Aardvark Weather creates forecasts faster and with less computational effort than current forecasting systems. (Image credit: BroadcastNews via Shutterstock)
Scientists suggest a new weather forecasting system based on artificial intelligence (AI) could change the way we make predictions
The system, called Aardvark Weather, produces forecasts tens of times faster than traditional models using a fraction of the computing power, researchers reported Thursday (March 20) in the journal Nature.
“The systems we rely on to predict the weather have been in development for many years, but in just 18 months we have been able to create a solution that can rival the best of these systems using just a tenth of the data on a typical computer,” said Richard Turner, an engineer at the University of Cambridge in the UK.
Modern weather forecasts are formed by feeding data into complex physical models, a multi-step process that requires several hours of work on a specialized supercomputer.
Aardvark Weather bypasses this labor-intensive process: the machine learning model uses raw data from satellites, weather stations, ships, and weather balloons to make forecasts, without relying on atmospheric models. Satellite data plays a key role in the model’s forecasts, the team emphasized.
This innovative approach could offer significant advantages in terms of the cost, speed and accuracy of weather forecasts, the researchers say. Instead of requiring a supercomputer and a dedicated team, Aardvark Weather can generate a forecast in minutes on a regular computer.
Replacing the Weather Forecasting Pipeline with AI
The team compared Aardvark's performance to existing systems that produce global forecasts. Using only 8% of the data required by traditional systems, Aardvark outperformed the US national Global Forecast System (GFS) and was comparable to forecasts provided by the US Weather Service.
However, Aardvark’s spatial resolution is somewhat lower than that of modern forecasting systems, which may make its initial forecasts less relevant for hyperlocal weather forecasting. Aardvark Weather operates at a resolution of 1.5 degrees, meaning that each block in its grid covers 1.5 degrees of latitude and 1.5 degrees of longitude. By comparison, GFS uses a 0.25 degree grid.
However, the researchers noted that because the AI learns from incoming data, it can be adapted to forecast weather in specific regions — for example, temperatures for agriculture in Africa or wind speeds for renewable energy in Europe. Aardvark can integrate higher-resolution regional data, if available, to improve local forecasts.
“These results are just the beginning of what Aardvark can achieve,” study co-author Anna Allen of the University of Cambridge said in a statement. “This learning approach could easily be adapted to other weather forecasting problems, such as hurricanes, wildfires and tornadoes. It could also be applied to a wider range of Earth system problems, including air quality, ocean dynamics and sea ice forecasting.”
The researchers believe that Aardvark could also support forecasting centres in regions where there are insufficient resources to translate global forecasts into high-resolution regional ones.
“Aardvark’s breakthrough isn’t just about speed, it’s about access,” Scott Hosking, an AI researcher at the Alan Turing Institute in the U.K., said in a statement. “By moving weather forecasting from supercomputers to desktops, we can make forecasting more accessible, bringing these powerful technologies to developing countries and data-poor regions around the world.”
Sourse: www.livescience.com
