THE SPECTRAL LAND FORM SIGNATURES FOR AUTOMATIC TERRAIN CLASSIFICATION (ON THE EXAMPLE OF SOUTH AMERICA)

Abstract

The way for computing some spectral landform’s characteristics (SLCs) is described on the example of the territory of the South America. SLCs can be potentially used for terrain classification with regard to the topographic dissection character, and then for a terrain thematic mapping. Five digital models in the small scale are designed: maximum of the wave’s magnitudes, the importance of the given share of waves, the general direction of height fluctuations, the severity of this direction, and the general wavelength. The distributions of some characteristics are largely correlated with the canonical geomorphometric variables; however, this relation is far from functional, other SLCs being completely independent. The Kohonen neural network dividing the South America territory into 225 separate neurons is constructed. The neurons with the hierarchical clustering are combined into 4 more general groups. Each group is defined by a typical combination of five SLCs. The scheme of the South America terrain clusters which reflect the types of topographic dissection is constructed.