Aspects of spatial autocorrelation in Landsat TM data to the inventory of waste-disposal sites in rural environment

This study concerns the attempt to identify and classify waste-disposal sites by means of Landsat TM data, mainly on the basis of the contrast of their spatial features with the surrounding rural landscape, rather than on their purely spectral responses. In fact, any waste-disposal site may be regarded as a land-cover unit characterized by high spatial variation of its pixel cluster, as compared with the neighboring pixels of the background, which, on the contrary, show an almost uniform pattern. The spatial autocorrelation was analyzed by means of the semivariogram function, which showed different shapes of the land-cover units selected The potentialities of spatial autocorrelation as a basis for image classification were confirmed. A synthetic descriptor, based on the ratio between sill and range, is proposed as a new tool for waste-disposal-site classification. The experiment was carried out over a portion of the Friuli plain, in northeastern Italy.