Self Organizing Maps for Multi-Scale Morphometric Feature Identification Using Shuttle Radar Topography Mission (SRTM) Data
2009 (English)In: Geocarto International, ISSN 1010-6049, Vol. 24, no 5, 335-355 p.Article in journal (Refereed) Published
This article presents a new procedure using artificial neural networks in the form of a self-organizing map (SOM), as a semi-automatic method for analysis and identification of morphometric features in the Man and Biosphere Reserve 'Eastern Carpathians' with nine spatial scales. The NASA Shuttle Radar Topography Mission (SRTM) has provided digital elevation models (DEM) for over 80% of the land surface on earth. The latest version 3.0 SRTM data provided by the Consultative Group for International Agricultural Research-Consortium for Spatial Information GeoPortal is the result of substantial editing effort on the SRTM digital elevation data produced by NASA. Easy availability of SRTM 3 arc second data has resulted in great advances in morphometric studies and numerical description of terrain surface features as shown by many literature references. The first derivative, slope steepness and second derivatives, minimum curvature, maximum curvature and cross-sectional curvature of elevation were derived by fitting bivariate quadratic surfaces with nine window sizes ranging from 5 to 55 cells to the processed SRTM DEM 90 m Version 3.0. These analyses represent landform entities with ground distances from 450 to 4950 m, which are local to regional features. The four morphometric parameters were used as input for the SOM algorithm. Forty-two SOMs with different learning parameter sets, e.g. initial radius, final radius and number of iterations were investigated. An optimal SOM with 10 classes based on 1000 iteration and a final neighbourhood radius of 0.01 provide a low average quantization error of 0.1780 and was used for further analysis. The effect of the random initial weights for optimal SOM was also studied. The results in this particular study are not sensitive to the randomization of initial weight vectors if many iterations are used. Feature space analysis, morphometric signatures, three-dimensional inspection and auxiliary data facilitated the assignment of semantic meaning to the output classes in terms of morphometric features. Results are provided as thematic maps of landform entities based on form and slopes. The result showed that a SOM is an efficient scalable tool for analysing geomorphometric features as meaningful landforms over different spatial extents, and uses the full potential of morphometric characteristics. This procedure is reproducible for the same application with consistent results.
Place, publisher, year, edition, pages
2009. Vol. 24, no 5, 335-355 p.
Landform; Morphometric feature; Multi-scale; Neural network; Self organizing map
Engineering and Technology
IdentifiersURN: urn:nbn:se:kth:diva-8594DOI: 10.1080/10106040802642577ScopusID: 2-s2.0-70349567236OAI: oai:DiVA.org:kth-8594DiVA: diva2:13959
QC 20100727. Uppdaterad från submitted till published (20100727).2008-06-022008-06-022010-07-27Bibliographically approved