Combining enormous data sets of past weather conditions with machine-learning algorithms, GraphCast can provide highly accurate weather predictions up to ten days in advance. Its calculations can be performed in under one minute, compared with about one hour for the supercomputers widely used for weather forecasting around the world up to this point. The technology, which DeepMind has made open source, was described in detail in research published in Science in November 2023.
Related content
According to the Royal Academy of Engineering (RAEng), which presents the MacRobert Award, GraphCast signifies a five-year leap forward in weather forecasting. By significantly improving the speed and reliability of weather predictions, the new technology has the potential to support critical decision-making across numerous industries, optimise resource allocation and help mitigate the impacts of severe weather events, potentially saving lives.
“The heroes behind the UK’s world-changing engineering innovations deserve to be celebrated, and the GraphCast team has made a revolutionary advance,” said Professor Sir Richard Friend FREng FRS, chair of the RAEng MacRobert Award judging panel.
“Modern weather forecasting depends on supercomputers to solve the many coupled equations that project forward in time and the only way to improve that methodology has been to use more computing power. But GraphCast, trained on the huge data set of past weather, speeds up what used to take hours to less than a minute and presents a new roadmap for improving both the timeliness and accuracy of weather forecasting.
“GraphCast is a great example of how engineering innovation can be an important driving force behind economic growth and a sustainable future. In an age of climate change and increasingly volatile weather events, accurate forecasting is an essential tool that can benefit society worldwide.”
Now in its 55th year, the MacRobert Award celebrates UK engineering skills and innovation that continue the UK’s distinguished engineering heritage. GraphCast was announced as the 2024 winner at the RAEng’s awards dinner at The Peninsula Hotel in London, where the Google DeepMind team behind the winning innovation was presented with the MacRobert Award gold medal and a prize of £50,000.
"We are deeply humbled and incredibly proud to receive the MacRobert Award,” said Remi Lam, staff research scientist at Google DeepMind. “Our passion for engineering has always been fuelled by the desire to build AI that helps solve some of the hardest scientific and engineering challenges of our time, and, ultimately, improve billions of people’s lives.
“This recognition reaffirms the crucial role engineering plays in that mission, and brings a spotlight to the nascent field of data-driven weather forecasting. We are particularly excited that GraphCast will enable a new generation of researchers to build on this technology, and we look forward to the breakthroughs they will achieve. Thank you to the Royal Academy of Engineering for sharing our vision."
The MacRobert Award winning Google DeepMind team in full:
- Ferran Alet, Research Scientist
- Peter Battaglia, Research Scientist
- Meire Fortunato, Staff Research Scientist
- Remi Lam, Staff Research Scientist
- Shakir Mohamed, Director of Research
- Alexander Pritzel, Senior Staff Research Scientist
- Alvaro Sanchez-Gonzalez, Staff Research Engineer
- Jacklynn Stott, Research Program Manager
- Matthew Willson, Staff Research Engineer
- Peter Wirnsberger, Staff Research Scientist
Amazon and Google bet big on SMRs to power AI
Yes. The factory is the key bit. But does that work for 500 MWe? It isn't just the core that you want to ship in and assemble on site, but most of the...