Terrain Segmentation and Classification using SRTM Data

Abstract

The main objective of this chapter is to segment and classify Shuttle Radar Topography Mission (SRTM) data into specific landforms. Based on the results of previous research (DRĂGUŢ and Blaschke 2006), a classification system of landform elements was improved and adapted for SRTM 3 arc second data. Terrain derivatives such as elevation, slope gradient, slope aspect, profile curvature, and plane curvatures were classified in a multi-resolution object-oriented approach comprising four different scale levels. We carried out object-based image analysis, using a software program called eCognition Professional 4.0, to segment terrain derivatives into relatively homogeneous objects, which were further classified using fuzzy logic rule sets. Special emphasis was put on the accuracy assessment of the results as well as on the transferability of the procedure between study areas. We classified two SRTM datasets comprising a rolling hill landscape, which covers small areas of the states of Arkansas, Missouri and Oklahoma, USA, and a high mountain area of 50 square km around Hochkalter Peak, Berchtesgaden National Park, Germany. Results were visually compared and accuracy assessments using fuzzy classification options and an error matrix were performed. The classification system proved to be transferable between hilly and high mountain areas, its outcomes being satisfactorily accurate