Over the past twenty years, a novel air pollutant -Ultrafine Particles (UFP), has attracted attention from scientific researches. UFP are particulate matters with a diameter of less than 100 nm. Currently, there is inadequate evidence to make a conclusion of the independent health impacts of UFP due to a lack of exposure data as UFP are not regulated or routinely monitored. Land Use Regression (LUR) models have been increasingly used to predict intra-city variations of UFP concentrations.
This thesis aims to build LUR models for UFP in urban areas in London and to evaluate the performance of different combinations of models and test their transferability within the city and to the city of Norwich in the UK. Three areas were selected at the neighbourhood level in London. A purposely designed short-term monitoring campaign was conducted to collect concentration data of UFP in these areas from 2016 to 2018. A series of models were developed for each individual and combined areas by using 100% of the site or 90% of the sites following a traditional approach (unrestricted approach). The issue of overfitting was addressed by using a limited number of variables (parsimonious approach). All developed models were transferred between the areas within London and to the city of Norwich to test transferability of models.
Results from this study suggest that the concentration of UFP in the West area of London was strongly affected by Heathrow Airport. In general, transferring the models from one area to another area or another city resulted in poor model performance, and transferring between areas within London performed better than transferring models to Norwich. Within the city of London, the models with a limited number of variables when transferred to other areas may perform better than the models where the number of variables was unrestricted. Calibration had more effects on external validation R2 when transferring to another city than transferring between areas within London, particularly when the number of variables was unrestricted.
To conclude, model transferability must be undertaken with caution. An improved strategy for future research on the development of UFP LUR models in megacities may be to develop models using a dense network of sites in selected neighbourhoods.