Regional-Scale Three-Dimensional Structure of the Western US from Seismic Waveform Inversion Using Salvus

Abstract:

Characterization of seismic sources by moment tensor estimation has proven successful at identifying earthquakes, explosions and mine collapses. However, moment tensor inversion suffers from uncertainties in waveform fitting related to inaccuracies in Greens functions for path-specific wave propagation effects. This problem is acutely period, wavelength and distance dependent. Moderately large events (M_W > 4.5) at regional distances reveal good signal-to-noise ratios (SNR) for long-period (> 20 seconds) waveforms and can be modeled with simple plane-layered, one-dimensional (1D) Earth models. However, smaller events observed at these ranges have weak long period energy and SNR is best for shorter periods. These events require analysis of shorter period waves which are strongly scattered by heterogeneous crustal and upper mantle structure. Source characterization of smaller events at regional distances requires three-dimensional (3D) Earth models in order to account for path-specific wave propagation effects.

We are developing 3D seismic models of the Western US (WUS) to investigate improvements in source characterization with moment tensor analysis, using the Salvus software package. Salvus is an end-to-end seismic imaging solution enabling estimation of sub-surface structure from waveform inversion based on the adjoint method. Salvus relies on the spectral element method to compute forward and adjoint 3D waveform simulations. It includes integrated workflow management that handles waveform and metadata for many events, launches simulations on local or remote computers (including massively parallel machines), computes waveform misfits and adjoint sources, combines and smooths model gradients and iterates for model updates by nonlinear optimization. We are running Salvus to infer 3D crustal and upper mantle structure for two domains: a large region of the WUS from the Pacific Ocean to the Great Plains and a smaller domain for California and surrounding areas. For both domains we are starting with available open-source broadband waveforms from moderately large earthquakes (M_W 5-6.5) and attempting to fit waveforms with periods down to 5-10 seconds. This is work in progress and preliminary results will be presented.