Air quality and environmental dynamics of fine, ultrafine, submicron particles and reactive atmospheric pollutants with machine learning based predictive insights

Abstract

This PhD thesis investigates urban air pollution dynamics in the United Kingdom by integrating
observational, statistical, spatial, and machine learning approaches to examine pollutant
sources, temporal behaviour, prediction, and health relevance. Long term, high resolution
measurements of ultrafine particles (UFPs), formaldehyde (HCHO), nitrous acid (HONO),
and related atmospheric composition changes were analysed at the Brighton Atmospheric
Observatory. Seasonal and diurnal patterns showed that smaller particles were mainly influenced
by local traffic and industrial activity, while larger particles reflected regional and long
range transport. HCHO increased during summer due to photochemical activity, whereas
HONO was higher in colder periods, highlighting the roles of emissions, meteorology, and
atmospheric processing. HCHO was also strongly linked with temperature, solar radiation,
ozone formation, and regional transport.

Machine learning models were developed to predict nitrogen dioxide (NO2) concentrations
at Marylebone Road, London, using meteorological and co-pollutant datasets. Linear
Regression, Random Forest, LightGBM, and Stacking Ensemble models were evaluated,
with the ensemble model showing the strongest performance. Titanium dioxide (TiO2)
nanocatalyst scenarios indicated a potential 28% reduction in NO2 levels. VOCs were also
analysed to identify traffic, biogenic, and temperature driven influences on ozone formation.
Overall, the thesis supports adaptive urban air quality management.

Date of Award	May 2026
Original language	English
Awarding Institution	University of Brighton
Supervisor	Maureen Berg (Supervisor), Kevin Wyche (Supervisor) & Kirsty Smallbone (Supervisor)