Floods in Spain. Fires in Patagonia. A heatwave in Rio de Janeiro. Open the news and you’ll be hit with stories about extreme weather events across the globe. As our climate becomes increasingly deadly, predicting the weather has become a lifesaving task.
Weather forecasts are also indispensable for society; from food production to safe transport to generating energy with wind or solar power. But, the supercomputers that create weather forecasts have a huge carbon footprint. The UK’s Met Office admitted to using a supercomputer that emits 14,400 tonnes of carbon a year. Paradoxically, these forecasting tools contribute to global warming, leading to more of the extreme weather events that they’re designed to predict.
In March 2025, researchers from the Alan Turing Institute, the University of Cambridge, Microsoft Research and the ECMWF published a blueprint for a new way of predicting weather. Known as ‘Aardvark’, the AI-model would allow weather to be forecast on a desktop computer. This would use thousands of times less computing power than the current supercomputers that create weather forecasts.
How can we ensure a green digital future?
Growing e-waste, carbon emissions from AI, data centre water usage––is rampant digitalisation compatible with a healthy planet? Our latest project explores how digital tools and services can be developed with sustainability in mind.
Numerical Weather Prediction vs AI-powered Aardvark Weather
Numerical Weather Prediction (NWP)—creating weather forecasts with supercomputers—has come a long way since its inception in the 1950s. Back then, it took 24 hours to create a single-day forecast. Accuracy and speed have greatly improved, but it still takes hours for NWP supercomputers to process data in order to tell us if tomorrow will bring rain or shine. And of course, all those hours running supercomputers requires massive amounts of energy from data centres.
Aardvark is much faster than NWP, taking just minutes to run on a desktop computer. The AI-model has been trained on data from weather stations around the world. With just 10 percent of input data, Aardvark is already able to generate some forecasts that are more accurate than the United States GFS forecasting system. And, it will be able to produce accurate eight day forecasts, compared to NWP which can only accurately predict up to five days of weather. The research team also outlines the potential for Aardvark models to be adapted to specific regions. This fine-tuning based on regional differences would improve accuracy when predicting weather.
The model has huge potential going forward
Users of NWP systems can easily switch to using Aardvark. And, areas of the Global South who don’t use NWP due to a lack of resources can also adopt the model. The research team wrote in their paper that, “The results presented in this study only scratch the surface of the potential of Aardvark Weather.” They are currently implementing the technology into existing studies at the Alan Turing Institute, while exploring how to deploy Aardvark in the Global South. While it’s not yet stated when Aardvark could replace NWP as our go-to method for predicting weather, the forecast for its future looks bright.
The post AI-Model Aardvark Forecasts Weather Without the Need for Carbon-Emitting Supercomputers appeared first on Digital for Good | RESET.ORG.
