Empirical ground motion prediction

Abstract

New methods of site-specific ground motion prediction in the time and frequency domains are presented. A large earthquake is simulated as a composite (linear combination) of observed small earthquakes (subevents) assuming Aki-Brune functional models of the source time functions (spectra). Source models incorporate basic scaling relations between source and spectral parameters. Ground motion predictions are consistent with the entire observed seismic spectrum from the lowest to the highest frequencies. These methods are designed to use all the available empirical Green’s functions (or any subset of observations) at a site. Thus a prediction is not biased by a single record, and different possible source-receiver paths are taken into account. Directivity is accounted for by adjusting the apparent source duration at each site. Our time-series prediction algorithm is based on determination of a non-uniform distribution of rupture times of subevents. By introducing a specific rupture velocity we avoid the major problem of deficiency of predictions around the main event's corner frequency. A novel notion of partial coherence allows us to sum subevents' amplitude spectra directly without using any information on their rupture times and phase histories. Predictions by this spectral method are not Jependent on details of rupture nucleation and propagation, location of asperities and other predominantly phase-affecting factors, responsible for uncertainties in time-domain simulations.