Rockfalls trajectography: 3D models predictive capability assessment and coefficients calibration using optimization-based processes

Abstract

This paper presents a method for assessing the predictive capability of three-dimensional (3D) trajectographic simulation models by back-analysis of real rockfall events. Relying on the Optimal Transport theory, we measure the difference between observed and simulated rock stop points distributions with the Wasserstein distance: this metric can be seen as a measure of the mean distance between observed an simulated stop points. We use the Wasserstein distance as a cost function in a Black Box optimization algorithm to calibrate soil restitution and energy dissipation coefficients. We test our methodology with the RocPro3D software to simulate a man-triggered boulder detachment, for which the final position of fragmented rocks is known. The calibrated simulation parameters enabled a 25% decrease in the mean prediction error.