Improving Seismic Moment-Tensor Inference Incorporating Uncertainty in Earth Structure

Abstract:

Seismic moment-tensor is the mathematical representation for energy radiation pattern of an earthquake approximated as a point source. To infer moment tensor of moderate earthquakes at regional scales, seismologists typically simulate waveforms using available structural models of the Earth to match the observed seismograms. The inverse problem is subjected to uncertainties present in the signal acquisition process, which is known as data noise, and in the theory underpinning the forward problem of full-waveform simulation. Recent advancements in the treatment of correlated seismic-noise in the source inversion have been made, but all-embracing treatment of theory uncertainties remains limited. It is reasonable to assume that an incomplete knowledge of Earth structure and its effect on seismic wavefield propagating from the source to the receiver is the most significant component of theory uncertainty. Here, we propose an improved inversion scheme under the Bayesian framework that incorporates both data and theory uncertainties under consideration. The hierarchical component of data noise is also considered in this inversion scheme, as the data noise level is treated as a free parameter to recover. In particular, the combined covariance matrix, defined as a sum of data noise and theory noise covariances, is instantaneously updated in both its shape and amplitude with the moment tensor and noise-strength proposals. The convergence and efficiency of the inference method is demonstrated though synthetic experiments with real noise as well as field data from a selected nuclear explosion. In future applications, a careful consideration of Earth model’s uncertainty as a part of the moment-tensor inversion will be necessary to obtain better understandings of complicated physics of earthquake sources.