Comparing Adjoint Waveform Tomography Models of California Using Different Starting Models

Abstract

Seismic tomography has been a vital tool for understanding Earth's interior structure since studies were first published in the late 1970s (e.g., Aki et al., 1977;Aki & Lee, 1976). Though the theory of Full waveform inversion (FWI) existed, ray-based travel time tomography methods were the primary method of velocity modeling for much of the field's history. Ray-based travel time tomography uses the difference between theoretical and observed arrival times of specified seismic waves to calculate velocity perturbations in the Earth (Dziewonski et al., 1977). Ray-based tomography can be computationally inexpensive but suffers in tectonically complex regions as the physics behind it is unable to resolve effects such as wavefront healing, focusing/defocusing, and other 3D wavefield effects (Liu & Gu, 2012;Rickers et al., 2012;Tarantola, 1984). Unlike ray-based methods, adjoint waveform tomography (AWT) accounts for 3D wavefield effects by solving the elastic wave equation through a given 3D volume, while also considering finite-frequency effects. AWT iteratively optimizes the synthetic data to best match the observed data for available source-receiver pairs. By using the recorded waveform instead of travel times, AWT includes more information (Romanowicz, 2003). AWT has a greater sensitivity to Earth structure than ray-based travel time tomography, enabling greater resolution of tectonic complexity.Uncertainty analysis and quantitative comparisons of FWI models continues to be a topic of active research (Flath et al.

CANVAS: An Adjoint Waveform Tomography Model of California and Nevada

“…As noted in Section 3, we inverted for source mechanisms before iterating at periods shorter than 20 s (Doody et al., 2023). The idea behind inverting for the source parameters is to reduce errors in the source mechanisms, better isolate path effects, and improve waveform fits (Lei et al., 2020; Liu et al., 2004; Sawade et al., 2022).…”

“…(2022) and Doody et al. (2023). An in‐depth discussion of the Salvus methodology is available in Rodgers et al.…”

“…The results of Stage 1 were presented as the CANV_WUS model in Doody et al. (2023). The maximum period was lowered in Stage 1c to improve the signal‐to‐noise ratio of the smaller events in the inversion dataset (M W < 5).…”

“…The description below provides a broad overview of the methodology used. We closely follow the methodology presented in Rodgers et al (2022) and Doody et al (2023). An in-depth discussion of the Salvus methodology is available in Rodgers et al (2022).…”

“…We used number and length of windows as a proxy for waveform fit as our window picking algorithm requires the cross-correlation coefficient between the observed and synthetic data to be over 0.75 and for phase delays to be within 0.25 wavelengths to pick a window. A full list of parameters considered for picking windows can be found in Doody et al (2023).…”