How to discriminate between area-source models for probabilistic seismic hazard assessment in a more objective manner?

Abstract

We propose a methodology to assist in the selection or definition of weights for components of seismic hazard models, which are combinations of seismic source models and ground motion models, in the context of probabilistic seismic hazard assessment (PSHA). The methodology uses Bayes's theory by optimally exploiting available observations that are the seismic catalogues and accelerometric databases. When compared to the current method of calculation, the proposed approach, simple to implement, allows a more exhaustive use of the data and discriminates between inputs without expert judgements to weigh branches of the PSHA logic tree.We implement the proposed methodology in a Python package called Phebus to process the seismic source models (the first of the two main ingredients of the seismic hazard model) as a first step. The main purpose of this package is to estimate earthquake recurrence parameters and confidence intervals using a full Bayesian approach, and perform a Bayesian model averaging (BMA) amongst multiple seismic source models. More particularly, we focused our study on area-source models, which consist of subdivisions of a particular region of interest into zones that are assumed homogeneous in terms of seismic activity rate, and that are systematically used in PSHA calculations for low-to-moderate seismic regions. We conducted sensitivity analyses on the selection performances and the adjustments performances of recurrence parameters for simplistic toy models against a synthetic model-generated seismicity catalogue. We also illustrate the application of Phebus to the metropolitan France, a low-strain region, where at least four national, competitive and published area-source models are used by engineers and researchers for seismic hazard evaluation.