5:00 PM - 6:30 PM
[S07P-03] Toward adjoint tomography of the large low seismic velocity provinces beneath the western Pacific Ocean
Introduction:It is well-known that there is a large low seismic velocity provinces (LLSVP) where the seismic velocity changes sharply above the core-mantle boundary (CMB), however it is unclear how it is distributed above the CMB. In recent years, the detailed S-wave velocity structure around the CMB was inferred from waveform inversion [1, 2]. In our previous study, we found an NE-SW running low S-wave velocity structure within the LLSVP lying beneath the western Pacific Ocean using analyses of travel times and the finite frequency inversion. To clarify the detail structure and its relationship with old subduction, we determine a three-dimensional (3D) P and S-wave velocity structure beneath the western Pacific Ocean based on waveform tomography using the combination of Spectral-element method (SEM) [3] and adjoint method [4].We show here some preliminary results with computational costs.
DATA and Method:We collected the seismic waveforms of three components obtained at seismic broadband stations from 102 earthquakes occurred from 2008 to 2015. These events are chosen so as to distribute evenly in space. Distribution of seismic stations used in this study is shown in Fig. 1. Removing the response of each instrument from the original records, we transferred into displacement waveforms with a bandpass filter (10-150 sec) to use the inversion. We constructed global Earth model based on S40RTS [5] as an initial 3D model in inversion to calculate synthetic displacements and adjoint waveforms using the SEM. Our SEM structural model was parameterized using 1.6 billion grid points to produce waveforms of a shortest period of 10 sec. The forward and adjoint simulation were conducted using a flat 600 Message Passing Interface (MPI) on Data Analyzer (DA) systemin Japan Agency for Marine-Earth Science and Technology (JAMSTEC). The model parameters Vp and Vs were updated iteratively by gradient method using the misfit kernels (Fig. 2) based on the adjoint sources to minimize the misfit between observed waveforms and synthetic waveforms. Here, we show some preliminary results: estimation of computational costs, forward modelling, and trial inversion.
Results and Discussion:This study requires large computational cost to calculate forward and backward waveforms in the 3D whole Earth. We estimated the calculation time on the DA system. The time step of the solver was 0.1045 sec and 30 min waveforms were calculated. As a result, the calculation time was required approximately 2 and 9 hours for conducting forward and adjoint simulation, respectively for each event. If the calculation is conducted as 4 independent parallel job, it takes 3 weeks to obtain one iterative results when ~200 events are used in inversion. We succeed construction of the iterative model using the misfit kernels of a few events as trial, synthetic waveforms by iterative model were improved clearly. We are planning to apply differential travel times of some core phases. we are expected to revise the LLSVP from initial 3D model more efficient.
Acknowledgements:We used seismic waveforms obtained by IRIS and NIED F-net stations and SPECFEM3D Globe package used in this study.
References:[1] Konishi, K., Kawai, K., Geller, R.J., and Fuji, N., Geophy. J. Int., 199, 1245-1267 (2014). [2] Yuan, K. and Romanowicz, B., Science, 357, 393-397 (2017). [3] Komatitsch, D. and Tromp, J., Geophy. J. Int., 149, 390-412 (2002). [4] Tromp, J., Tape, C., and Liu, Q., Geophy. J. Int., 160, 195-216 (2005). [5] Ritsema, J., Deuss, A., van Heijst, H. J., and Woodhouse, J. H., Geophy. J. Int., 184, 1223-1236 (2011).
DATA and Method:We collected the seismic waveforms of three components obtained at seismic broadband stations from 102 earthquakes occurred from 2008 to 2015. These events are chosen so as to distribute evenly in space. Distribution of seismic stations used in this study is shown in Fig. 1. Removing the response of each instrument from the original records, we transferred into displacement waveforms with a bandpass filter (10-150 sec) to use the inversion. We constructed global Earth model based on S40RTS [5] as an initial 3D model in inversion to calculate synthetic displacements and adjoint waveforms using the SEM. Our SEM structural model was parameterized using 1.6 billion grid points to produce waveforms of a shortest period of 10 sec. The forward and adjoint simulation were conducted using a flat 600 Message Passing Interface (MPI) on Data Analyzer (DA) systemin Japan Agency for Marine-Earth Science and Technology (JAMSTEC). The model parameters Vp and Vs were updated iteratively by gradient method using the misfit kernels (Fig. 2) based on the adjoint sources to minimize the misfit between observed waveforms and synthetic waveforms. Here, we show some preliminary results: estimation of computational costs, forward modelling, and trial inversion.
Results and Discussion:This study requires large computational cost to calculate forward and backward waveforms in the 3D whole Earth. We estimated the calculation time on the DA system. The time step of the solver was 0.1045 sec and 30 min waveforms were calculated. As a result, the calculation time was required approximately 2 and 9 hours for conducting forward and adjoint simulation, respectively for each event. If the calculation is conducted as 4 independent parallel job, it takes 3 weeks to obtain one iterative results when ~200 events are used in inversion. We succeed construction of the iterative model using the misfit kernels of a few events as trial, synthetic waveforms by iterative model were improved clearly. We are planning to apply differential travel times of some core phases. we are expected to revise the LLSVP from initial 3D model more efficient.
Acknowledgements:We used seismic waveforms obtained by IRIS and NIED F-net stations and SPECFEM3D Globe package used in this study.
References:[1] Konishi, K., Kawai, K., Geller, R.J., and Fuji, N., Geophy. J. Int., 199, 1245-1267 (2014). [2] Yuan, K. and Romanowicz, B., Science, 357, 393-397 (2017). [3] Komatitsch, D. and Tromp, J., Geophy. J. Int., 149, 390-412 (2002). [4] Tromp, J., Tape, C., and Liu, Q., Geophy. J. Int., 160, 195-216 (2005). [5] Ritsema, J., Deuss, A., van Heijst, H. J., and Woodhouse, J. H., Geophy. J. Int., 184, 1223-1236 (2011).