Microsoft has developed a cutting-edge AI model that can accurately predict various weather-related phenomena, including air quality, hurricanes, and typhoons, according to the company.

In a recently published research paper in the journal Nature and a corresponding blog post, Microsoft provided details about its new AI model, Aurora, which is claimed to forecast atmospheric events with greater precision and speed than traditional methods. With training data exceeding one million hours from satellites, radar, weather stations, simulations, and forecasts, Aurora can be fine-tuned with additional data to make predictions for specific weather events.

The concept of AI-powered weather models is not new, as companies like Google DeepMind have released several models in recent years, including WeatherNext, which has been reported to outperform some of the world’s top forecasting systems. Microsoft is positioning Aurora as a top performer in the field, with potential benefits for research laboratories studying weather science.

According to Microsoft, Aurora has demonstrated impressive predictive capabilities in experiments, including predicting the landfall of Typhoon Doksuri in the Philippines four days in advance, surpassing some expert predictions. Additionally, the model outperformed the National Hurricane Center in forecasting five-day tropical cyclone tracks for the 2022-2023 season and successfully predicted the 2022 Iraq sandstorm.

Although training Aurora required significant computing infrastructure, Microsoft notes that the model is highly efficient to run, generating forecasts in seconds compared to the hours required by traditional systems using supercomputer hardware.

Microsoft has made the source code and model weights for Aurora publicly available and plans to incorporate the AI modeling into its MSN Weather app through a specialized version of the model, which will provide hourly forecasts, including cloud cover.