A Bootstrapped Modularised method of Global Sensitivity Analysis applied to Probabilistic Seismic Hazard Assessment

Probabilistic Seismic Hazard Assessment (PSHA) evaluates the probability of exceedance of a given earthquake
intensity threshold like the Peak Ground Acceleration, at a target site for a given exposure time. The stochasticity
of the occurrence of seismic events is modelled by stochastic processes and the propagation of the earthquake
wave in the soil is typically evaluated by empirical relationships called Ground Motion Prediction Equations. The
large uncertainty affecting PSHA is quantified by defining alternative model settings and/or model parametri zations. In this work, we propose a novel Bootstrapped Modularised Global Sensitivity Analysis (BMGSA) method
for identifying the model parameters most important for the uncertainty in PSHA, that consists in generating
alternative artificial datasets by bootstrapping an available input-output dataset and aggregating the individual
rankings obtained with the modularized method from each of those.
The proposed method is tested on a realistic PSHA case study in Italy. The results are compared with a
standard variance-based Global Sensitivity Analysis (GSA) method of literature. The novelty and strength of the
proposed BMGSA method are both in the fact that its application only requires input-output data and not the use of a PSHA code for repeated calculations.