Introduction
Air pollution is one of the leading environmental risks to human health, yet understanding how pollutants vary across cities remains a major scientific challenge.
Air quality measurements are typically collected by sparse networks of sensors that differ in accuracy and reliability, making it difficult to estimate pollutant concentrations and personal exposure at locations where no measurements exist. New data-driven approaches that combine physical models with machine learning offer exciting opportunities to extract more information from these incomplete datasets.
This project will investigate novel methods for analysing urban air quality data collected across Hampshire. The work will make use of measurements from a growing network of air quality sensors operated by the University of Southampton and local authorities, including Southampton and Winchester City Councils, New Forest District Council and Hampshire County Council. This project is linked with the Centre of Excellence for Low Carbon Comfort and aligns with the University’s strengths in Sustainable AI and the University Civic Agreement to tackle social challenges in the region.
The project will address questions such as:
How do meteorological conditions, traffic, and proximity to emission sources influence measured pollutant concentrations?
How can high-fidelity computational fluid dynamics simulations or laboratory experiments be used to improve data-driven models of pollutant transport?
How can physics-informed machine learning and reduced-order modelling be combined to predict pollutant dispersion from sparse, multi-fidelity measurements?
What are the fundamental limitations of sparse monitoring networks, and how can we estimate air quality and personal exposure in locations where no measurements are available?
Methodology
Depending on the student's interests, the project may involve computational fluid dynamics, machine learning, reduced-order modelling, statistical analysis and physics-informed artificial intelligence. The successful candidate will gain experience working with real-world environmental datasets while developing advanced modelling techniques with applications to urban planning, public health and environmental policy.
If you enjoy working with complex datasets and are also motivated by research that has the potential to improve our environment, this could be a great project for you!
References:
Rawden, J. I., Vanderwel, C., & Symon, S. (2026). Physics-informed neural networks for passive scalar emission and transport. Physical Review Fluids, 11(2), 024501.
Vanderwel, C., Simmonds, O., & Peng, Z. X. (2024). Identifying regional air quality trends from sensor network data: An analysis of PM2.5 measurements in Hampshire. Public Policy Brief, University of Southampton, October 2024, DOI: 10.5258/SOTON/PP0068
Clements, D., Coburn, M., Cox, S. J., Bulot, F. M., Xie, Z. T., & Vanderwel, C. (2024). Comparing large-eddy simulation and Gaussian plume model to sensor measurements of an urban smoke plume. Atmosphere, 15(9), 1089.