The composite source model for generating synthetic strong ground motions is tested for its ability to predict the statistical characteristics of Northridge accelerograms recorded in or adjacent to the San Fernando valley. The general problem is prediction of strong motions at a site of engineering interest with sufficient realism to be useful for engineering applications. The strongest test of any proposed method is a blind prediction. For this study, a completely blind test was not possible. Our objective was to use only a preliminary description of fault geometry and magnitude and previously published velocity models and, without iteration to improve the quality of fit, to evaluate the differences between predicted and observed accelerograms.
The parameters that we predict are peak acceleration, peak velocity, peak displacement, Fourier spectra at seven frequencies, and pseudorelative velocity response (5% damping) at seven periods. Our results are given for 14 stations. For the horizontal components, these parameters are all predicted with a maximum bias of under 50% and an average bias of observations exceeding predictions by 6%. For peak acceleration and some response spectral periods, the bias for this model is smaller than at least some regressions, when applied to this specific earthquake. On the vertical component, the maximum bias is a factor of 2, and the average gives predictions exceeding observations by 25%. Standard deviations of the common logarithm of the ratio of observed-to-predicted parameters are typically about 0.3, which is perhaps 50% greater than the standard deviations typical of regressions but comparable to standard deviations of observations from this earthquake compared to regressions. In the future, it is likely that, in some cases, traditional regressions will be replaced with synthetic calculations of some type, such as the method used here. Based on the results of this study, the amount of progress that has been made in obtaining that goal is very encouraging.