Delineation and identification of neighborhoods from high spatial resolution imagery

Urban neighborhoods exhibit distinctive spatial expressions in terms of their architectural, structural, and morphological composition. By employing spatial metrics to quantify these attributes it is possible to demonstrate how individual urban neighborhoods may be distinguished and delineated from second order imagery. On-going research is exploring an agenda for building disaggregated urban models that infer spatial urban structural configurations within spectral limitations. The disaggregated models are based on point-based GIS data, from both the United Kingdom (postal records) and the United States (parcel records). Knowing the spatial distribution of these point data introduces a number of key indicators that measure parameters such as density (compactness versus sparseness) and arrangement (linearity versus randomness). These are measured using spatial metrics, adjusted by the contagion index, a measure of fragmentation, and fractal dimensions, used to measure the degree of space filling and the level of irregularity within neighborhoods.

Within the urban environment there are a number of different neighborhoods that are distinguishable in architectural, structural, economical, and spatial terms. The complex assemblage of different land covers (bare soil, concrete, tarmac, grass, water etc.) within these neighborhoods give rise to unique spatial expressions that can be quantified through the use of spatial metrics – measures originating from landscape ecology to describe the structure and pattern within landscapes. This study demonstrates how quantifying the spatial arrangement (pattern) of buildings, through the use of spatial metrics, provides a platform from which second-order urban land use may be inferred from classified high spatial resolution IKONOS imagery. Commercial neighborhoods exhibit different levels of complexity and irregularity to residential neighborhoods, so too does high density residential from low density residential. Further still residential housing ‘eras’ are made distinguishable as different architectural and structural styles are reflected within their morphology. This can be seen within even the most elementary metrics such as area, density, and percent land cover. More stringent metrics used include the fractal dimension (D), the contagion index, and Lacunarity. Fractal geometry is well suited to measuring the morphology of urban neighborhoods where increasing irregularity is reflected in dimension. A useful compliment to the D is the contagion index, which measures the degree of fragmentation within the neighborhood.

By establishing relationships between image pixels and building spatial distributions, the long-term research goal is to facilitate reliable and accurate spatial pattern recognition and object-based multispectral classification methodologies to a level that renders resulting output irresistible to planners and policy makers. Encouraging results are documented from preliminary empirical testing on IKONOS imagery using aerial photography at 15cm spatial resolution. This presentation will give examples of an iterative computational procedure that links the spatial delineation of buildings from high resolution sensor data with the functional characteristics from postal records. Also, using the software e-Cognition, a spectra-spatial classification based on nearest neighbor contextual rules produced accuracies of 95.4% compared to 90.7% from a multispectral-only classification. Further, more extensive testing is continuing that allows the temporal dimension to calibrate and validate the spatial-spectral links, as well as provide measures of urban change.