Microsoft’s Aurora AI Now Predicts Air Quality with High Accuracy
Microsoft has announced that its AI weather model, Aurora, can now accurately forecast air quality in addition to predicting severe weather events. Developed by Microsoft Research, Aurora has been designed to outperform traditional meteorological systems in terms of speed and precision, and is now being used to generate hourly weather updates, including cloud coverage, on the MSN Weather app.
A research paper detailing Aurora’s capability was recently published in the journal Nature, and Microsoft has made both the source code and model weights publicly accessible for further research and innovation.
What Sets Aurora Apart?
Aurora was built as a foundational model, meaning it was pre-trained on vast datasets before being fine-tuned for specialized tasks such as air pollution prediction. According to Microsoft, this architecture allows Aurora to learn from over a million hours of satellite, radar, and weather station data, making it adaptable with relatively smaller datasets for specific applications.
Its encoder-based architecture translates massive volumes of data from various sources into a unified format for forecasting. “This is the power of deep learning in simulation problems,” said Megan Stanley, Senior Researcher at Microsoft Research.
Aurora’s predictive prowess was demonstrated when it accurately forecasted Typhoon Doksuri’s landfall in the Philippines four days in advance and predicted a sandstorm in Iraq two years ago. It also outperformed the US National Hurricane Centre in five-day cyclone path forecasting during 2022 and 2023.
Also Read: Google Adds ‘Deep Think’ to Gemini 2.5 Pro, Boosting AI Reasoning Power
Faster, Cheaper, And Open Source
Unlike traditional supercomputer-powered systems, Aurora uses GPUs to deliver forecasts within seconds, significantly reducing operational costs despite its high initial training expenditure. Microsoft believes Aurora’s open-source release will spur collaborative innovation in AI-based weather forecasting and environmental monitoring worldwide.
