Tree Geometry as a Potential Proxy for Landslide Activity in Densely Forested Slopes: A LiDAR-Based Digital Dendrogeomorphology Approach

Despite significant advancements in landslide monitoring, landslides occurring on densely forested slopes remain largely unexplored. While conventional subsurface characterization methods (e.g., DPH, CPT, percussion drilling) are often impractical due to limited accessibility and steep rugged terrain, surficial analyses using remote sensing techniques frequently face challenges in capturing high-resolution ground surface data due to occlusion caused by dense vegetation cover as well as technical limitations.
Although trees and forests are generally acknowledged to reduce the probability of landslide occurrence, they are unlikely to prevent or substantially mitigate deep-seated landslides or failures on very steep slopes. Instead, trees may serve as proxies of landslide activity, potentially improving the understanding and monitoring of densely forested slopes. Affected by slope movements, trees experience external growth disturbances and develop characteristic growth anomalies that can be partly attributed to underlying landslide processes.

Multiple studies have demonstrated the feasibility of extracting such external growth disturbances, primarily stem tilting, by assessing the inclination and curvature of tree stems in LiDAR point clouds, greatly building upon previous forestry-related studies exploring the mapping, classification, and derivation of stem parameters such as height and diameter from digital twins. However, the potential to extract externally visible eccentric growth patterns in stem cross-sections at heights of maximum bending, analogous to dendrogeomorphologic tree-ring analyses, as a proxy for landslide activity has not yet been explored. Additionally, the classification of overall tree shape may provide valuable insights into the characteristics of underlying slope movements, but, to the best of the author’s knowledge, this has not been addressed in previous research.

To investigate the potential of automatically extracting tree shape and stem eccentricity from LiDAR data, and to evaluate their suitability as proxies of landslide activity, we introduce an improved two-stage processing pipeline for tree identification and extraction, along with a dedicated framework for digital dendrogeomorphology. Building upon previous work, we compute normal vectors of locally fitted planes and projected point densities to separate trees from the point cloud. To enhance the extraction of complex shaped trees (e.g., S-shaped or pistol-butted) characteristic of landslide-prone slopes, we introduce dynamically adjusted normal vector thresholds derived from estimated stem inclination. After segmenting tree stems from the point cloud, ellipses are fitted at configurable height intervals to determine cross-section centroids. These centroids are then connected as vertices of a 3D polyline, which is subsequently smoothed using a natural spline to represent the generalized stem geometry. Based on the curvature of the resulting polyline, the height of maximum bending is identified, and the corresponding cross-section eccentricity is extracted. In addition, the curvature of the polyline is used to categorically classify overall tree shape.

Our digital dendrogeomorphology approach applied to 3D point clouds enables accurate extraction of stem eccentricity, even for complex tree shapes typical of landslide-prone slopes. When paired with automated tree-shape classification, these data offer insights into slope movement and improve understanding of landslide processes in densely forested environments.