Microsoft has introduced a cutting-edge AI model named Aurora, which is designed to accurately predict various atmospheric events, including air quality, hurricanes, and typhoons. The company asserts that Aurora surpasses traditional meteorological methods in both precision and speed. In a recent paper published in the journal Nature, along with a detailed blog post, Microsoft elaborated on Aurora’s capabilities.
The model has been trained on a vast dataset exceeding a million hours, encompassing information from satellites, weather stations, radar, and simulations. With the ability to be fine-tuned for specific weather phenomena, Aurora aims to enhance the forecasting process. AI weather models have been emerging for several years, with notable contributions from companies like Google DeepMind.
Their weather model, WeatherNext, claims to outperform some of the existing top forecasting systems. Positioning Aurora as a leading competitor, Microsoft believes it could significantly benefit laboratories engaged in weather science research. In practical applications, Aurora has demonstrated its prowess by predicting the landfall of Typhoon Doksuri in the Philippines a remarkable four days ahead of the actual event.
Additionally, the model outperformed the National Hurricane Center in forecasting tropical cyclone paths during the 2022-2023 season and accurately predicted a sandstorm in Iraq in 2022. Although training Aurora required extensive computing resources, Microsoft emphasizes that it is efficient to operate. The model can produce forecasts in mere seconds, unlike traditional systems that can take hours relying on supercomputer infrastructure.
Furthermore, Microsoft is integrating Aurora’s capabilities into its MSN Weather app, offering specialized features that include hourly forecasts and cloud predictions, while also making the source code and model weights publicly accessible.