Search for: All records

Accurately predicting the seismic wavefield is important for physics-based earthquake hazard studies and is dependent on an accurate source model, a good model of the subsurface geology, and the full physics of wave propagation. Here, we conduct numerical experiments to investigate the effect of different representations of the Southern California Earthquake Center and Harvard community velocity models on seismic waveform predictions in the vicinity of the San Andreas fault in Salton Trough. We test general preconceptions about the importance of topography, near-surface geotechnical layering, and anelastic attenuation up to a maximum frequency of 0.5 Hz. For the Southern California Earthquake Center model developed without topography, we implement 1D and linear model extensions that preserve the geologic structure and a pull-up approach that adapts the original model to topographic variations and distorts the subsurface. The Harvard model includes an elevation model, so we test the squashed topography representation, which flattens it. For both community models, we modify the top 350 m by partially applying the Ely geotechnical layer using a minimum shear wave velocity of 600 m/s and incorporate an Olsen attenuation model using a ratio of 0.05. We evaluate the resulting 24 model representations using the classical waveform misfit and five moderate-magnitude earthquakes. Only the inclusion of attenuation consistently improves the wavefield predictions. It becomes more impactful at higher frequencies, where it significantly improves the performance levels of the crude 1D and linear extension models close to that of the original version. The pull-up topography representation also enhances the waveform prediction ability of the original model. Squashing the topography of the elevation-referenced Harvard model produces better seismogram fits, suggesting that seismic imagers construct community tomographic models without topography to avoid issues related to missing model parameters near the free surface or discrepancies with a different elevation model. Although full implementation of the Ely geotechnical layer that would permit shear wave velocities as low as 90 m/s proves computationally expensive, our partial implementation provides slightly better results in some cases. Our results can serve as recommendations for implementing these community models for future validation or optimization studies. 

Abstract The Indo‐Burman subduction zone represents a global endmember for extreme sediment accretion and is a region characterized by ambiguous tectonic structure. The recent collection of broadband seismic data across the Indo‐Burman accretionary margin as part of the Bangladesh‐India‐Myanmar Array (BIMA) experiment provides an opportunity to investigate the subsurface velocity structure across the incoming plate of an endmember subduction system. We construct a three‐dimensional model for seismic shear velocity using a joint inversion of surface‐ and scattered‐wave constraints. Rayleigh‐wave phase velocities measured from ambient‐noise (12–25 s) and teleseismic earthquakes (20–80 s) constrain absolute shear velocities, while we constrain the locations of and relative contrasts across significant discontinuities in the subsurface using observations from scattered‐wave imaging. From the resulting inversion, we observe two model classes that characterize the evolution of consolidation within the markedly slow uppermost sediments and metasediments along a predominantly southwest‐to‐northeast trend. We interpret variations in deeper seismic structure under two proposed scenarios: (a) a Moho of ∼21–26 km depth underlying a package of metasediments and a thinned basement component, with a slow mantle lithosphere (4.2 km/s) that may contain retained melt from the onset of India‐Antarctica seafloor spreading; or (b) a Moho of ∼51–59 km depth underlying a package of metasediments, basement, and a thick slug of mafic material, which may correspond to significant Kerguelen‐plume‐related underplating. By combining constraints from highly resolved phase‐velocity estimates and scattered‐wave images, we successfully characterize the lateral transitions across the Indo‐Burman forearc margin. 

Abstract Cost-effective strategies for enhancing seismic velocity models are an active research topic. The recently developed hybridization technique shows promise in improving models used for deterministic earthquake hazard evaluation. We augment the results of Ajala and Persaud (2021) by exploring other hybrid models generated using 13 sets of embedding parameters—taper widths and subvolumes—and summarize their effect on waveform predictions up to a minimum period of 2 s. Our results introduce the notion of compatibility as a consideration by showing that the same basin models embedded into two different regional models can produce notably different outcomes. In contrast to most of our hybrid Harvard models that produce better matching ground motions, only one of the hybrid models generated using the Southern California Earthquake Center model as a regional model gives a closer match to the waveforms. Similar results are obtained at higher frequencies; however, improvements due to hybridization are reduced. A potential explanation for these results may be the limited high spatial frequencies in the travel time tomography basin models and the >5–6 s wavefield-dominated adjoint regional models. Although the strongly tapered compatible hybrid models tend to produce better results, we find instances of improvements even with merging artifacts.