UC Berkeley

Accurate seismic velocity models are a necessary parameter for seismic hazard analysis in high hazard regions. Therefore, we present the California-Nevada Adjoint Simulations (CANVAS) model, an adjoint waveform tomography model of California and Nevada that is optimized to fit waveforms. We began by testing different starting models for CANVAS to determine what effect starting model choice plays on final inversion results. Though studies have compared synthetic seismograms to compare starting models (e.g., Zhou et al., 2021), this is the first study to compare results after inversion. We chose CSEM_NA (Krischer et al., 2018), SPiRaL (Simmons et al., 2021), and WUS256 (Rodgers et al., 2022) as starting models and iterated each model to a minimum period of 20 seconds. We then compared the results of the final models using five comparison metrics. All five of the comparison metrics showed that the final models all resolved tectonic structure and fit observed data similarly, regardless of the structure or fit of the starting models. Therefore, we conclude that the choice of starting model has minimal effect on the final model results.

Errors in source mechanisms can produce bad synthetic seismograms, which limit the data available to use in adjoint waveform tomography inversions. Many researchers use source mechanisms from the Global Centroid Moment Tensor (GCMT) catalogue. However, the GCMT catalogue has known issues with resolving shallow crustal earthquakes. Since shallow crustal earthquakes dominate CANVAS’s dataset, we inverted for source mechanisms using MTTime, a time-domain moment tensor inversion code. We calculated 3D Green’s functions using an intermediate version of CANVAS that was iterated to a minimum period of 20 seconds. We showed that the inverted source mechanisms greatly improved waveform fit compared to the GCMT solutions, particularly at distant stations. We also demonstrated that 3D Green’s functions have better and more consistent waveform fits compared to solutions inverted using 1D Green’s functions.

Before introducing shorter period data in CANVAS, we replaced the original GCMT solutions with the inverted solutions. Synthetics calculated using the inverted source mechanisms better matched the amplitude of the observed data, and greatly increased the amount of windowed data that could be included in the inversion of CANVAS. Therefore, we advocate for the importance of regularly updating source mechanisms in adjoint waveform tomography workflows. We conclude by proposing CANVAS, an adjoint waveform tomography model of California and Nevada iterated to a minimum period of 12 seconds. Though CANVAS has coarse resolution compared to other crustal tomography models, it can resolve the geographic extents and velocities of well-known tectonic features throughout California and Nevada. The model particularly excels at fitting the dispersed surface waves, showing significant improvement compared to its starting model, WUS256 (Rodgers et al., 2022). Finally, CANVAS also accurately resolves the depth to basement of large basins in California without a priori basin information. This highlights the inherent information contained in seismic waveforms, which full waveform inversion tomography is exceptional at harnessing. We hope that CANVAS can be used as a starting model for smaller regional tomography studies that could eventually become input velocity models for stress inversions or rupture propagation models.