[S-CG45] Science of slow-to-fast earthquakes

[SCG45-10] Adjoint tomography of the Nankai and Kyushu subduction zones - Waveform misfits and assessment of existing 3D velocity models

*Samriddhi Prakash Mishra¹, Yoshihiro Kaneko¹, Bryant Chow², Shun Adachi¹, Yusuke Yamashita³, Masanao Shinohara⁴ (1.Graduate School of Science, Kyoto University, 2.University of Alaska – Fairbanks, Alaska, USA, 3.Disaster Prevention Research Institute, Kyoto University, Kyoto, Japan, 4.Earthquake Research Institute, University of Tokyo, Tokyo Japan)

Keywords:seismic tomography , full waveform inversion, adjoint methods, subduction zones, Nankai Trough

The Nankai-Kyushu subduction system is a complex seismotectonic region marked by an abrupt transition in the inter-plate coupling, with a fully coupled megathrust to the northeast. In this study, we utilize an earthquake-based, full-waveform inversion technique, termed Adjoint Tomography (Tromp et al., 2005; Fichtner et al., 2006; Tape et al., 2007), to develop an accurate and high-resolution P-wave and S-wave velocity models of the Nankai-Kyushu subduction zone. We aim to reveal the crustal structure responsible for the origin of the inter-plate coupling transition. Our target region includes ~100 regional earthquakes (4.5 < Mw <6) with ~ 400 permanent receivers, DONET, and ~20 temporary Ocean Bottom Seismometers (OBS) stationed in the Hyuga-nada region. We first assess the accuracy of two candidate initial velocity models (Koketsu et al., 2012; Bassett et al., 2022). For the Bassett et al. (2022) model, we use empirical relations to convert P-wave velocities to obtain the corresponding S-wave velocity model. Using a spectral element solver, we simulate the earthquakes and compare the resulting synthetics with recorded waveforms at 6 – 40 s period ranges. Our results suggest that seismic waves are systematically amplified by accretionary sediments. We also perform a quantitative misfit analysis, incorporating parameters such as time shifts, amplitude ratios, peak cross-correlation, and apparent velocities. Our analysis reveals that large (> 3 s) travel-time misfits emerge in these initial models. Nevertheless, the overall time shifts and amplitude ratios, high peak cross-correlation values, and even distribution of phase measurements among components suggest that both models are reasonable starting models for computationally intensive adjoint tomography. We will present our ongoing efforts towards developing an updated velocity model of the study region that is more accurate and has higher resolution than the existing velocity models.