Abstract
Today, more people live in urban areas than do not, and this number is only predicted
to increase in the future; therefore, urban areas must be monitored carefully to avoid negative
impacts on the environment and inhabitants. Satellite-derived land cover maps are frequently
the source of information for researchers; however, they are prone to errors, especially when
quantifying land cover change. Therefore, there is a requirement for high-accuracy land cover
classification and change detection methods, where errors are understood and managed.
This thesis aims to classify and understand urban growth while understanding and
managing misclassification errors. A new change detection algorithm that uses all available
Landsat images to detect and classify land cover change was developed that dates and classifies
land cover change. At all stages, to be tailored to the detection of urban growth specifically.
Rigorous accuracy assessment using randomly selected pixels was performed and integrated
into the analysis at all stages. The change detection achieved an overall accuracy of 91.2%,
outperforming a high-quality post-classification comparison (72.6%).
The Landscape Expansion Index characterises the type of urban growth; however, this
metric is currently only usable with a single class of urban density. However, a single class of
urban density is a vast oversimplification of the heterogeneity found in urban areas. Secondly,
this metric has been untested for its sensitivity to misclassification errors. Therefore, this thesis
extended the Landscape Expansion Index to apply to two urban density classes. The new
metric characterises the location and type of growth. This new metric was applied to three
cities in the Midlands to demonstrate its potential and to characterise urban growth in the
Midlands. It found that Leicester (which lacks a greenbelt) has grown more diffusely than both
Nottingham and Birmingham (which are constrained by them). Finally, a new systematic
methodology of testing the sensitivity of landscape metrics to misclassification errors is
proposed and used to test the sensitivity of the newly developed metric. This provides a deeper
and more rigorous understanding of sensitivity than has been applied to any metric to date.
This analysis found the metric generally insensitivity to errors; however, sensitivity varied
between the specific misclassification type.