Seismic tomography is instrumental in mapping 3D velocity structures of the Earth’s interior based on travel-time measurements and waveform differences. Although both ray theory and other asymptotic methods have been successfully employed in global tomography, they are less accurate for long-period waves or steep velocity gradients. They also lack the ability to predict “non-geometrical” effects such as those for the core diffracted phases (Pdiff, Sdiff) which are crucial for mapping heterogeneities in the lowermost mantle (D00 layer). On the other hand, sensitivity kernels can be accurately calculated with no approximations by the interaction of forward and adjoint wavefields, both numerically simulated by spectral element methods.

Using a combination of these pre-computed sensitivity kernels and picks of over 470 000 S and 660 000 P phases, we perform a parallel LSQR inversion which generates an updated tomographic model with 1 degree resolution. Through this improved coverage, we investigate differences between global models inverted based on ray theory and finite-frequency kernels.