From Field to Simulation: 3D Segmentation of Precariously Balanced Rocks and Dynamic Simulation of Their Response to Earthquake Ground Motions

Ramon Arrowsmith, Zhiang Chen, Deep Rodge, Akshay Mahalle, M. Khalid Saifullah, Jnaneshwar Das, Christine Wittich, Albert R. Kottke, & Christopher M. Madugo

Submitted September 7, 2025, SCEC Contribution #14881, 2025 SCEC Annual Meeting Poster #TBD

Precariously balanced rocks (PBRs) provide natural geological indicators for constraining the upper bounds of earthquake ground motions over long timescales. We present an unified workflow that combines high-resolution 3D mapping, point cloud segmentation, and physics-engine simulation to enable rapid, accurate, and scalable PBR fragility modeling. In the field-mapping stage, we have explored three approaches for acquiring high-resolution 3D datasets of PBR sites: UAV photogrammetry, UAV lidar, and iPhone lidar scanning. For UAV photogrammetry and UAV lidar surveys, we have developed a detection workflow that combines orthomosaic analysis with principal geometric features extracted from rock-scale point clouds to identify PBR candidates. A custom 3D segmentation tool then separates rock bodies from their pedestals using an improved region-growing algorithm guided by seed points and basal contact lines. The resulting watertight rock meshes are used to compute key fragility metrics, including the angles between the center of mass and rocking points, produce ensemble fragility distributions across surveyed sites, and serve as direct geometric inputs for subsequent dynamic simulations.

In the simulation stage, we have developed Virtual Shake Robot 2 (VSR2), a next-generation PBR dynamics platform built with the PyBullet physics engine and Robot Operating System 2. VSR2 features a modular architecture that supports flexible contact physics configurations, multiple ground motion modes, and parallel high-throughput simulations. Calibration against 582 real earthquake records and large-scale physical shake table experiments yields overturning prediction accuracy comparable to well-studied discrete element models, with 102-105x greater computational efficiency. Using site-derived 3D geometries as direct simulation inputs, we perform large-scale virtual shake testing to generate probabilistic fragility curves for individual PBRs and regional populations. This combined field-to-simulation framework enables end-to-end, reproducible PBR fragility modeling, providing a powerful tool for constraining earthquake ground motions and informing seismic hazard models.

Key Words
Fragile Geoloogic Features, Ground Motions

Citation
Arrowsmith, R., Chen, Z., Rodge, D., Mahalle, A., Saifullah, M., Das, J., Wittich, C., Kottke, A. R., & Madugo, C. M. (2025, 09). From Field to Simulation: 3D Segmentation of Precariously Balanced Rocks and Dynamic Simulation of Their Response to Earthquake Ground Motions. Poster Presentation at 2025 SCEC Annual Meeting.

Related Projects & Working Groups
Earthquake Geology