Two topics shall be covered: (1) a technique for simulation strong motion at a moderate distance from an earthquake source, including an approach to bracket uncertainty of parameters of such ground motion; and (2) a study of earthquake body waves aimed at estimating important parameters that are needed for such a simulation to be reliable.

The proposed simulation technique is applicable mainly for sites with rock ground. It combines a multiple-point-source version of a stochastic earthquake fault model and a suite of Green’s functions calculated for layered weakly anelastic medium. The source model consists of a grid of point subsources with appropriate random time histories. A large number of properties/parameters of the simulated earthquake fault can be adjusted in order to: (1) tune the model to a particular seismological situation; and/or (2) to analyze the variability and uncertainties of strong motion prediction.

With an example of Northrige (1994) earthquake simulation, analysis of uncertainty was performed with respect to a number of factors. It was made in terms of standard deviations of log₁₀(pseudo response acceleration) or log(PSA). For the first, “aleatoric” group of factors, variances of log(PSA) were directly estimated. This group included: the random choice of 2D final-slip function; the random choice of detailed subsource time histories; random rupture velocity history (that varies during rupture propagation); random nucleation point position, random mean rupture velocity. For other factors, sensitivities to parameters were estimated. These parameters included source-intrinsic group (scalar seismic moment, stress drop, range of variations for rupture velocity, etc; and “geometry” group (dip, slip, and rake angles; depth and epicenter). For the subset of important parameters, a trial value of rms deviation was selected, and corresponding contribution to the variance of log(PSA) was estimated through propagation of errors.

One of poorly understood factors that is to be known for a realistic broad-band simulation of the source/fault is the degree of correlation between local slip rate at a certain spot of a fault, and the high-frequency radiation capability (HF seismic luminosity) that is specific for this spot. Most source models predict relatively high correlation, whereas the results of inversions performed for a few faults seem to indicate the opposite. To study this question in more detail, a systematic analysis is performed that uses teleseismic P-waves of 23 intermediate-depth earthquakes of magnitude above 6.8. From each of 344 broadband records we determine two time histories: (1) displacement and (2) squared, 0.5-2.5Hz band-passed, velocity, or "power", and then calculate the correlation coefficient, ρ, between them. The distortion related to formation of P coda is partly suppressed. We estimated the average value ρ = 0.52 for the correlation coefficient between the radiated time histories for displacement and "power". This value can be ascribed to the similar correlation coefficient between slip rate and HF seismic luminosity over the source area.

Two different factors cause decorrelation: random fluctuations and genuine mismatch of slip rate and mean luminosity. They were isolated, and we found that fluctuations produce ρ = 0.72 and the mismatch produces ρ = 0.83. Thus the observed values of ρ indicate genuine differences between the distributions of the slip rate and the seismic luminosity over the fault area. These results provide important constraints both for the accurate wide-band simulation of strong ground motion and for theoretical dynamic source models.