UK researchers use AI to forecast air pollution nationwide
University of Manchester researchers have adapted generative AI models to forecast air pollution across the UK.
Researchers at the University of Manchester have adapted generative AI models developed for weather and climate forecasting to model air pollution across the UK, using the country’s Isambard-AI supercomputer to train the system.
The researchers worked with NVIDIA’s Earth-2 team to train the Earth-2 CorrDiff model using data generated from existing chemistry-climate simulations. According to NVIDIA, a year of hourly UK pollution data was used to produce a nationwide model with a spatial resolution of around 2–3 square kilometres, with the training process taking two days on a single eight-GPU node of Isambard-AI.
The team has also incorporated Earth-2 StormCast to support time-dependent forecasts using air-quality observations. Researchers say the system could be used to examine how pollution levels might change under different government policies, while future applications could include advance air-quality information for healthcare services and the integration of real-time observations from edge AI devices during events such as wildfires.
The researchers plan to incorporate additional open data to increase the model’s resolution towards street level. They also intend to release training data and workflows as open source, potentially allowing researchers in other countries and cities to develop similar pollution models using local environmental data.
Why does it matter?
Air pollution can vary substantially between locations and over time, so useful forecasts require models capable of combining weather conditions with the chemical processes that determine how pollutants form and move through the atmosphere. Conventional chemistry-based modelling can require substantial computing resources, limiting how frequently or at what resolution some simulations can be run.
The Manchester research explores whether generative AI can make that modelling faster and more accessible. A system capable of producing detailed forecasts could help authorities identify where pollution is likely to increase, examine possible effects of pollution-control policies, and potentially provide earlier warnings to people particularly vulnerable to poor air quality.
Would you like to learn more about AI, tech, and digital diplomacy? If so, ask our chatbot!