Natural disasters—whether hurricanes, tsunamis, or severe smog —threaten many regions around the world. Microsoft’s Aurora AI reportedly accurately predicts natural disasters and air quality.

AI Disaster Prediction

There are several weather prediction models, but a recent study published in Nature found that Microsoft’s Aurora AI goes beyond anticipating weather. As a foundational model, Aurora uses the latest advances in AI to accurately predict a wide range of environmental events, including “hurricanes and typhoons, air quality, and ocean waves.”

Initially trained as a foundational model, researchers fine-tuned Aurora to go beyond weather prediction, expanding into air pollution prediction and predicting ocean waves and tropical cyclones.

Microsoft says Aurora learns how to generate forecasts in seconds. Aurora collected over one million hours of data from satellites, radar and weather stations, simulations, and forecasts. Then, researchers “fine-tune” Aurora to perform specific tasks based on “a modest amount” of additional data. These tasks include predicting wave height or air quality.

Aurora uses its size and training with diverse data to forecast weather and environmental events. “This helps Aurora beat existing numerical and AI models across 91 percent of forecasting targets when fine-tuned to medium-range weather forecasts at a resolution of .25 degrees,” researchers reported in the study.

Using a diverse range of data results in “not only greater accuracy in general, but it also means we are better at forecasting extreme events,” says Megan Stanley, a senior researcher with Microsoft Research who was part of the core Aurora project team.

Accurately Predicting Weather Events

The Nature study reports that Aurora predicted Typhoon Doksuri in the Philippines in July 2023, four days before landfall. Aurora was reportedly more accurate than expert models, which predicted that the storm would be off the coast of Northern Taiwan.

Most recently, Aurora reportedly beat out the National Hurricane Center in forecasting 5-day tropical cyclone tracks. Microsoft says this breakthrough is a first for a machine-learning model. The tech giant also said Aurora outperformed seven major forecasting centers globally on all cyclone forecasts for the 2022-2023 season.

Beyond weather events, Aurora showcased accuracy on June 13, 2022, when the model predicted a devastating sandstorm in Iraq. As the study reports, Aurora predicted the sandstorm a day in advance.

Microsoft made Aurora’s source code and model weights publicly available “to help advance the field of atmospheric forecasting.” This allows developers to download, run, or build on Aurora. Microsoft says MSN Weather now incorporates Aurora’s advanced AI modeling. The MSN Weather team also developed a specialized model of the system to include hourly forecasts and more weather parameters, including cloud and precipitation.