March 21st.MicrosoftThe Institute, in conjunction with the University of Cambridge and the Alan Turing Institute, is developing a program called Aardvark Weather of AI weather systemwithout resorting to supercomputers for complex physical simulations.Through deep learning technology, future weather can also be accurately predicted.
Unlike relying on traditional numerical models such as the U.S. Global Forecast System (GFS), Aardvark Weather eschews physical simulations altogether in favor of deep learning techniques.
The system generates highly accurate forecasts at a fraction of the computational cost by analyzing historical and real-time observations. "Aardvark Weather's forecasts can be tens of times faster and consume one-thousandth of the computational resources of traditional methods," the researchers said.
Traditional weather forecasting models require supercomputer support, limiting their application to high-resource weather organizations. Aardvark can be run on standard desktop computers, significantly lowering the technical barrier. This is especially important for developing countries and remote areas that lack high-performance computing infrastructure.
1AI previously reported that the Google DeepMind team launched GenCast in December 2024, which generates a 15-day ensemble forecast in less than 8 minutes. Benchmark tests showed that it outperformed traditional models in a 97.2% scenario, making it one of the most reliable AI forecasting tools available.
Microsoft, on the other hand, focuses on hyper-localized short-term weather forecasting, with its AI models updating precipitation forecasts every two minutes, combining radar and satellite data to provide highly accurate, instantaneous forecasts.
