On the separability of urban land-use categories in fine spatial scale land-cover data using structural pattern recognition

It has been widely asserted that the morphology of urban areas is a result of the interactions of urban function and urban form. This has led a number of studies to postulate, either explicitly or implicitly, that a mapping exists between the physical form (land cover) of the urban fabric and its corresponding function and activity (land use), although relatively little quantitative evidence has been presented to support this assertion. This paper presents the results of an investigation into the relationship between urban form and urban function using fine spatial scale digital map data. These are used to derive quantitative information on the morphological properties and spatial structure of the buildings present in a series of urban land-use categories identified in two urban areas (Cardiff and Orpington) in the United Kingdom. A statistical separability analysis of these land-use samples suggests that a mapping exists between urban form and function, which, if replicated for other urban areas, would allow urban land use to be inferred from an analysis of the spatial disposition of land-cover parcels, particularly buildings. DOI:10.1068/b3016 }Current address: School of Civil Engineering and Geosciences, University of Newcastle upon Tyne, Newcastle upon Tyne NE1 7RU, England; e-mail: [email protected]. }Current address: European Space Research and Technology Centre, Keplerlaan 1, Postbus 299, 2200 AG, Noordwijk, The Netherlands. it has been argued that this can best be achieved by means of a two-stage procedure: the first stage involving some form of mapping from spectral radiance to land cover at the level of individual pixels, typically based on standard, multispectral, image classification techniques; the second involving the identification and subsequent structural analysis of discrete land-cover parcels (that is, multiple pixel regions), in terms of their spatial and semantic relations, as well as their morphological properties (Barnsley and Barr, 1997; Barr and Barnsley, 1997; 1999). The potential of this procedure has been explored with some success in several preliminary studies (Barnsley and Barr, 2001; Barr and Barnsley, 2000; Herold et al, 2002). The notion of a relationship between form and function in urban areas, which is implicit in the second stage of the procedure outlined above, is hardly new. Indeed, it underpins much research in the field of computational urban morphology (Batty and Howes, 2001; Batty and Longley, 1994; Hillier and Hanson, 1984; Kru« ger, 1979a; 1979b; Longley and Harris, 1999; Longley and Mesev, 2000; Mesev et al, 1995). Nevertheless, some researchers remain skeptical, not only about its potential for practical application to determine urban land use from automated analyses of remotely sensed images, but also its conceptual basis (Bibby and Shepherd, 1999). In this paper we therefore examine the function-from-form hypothesis in some detail. More specifically, we explore the success with which different types of urban land use can be distinguished, solely on the basis of an analysis of the spatial disposition of their constituent land-cover parcels. In doing so, we employ a combination of structural and statistical pattern-recognition techniques (Fu, 1986; Schalkoff, 1992). These are embodied in the Structural Analysis and Mapping System (SAMS), which we have developed (Barr and Barnsley, 1997; 1999). To remove the potential effects of errors and uncertainty in the initial land-cover map, we make use of rasterized digital map data in this study, rather than a classified multispectral image. This allows us to focus exclusively on the mapping from land cover to land use, in the absence of the sorts of misclassification that inevitably result from, among other things, shadowing, occlusion, and mixed pixels present in remotely sensed images. Although digital map data are themselves a model of the real world, encompassing the processes of feature selection, abstraction, and thematic and spatial generalization, the results presented here can perhaps be thought of as presenting the `best case' scenario in terms of the ability to distinguish different categories of urban land use solely on the basis of an analysis of the structural disposition of land-use parcels. 2 Study areas Two study areas are examined in detail in this paper, both of which are drawn from the United Kingdom. The first covers part of the city of Cardiff,Wales; the second covers a section of the town of Orpington in the London Borough of Bromley. These areas have been selected largely for pragmatic reasons, associated with the availability of appropriate digital datasets. Nevertheless, each contains a diverse set of land-use categories typical of many other UK towns and cities. By using both sites, we can increase the number of independent samples of these land uses. The Cardiff study area includes a number of residential districts built during several of the major periods of house construction that have taken place in the United Kingdom, namely the Victorian 1890s/1900s (dominated by `back-to-back' terraced houses; hereafter referred to simply as 1890s), the 1950s, 1960s, and 1970s (each consisting of a mixture of semidetached houses and small, low-rise, blocks of town houses and flats), and the 1990s (dominated by tightly clustered estates of small, detached, and semidetached, houses). There is also a single industrial zone within this study area. The Orpington site, on the other hand, contains three distinct areas 398 S L Barr, M J Barnsley, A Steel