We considered the effects of source, propagation path, and local site conditions on strong ground motion from the Loma Prieta earthquake in the following steps. First, the local site effect was removed from the strong motion records applying the weak-motion amplification factors estimated from nearby stations of USGS regional network using the coda method. Then, we confirmed the approximate validity of the 1/R law of geometrical spreading factor for the crustal model applicable to the region, and we determined the attenuation parameter Q(f) and source spectrum S(f) empirically from the site-effect — corrected spectral amplitude. Both Q(f) and S(f) are consistent with the results obtained earlier for major California earthquakes.
We then synthesized the time history of ground acceleration by applying the stochastic simulation technique. The agreement between the observed and predicted records was good for duration and spectral content. We found, however, a strong systematic discrepancy with regard to peak acceleration; namely, for distances less than 50 km, the predicted peak acceleration consistently overestimated the observed for sediment sites and underestimated it for the Franciscan formation sites. Since the above systematic discrepancy applies to all the available stations within about 50 km from the hypocenter, it is unlikely that the effect is due to the radiation pattern, near-source structure, or topography. It appears that the strong difference in the amplification factor between the Franciscan and the sediment sites observed for weak motion disappears at acceleration levels higher than about 0.1 to 0.3 g. If this is due to the nonlinear response of the sediment sites, it is in the range expected from geotechnical engineering studies.