EGU24-15896, updated on 09 Mar 2024
EGU General Assembly 2024
© Author(s) 2024. This work is distributed under
the Creative Commons Attribution 4.0 License.

Classification and segmentation of 3D point clouds to survey river dynamics and evolution 

Laure Guerit1, Philippe Steer1, Paul Leroy1, Dimitri Lague1, Dobromir Filipov2, Jiri Jakubinsky3, Ana Petrovic4, and Valentina Nikolova5
Laure Guerit et al.
  • 1CNRS - Université de Rennes, UMR 6118, Geosciences Rennes, Rennes, France (
  • 2University of Architecture, Civil Engineering and Geodesy, Bulgaria
  • 3Global Change Research Institute, the Czech Academy of Sciences, Czech Republic
  • 4Geographical Institute „Jovan Cvijić“ SASA, Serbia
  • 5University of Mining and Geology “St. Ivan Rilski” Sofia, Bulgaria

3D data for natural environments are now widely available via open data at large scales (e.g., OpenTopography) and can be easily acquired on the field by terrestrial LiDAR scan (TLS) or by structure-from-motion (SFM) from camera or drone imagery. The 3D description of landscapes gives access to an unprecedented level of details that can significantly change the way we look at, understand, and study natural systems. Point clouds with millimetric resolution even allow to go further and to investigate the properties of riverbed sediments: dedicated algorithms are now able to extract the sediment size distribution or their spatial orientation directly from the point cloud. 

Such data can be real game changers to study for example torrential streams prone to flash floods or debris flows. Such events are usually associated with heavy rainfall events, while conditioned by the geomorphological state of a stream (e.g., channel geometry, vegetation cover). The size and the shape of the grains available in the river also strongly influence river erosion and sediment transport during a flood. 3D data can thus help to design prevention and mitigation measures in streams prone to torrential events. 

However, it is not straightforward to go from data acquisition to river erosion or to grain-size distributions. Indeed, isolating and classifying the areas of interest can be complex and time-consuming. This can be done manually, at the cost of time and absence of reproducibility. We rather take advantage of state-of-the-art classification method (3DMASC) to develop a general classifier for point clouds in fluvial environments designed to identify five classes usually found in such settings: coarse sediments, sand, bedrock, vegetation and human-made structures. We also improved the G3Point sediment segmentation algorithm, developed by our team, to make it more efficient and straightforward to use in the CloudCompare software, which is dedicated to point cloud visualization and analysis. We apply it to the coarse sediments class identified by 3DMASC to provide a more accurate description of grain size and orientation. We also make a profit of the sand class to estimate its relative areal distribution that can then be compared to the coarse sediment class. This provides valuable information about the type of flows which are also important for planning torrential events mitigation measures.

We illustrate this combined approach with two field examples. The first one is based on SFM data acquired along streams prone to torrential events in Bulgaria and in Serbia where we documented sediment size and orientation. The second one is based on TLS data acquired along a bedrock river in France that experienced a major flood which induced dramatic changes in the river morphology. 

This work has been partially funded by PHC Danube n° 49921ZG/ n° KP-06-Danube/5, 14.08.2023 (National Science Fund, Bulgaria) and the H2020 European Research Council (grant no. 803721). 

How to cite: Guerit, L., Steer, P., Leroy, P., Lague, D., Filipov, D., Jakubinsky, J., Petrovic, A., and Nikolova, V.: Classification and segmentation of 3D point clouds to survey river dynamics and evolution , EGU General Assembly 2024, Vienna, Austria, 14–19 Apr 2024, EGU24-15896,, 2024.