Understanding and reducing uncertainties in ground‐motion prediction are high priorities for seismic‐hazard analysis. This article examines forumla, the variability in synthetic ground motions at rock sites caused by the variability in randomly generated velocity profiles of the geological column from 5 km depth to the surface. Only sites with VS30 of 500  m/s or higher are considered, and linearity is assumed. These synthetic estimates of the mean value of forumla are a complicated but understandable function of magnitude, period, and VS30. The distribution of modeled residual response spectral amplitudes at several oscillator periods is not lognormal, but the deviations are in the central part of the distribution, in which the effect on probabilistic seismic‐hazard analysis may not be very large. Adding another constraint to the velocity profile, namely that the shear‐wave velocity of the uppermost layer should be at least 70% of VS30, greatly reduces the uncertainty at high frequencies. We tentatively identify sites with this property as nonresonant rock, because it excludes sites with a strong resonance in a thin shallow layer. The reduction in uncertainty that this allows might reduce or eliminate the contradiction between the U.S. National Seismic Hazard Map and precarious rocks in southern California. Furthermore, for nonresonant rock sites, the residual impact of the ergodic assumption might be reduced to more tolerable levels.

Online Material: Tables of variability values.

You do not currently have access to this article.