Scholarly record

USING OF THE SPECTRAL GEOMORPHOMETRIC CHARACTERISTICS FOR AUTOMATIZED CLASSIFICATION OF LANDFORMS (ON THE EXAMPLE OF AUSTRALIA)

Sergey Kharchenko

Abstract

The algorithm for computation of the spectral characteristics of terrain was prepared. The periodicity of Earth surface is specific В«fingerprintВ» of the cooperation of geological structure, exogenic processes and history of nature development for geomorphologically detached areas. For computing we used DEMs and 2D discrete Fourier transform with moving quadratic window of different sizes. Matrix of elevations decomposes in the Fourier image which defines harmonic waves set. All set of waves determine height field with absolute accuracy. However fixed part of waves allows reconstructing DEM with the different accuracy according to specific topographical situation. We can use this effect for landform classification. The follow characteristics was calculated: maximum of the waves magnitudes, the importance of given share of waves, the general direction of heights fluctuation, the severity of this direction, frequencies of the biggest wave by the orthogonal directions (W-E and S-N) and the general wavelength. The parameters of frequencies depends of window size (they are expressed in units/parts of window: 1/1, 1/2, 1/3 of window size and others). So we use only five parameters without frequencies for the classification. The used method of the landforms delineation on the Australia territory is Kohonen selforganizing maps (neural network). On the quadratic template net (15*15 neurons) we projected five-dimensional data and visualize the results as map of Sammon. On the finish we created the map of landforms types in Australia in the context of terrain periodicity.

Publication Impact Profile

• Citations
• CrossRef - Citation Indexes: 1
• Scopus - Citation Indexes: 1

• Captures
• Mendeley - Readers: 4

PLUMX see details

DOI resolver: doi.org/10.5593/sgem2018/2.3/s11.091

Publication details

Back to publication list