Manchester trains UK-wide air pollution model on NVIDIA Earth-2 in just two days
NVIDIA Blog · rss · 2026-09-16
Air pollution contributed to an estimated 30,000 UK deaths last year, but chemistry-based air quality models are computationally expensive. University of Manchester's David Topping partnered with the NVIDIA Earth-2 team to apply generative weather/climate frameworks to pollution fields.
- Training data was generated from existing chemistry-climate simulations; on Isambard-AI (5,448 GH200 chips, 21 exaflops), a single 8-GPU node trained the CorrDiff generative downscaling model in two days, succeeding on the first attempt and producing a 2-3 km resolution UK-wide pollution model.
- Earth-2 StormCast was added for time-dependent forecasts using real air quality observations, and the full workflow runs inference and small retraining on a desktop DGX Spark (GB10 chip).
- Applications include proactive alerts for asthma patients, what-if policy modeling, and real-time decisions with edge AI devices during wildfires.
- The team will open-source training data and workflows so any country or city can build local pollution models with modest supercomputer time; Topping envisions an agentic interface within five years where clinicians or agencies just ask a question.
More from Infra
- OpenRouter crosses 10 quadrillion annualized token run rate, doubling every 10.5 weeks — deedydas · 2026-11-02
- Palmer Luckey asks: can you use datacenter RAM for things prohibited by the operator's terms of service? — avaitopiper · 2026-11-02
- Disable MTP for 20-30% faster local LLM inference on limited RAM — InternalMode8159 · 2026-11-02
- Kioxia up 216% in a year but down 68% from its June 2026 peak — zephyr_z9 · 2026-11-02
- Intel plunges 29%, worst day since the 1970s, wiping $12B off US government stake — emmanuelvivier · 2026-11-02
- Researcher hoards nearly 1TB of DRAM, refuses to sell a stick to data centers — kevinnbass · 2026-11-02