Microsoft has represented Aurora's artificial intelligence capable of predicting weather and natural disasters with high accuracy. Aurora became the world's first machine -based system to exceed the US National Hurricane Center for the precision of a five -day tropical cyclone forecast.
It is a large -scale neural network trained for over a million hours of satellite, radar and meteorous data, as well as climatic simulations. The developers point out that this is the largest array of atmospheric data ever used in meteorological AI.
Aurora works without predetermined physical rules - the model independently reveals the relationships between climatic variables. At first, she studies on general patterns, and then she is prevented for specific tasks - such as predicting the quality of air, height of waves or trajectories of hurricanes.
During the testing, Aurora accurately predicted the Typhoon's way four days before his fall on the Philippines, while official models indicated a different direction. In the 2022–2023 season, the model exceeded seven leading world weather centers to the precision of tropical cyclones forecasting.
The model is also effective in predicting atmospheric pollution. In June 2022, she was able to warn in advance about a large -scale sand storm in Iraq, which paralyzed Baghdada airport. In this case, Aurora used much less computing resources than traditional weather systems.
Thanks to optimized architecture, the system generates forecasts tens of thousands of times faster than supercomputers - within seconds, not hours.
Microsoft has opened the model and weight code for researchers, which will promote the development of weather forecasting technologies. Aurora is already available through the Azure Ai Foundry Labs cloud platform and is used in MSN Weather service.
