Oral Presentation

From a dark desert highway, towards a shimmering light, we journey through past and future directions in ground motion modeling and seismic hazard analysis. Ground motion models (GMMs) prescribe the distribution of ground motions that is associated with event occurrence within the framework of probabilistic seismic hazard analysis (PSHA). I overview recent trends that impact three attributes of GMMs that drive PSHA results and discuss their interactions. The attributes are: the median model (i.e. median amplitudes of response spectral ordinates and peak ground motions as functions of magnitude, distance and site parameters); epistemic uncertainty (i.e. alternative models that reflect lack of knowledge in the correct median); and aleatory variability (i.e. random scatter about the median).

I focus on the importance of data in developing and implementing promising recent approaches. An important recent trend stems from advancements in GMMs and associated PSHA approaches for site-specific applications. The median, epistemic and aleatory components of GMMs can be refined by using site-specific ground motion recordings to modify generic regional models–thereby moving from ergodic to partially or fully non-ergodic GMMs. Site-specific GMMs increase the reliability of the median and better define its epistemic and aleatory uncertainties. Adequate treatment of site response is an important component of a non-ergodic approach, which can only be truly realized at sites with sufficient ground motion recordings. In addition to reducing bias (which may increase or decrease computed hazard), the site-specific approach may significantly reduce computed PSHA result through reduction of aleatory uncertainty. The reduction in computed hazard often more than justifies the expense of the recording program for critical sites.

I review the modeling of epistemic uncertainty in GMMs and consequent interactions between epistemic and aleatory uncertainty. The total distribution of ground motions input to the PSHA should be considered to avoid overestimation of the hazard through over-counting of uncertainties. With sufficient data, a self-consistent PSHA model of median inputs and their epistemic and aleatory uncertainties can close the loop between observed and estimated hazard on a regional basis.