From Efficient Geomorphological Mapping to Semi-Automated Landslide Detection

Detailed geomorphological mapping is a fundamental tool for understanding landform evolution and landslide processes; however, conventional field-based approaches remain time-consuming, expert-dependent, and spatially limited, particularly in complex or densely forested terrain. Although high-resolution remote sensing data and semi-automated methods have improved mapping capabilities, challenges remain regarding scale dependency, sensitivity to parameter settings, and limited transferability across geomorphologically distinct regions. In addition, data quality and acquisition timing further influence the reliability of derived geomorphological products. These limitations highlight the need for robust, reproducible, and scalable workflows that integrate multi-source geospatial data while maintaining geomorphological interpretability. This cumulative dissertation, "From Efficient Geomorphological Mapping to Semi-Automated Landslide Detection," addresses these challenges by developing and evaluating an integrated framework for geomorphological mapping and landslide detection based on high-resolution Digital Terrain Model (DTM) data and Geographic Object Based Image Analysis (GEOBIA) approaches. The main objective is to improve the reliability and transferability of detailed geomorphological mapping and landslide detection approaches in complex, forested middle-mountain environments. The first part of the research demonstrates how the integration of LiDAR-derived terrain models, land surface parameters, orthophotos, and geological information significantly enhances the quality and consistency of detailed geomorphological mapping. A structured GIS-based workflow enables systematic feature extraction and the delineation of geomorphological units, providing a high-quality reference inventory for subsequent analysis and model development. Building on this foundation, the second part investigates semi-automated GEOBIA-based landslide detection with a focus on forest-covered old landslides. The results show that detection performance is strongly influenced by scale-related parameters, particularly moving-window sized used in terrain-related landform calculation. Optimised parameterisation improves the identification of landslide components and reduces classification errors compared to default settings, demonstrating the importance of scale-sensitive model design. The third part evaluates the transferability of the developed approach across geomorphologically contrasting regions by applying the model from the Jena region to the Swabian Alb escarpment. The results confirm that the framework is broadly transferable, although performance is affected by terrain differences, land cover conditions, and variations in data quality and acquisition timing. Higher-resolution datasets significantly improve detection reliability, particularly in forested environments. Overall, this dissertation demonstrates that the integration of high-resolution DTM data, scale-optimised GEOBIA methodologies, and cross-regional validation enables robust, transferable, and scalable geomorphological mapping and landslide detection. The findings highlight the critical role of data quality, parameter optimisation, and multi-source integration for advancing semi-automated geomorphological analysis in complex landscapes.