Tech insider: Aardvark Weather Helen Norman AI REVOLUTION I Researchers in Europe have developed a fully AI-driven weather prediction system that they believe could start a revolution in forecasting. MTI speaks to one of the professors involved to find out more n late March 2025, researchers unveiled a new AI weather prediction system that is claimed to deliver accurate forecasts tens of times more quickly and using thousands of times less computing power than current AI and physics-based forecasting systems. Aardvark Weather has been developed by researchers from the University of Cambridge, supported by the Alan Turing Institute, Microsoft Research and the European Centre for Medium Range Weather Forecasting. According to the researchers, it provides a blueprint for a completely new approach to weather forecasting with the potential to transform current practices. The team has revealed that Aardvark Weather goes further than current AI weather prediction models, such as those developed by Huawei, Google and Microsoft, by replacing the entire weather prediction pipeline with a single, simple machine learning model. “Big tech companies and forecasting agencies have made significant advances in applying AI to weather prediction in the last couple of years, demonstrating the potential of AI-powered weather models to complement traditional physics-based numerical weather prediction models,” explains Prof. Richard Turner, lead researcher for weather prediction at the Alan Turing Institute and professor of machine learning in the department of engineering at the University of Cambridge. “All of these AI systems just focused on the second step of this process, replacing the simulator. They rely on traditional methods to process the raw observed data, which means they are still computationally expensive. Like numerical weather prediction, Aardvark uses raw data from satellites, weather stations and RIGHT: Prof. Richard Turner, lead researcher for weather prediction at the Alan Turing Institute and professor of machine learning in the department of engineering at the University of Cambridge, testing the new Aardvark Weather model A BACKING UP CLAIMS t the launch of Aardvark Weather, the research team claimed that the new solution is “thousands of times faster than all previous weather forecasting methods”. To determine this, the team compared Aardvark Weather’s performance against two of the most widely used deterministic operational global NWP systems: ECMWF’s Integrated Forecast System (IFS) in its high-resolution (HRES) configuration, and the Global Forecast System (GFS) from the National Centers for Environmental Prediction. “Aardvark generates a full forecast from observational data in approximately one second on four Nvidia A100 GPUs, compared with the approximately 1,000 node-hours required by HRES to perform data assimilation and forecasting alone, before accounting for downstream local models and processing,” Turner explains. 8 • www.meteorologicaltechnologyinternational.com • April 2025
Meteorological Technology International Meteorological Technology International April 2025: Page 8
