[en] In today’s urbanising world, effective urban management and planning strategies are needed to temper the impact of urban change processes on the natural and human environment. To develop and monitor such strategies, and to assess their spatial impact, analysing changes in urban structure is essential. Data from earth observation satellites provide regular information on urban development and, as such, may contribute to the mapping and monitoring of cities and the modelling of urban dynamics. Especially images of medium resolution (Landsat, SPOT, …), which are cheap, widely available and often part of extensive historic archives, offer a wealth of information that may be useful for urban monitoring purposes. The lower resolution of this type of imagery, however, hampers the study of urban morphology and change processes at a more detailed, intra-urban level. Spectral unmixing approaches, which allow characterising land-cover distribution at sub-pixel level, may partly compensate for this lack of spatial detail, and may render medium-resolution imagery more useful for urban studies. The main research question addressed in this paper is how medium-resolution imagery could be used to describe urban morphology, by combining spectral unmixing approaches with spatial metrics. Spatial metrics derived from satellite imagery may be useful to quantify structural characteristics of expanding cities, and may provide indications of functional land use. In this study, we develop a set of urban metrics for use on continuous sealed surface data produced by sub-pixel classification of Landsat ETM+ imagery. Two sub-pixel classification approaches are examined for that purpose. In a first approach, we use a linear spectral mixture model with a vegetation and a non-vegetation endmember to deconvolve each pixel’s spectrum into fractional abundances of the two end member spectra, which are determined by visualising mixture space with principal component analysis. In a second approach, we use a linear regression model to estimate the proportion of vegetation cover within each Landsat pixel. In both approaches, an urban mask is used to indicate pixels belonging to urban land cover. Only pixels within the urban mask are subjected to sub-pixel classification. We hereby assume that the urban area does not contain bare soil and that the area of a pixel not covered by vegetation fully consists of sealed surface cover. The resulting sealed surface proportion map is then used to characterise urban morphology and land use by means of the shape of the cumulative frequency distribution of the estimated sealed surface fractions within a building block. A transformed logistic function is fitted to this distribution with a least-squares approach to obtain function parameters that are used as variables in a supervised classification approach, together with spatially explicit metrics (spatial variance and Moran’s I). Our study demonstrates that images from medium resolution sensors can be used to characterise intra-urban morphology, and that the structure of a building block as described by the proposed metrics gives an indication of its membership to certain morphological/functional urban classes. In future research we will incorporate socio-economic data in the metric analysis to further improve the distinction of urban land-use categories. The spatial metrics approach developed in this study will be used in experiments to improve the calibration of the MOLAND urban growth model, which is currently calibrated with historical land-use maps available for approximately 10-year intervals.
Politique Scientifique Fédérale (Belgique) = Belgian Federal Science Policy