Aurora is an AI model trained on over a million hours of geophysical data. Its creation had two main goals: to provide a forecasting tool that is both more accurate and computationally much more efficient. Adopting a fundamentally different approach from traditional weather prediction models, Aurora learns patterns directly from data, identifying complex relationships in historical Earth system data to make forecasts.
As noted in the publication, Aurora outperforms existing models for air quality, ocean wave patterns, tropical cyclone tracking, and high-resolution weather forecasting—all while operating at a lower computational cost than current forecasting methods. According to the reported figures, for air quality prediction, Aurora met or exceeded the performance of the Copernicus Atmosphere Monitoring Service in 74% of targets, while being about 50,000 times faster. Additionally, for high-resolution weather forecasting, the model outperformed the top numerical weather model, IFS HRES, in 92% of targets at a resolution of 0.1°, showing particularly strong performance in extreme events.
“Aurora represents a major innovation in environmental system forecasting, as it is the first AI model to function as a unified foundational model capable of adapting to various applications—from high-resolution weather forecasting and air quality prediction to tropical cyclone tracking and ocean wave monitoring. This approach achieves high accuracy at a fraction of the computational cost, making advanced environmental forecasting accessible to broader communities globally,” said Mr. Perdikaris in a statement to the Athens-Macedonian News Agency.
A key innovation of the model is that it can be trained on massive volumes of diverse geophysical data and then fine-tuned for specific forecasting tasks—like a powerful brain that can specialize in performing different prediction-related functions.
As Mr. Perdikaris observes,
“The Aurora project, during my time at Microsoft Research, was part of my broader research vision to create foundational models for scientific applications that can generalize across disciplines and accelerate discovery.”
He adds that at the University of Pennsylvania,
“My team is expanding this vision beyond Earth sciences to a range of science and engineering applications, building AI systems that not only predict but also help us understand complex physical phenomena across multiple fields.”
Similar model-based approaches are also being applied by his team in other scientific areas, ranging from materials engineering to biomedical applications.
Ask me anything
Explore related questions
