Effect of Merging Multiscale Models on Seismic Wavefield Predictions Near the Southern San Andreas Fault

Abstract

Physics-based probabilistic seismic hazard analysis (PSHA) depends on a good understanding of earthquake processes that are described using rupture models and accurate structural models of the earth, used in 3D deterministic wave propagation simulations to predict ground motions (Milner et al., 2021;B. E. Shaw et al., 2018). The hazard information is utilized in tandem with engineering data to produce risk assessments that form an integral part of building codes and can help preserve the structural integrity of existing infrastructure in the occurrence of a large earthquake (Maechling et al., 2007). In general, areas with low seismic velocities, such as sedimentary basins, are prone to more vigorous and prolonged shaking (Bijelic et al., 2019;Brissaud et al., 2020;Graves et al., 1998;Lovely et al., 2006) and are thus sites where improvements in the shallow parts of structural models can be most beneficial. According to the latest Uniform California Earthquake Rupture Forecast (UCERF3) model of California, the southern San Andreas fault poses the greatest threat in southern California, with the highest probability of a large-magnitude event in the next few decades (Field et al., 2013). Several scenario simulations of the San Andreas fault rupture show significant amplification in the sedimentary basins in the nearby Salton Trough and the densely populated Los Angeles region located further north along the fault (Day et al., 2012;Jones et al., 2008;Porter et al., 2011). Recent modeling also suggests additional complexity through the San Andreas fault's potential simultaneous rupture with the nearby San Jacinto fault (Lozos, 2016) (Figure 1). To be better prepared,

Earth model-space exploration in Southern California: Influence of topography, geotechnical layer, and attenuation on wavefield accuracy

“…From our results, we are reminded that beyond simple models with analytical solutions, predictions about the performance of 3D heterogeneous Earth models should generally be avoided unless the claims are explicitly validated. Due to the complexity of model validation, the results presented are strictly valid only for the earthquakes-station distribution we have used, and our choice of the normalized classical waveform misfit function (Ajala & Persaud, 2021). Other error quantifiers such as Goodness-of-Fit (Olsen & Mayhew, 2010) favored by engineers or time-frequency misfit (Kristekovà et al, 2009) can and should be explored to check if the results are globally equivalent.…”

“…The model space can be equipped with a norm m(x) p that measures the size of each model in the space. Ajala and Persaud (2021) illustrate a step in the direction of showing the convexity of the model space, as the linear interpolation between two elements of the space yields another member. The model space is infinite.…”

“…Our focus is on matching the observed waveforms. We refer to Ajala and Persaud (2021) for more details regarding the simulation setup.…”

“…To this end, we validate several representations of two Earth models hosted by the Southern California Earthquake Center (SCEC) by measuring local full-waveform misfits between synthetic and observed seismograms at sites with broadband seismometers. The validation methodology in this research follows Ajala and Persaud (2021), including a subset of their earthquakes. Compared to Aagaard et al (2008), we do not examine source effects and use a single wave propagation solver to investigate all model representations.…”

Parsimonious Velocity Inversion Applied to the Los Angeles Basin, CA

“…The Love wave data set then updates the CVM prior into an approximate posterior model which includes the influence of both the new data and the data that went into the CVM via its expression in the CVM model. At a global scale, a similar scheme of local quasi-Bayesian model refinement has been proposed by Fichtner et al (2018), and within the SCEC CVM framework Ajala and Persaud (2021) have proposed a means of blending local updates into existing regional models -this work differentiates itself by its data-driven choice of model updating region, consistent with estimated data uncertainty. Integration of local models within the SCEC CVM framework will become an important part of hazard modeling within Southern California as high-density arrays allow access to the fine scale detail of path effects.…”