1-D and 2-D Probabilistic Inversions of Fault Zone Guided Waves

Abstract:

Fault Zone Guided Waves (FZGWs) are seismic coda that are trapped by the low velocity damage zone of faults. Inversions of these phases can be carried out using their measured dispersion and a Bayesian probability approach. This method utilises a Markov chain Monte Carlo which allows uncertainties and trade-offs to be quantified. Accordingly we have developed a scheme that estimates the dispersion curve and amplitude response variability from a FZGW record. This method allows the computation of both the point estimates and the covariance of the dispersion curve. The subsequent estimation of fault zone parameters is then based on a Gaussian model for the dispersion curve. We then show that inversions using FZGW dispersion data can only resolve fault zone velocity contrast and fault zone width - it leaves densities, absolute country rock velocities and the earthquake location unresolved. We show that they do however significantly affect the estimated fault zone velocities and widths. As these parameters cannot be resolved, we allow for their effects on the estimates of fault zone width and velocity contrast by using the Bayesian approximation error method. We show that using this method reduces computational time from days to minutes and the associated loss of accuracy is insignificant compared to carrying out the inversion on all parameters. We have extended our scheme to 2-D using 1-D slices. The Bayesian approximation error methodology is further employed to provide a ‘correction term’ with uncertainty for the 1-D slice approximation. We investigate these features with both synthetic data and FZGW data from the Alpine Fault of New Zealand.