Capturing epistemic uncertainty in site response

Abstract

The incorporation of local amplification factors (AFs) determined through site response analyses has become standard practice in site-specific probabilistic seismic hazard analysis (PSHA). Another indispensable feature of the current state of practice in site-specific PSHA is the identification and quantification of all epistemic uncertainties that influence the final hazard estimates. Consequently, logic trees are constructed not only for seismic source characteristics and ground-motion models (GMMs) but also for the site AFs, the latter generally characterized by branches for alternative shear-wave velocity ( VS) profiles. However, in the same way that branch weights on alternative GMMs can give rise to unintentionally narrow distributions of predicted ground-motion amplitudes, the distribution of AFs obtained from a small number of weighted VS profiles will often be quite narrow at some oscillator frequencies. We propose an alternative approach to capturing epistemic uncertainty in site response in order to avoid such unintentionally constricted distributions of AFs using more complete logic trees for site response analyses. Nodes are included for all the factors that influence the calculated AFs, which may include shallow VS profiles, deeper VS profiles, depth of impedance contrasts, low-strain soil damping, and choice of modulus reduction and damping curves. Site response analyses are then executed for all branch combinations to generate a large number of frequency-dependent AFs. Finally, these are re-sampled as a discrete distribution with enough branches to capture the underlying distribution of AFs. While this approach improves the representation of epistemic uncertainty in the dynamic site response characteristics, modeling uncertainty in the AFs is not automatically captured in this way, for which reason it is also proposed that a minimum level of epistemic uncertainty should be imposed on the final distribution.