Land-surface segmentation to delineate elementary forms from Digital Elevation Models

TitleLand-surface segmentation to delineate elementary forms from Digital Elevation Models
Publication TypeConference Paper
Year of Publication2013
AuthorsDrăguţ, Lucian, Ovidiu Csillik, Jozef Minár, and Ian Evans
Conference NameGeomorphometry 2013
Date Published2013
Conference LocationNanjing, China
Abstract

In this paper, we introduce an algorithm to delineate elementary forms on Digital Elevation Models (DEMs). Elementary forms are defined by constant values of fundamental morphometric properties and limited by discontinuities of these properties. A multiresolution segmentation technique was customized to partition the layers of altitude derivatives into homogeneous divisions through a self-scalable procedure, which reveals the pattern encoded within the data. Layers were segmented successively, following the order of elevation derivatives, i.e. gradient, aspect, profile curvature, and plan curvature. Each segmentation was followed by extraction of elementary forms, thus leaving only heterogeneous surfaces for further segmentation steps. The sequential selection of elementary forms based on dynamic thresholds of: 1) the inner variance of the respective land-surface variable (LSV); 2) the difference between the mean LSV value of the target segment and the mean LSV values of its neighbor segments; and 3) the shape indices of segments. The results were compared with an existing manually-drawn geomorphological map to evaluate the potential of the algorithm of producing morphologically meaningful land-surface divisions. The evaluation showed that the most segments are either directly comparable with manual delineations, or have a clear morphological meaning. We conclude that algorithmic delineation of elementary forms from real DEMs is feasible; more work is needed, however, to design a fully operational process.

AttachmentSizeHitsLast download
Dragut2013geomorphometry1.pdf945.37 KB17045 days 16 hours ago
4.166665
Your rating: None Average: 4.2 (6 votes)