3D Lithospheric Imaging by Time-Domain Full-Waveform Inversion of Teleseismic Body-Waves

Abstract:

With the deployment of dense seismic arrays and the continuous growth of computing facilities, full-waveform inversion (FWI) of teleseismic data has become a method of choice for high-resolution lithospheric imaging. FWI can be recast as a local optimization problem that seeks to estimate Earth's elastic properties by iteratively minimizing the misfit function between observed and modeled seismograms.
In passive teleseismic configurations, the seismic source no longer corresponds to a point source embedded in the targeted medium but rather corresponds to a wavefront incoming from the outside of the model. We develop a 3-dimensional time-domain full-waveform inversion program that is more designed for this configuration. The gradient of the misfit function is efficiently computed with the adjoint-state method. A velocity-stress finite-difference time-domain modeling engine, which is interfaced with the so-called total-field/scattered-field method, is used to propagate in the targeted medium the incident wavefield inferred from a global Earth simulation (AxiSEM). Such interfacing is required to account for the multiple arrivals in the incoming wavefield and the sphericity of the Earth. Despite the limited number of nearly plane-wave sources, the interaction of the incident wavefield with the topography (P-Sv conversions and P-P reflections acting as secondary sources) provides a suitable framework to record both transmitted wavefields and reflected wavefields from lithospheric reflectors. These recordings of both transmitted and reflected waves makes FWI amenable to a broadband-wavenumber (i.e., high resolution) reconstruction of the lithosphere.
Feasibility of the method is assessed with a realistic synthetic model representative of the Western Alps. One key issue is the estimation of the temporal source excitation, as there might be some trade-off between the source estimation and the subsurface update. To avoid being trapped in a local minimum, we follow a multiscale approach, which hierarchically proceeds from the low frequencies to the higher ones and from the transmission regime to the reflection one. A real data case study will focus on the inversion of the CIFALPS array data up to a maximum frequency of 1 Hz to investigate the lithospheric structure of the Alps.