Global upper-mantle tomography with the automated multimode inversion of surface and S-wave forms

We apply the Automated Multimode Inversion of surface and S-wave forms to a large global data set, verify the accuracy of the method and assumptions behind it, and compute an S-v-velocity model of the upper mantle (crust-660 km). The model is constrained with similar to 51 000 seismograms recorded at 368 permanent and temporary broadband seismic stations. Structure of the mantle and crust is constrained by waveform information both from the fundamental-mode Rayleigh waves (periods from 20 to 400 s) and from S and multiple S waves (higher modes). In order to enhance the validity of the path-average approximation, we implement the automated inversion of surface- and S-wave forms with a three-dimensional (3-D) reference model. Linear equations obtained from the processing of all the seismograms of the data set are inverted for seismic velocity variations also relative to a 3-D reference, in this study composed of a 3-D model of the crust and a one-dimensional (1-D), global-average depth profile in the mantle below. Waveform information is related to shear- and compressional-velocity structure within approximate waveform sensitivity areas. We use two global triangular grids of knots with approximately equal interknot spacing within each: a finely spaced grid for integration over sensitivity areas and a rougher-spaced one for the model parametrization. For the tomographic inversion we use LSQR with horizontal and vertical smoothing and norm damping. We invert for isotropic variations in S- and P-wave velocities but also allow for S-wave azimuthal anisotropy-in order to minimize errors due to possible mapping of anisotropy into isotropic heterogeneity. The lateral resolution of the resulting isotropic upper-mantle images is a few hundred kilometres, varying with data sampling. We validate the imaging technique with a 'spectral-element' resolution test: inverting a published global synthetic data set computed with the spectral-element method using a laterally heterogeneous mantle model we are able to reconstruct the synthetic model accurately. This test confirms both the accuracy of the implementation of the method and the validity of the JWKB and path-average approximations as applied in it. Reviewing the tomographic model, we observe that low-S-v-velocity anomalies beneath mid-ocean ridges and backarc basins extend down to similar to 100 km depth only, shallower than according to some previous tomographic models; this presents a close match to published estimates of primary melt production depth ranges there. In the seismic lithosphere beneath cratons, unambiguous high velocity anomalies extend to similar to 200 km. Pronounced low-velocity zones beneath cratonic lithosphere are rare; where present (South America; Tanzania) they are neighboured by volcanic areas near cratonic boundaries. The images of these low-velocity zones may indicate hot material-possibly of mantle-plume origin-trapped or spreading beneath the thick cratonic lithosphere.