The Met Office and the Alan Turing Institute provide an update on developments around AI's future role in weather forecasting and understanding climate change.
AI technology is playing an increasingly crucial role in addressing environmental challenges and improving our understanding of weather systems and the changing climate, according to leading AI, climate and weather experts at a recent event in London. The event at County Hall in London showcased FastNet, a new machine learning model developed by the Alan Turing Institute and the Met Office which aims to deliver delivering next-generation UK weather forecasts.
Weather prediction has always been complex with ever expanding data availability from satellites and surface sensors. And the last few decades have seen significant improvements in the UK’s ability to make accurate predictions. But now AI has the potential to drastically improve weather forecasting, making it even more accurate and faster.
“We’re in the midst of an AI revolution and it’s happening at just the right time”, according to Professor Kirstine Dale, Chief AI Officer at the Met Office. “There are some major environmental challenges facing us and we’re increasingly aware of our vulnerability to extremes in the weather.”
AI’s ability to process vast amounts of data at unprecedented speed can help to transform the sector. While traditional models can provide accurate forecasts, they rely on immense computational power which limits their use.
Integrating AI and data-driven approaches with well-established physics-based numerical models presents technical challenges, particularly in ensuring robustness, reliability, and trustworthiness leading to the ongoing work between the Met Office and the Turing.
Referring to the partnership Dr Scott Hosking, Interim Director of Science and Innovation of Environment and Sustainability at the Turing, said, “In just a few months the partnership between the Met Office and Turing has built something that matches the performance of the traditional models, and we are really pleased with our progress but there’s a lot more to do. The next step will be to improve precision of our AI model, particularly at a local level to quantify regional impact.”
Despite impressive advances in AI, the experts agreed that there will always be a need for people to be involved in the process. “I’m a massive believer that the compute should follow the subject matter experts into the field. The job of the meteorologist might change but if you just try to solve a problem with a computer alone, you’re going to miss a huge amount of insight,” said Dan Travers, co-founder of Open Climate Fix.
Dr Florence Rabier from the European Centre for Medium-Range Weather Forecasts added, “We are working closely with meteorologists and users from our Member States and rely on their feedback to improve our forecasts. They have the expertise to understand how different models work and to determine which ones are most effective in a particular situation. Only meteorologists can truly assess the quality of the models to help us improve them.”
Rabier also highlighted the importance of collaborating with the tech sector, with many companies developing open-source code which can be used by meteorological institutes and organisations around the globe.
While machine learning and AI holds great potential for analysing weather and climate data, experts discussed the need to find the right balance between using them alongside existing physics-based weather prediction models.
Professor Stephen Belcher, Chief Scientist at the Met Office, said: “What we’d like to do is have information on climate change on a resolution around the globe at a kilometre scale because that’s where the extreme weather events are happening. That’s unlikely to be feasible with current physics-based models for 10-20 years. The question is, can we do that more cheaply and easily using AI tools?”
Professor Penny Endersby, CEO of the Met Office, highlighted the critical importance of continuing to improve weather and climate understanding to “keep people safe, protect businesses and improve our health”. She also warned that “extreme heatwaves have a severe impact on people’s health."
As AI continues to revolutionise society, the partnership between the Met Office and the Turing is essential for maintaining the UK’s leadership in weather prediction while applying AI for public good. Dr Jean Innes, CEO of the Alan Turing Institute, noted that the aim is to put an AI weather prediction model in the hands of Met Office forecasters, to evaluate alongside current methods, within 12 months.
Travers discussed the practical applications, with Open Climate Fix using AI models to help predict how much renewable energy can be generated. This helps grid operators to plan backup power more effectively and to handle uncertainty or unexpected changes in energy supply, saving both money and reducing carbon.
Beyond weather forecasting, AI’s potential extends across many sectors. Dr Kathryn Magnay, Deputy Director at EPSRC highlighted the potential applications of these technologies: “Building on these environmental AI models, there are huge opportunities in energy, the planning of renewables, agri-tech and precision agriculture, biodiversity and then layering on behavioural modelling and datasets and social science research to see what that tells us about the way we behave in a changed climate.”
As AI continues to evolve, its potential to revolutionise weather forecasting and address important environmental challenges remains a key focus for experts across a range of sectors. With AI models set to be integrated into national weather forecasting systems, the future looks bright.
The ultimate goal of The Turing and Met Office partnership is to operationalise the FastNet machine learning model so that meteorologists can use the optimal blend of physics-based and machine learning-based modelling for UK weather prediction. Find out more about the project.