© IE Online Media Services Pvt Ltd
Tags:
Journalism of Courage
Researchers at Cambridge University have developed a new weather prediction system that “is thousands of times faster than all previous weather forecasting methods.” Named Aardvark Weather, researchers say the new weather forecasting system is tens of times faster compared to current AI and physics based systems while consuming “thousands of times less computing power.”
According to the lead researcher Richard Turner, Aardvark has the potential to “make weather forecasts faster, cheaper, more flexible and more accurate than ever before.”
For those unfamiliar, the current weather forecasting system involves a lot of complex stages, each of which take several hours to process even on powerful supercomputers. Unsurprisingly, they also require a lot of maintenance, a large team of experts and have high development costs.
With Aardvark, researchers say they have replaced the entire traditional system with a single and simple large language model, which takes in information from satellites, weather stations and other sensors for both local and global forecasts.
Also, predictions that were once developed using several models and backed by several supercomputers and teams of experts can now be produced “in minutes on a desktop computer.” When Aardvark used just ten per cent of the input data of existing systems, it was able to outperform the United States’ national GFS forecasting systems in various parameters.
According to Anna Allen, the research’s first author, “this end-to-end learning approach can be easily applied to other weather forecasting problems” like hurricanes, tornadoes and wildfires and even be extended to air quality, ocean dynamics and sea ice prediction.
