University of Manchester Advances Air Pollution Forecasting with NVIDIA

Air pollution is a major public health issue, with an estimated 30,000 deaths in the UK alone last year. Traditional chemical models for air quality measurement are costly, limiting the frequency and detail of analyses. David Topping, professor at the University of Manchester, discovered that NVIDIA Earth-2 AI models and tools had provided solutions for weather forecasting and wondered if these generative frameworks could also be applicable for predicting air pollution.

In collaboration with the NVIDIA Earth-2 team, Topping generated training data based on existing chemical climate simulations and trained the Earth-2 CorrDiff, a generative downscaling model, on Isambard-AI, the UK’s national AI supercomputer in Bristol. This initial attempt was successful, significantly accelerating and improving the development of the air pollution model. The addition of Earth-2 StormCast, which enables time-dependent predictions, further refined the model.

The UK-wide pollution models allow researchers to model future scenarios and simulate the effects of policy changes. An interesting application of this model is the ability for healthcare institutions to proactively respond to air pollution levels, enabling, for example, timely notifications to asthma patients about increased pollution in their region. The team is also exploring the integration of real-time data from edge AI devices to facilitate quicker decision-making, especially in emergencies such as wildfires.

Thanks to the efficient execution of the model on powerful NVIDIA hardware, it was possible to create a detailed air pollution model for the UK with limited computing power. The team plans to release open-source training data and workflows so that similar models can be developed for other countries and regions. The ultimate goal is a user-friendly interface where, for example, health authorities can inquire about air quality predictions in specific neighborhoods.

Read the full article from Nvidia.