One of Microsoft’s latest AI models can accurately predict air quality, hurricanes, typhoons and other weather-related phenomena, the company claims.
In a paper published in Nature and a blog post this week, Microsoft said Tech Giant can predict atmospheric events with greater accuracy and speed than traditional weather approaches. Trained with over a million hours of data from satellites, radars, weather stations, simulations and predictions, Aurora can fine-tune additional data to predict specific weather events.
The AI weather model is nothing new. Google Deepmind has released only a handful over the past few years, including Weathernext. This claims that the lab is breaking some of the world’s best prediction systems. Microsoft positions Aurora as one of the top performers in the field, and is potentially beneficial for labs studying weather science.
In the experiment, the Aurora predicted a typhoon landing in the Philippines four days before the actual event, breaking predictions from several experts, Microsoft says. The model also defeated the National Hurricane Center in predicting the five-day tropical cyclone track for the 2022-2023 season, successfully predicting the 2022 Iraxand Storm.
Aurora required substantial computing infrastructure to train, but Microsoft said the model is very efficient to run. Generate predictions in seconds compared to the time spent using a traditional system using supercomputer hardware.
Microsoft, which has published source code and model weights, says it is incorporating Aurora’s AI modeling into the MSN weather app via a special version of the model that generates hourly predictions, including the cloud.