Cambridge University researchers have demonstrated an advanced AI-driven weather forecast system that outperforms current supercomputers. The Aardvark Weather technology is characterised as thousands of times faster than traditional weather forecasting methods. Scientists claim it runs tens of times quicker than current AI and physics-based models while using far less computing power.
According to main researcher Richard Turner, Aardvark has the potential to transform weather forecasting by making it faster, cheaper, and more accurate. Turner claimed the system could provide solid weather forecasts up to eight days in advance, three days more than current models. The device can also make specialised projections, such as wind speeds for renewable energy installations or temperature forecasts for agricultural land.
Unlike traditional forecasting, which requires numerous models and supercomputers, Aardvark streamlines the process by employing a single, massive language model. This artificial intelligence algorithm collects data from satellites, weather stations, and sensors to generate highly accurate predictions in minutes—all from a desktop computer. Even with only 10% of the data used by previous systems, Aardvark surpassed the US national GFS forecasting system across multiple parameters.
Aardvark’s technique may be used to predict extreme weather occurrences such as hurricanes and wildfires and potentially improve forecasts for air quality, ocean patterns, and sea ice dynamics.
