Abstract. Landslide inventories are crucial for the assessment of landslide susceptibility and hazard. An analysis of old landslides can reveal periods of intensified landslide activity, but the features of these landslides may have diminished over time, particularly in the context of human impact. However, landslide features are often preserved well under forest cover and are thus valuable for compiling or updating landslide inventories. However, the mapping of these features remains challenging. Light detection and ranging (lidar) analysis and its derivatives are essential in landslide research, particularly in landslide identification and mapping. Unlike the expert-based analysis of lidar derivatives, the use of object-based approaches to map landslides from lidar data (semi)automatically requires further studies. This study adopts geographic-object-based image analysis based solely on lidar derivatives for the inventory mapping of forest-covered old landslides within a middle-mountain region in Jena, Germany, and surrounding areas. A manually prepared expert-based inventory map was used for model training and validation. Lidar derivative data were processed using (a) a default moving-window size (3 × 3; model I) and (b) an optimal window size (model II). Multi-resolution segmentation and support vector machine classification with distinct rule sets were implemented for each model, followed by refinement and accuracy assessment against the inventory map for model performance evaluation. The proposed approach achieved a 70 % detection of existing landslides compared with the inventory. Model II outperforms model I in accuracy, as indicated by its superior performance in scarp area detection (15 % improvement) and significantly lower false positives (30 % reduction). However, although this method excellently identifies and maps forest-covered old landslides, its applicability is currently limited to large and medium landslides (area > 0.5 ha). Overall, our findings suggest that landslides worldwide with clear geomorphological signatures in lidar data can be identified using this approach.
Drăguţ et al. (Mon,) studied this question.