Evaluation of models for earthquake source spectra in eastern North America

Abstract There have been several relations proposed in the last few years to describe the amplitudes of ground motion in eastern North America (ENA). These relations differ significantly in their assumptions concerning the amplitude and shape of the spectrum of energy radiated from the earthquake source. In this article, we compare ground motions predicted for these source models against the sparse ENA ground-motion database. The source models evaluated include the two-corner models of Boatwright and Choy (1992), Atkinson (1993a), Haddon (1996), and Joyner (1997a,b), and the one-corner model of Brune as independently implemented by Frankel et al. (1996) and by Toro et al. (1997). The database includes data from ENA mainshocks of M 4 and historical ENA earthquakes of M 5.5, for a total of 110 records from 11 events of 4 ≦ M ≦ 7.3, all recorded on rock. We also include 24 available rock records from 4 large earthquakes in other intraplate regions; conclusions are checked to determine whether they are sensitive to the addition of these non-ENA data. The Atkinson source model, as implemented in the ground-motion relations of Atkinson and Boore (1995), is the only model that provides unbiased ground-motion predictions over the entire period band of interest, from 0.1 to 10 sec. The source models of Frankel et al. (1996), Toro et al. (1997), and Joyner (1997a,b) all provide unbiased ground-motion estimates in the period range from 0.1 to 0.5 sec but overestimate motions at periods of 1 to 10 sec. The Haddon (1996) source model overpredicts motions at all periods, by factors of 2 to 10. These conclusions do not change significantly if data from non-ENA intraplate regions are excluded, although the tendency of all models toward overprediction of long-period amplitudes becomes more pronounced. The tendency of most proposed ENA source models to overestimate long-period motions is further confirmed by an evaluation of the relationship between Ms, a measure of the spectrum at 20-sec period, and moment magnitude. A worldwide catalog of shallow continental earthquakes (Triep and Sykes, 1996) is compared to the Ms-M relations implied by each of the source models. The Atkinson source model is consistent with these data, while other proposed ENA models overpredict the average Ms for a given M. The implications of MMI data from historical earthquakes are also addressed, by exploiting the correlation between felt area and high-frequency source spectral level. High-frequency spectral amplitudes, as specified by the Atkinson and Boore (1995), Frankel et al. (1996), Toro et al. (1997), and Joyner (1997a,b) source models, equal or exceed the levels inferred from the felt areas of most of the large ENA events, with the noteable exception of the Saguenay earthquake. By contrast, high-frequency spectral amplitudes specified by the Haddon (1996) source model agree with the felt area of the Saguenay earthquake but overpredict the felt areas of nearly all other large events. In general, models that fit the Saugenay data—be it intensity data, strong-ground-motion data, regional seismographic data, or telescismic data—will not fit the data from the remaining earthquakes. A source model derived from the California database, suitably modified for regional differences in crustal properties, is also evaluated. This model is not significantly different from the Atkinson model for ENA. There is an important practical application of this similarity, which we develop as an engineering tool: Empirical ground-motion relations for California may be modified to predict ENA ground motions from future large earthquakes.