Assessing the predictability of rockfall simulations constrained to simple objective input parameters

Abstract. Rockfall simulations are used to predict runout distances in case potentially unstable rock compartments would eventually fail. Transit simulated values such as the bounce heights and involved energies are also useful for hazard and risk assessments and for mitigation design tasks. However, it has been shown that the predictions from simulation results can vary significantly from user to user and from site to site. This highlights the need for simulation models with quantified accuracy and precision, low parametric subjectivity, and with good performance at predicting the transit values. The objective of this work is to present a validation methodology for rockfall simulation models and to objectively evaluate the predictive performance of stnParabel freeware simulation model when used with the Rolling friction rebound model. For this purpose, numerous mapped observations from a combination of back analyses of rockfall experiments and real events involving different remote sensing techniques were gathered. They cover twelve sites of various characteristics and geometries. The extensive collected observations include several hundred mapped deposited rock fragments of known dimensions and respective source locations. Each individual rock’s dimensions and masses were repetitively simulated without any other parameter adjustments in order to minimize the subjectivity of the simulation approach. In complement to the systematic objective process-based simulations, the runouts were also predicted for all sites with simulated trajectories from two additional process-based models for comparisons. Moreover, runout extents were also obtained geometrically with a commercial software and with a common geometrical approach for comparison. The results showed that the runout prediction accuracy from our process-based simulated trajectories is generally stable from site to site. Moreover, the runout precision of the simulations with stnParabel is improved by 2× to 3× compared to those of all other methods tested. And this is achieved with limited errors on the predicted transit values such as the bounce heights and translational velocities.