Spatial analysis of hind cover patterns and corresponding remotely-sensed image brightness

Abstract

Abstract Existing land use and cover maps and the aerial photographs they were derived from were studied for a number of different landscape settings using spatial analysis techniques. The purpose of this study was to (1) determine if spatial variations in land use and cover could be quantified and characterized, (2) investigate the spatial scale relationships between land use and cover parcels and their image brightness composition on remotely-sensed images and (3) assess the relevance of these findings to the selection of appropriate remote sensor spatial resolution for image classification of land use and cover types. Analysis of land use and cover parcel size data indicated that the mode and the ‘85 per cent cumulative frequency’ statistics were useful measures for characterizing and comparing surface cover variability for different landscape types. Results of Kolmogorov-Smirnoff tests showed that differences in parcel sizes between urban and rural landscapes could be quantified and that the size of urban type land use classes were more variable within rural scenes than for urban landscapes. Spatial frequency analyses of digital air photo data indicated that surface cover brightness variations could be characterized for different landscapes. Two prominent scales of maximum variance were observed in the spectral analysis results, ranging from 350 m to 750 m and from 50 m to 125 m. The larger scale range was often similar to the mode of the corresponding land use and cover parcel size distribution. The smaller range appeared to relate to the scale of maximum internal brightness variation within land use and cover parcels. The most useful measure for characterizing image brightness variability was the 90 per cent cumulative power frequency (CPF) of the Fourier power spectrum. Block size variance analysis was found to be less sensitive and reliable for assessing spatial frequency patterns of scene brightness.