Comment on "Evaluation of Models for Earthquake Source Spectra in Eastern North America" by Gail M. Atkinson and David M. Boore

Abstract

In their article, Atkinson and Boore (1998) (AB) attempt to evaluate a number of proposed source models for eastern North America (ENA) by comparing “predicted” ground motions for the source models against certain ground-motion data (as listed in their Table 1). The reliability of their conclusions evidently depends both on the reliability of their predicted ground motions and on the reliability of their assumed data. Although the extent to which the data used are representative of ground motions in ENA is also questionable, in this comment only the question of the reliability of AB's predicted ground motions and their conclusions concerning the source models they have considered will be discussed. The ground motions were calculated for all of the source models using the stochastic ground-motion model of Atkinson and Boore (1995). AB assert that this model “provides a sound [emphasis added] basis for a test of the implications of the source models.” It will be shown that this assertion is unfounded because the model is inconsistent with basic seismological principles, resulting in errors of up to about an order of magnitude in ground-motion estimates. Their source-model evaluations and all of their main conclusions are therefore unfounded. The stochastic method for wave-train synthesis is just one of many possible ways of generating a time series (or synthetic wave train) from an amplitude spectrum (for another see e.g., Haddon, 1996a). In order to generate any such time series, a time interval (or waveform envelope) over (or within) which the energy is to be distributed has to be assigned. The root mean square (rms) amplitudes of the time series are uniquely determined, for each frequency, by the spectrum and the assigned duration. No information about phase is involved. The stochastic method simply assumes that the phases of the different frequencies differ randomly from …