A revolutionary new weather forecasting system developed at the University of Cambridge could upend decades of meteorological tradition by replacing complex supercomputer-based models with fast, accurate predictions generated on a desktop PC.
The system, known as Aardvark Weather, was developed by researchers in Cambridge’s Department of Engineering and the Department of Computer Science and Technology, with support from the Alan Turing Institute, Microsoft Research and the European Centre for Medium Range Weather Forecasts (ECMWF). The findings were recently published in Nature.
“Aardvark reimagines current weather prediction methods,”
Professor Richard Turner
“It’s thousands of times faster than all previous methods, yet it’s also cheaper, more flexible and more accurate.”
Unlike traditional forecasting, which uses numerous interdependent models and requires hours of processing time on powerful supercomputers, Aardvark uses a single machine learning model. It processes data from satellites, weather stations and other sensors to produce global and local forecasts within minutes.
Speed, Accuracy and Accessibility
The new system already outperforms the US National GFS forecasting system using just 10% of the data. It’s even competitive with the highly refined forecasts of the US Weather Service, which combine dozens of models with human analysis.
Researchers believe Aardvark could have huge implications for developing countries, where access to forecasting infrastructure is limited. “It’s about access,” said Dr Scott Hosking of the Alan Turing Institute. “Aardvark shifts forecasting from supercomputers to desktops, putting cutting-edge prediction tools in the hands of those who need them most.”
Its versatility is also key. Aardvark can be quickly adapted for specific industries or regions — from African agriculture to European wind farms — without the years of development typical of traditional systems.
The team now plans to expand their work through a new group at the Alan Turing Institute, focusing on deploying Aardvark in data-sparse regions and extending its capabilities to broader environmental forecasting.
