1 Scale Analysis for Remotely Sensed Images and Areal Census Data: Stressing Objects

Abstract – Spatial scale analysis for disparate geospatial data is facilitated by object-based reasoning and analysis. This paper introduces a wavelet transform-based approach for remotely sensed images and a variogram approach for areal census data; both stress objects as the core for performing scale analysis. By calculating a series of statistics from wavelet transform-based sub-images and energy signature images, it was found that with energy signature images, the approximate characteristic scale of scene variation can be suggested by the change rate of the standard deviation over two successive decomposition resolutions (∆SD/∆R). Four simulated fine spatial resolution images having different object specifications were analysed and in each case the method identified the approximate characteristic size of the salient objects in the image. An accurate result can be obtained more easily using the range of geostatistical variograms. For areal census data, we correlate census dwelling data and image structural information. By calculating variograms at individual areal units rather than at a whole scene level, high correlations exist between variogram sills and dwelling densities of individual areal units. Thus, dwelling attributes across arbitrary areal units can be interpolated and scaled. This paper concludes that such scale analyses in a modern, digital spatial analysis environment should still conform to the classical principle of cartographic scale transform; that is, scale association with object representation. Using geostatistics for examining the cognate relations between scale and objects holds promise