Inferring urban morphology for the greater Dublin area from continuous sealed surface data: a metric based approach

Urban land-use change models are valuable tools for assessing the spatial impact of alternative planning strategies via scenario analysis. A trustworthy output can only be obtained if the models are correctly calibrated and validated, which requires historic land-use data. Land-use information is often inconsistent or unavailable at the desirable time interval. Earth observation satellites supply regular information on urban development and can provide additional data to improve calibration. To study urban growth patterns with a time-span that exceeds the availability of high resolution imagery, a mapping approach is required that succeeds in capturing structural information from medium resolution images. A rather novel approach in this research area is to describe urban structure by means of spatial metrics, which are usually applied on categorical data such as land-cover maps. The main objective of our research was to quantify the spatial structure of the Greater Dublin Area by applying metrics on continuous sealed surface data. Three types of urban metrics were derived at the level of spatial units (blocks): average building density, distribution characteristics of sub-pixel sealed surface fractions and spatial variability. Using these spatial metrics as quantitative variables describing urban structure, a supervised classification based on a multi-layer perceptron classifier was used to infer basic land-use categories. With this approach, we were able to distinguish broad land-use types such as residential areas and employment zones (e.g. industrial, commercial). This distinction is relevant for urban modelling as these classes represent important driving factors. Results of applying the proposed methodology on a timeseries of Landsat images are currently being used for improving the calibration of the MOLAND land-use model for Dublin.