Quantifying slope movements

• Verfasst von: Tabernig, Ronald [VerfasserIn]
• Verfasst von: Anders, Katharina [VerfasserIn]
• Verfasst von: Fritzmann, Patrick [VerfasserIn]
• Verfasst von: Weiser, Hannah [VerfasserIn]
• Verfasst von: Zahs, Vivien [VerfasserIn]
• Verfasst von: Höfle, Bernhard [VerfasserIn]
• Verfasst von: Rutzinger, Martin [VerfasserIn]
• Titel: Quantifying slope movements
• Titelzusatz: tree trunk tracking using long-range stationary 4D laser scanning point clouds
• Verf.angabe: Ronald Tabernig, Katharina Anders, Patrick Fritzmann, Hannah Weiser, Vivien Zahs, Bernhard Höfle & Martin Rutzinger
• E-Jahr: 2023
• Jahr: November 2023
• Umfang: 2 S.
• Illustrationen: Illustrationen
• Fussnoten: Gesehen am 11.04.2024
• Titel Quelle: Enthalten in: Virtual Geoscience Conference (5. : 2023 : Dresden)VGC 2023 - Unveiling the dynamic Earth with digital methods
• Ort Quelle: Freiberg : Helmholtz-Institut Freiberg für Ressourcentechnologie, 2023
• Jahr Quelle: 2023
• Band/Heft Quelle: (2023) vom: Nov., Seite 114-115
• Datenträger: Online-Ressource
• Sprache: eng
• K10plus-PPN: 1885853831

Abstract

Large-scale landslides and secondary processes threaten human lives and infrastructure. Detecting, monitoring and understanding Alpine mass movements lay the foundation for the establishment of early warning systems and construction of protective structures. Challenges of point cloud based change quantification (e.g., with M3C2) arise when point cloud density is generally low for the single epochs due to long measurement ranges. Additionally, in such monitoring setups with dense forest cover the slope surface is only sparsely represented by points. Thus, our novel approach uses temporal aggregation of point clouds to increase point density and uses 3D displacements of tree trunks as proxy for surface slope movements. Specifically, we are applying this technique to trees located above the Leckgalerie (Obergurgl, Austria), a stretch of an important connecting road. Terrestrial laser scans were acquired every three hours from June to October in 2021 using a long-range stationary setup for the Riegl VZ-2000i. The segmentation and matching of tree trunks through all available point clouds allow to derive and compare displacement vectors of corresponding tree trunk objects throughout the scanning period. Three-dimensional distance measurements of three different methods are investigated: center of gravity (COG), iterative closest point (ICP), and trunk-ground-intersection (Intersection). The results of all three methods are compared to manual point-to-point measurements in the raw data to assess the accuracy of the developed method. Additionally, a simulation-based error margin is established based on more than 5000 simulated trunk displacements under varying conditions. These trunks are derived from laser scanning simulations on artificially generated scenes with the open-source scientific software HELIOS and Blender. Based on the results the area of interest can be classified into an unstable and a stable area. Vertical displacements are prone to disproportional error margins and thus the displacements are inspected in 2D. During the active period of 56 days, the unstable area shows average displacements of 1.8 m (COG), 1.81 m (ICP), and 1.6 m (Intersection), with a weekly mean of 0.22 m, 0.23 m, and 0.2 m respectively. The simulation-based error margins are derived by the Mean Absolute Percentage Error (MAPE). These errors are 23.7 % (COG), 28.5% (ICP), and 39.2 % (Intersection). The calculated MAPE, when comparing the methods results to the manual point-to-point measurements from the first to the last epoch, are found to be consistent for both the COG and ICP methods at 9.5 %, and 3.4 % for the Intersection method. The developed approaches deliver reasonable results demonstrating the potential of the utilisation of tree trunks as common and reoccurring objects for change detection. This provides additional information for forested areas on landslides that would otherwise be difficult to obtain.