Multi-parameter high-resolution lithospheric imaging by source-independent full-waveform inversion of teleseismic data

Abstract:

Building broadband multi-parameter lithospheric models is one of the quest of earthquake seismology. Nowadays, deployment of dense arrays of broadband stations and advances in high-performance computing open new perspectives to achieve this goal by full waveform inversion (FWI) of teleseismic data. Compared to traveltime tomography, broadband images can be obtained by FWI when wavefields that are forward-scattered (i.e., transmission regime) and backward-scattered (reflection regime) by lithospheric heterogeneties to be imaged are involved in the inversion. In teleseismic setting, incident wavefields impinge the boundaries of the lithospheric target and propagate up to the surface where they are recorded by the stations, giving rise to the transmitted part of the recorded wavefield. The incident wavefield is also reflected back into the lithospheric target by the free surface acting as P- and S-waves secondary sources. The resulting wavefield is reflected by the lithospheric reflectors before being recorded by the stations, giving rise to the second-order reflection part of the recorded wavefield. While the transmitted part of the wavefield allows one to achieve a resolution close to that obtained by traveltime tomography, involving the reflected part of the wavefield in the FWI is amenable to the short-wavelength updates, hence broadaning the wavenumber spectrum of the lithospheric models toward high wavenumbers. Another benefit to involve the reflection regime in FWI is to increase the sensitivity of the FWI to the density parameter.

In this study, we first discuss the feasibility of the density reconstruction in addition to that of the P- and S-waves velocities by FWI of teleseismic wavefields with a realistic synthetic study representative of the western Alps. The density reconstruction implies the extraction of information given by small amplitude secondary wavefields from the data that may be drastically affected by noise and trade-off between model parameter update and inaccurate temporal source signature. To overcome this latter issue, we describe a method that makes FWI independent to source signature. This approach, originally designed for exploration geophysics applications, is adapted to earthquake seismology.