Morphometric classification of the biosphere reserve “Eastern Carpathians” (Central Europe) with self-organizing map

Abstract

In this paper, we propose a semiautomatic method for landscape analysis of biosphere reserve Eastern Carpathians with both spectral and morphometric constituents. The Shuttle Radar Topography Mission (SRTM) has provided digital elevation models for approximately 80 % of the earth’s land surface. SRTM data are used to calculate first derivatives (slope) and second derivatives of elevation (such as minimum curvature, maximum curvatures, and cross-sectional curvature) by fitting a bivariate quadratic surface with a window size of 9 by 9. Together with multispectral remote sensing data like Landsat 7 ETM+ with 28.5 m raster elements, these data provide comprehensive information for the analysis of the landscape in the study area. Unsupervised neural network algorithm—self-organizing map—divided all input vectors into inclusive and exhaustive classes on the basis of similarity between attribute vectors. An optimal self-organizing map with 21 classes using 1,000 iterations and a final neighborhood radius of 0.05 provided a low average quantization error of 0.3394 and was used for further analysis. Morphometric analysis, spectral signature analysis, and feature space analysis are used to assign semantic meaning to the classes as landscape elements according to form, cover, and slope, e.g., deciduous forest on ridge (convex landform) with steep slopes. The results revealed the efficiency of self-organizing map to integrate SRTM and Landsat data for landscape classification. This makes it possible to develop an alternative method for fast assessment and comparison of landscapes over large areas. This procedure is reproducible for the same applications with consistent results.