Developing a Semantic Model of Glacial Landforms for Object- Based Terrain Classification – the Example of Glacial Cirques

Earth surface processes operate on various spatio-temporal scales and produce landforms that are structured in a nested hierarchical manner. Semi-automated extraction of selected landform types from digital elevation models (DEMs) is of great importance in geomorphology. Object-based image analysis (OBIA) provides an attractive framework for multi-scale terrain segmentation and classification. However, classification in OBIA requires a priori knowledge and conceptualization approaches about candidate landforms as well as thorough modelling of their semantics. In our study we developed a semantic model for object-based classification of glacial cirques. The real-world feature ‘glacial cirque’ was conceptualized based on pre-existing knowledge about morphology, morphometry, and spatial context. From a 10 m DEM we derived curvatures. Filtered mean curvature served as input for the multi-resolution segmentation. Appropriate scale levels were identified by the method of local variance. In a first classification, ridges as features bordering cirques were extracted and masked out from mean curvature for a second segmentation process. Through semantic modelling we were able to link the concept of cirque and the image object domain to derive proper classification rules. The proposed model formalizes knowledge about glacial cirques and guides the user in the operational implementation of cirque classification in OBIA software. Since the classification process solely integrates shape and context information, the system becomes more objective and transferable. Extracted cirque objects contribute to a better understanding of cirque distribution and glacial processes. In future we will further improve our system to break new ground in automated mapping of glacial landforms.