← Back to Sessions

LLNLGlobeFWI: First Iterations Using a Semi-Automatic FWI Framework Applied to the Globe With Spiral as the Starting Model

Description: 

Building on the global anisotropic travel time tomography model, SPiRaL (Simmons et al., 2021), we present the first global full waveform inversion (FWI) updates to SPiRaL using an initial set of 103 events distributed throughout the globe at a period band from 40 s to 200 s. We employed LLNLGlobeFWI which is a python-based framework for semi-automated global full waveform inversion using the adjoint method. LLNLGlobeFWI was developed at Lawrence Livermore National Laboratory (LLNL) and used SeisFlows (Modrak et al., 2018) as a starting point for the development. LLNLGlobeFWI was executed on LLNL’s Quartz supercomputer and 90-minute seismograms were computed for the events to capture seismic phases at global distances. Seismograms were windowed with Pyflex, multitaper misfits and associated adjoint sources were computing using Pyadjoint, and the L-BFGS method was used for the optimization within the custom semi-automated workflow. Accurate prediction of waveforms and onset arrival times at all distances with a single global model is the overall goal of this project. Here, we present the first full waveform inversions aimed at reaching that goal starting with the SPiRaL model and an initial set of global events. Future work will include the incorporation of a larger data set, period band reduction, and benchmarking with SPiRaL to ensure the onset times are improving as well.

This work was performed under the auspices of the U.S. Department of Energy by Lawrence

Livermore National Laboratory under Contract DE-AC52-07NA27344.

Session: 3D Wavefield Simulations: From Seismic Imaging to Ground Motion Modelling - III

Type: Oral

Date: 5/2/2024

Presentation Time: 02:30 PM (local time)

Presenting Author: Luis

Student Presenter: Yes

Invited Presentation: 

Authors