Environmental mapping based on spatial variability.

Environmental maps show the probable environmental states of different types of land use or development of landscape in a geographic context. Remotely sensed data are particularly efficient for environmental mapping in order to outline major environmental types. Multiple schemes of image classification used in environmental mapping are either traditionally statistical or heuristic. While the former methods do not take account of spatial variability in space and aerial data, the latter ones does not lend themselves to optimal solutions we present. Novel probabilistic models of piecewise-homogeneous images are used in environmental mapping to segment real images. The models consider both an image and a land cover map. Such a pair constitutes an example of a Markov random field specified by a joint Gibbs probability distribution of images and maps. Parameters of the model are estimated by using a stochastic approximation technique. Its convergence to the desired values is studied experimentally. Addition of spatial attributes appears to be necessary in most areas where the differences in spatial data between regions in the image occur. Experiments in generating the pairs of images and environmental maps and in segmenting the simulated as well as real images are discussed.