Weather forecasting has entered an exciting new era with the development of GenCast, an artificial intelligence-based tool created by Google DeepMind.
Demonstrating remarkable accuracy, GenCast has outperformed ENS, the highly respected forecasting system from the European Centre for Medium-Range Weather Forecasts (ECMWF), in predicting both daily weather and the trajectories of extreme events like hurricanes and cyclones.
In tests, GenCast delivered forecasts up to 20 per cent more accurate than ENS, particularly excelling in its 15-day predictions of weather patterns and the paths of tropical storms.
“Outperforming ENS marks something of an inflection point in the advance of AI for weather prediction. At least in the short term, these models are going to accompany and be alongside existing, traditional approaches,” said Ilan Price, a research scientist at Google DeepMind.
Published this week in Nature, the findings showcase AI’s transformative potential in weather forecasting, and the crucial role tools like GenCast could play in saving lives and mitigating economic losses, as the climate crisis drives more frequent and severe weather patterns.
It could also play a role in renewable energy planning, which could help in the adoption of sustainable energy sources like wind power.
The power of AI in weather prediction
Traditional weather forecasting relies on solving complex physical equations using supercomputers, a process that can take hours. In contrast, GenCast operates on a radically different approach. By training on 40 years of historical weather data, including wind speeds, temperature, and humidity at multiple altitudes, the AI learned to predict weather changes across the globe. The system processes this information in minutes using a single Google Cloud TPU, a specialised machine-learning chip.
GenCast doesn’t just produce a single prediction; it generates ensembles—multiple forecasts with assigned probabilities for different events. This approach allows for better risk assessment, particularly useful for predicting extreme weather, which often carries life-or-death consequences.
Practical implications
While GenCast won’t replace traditional forecasts anytime soon, its assistive role has significant potential. For instance, energy companies could use its data to optimise windfarm operations, and emergency services could better prepare for severe weather events like hurricanes, including their landfall locations.
The advancements build on Google’s previous AI-powered forecasting efforts, such as GraphCast and NeuralGCM. GraphCast provides single “best-guess” forecasts, while NeuralGCM combines AI with traditional physics for extended climate and weather modelling. GenCast’s innovation lies in its ability to synthesise and expand on these systems, offering faster, more detailed, and more reliable predictions.
Accelerating the UK’s journey to net zero
Discover how Infinis and Ethical Power are partnering to convert untapped brownfield land into solar energy projects, helping to accelerate the UK’s journey to net zero.
