University of Manchester Uses NVIDIA Earth-2 to Forecast Air Pollution Across the UK
nvidia ·
Physicists David Topping and doctoral student Hao Zhang at the University of Manchester have applied NVIDIA Earth-2 generative models to build a UK-wide air-pollution forecasting workflow. Air pollution contributed to an estimated 30,000 deaths in the UK last year, while traditional chemistry-based air-quality models are costly and slow to run. Using training data generated from existing chemistry-climate simulations, the team trained Earth-2 CorrDiff, a generative downscaling model, on Isambard-AI, and it worked on the first attempt. With a year of UK pollution data simulated hourly, it produced a model at 2–3 square kilometers resolution in two days on a single eight-GPU node; Isambard-AI contains 5,448 NVIDIA GH200 Grace Hopper Superchips and delivers 21 exaflops of AI performance. The team later added Earth-2 StormCast for time-dependent forecasts that directly use air-quality observations and demonstrated the training and inference workflows on an NVIDIA DGX Spark desktop AI supercomputer. Potential applications include testing policy scenarios, providing advance pollution alerts to healthcare services and patients, and combining the model with edge-AI data for real-time decisions such as during wildfires; the team also wants to improve it to street-scale resolution using more open data. The researchers plan to release open-source training data and workflows so other countries and cities can build local models, while Topping envisions in roughly five years an agentic interface that could answer neighborhood-level questions about tomorrow’s pollution by chaining the models together.
nvidia