Pattern Recognition in the Tasks of Landform Mapping

Abstract

The article aims to show the modern state of pattern recognition techniques for automatic and semi-automatic geomorphological mapping. There is opinion among the geomorphometrists about the expert rules for traditional landform mapping can be quantitated. The general unsolved tasks of automatic landform mapping are: recognition of origin for morphologically similar Earth’s surface forms; criteria development for transfer from morphological to genetic and age landform’s characteristics; preventive choosing the optimal resolution of the remote sensing data; the choosing and rationale of predictor’s weights in statistical modeling procedures. Some cases of the pattern recognition techniques using in geomorphology and landform mapping are given: generalized linear models; classification trees; random forest; artificial neural networks; and computer vision methods. The overall accuracy of the different models according to planar continuous landform recognition (and recognition of lithology types too) is about 50–70% and more. At the same time, specific landform type’s (craters, volcanic cones and others) recognition can reach 90–100%.