Researchers from the University of Cambridge developed a new AI weather prediction system that reportedly forecasts 10 times faster and more accurately than current models. The system,  Aardvark Weather, is a so-called “blueprint” for a new approach to weather forecasting and could potentially shift the current approach to something entirely new. 

AI Weather Prediction

Photo: University of Cambridge

“Aardvark reimagines current weather prediction methods, offering the potential to make weather forecasts faster, cheaper, more flexible, and more accurate than ever before, helping to transform weather prediction in both developed and developing countries,” said Professor Richard Turner from Cambridge’s Department of Engineering, who led the research. “Aardvark is thousands of times faster than all previous weather forecasting methods.”

Current weather forecasts are generated on supercomputers through complex stages, which take several hours to run. In addition, developing and maintaining the systems is time-consuming and requires a large team of experts. 

Research by Huawei, Google, and Microsoft shows that the numerical solver, which calculates how the weather evolves, could be replaced with AI. However, the Cambridge researchers replaced the whole weather forecasting pipeline with a “single, simple machine learning model.” Aardvark’s model gathers data from satellites, weather stations, and other sensors to produce local and global weather predictions. 

Predictions that were once made on several supercomputers using many models with a large team, can now be made within minutes on a desktop computer. 

The researchers say that Aardvark’s forecasts are apparently competitive with those of the United States Weather Services, which use dozens of models and expert forecasts.

“These results are just the beginning of what Aardvark can achieve,” said first author Anna Allen from Cambridge’s Department of Computer Science and Technology. “This end-to-end learning approach can be easily applied to other weather forecasting problems, for example, hurricanes, wildfires, and tornadoes.”

She continued, “Beyond weather, its applications extend to broader Earth system forecasting, including air quality, ocean dynamics, and sea ice prediction.”