An AI weather prediction model is set to transform global forecasting, offering predictions tens of times faster and using thousands of times less computing power than current AI and physics-based forecasting systems.
Developed by researchers from the University of Cambridge in collaboration with the Alan Turing Institute, Microsoft Research, and the European Centre for Medium-Range Weather Forecasts (ECMWF), the Aardvark Weather model aims to democratise weather forecasting, making it more accessible to developing nations.
Professor Richard Turner from Cambridge’s Department of Engineering, who led the research, described Aardvark as “thousands of times faster than all previous weather forecasting methods.”
Unlike traditional physics-based models that require extensive computational power, Aardvark operates on a standard desktop computer, cutting down processing time from hours or even days to mere seconds.
“Aardvark reimagines current weather prediction methods, offering the potential to make forecasts faster, cheaper, more flexible, and more accurate than ever before,” Turner said. “This could be a game-changer for both developed and developing countries.”
Current weather prediction systems rely on a multi-stage process involving physics-based numerical weather prediction (NWP) models that demand significant computing power. The introduction of AI in recent years by companies such as Google DeepMind, Huawei, and Microsoft has shown that certain components of this pipeline can be replaced with machine learning, improving efficiency. However, Aardvark takes this a step further by replacing the entire pipeline with an AI-driven approach.
The new model ingests observational data from satellites, weather stations, balloons, ships, and aircraft, generating both global and hyper-local forecasts without the need for expensive supercomputers. According to the researchers, Aardvark already outperforms the US national Global Forecast System (GFS) on many forecasting variables, even using just 10% of the input data.
Dr Scott Hosking, Director of Science and Innovation for Environment and Sustainability at the Alan Turing Institute, said the breakthrough could “democratise forecasting by making powerful technologies available to developing nations around the world, as well as assisting policymakers, emergency planners, and industries that rely on accurate weather forecasts.”
Dr Anna Allen, lead author of the study, noted that Aardvark’s end-to-end learning approach could be extended beyond standard weather forecasting to enhance predictions for hurricanes, wildfires, tornadoes, air quality, ocean dynamics, and sea ice conditions.
Despite its promising results, Aardvark still faces challenges before it can replace traditional meteorological models entirely. It has been cautioned that while the model is impressive, further development is needed to generate all necessary variables at high spatial resolution.
Microsoft, a key industrial partner in the project, has confirmed that Aardvark will remain completely open-source, ensuring that its benefits can be widely shared and developed collaboratively by the global scientific community.
The Alan Turing Institute is now establishing a dedicated research team led by Turner to explore ways to integrate Aardvark into high-precision environmental forecasting efforts for weather, oceans, and sea ice.
The team aims to roll out Aardvark across data-sparse regions worldwide, unlocking new possibilities in AI-driven meteorology.
