Extreme weather events are becoming more frequent and intense. Researchers are seeking faster and more accurate prediction methods, and AI brings new possibilities.
In May this year, Microsoft released the weather prediction tool Aurora. Paris Perdikaris, a Microsoft researcher involved in the Aurora project, said: "These AI tools are good at identifying patterns."
To train Aurora, Microsoft provided over 100 million hours of climate data, about 16 times the amount of the latest GPT model. Aurora can now predict global air pollution for the next 5 days and weather conditions for the next 10 days 5000 times faster than traditional methods.
After collaborating with NVIDIA, The Weather Company's more powerful computing capabilities make AI predictions faster and results more accurate and detailed.
A team from Villanova University focuses on storm prediction. Their model judges the impact of storms by identifying their scale and shape, such as whether they will form tornadoes or hail. With the help of machine learning, the warning time has been extended from 15 minutes to 1 hour before occurrence.
"Speed" is the most significant advantage of AI tools. Traditional General Circulation Models (GCMs) require large amounts of climate data and supercomputers, consuming time and energy. In comparison, new AI weather prediction tools can potentially run on laptops, but their accuracy remains to be seen.
Microsoft says Aurora will be open to the public in the coming months. Perdikaris predicts that AI may be integrated into meteorological workflows in the next 2-5 years.
Google DeepMind's new model "NeuralGCM" takes a comprehensive approach. It is more accurate in 1-10 day climate predictions than pure machine learning models and some currently used models. NeuralGCM combines AI and traditional fluid dynamics calculations, significantly reducing computational requirements while maintaining prediction accuracy.
Aaron Hill, Assistant Professor of Meteorology at the University of Oklahoma, believes that the most meaningful aspect of such AI tools is reducing computational burden, with the potential to build and compute long-term, large-scale climate models.
Under the climate crisis, besides meteorological prediction agencies, commodity traders, agricultural planning industries, and insurance industries are willing to pay for faster and more accurate weather prediction models. This field is developing rapidly.