Global 3D model of mantle attenuation using normal modes

Seismic tomographic models based only on wave velocities have limited ability to differentiate between a compositional or temperature origin for the Earth's 3D structure variations. Complementing wave velocities with attenuation (conversion of energy to heat) can help make that distinction, which is fundamental to understand mantle convection evolution. For example, a thermal origin for the lower mantle large low shear velocity provinces (LLSVPs) will point to them being short-lived anomalies, whereas a compositional origin will point to them being long-lived, forming stable 'anchors' and influencing the pattern of mantle convection. So far, only global 3D attenuation models built using seismic body waves and surface waves have been available for the upper mantle. Here, we use whole Earth oscillations or normal modes to measure 3D variations in mantle attenuation, which allow us to include focussing and scattering without the need for approximations. We achieve this by jointly measuring 3D variations in velocity and attenuation using splitting functions, which are depth-averaged models of how a mode 'sees' the Earth. 

Splitting functions are linearly dependent on heterogeneous structure and can be easily incorporated in tomographic models. We measured 14 anelastic splitting functions and used those to build a 3D global model of attenuation for the whole mantle. For comparison purposes, we have also constructed a 3D shear-velocity model using the same number of modes and model parametrization. In the upper mantle, we find high attenuation in the low velocity spreading ridges, which suggests a thermal origin and agrees with previous surface wave studies. In the lower mantle, we find the highest attenuation in the 'ring around the Pacific' high velocity region, which is thought to be the 'graveyard' of subducted slabs, and not in the LLSVPs beneath Africa and the Pacific. We compare our 3D attenuation model to the wave-speeds and attenuation predictions of a laboratory-based viscoelastic model. Our comparison indicates that the higher attenuation seen in the slab regions can be explained by a small grain-size in combination with cold temperatures, while the lower attenuation in the LLSVPs can be explained by a large grain-size in combination with high temperatures. Grain-size is related to viscosity in diffusion creep, which would mean that the LLSVPs have larger viscosity making them long-lived stable features, while the graveyard of slabs would have a lower viscosity making them shorter lived.