Incorporating near-surface attenuation in empirical ground motion models

Abstract

This thesis addresses empirical ground motion modelling in the context of seismic hazard assessment. In particular, the work focuses on empirical modelling of high frequency ground motions at rock sites. The first section uses data from the Canterbury earthquake sequence to improve understanding of the physical mechanisms behind high frequency Fourier spectral amplitudes and develop robust methods of estimating the high frequency spectral decay parameter, K. The improved estimation method is then used to calculate the site attenuation parameter K0, at GeoNet rock sites across New Zealand. This analysis reveals strong variation in site attenuation between the high and low seismicity regions of New Zealand. Using the geostatistical technique of ordinary kriging, a continuous 0 map of New Zealand is developed. The second half of the thesis utilises the K0 map to derive an empirical model of the earthquake Fourier amplitude spectrum, using K0 as a site-specific predictor. The model is derived by fitting previously-recorded New Zealand ground motion data. It is found that using K0 as a predictor in the empirical model is largely unsuccessful, and the model performs better if K0 is excluded from the model. A model is instead derived without explicit specification of site attenuation effects. To complement the Fourier amplitude spectrum model, a model for the significant duration of ground motion is derived. These two models are used together in a random vibration theory analysis, to predict response spectra. In deriving the duration model, conceptual improvements are made to random vibration theory calculations in the context of engineering seismology. It is shown that the subsequent empirical model of response spectra is generally unbiased with respect to New Zealand response spectral data, for damping levels of 5% and 20%.