Full-3D Tomography of the Crustal Structure in Southern California Using Earthquake Seismograms and Ambient-Noise Correlagrams

Tuesday, 16 December 2014
En-Jui Lee1, Po Chen2, Thomas H Jordan1,3, Philip J Maechling1, Marine Denolle4 and Gregory C Beroza5, (1)Southern California Earthquake Center, Los Angeles, CA, United States, (2)University of Wyoming, Laramie, WY, United States, (3)University of Southern California, Los Angeles, CA, United States, (4)Scripps Institution of Oceanography, IGPP, La Jolla, CA, United States, (5)Stanford University, Stanford, CA, United States
We have constructed a high-resolution model for the Southern California crust, CVM-S4.26, by inverting more than half-a-million waveform-misfit measurements from about 38,000 earthquake seismograms and 12,000 ambient-noise correlagrams. The inversion was initiated with the Southern California Earthquake Center’s Community Velocity Model, CVM-S4, and seismograms were simulated using K. Olsen’s staggered-grid finite-difference code, AWP-ODC, which was highly optimized for massively parallel computation on supercomputers by Y. Cui et al. We navigated the tomography through 26 iterations, alternating the inversion sequences between the adjoint-wavefield (AW) method and the more rapidly converging, but more data-intensive, scattering-integral (SI) method. Earthquake source errors were reduced at various stages of the tomographic navigation by inverting the waveform data for the earthquake centroid-moment tensors. All inversions were done on the Mira supercomputer of the Argonne Leadership Computing Facility. The resulting model, CVM-S4.26, is consistent with independent observations, such as high-resolution 2D refraction surveys and Bouguer gravity data. Many of the high-contrast features of CVM-S4.26 conform to known fault structures and other geological constraints not applied in the inversions. We have conducted several other validation experiments, including checking the model against a large number (>28,000) of seismograms not used in the inversions. We illustrate this consistency with the excellent fits at low frequencies (≤ 0.2 Hz) to three-component seismograms recorded throughout Southern California from the 17 Mar 2014 Encino (MW4.4) and 29 Mar 2014 La Habra (MW5.1) earthquakes, and we show these fits to be much better than those obtained by two community velocity models in current use, CVM-S4 and CVM-H11.9. We conclude by describing some of the novel features of the CVM-S4.26 model, which include unusual velocity reversals in some regions of the mid-crust.