Towards the Next Generation Upper-Mantle 3D Anelastic Tomography

In order to distinguish the thermal and compositional heterogeneities in the mantle, it is crucial to resolve the lateral variations not only in seismic velocities but also in intrinsic attenuation. Indeed, the high sensitivity of intrinsic attenuation to temperature and water content, governed by a form of Arrhenius equation, contrasts with the quasi-linear dependence of velocities on both temperature and major element composition. The major challenge in imaging attenuation lies in separating its effects on seismic waves from the elastic ones. The latter originate from the wave propagation in media with strong lateral elastic gradients causing (de)focusing and scattering. We have previously developed a 3D upper-mantle shear attenuation model based on time domain waveform inversion of long period (T > 60s) fundamental and overtone surface wave data (Gung & Romanowicz, 2004). However, at that time, resolution was limited to very long wavelength structure, because elastic models were still rather smooth, and the effects of focusing could only be estimated approximately, using asymptotic normal mode perturbation theory.With recent progress in constraining global mantle shear velocity from waveform tomography based on the Spectral Element Method (e.g. SEMUCB_WM1, French & Romanowicz, 2014), we are now in a position to develop an improved global 3D model of shear attenuation in the upper mantle. In doing so, we use a similar time domain waveform inversion approach, but (1) start with a higher resolution elastic model with better constraints on lateral elastic gradients and (2) jointly invert, in an iterative fashion, for shear attenuation and elastic parameters. Here, we present the results of synthetic tests that confirm our inversion strategy, as well as preliminary results towards the construction of the next generation upper-mantle anelastic model.