Stochastic Descriptions of Small-Scale, Near-Surface Velocity Variations in the Los Angeles Basin for Modeling Earthquake Ground Motions

Abstract:

Simulations of earthquake ground motions at high frequencies (> 1 Hz) require high-resolution velocity models to quantify the effects of wave scattering, attenuation, and anisotropy. Valuable information about the statistical variations of the velocity structure in the upper few kilometers of the crust can be obtained from well logs. We apply geostatistical methods to characterize the one-point and two-point statistics of velocity variations observed at vertical scale lengths less than 200 m in an ensemble of vertical sonic logs from the central Los Angeles basin. The stochastic variability at these scales can be separated into two components, one with a correlation length of about 10-60 m and a second with a correlation length of about 1-4 m. The one-point statistics of both components are distinctly non-Gaussian, with those of the first skewed to low velocities and those of the second skewed to high velocities. Assuming that the horizontal correlation lengths are much greater than the vertical correlation lengths, we obtain a long-wavelength polarization anisotropy of around 6-8% from Backus averaging of the profiles. We also investigate the magnitude of wave scattering caused by the small-scale heterogeneity and speculate on its role in explaining the anomalous near-surface attenuation of high-frequency seismic waves.