Increasing resolution of lithospheric images by full-waveform inversion of teleseismic data

Abstract:

Building high-resolution models of elastic properties from the Earth's lithosphere is essential to improve our understanding of tectonic processes in the deep crust and the upper mantle. Up to now, conventional imaging methods such as traveltime tomography and receiver function analysis suffer either from a lack of spatial resolution in the tomographic velocity model or from inaccuracies of the reflectivity images. Nowadays, deployment of dense networks of broadband stations and advances in high-performance computing open new perspectives to overcome these limitations by applying full-waveform inversion (FWI) to teleseismic wavefields. FWI consists in using the full information content in the data by fitting the full seismograms, namely, the phase and amplitude of each arrival transmitted across or reflected from the lithospheric heterogeneities. Combining the transmission and reflection propagation regimes (or the forward- and backward scattering regimes) in the FWI is a key issue to achieve the desired spatial resolution in lithospheric imaging and to uncouple the signature of multiple classes of parameters such as P and S wavespeeds and density in the seismic wavefields.

In this study, we illustrate the promises and pitfalls of teleseismic FWI for high-resolution lithospheric imaging with a synthetic case study representative of the South-Western Alps. Three-dimensional elastic full waveform modeling in the lithospheric target located below the receiver array is performed with the spectral element method. A grid injection method is used to inject the teleseismic wavefield computed in a global axisymmetric earth at the boundaries of the lithospheric target. Model parameter updates involving the P and S wavespeeds as well as density are performed in the lithospheric target by quasi-Newton local optimization where the gradient of the misfit function is computed with the adjoint-state method. The footprint of several experimental factors such as noise, parameter leakage, unknown temporal source function, inaccurate initial model and non-even source/receiver acquisition on the FWI results will be discussed. We shall also assess a preliminary lithospheric FWI model of the South-Western Alps inferred from CIFALPS dataset against previous geophysical investigations of the area.