Scholarly record

AUTOMATIC RECOGNITION OF EXOGENIC LANDFORM TYPES ON THE ARCTIC TERRAIN USING SPECTRAL GEOMORPHOMETRIC VARIABLES (EXAMPLE OF THE EUROPEAN PART OF THE RUSSIA)

Sergey Kharchenko

Abstract

Possibility of using of the spectral geomorphometric variables for automatic landform types classification was estimated. The area of our interest - European part of the Russian sector of Arctic region. There are four main types of landforms with exogenic origin on the territory: 1) glacial and 2) glaciofluvial accumulative terrains, 3) ridges and hills terrain from glacial exaration, 4) wave terrain from glacial exaration. These landforms types cover from 55% to 80% in the different parts of the area of interest. A few of spectral geomorphometric variables were computed by sampled 2D discrete Fourier transform of DEM by 10*10 km moving window. We use follow spectral variables for landform type recognition: 1) magnitude of the main wave in the heigth field, 2) wavelength of the main wave, 3) importance (share of the heigth variation) of the fix pool (often 1%) of biggest harmonic waves, 4-5) general direction (and its significance) of the height oscillations, 6-7) coefficients of the exponential trend equation for approximation wave magnitudes distribution. We compare five techniques for classification: 1-2) linear (LDA) and quadratic (QDA) discriminant analysis, 3) support vector machine (SVM), 4-5) classification tree (CT) and random forest (RF). It has been get that the efficiencies of the techniques (on the initial training data) are sorted respectively: 0.99 (RF), 0.62(SVM), 0.58 (CT), 0.5 (LDA), 0.45 (QDA) . This values are given versus random guessing (around 0.25). That mean the RF (for example) explained 99% of multidimensional variance. For the control data (new 20 samples of each landform types) resulting accuracies are comparable.

Publication Impact Profile

• Citations
• Scopus - Citation Indexes: 2

• Captures
• Mendeley - Readers: 2

PLUMX see details

DOI resolver: doi.org/10.5593/sgem2019/2.1/s08.102

Publication details

Back to publication list