Bayesian estimation of moment tensors and slip history based on prior knowledge from deterministic source inversion

Abstract:

Determination of point source solutions is a necessity for numerous applications in seismology (e.g., waveform tomography), as well as applications where the focal mechanisms and locations of earthquakes are used to infer tectonic models. We present a probabilistic framework for inversion of depth, moment tensor and moment rate (source time function) from teleseismic P and SH waveforms. The results are to be compiled into a global Bayesian earthquake catalogue.

Bayesian inference from waveforms crucially depends on the noise model. Since we consider the assumption of sample-wise normally distributed noise unrealistic, we opt for a waveform misfit criterion. In order to establish its statistical properties in the presence of noise and realistic modelling errors, we harness a database of 1800 deterministic source solutions to derive the parameters of the Likelihood function needed in Bayesian inference.

The inversion scheme itself uses the Neighbourhood Algorithm to infer a statistical ensemble of source solutions. These can for example be used to infer the influence of source uncertainty on the input parameters for seismic waveform tomography, such as cross-correlation traveltimes.