Every hour, ambient concentration data for dozens of air pollutants is collected
from hundreds of monitoring sites across the UK, adding to a repository consisting
of more than 370 million observations going back 47 years. And yet, due to the
difficulty of extracting meaningful information from this data, it is principally
used for monitoring compliance with air pollution limits. This thesis aims to
develop new statistical techniques and apply them to the air quality network data
to derive additional insights concerning changes in air quality in the UK over the
last twenty years.
The rolling change method is a new way of conducting robust long term trend
analysis across multiple sites within monitoring networks that are subject to biases
caused by site flux. It is used to analyse the long term trends in NOx and NO2
concentration, and in NO2/NOx ratio in London, Scotland, and the UK between
2000 and 2017. At each scale, the results are consistent, showing declines in NOx
and NO2 concentration, and a peak in NO2/NOx around 2010, followed by a fall.
The 'meteorological normalisation' method using random forest is applied
to remove the effect of meteorology from air pollutant concentrations at London
sites to enable clearer visualisation of the trend due to changes in emissions. The
method is also used to evaluate the efficacy of the London Low Emission Zone
through the generation of counterfactual scenarios that are compared to the true
normalised trend. The results suggest a mild improvement in air quality.
The influence of inter-annual meteorological variation on annual average
concentrations of NOx , NO2 and O3 is estimated for a large number of UK sites
using the novel tools of heatmaps and cumulative sum plots. This influence is
shown to be considerable: the range of the annual average concentration due to
meteorological variation is 2.9 μg m −3 (8.2%) for NO2 , 9.9 μg m −3 (12.6%) for NOx
and 3.3 μg m −3 (7.5%) for O3 . The implications of these findings for the use of the
annual average metric in compliance monitoring within the EU are considered.
