Multiple comparisons of point clouds acquired by a permanent LiDAR (PLS) to improve the reliability of a rockfall event catalogue

The ANR C2R-IA project (www.anrc2ria.fr) aims to develop reliable decision-support tools for the dynamic management of rockfall hazard. Its goal is to understand how meteorological forcing influences rockfall occurrence and to anticipate temporary increases in hazard in order to implement risk reduction measures. To this end, a predictive model of rockfall occurrence as a function of meteorological conditions is being developed using artificial intelligence tools (neural network training), which requires a comprehensive and well-labelled dataset. Several monitoring instruments have been deployed at the Saint-Eynard site (Grenoble, France). Among them, a permanent LiDAR scanner (PLS) acquires point clouds continuously, with one acquisition per hour, providing high temporal resolution representative of what could be used for operational monitoring or crisis management. An automated data-processing workflow has been developed in Python. It is based on a pairwise comparison of the clouds (Manceau et al., 2025) and includes the alignment of successive point clouds, filtering of points outside the cliff area, change detection using M3C2 distances computation, clustering with DBSCAN, and volume quantification of rockfalls using alphashapes. This well-structured processing has significantly reduced the detection threshold, identifying relief change of only 10 cm deep (compared to 40 cm previously; Le Roy et al, 2020) and 10 liters in volume, while the scanner is located approximately 1 km from the cliff. Depending on acquisition quality, the effective temporal resolution of detected rockfall events may range from one hour to several days. Combining relief-change detections with simultaneously deployed seismic monitoring should further refine event timing. The completeness of the event catalogue has therefore improved, increasing from fewer than 10 detected rockfalls per month to around 30. However, some false positives remain, mainly related to recurring artifacts despite preprocessing. To mitigate these errors, the previous pairwise comparison of the clouds has been refined to a multiple point-cloud comparison strategy, enabling the tracking of the temporal persistence of changes. This allows distinguishing changes corresponding to real rockfalls, which persist over time, from transient artifacts. This improvement leads to a more reliable and complete rockfall event database. It includes block shape ratios, identified failure mechanisms, and free-fall heights under overhanging sections, providing a suitable basis for future fusion with seismic data.

Manceau, L., Chanut, M.-A., Levy, C., Dewez, T., and Amitrano, D.: Enhancing Rockfall Detection Using Permanent LiDAR Scanner (PLS) Data and Automated Workflows at St. Eynard Cliff (Grenoble, France), EGU General Assembly 2025, Vienna, Austria, 27 Apr–2 May 2025, EGU25-6312, https://doi.org/10.5194/egusphere-egu25-6312

Le Roy, G., Helmstetter, A., Amitrano, D., Guyoton, F., & Le Roux-Mallouf, R. (2019). Seismic analysis of the detachment and impact phases of a rockfall and application for estimating rockfall volume and free-fall height. Journal of Geophysical Research: Earth Surface, 124, 2602-2622. https://doi.org/10.1029/2019JF004999