Probabilistic Seismic Hazard Analysis Using Non-Ergodic Ground-Motion Prediction Equations
UCLA Garrick Institute for the Risk Sciences
- Date & Time
- Building 3, Rambo Auditorium
- Grace Parker
With an increasing number of strong-motion records over the last decade, it has become clear that there are significant differences in ground-motion scaling even within relatively small regions such as California. These differences are typically not modeled in ground-motion prediction equations (GMPEs); they are based on the so-called ergodic assumption. By including the systematic source, path and site effects in fully non-ergodic GMPEs it is possible to reduce the value of the aleatory variability by about 30-40%. This requires the estimation of the systematic effects for every possible source/site combination, together with their epistemic uncertainty. If a non-ergodic GMPE is used in seismic hazard analysis, it is very important to propagate the epistemic uncertainty of the systematic effects to obtain the full hazard distribution.
In this talk, I will describe how non-ergodic adjustment terms can be estimated for California. The adjustment terms are estimated using a Gaussian process framework, which allows one to keep track of the associated uncertainties. The underlying assumption is that the systematic effects are spatially correlated, which can be exploited to estimate the distribution of these terms at non-observed locations. The impact of incorporating the non-ergodic adjustment into seismic hazard calculations is shown for a few example sites. For sites that have abundant data in their vicinity, the non-ergodic hazard changes compared to the ergodic one; while for sites with sparse data the mean hazard stays the same, but there is a large increase in the epistemic uncertainty range of the hazard distribution. Finally, I outline how one can apply non-ergodic hazard analysis in regions with very sparse data.