Simulating the scale dependence of urban land cover inference for the region-based derivation of fine scale urban land use information

Images acquired by very high spatial resolution multispectral satellite sensors are one means by which urban land use information can be derived. It has been argued that this may be achieved using a twostage region-based approach, where initially derived land cover parcels (regions) are subsequently analysed, in terms of their spatial and morphological structure, to infer land use. Although initial results of the above approach have been promising, little work to date has been conducted into quantifying its scale dependence. This is particularly the case in relation to how inferred region morphology and spatial composition changes as a function of image spatial resolution. In order to address this issue this paper employs fine scale land cover digital map data to generate simulated classified images at resolutions of between 2–6m. Subsequent regionbased structural analysis shows that inferred region morphology, particularly for man-made objects (buildings and roads), is severely degraded at spatial resolutions below 2m, making the inference of certain residential land use types problematic. The implication of these findings are that for certain spatially fragmented residential types, characterised by a relatively large number of small detached houses, region-based structural inference of urban land use using multispectral images acquired by sensors such as IKONOS-1 (4m) is likely to be problematic.