AI Forecast: The Future of Weather Prediction Is About to Change

Meteorologists, Cambridge researchers and industry professionals at the UK AI for Weather Forecasting Workshop, co-organised by Cambridge Zero and Institute of Computing for Climate Science (ICCS), said rapid advances have already opened new possibilities for AI weather forecasting, but it still lags behind existing methods based on physical models. 

"Whilst there have been great advances in this space there was an appreciation that there is still a long way to go to make these truly on par with our state-of-the-art physics-based models," Met Office Chief Operational Meteorologist Steve Ramsdale told conference participants. 

Weather forecasting underpins many of the decisions that shape daily life, with temperature and rainfall extremes driving more frequent floods, droughts and storms across the country. These shifts have direct implications for food security, from disrupted planting seasons to increased crop losses, and for nature, as ecosystems struggle to adapt to rapid swings between very wet and very dry conditions. 

As the climate system changes and weather patterns become more complex, existing forecasting approaches are being pushed to their limits in both accuracy and the speed at which information can be delivered and acted upon.

Attendees agreed that the need for better tools to help society navigate a world where weather extremes are becoming more frequent, more disruptive and more deeply felt has never been clearer.

In response to these challenges, some of the workshop conversations centred around how the ability of AI forecasting models to predict extreme weather events can be improved. Currently, these AI models have typically been developed to skilfully reproduce the "usual weather" because that helps keep the model reliable and stable.

Discussions focused on how these approaches can responsibly complement physics-based models, while ensuring forecasts remain reliable, interpretable and useful in a rapidly changing world.

Participants agreed that the next phase of progress in AI weather forecasting is likely to come from making better use of the growing volume of remote sensing and observational data now available, with the potential to deliver significant improvements in forecast accuracy.

“It's important that the problems machine learning researchers are solving are actually the right problems to solve, because when our systems fail, there are lives at stake,” said Cambridge PhD student Jonas Scholz. 

As the workshop drew to a close, participants reflected on the importance of ensuring that advances in AI translate into real-world value. Beyond improving skills, the right models must be developed based on the right data to deliver insights that genuinely support decision-making for communities, businesses, and government. 

This shift from raw prediction to meaningful interpretation will shape how effectively new forecasting tools can help society navigate a more volatile climate.

“We at the Met Office refer to the ‘final mile’,” Ramsdale said. “That is how we turn this raw weather model output into something that is of value to our customers and the world.”