Bayesian analysis of earthquake seismology models under uncertainty

In this paper we apply a rigorous Bayesian methodology to quantitatively assess and compare, under uncertainty, rupture modeling hypotheses for computational earthquake seismology and advanced ground-motion simulations. Reliable ground motion predictions, with quantified uncertainty, are critical for designing large civil structures and response plans, thus helping to mitigate human and economical losses during earthquakes. With the advent of HPC and state-of-the-art parallel statistical algorithms, comprehensive uncertainty quantification of the expected shaking levels is now becoming possible. To demonstrate the potential of Bayesian methodology for the analysis of earthquake ground-motion simulations, this feasibility study proposes just two "simple" rupture modeling hypotheses, denoted M₁ and M ₂, which treat as random some of the physical parameters describing the fault properties. For each modeling hypothesis, ground-motions are computed at a dense virtual seismic network. Such ground motions are compared against a chosen empirical ground-motion prediction equation (GMPE, the reference data d). The ranking returns the hypothesis that best reproduces, among the proposed candidates M₁ and M₂, peak ground velocities with respect to d.