EGU2020-21880, updated on 12 Jun 2020
EGU General Assembly 2020
© Author(s) 2020. This work is distributed under
the Creative Commons Attribution 4.0 License.

Landslide and Rockfall failures Characterization with Object-Based 3D Analysis

Efstratios Karantanellis1, Vassilios Marinos1, and Emmanouel Vassilakis2
Efstratios Karantanellis et al.
  • 1Aristotle University of Thessaloniki, Geology, Engineering Geology and Hydrogeology, Thessaloniki, Greece (
  • 2School of Geology, National and Kapodistrian University of Athens, Athens, Greece

Geological failures from massive rockfall failures to small landslides of few cubic meters are a major geological hazard in many parts of the world. Based on the latest developments, close-range photogrammetry and individually UAV photogrammetry and Light Detection and Ranging systems have become indispensable tools for geo-experts in order to provide ultra high-resolution 3D models of the failure site. TLS suffers from the fact that is sometimes tricky to capture the holistic area of interest from the ground, while some areas may often be obscured by vegetation or negative inclinations. The science of photogrammetry has long been used to accurately detect and characterize landslide and rockfall failures. Due to the continuously increasing spatial resolution capability of new generation sensors, traditional pixel-based approaches are not capable to cope with the level of detail resulted from those sensors. Mostly, landslides present complex and dynamic geomorphological features with great heterogeneity in their spatial, spectral and contextual properties dependent on the specific failure mechanism. In the current study, an object-based 3D approach for the automated detection of landslide and rockfall hazard is presented based on detailed topographic photogrammetric point clouds and 3D analysis. Recent trends show that close photogrammetry will play a vital role on the geological and engineering geological assessments concerning geo-failures. The results show that object-based approach is closer to human interception due to integration of contextual and semantic, spectral and spatial information rather than translating pixel’s spectral information solely. The current procedure provides a detailed objective quantification of landslide characteristics and automated semantic landslide modelling of the case site.

How to cite: Karantanellis, E., Marinos, V., and Vassilakis, E.: Landslide and Rockfall failures Characterization with Object-Based 3D Analysis , EGU General Assembly 2020, Online, 4–8 May 2020, EGU2020-21880,, 2020

Display materials

Display file

Comments on the display material

AC: Author Comment | CC: Community Comment | Report abuse

Display material version 2 – uploaded on 08 May 2020, no comments
Including some more information and figures
Display material version 1 – uploaded on 07 May 2020
  • CC1: Comment on EGU2020-21880, Rafaela Niemann, 08 May 2020

    Hello! I would like to know two things about your work: a) How object parameters are defined for your segmentation, and b) if you have already used a multispectral or hyperspectral camera to help with the segmentations.

    • AC1: Reply to CC1, Efstratios Karantanellis, 08 May 2020

      Dear Rafaela,

      Thank you very much for your interest in the current study.

      a) How object parameters are defined for your segmentation,

      If you are considering the parameters of scale and shape/comapctness, scale has been used in 3 different levels in order to identify multidimensional primitives but at the same time with interelationships among each other. Shape and comapctness have been kept as constant values after some trial and errors tests. In addition, Object parameters have been selected based on their relation with landslide features. For example, there are numerous features that they are able to work as precursor of disnstict features. i.e. Scarp zone or depletion zone. Unfortunetly, I can not provide specific values and thresholds due to the current study is under review in a journal but i would be happy to discuss more on that with you if you are interested. 

      b) if you have already used a multispectral or hyperspectral camera to help with the segmentations.

      No we did not use any other remote sensing dataset except the UAV derivatives. RGB values, Curvature, Slope, DSM and Aspect have been included as primary datasests for the segmentation process. The latter enable the proposed study to be cost efficient and effective for site specific landslide and rockfall detection cases.