1. Abstract of the G-Cubed article

We conduct a comprehensive and quantitative analysis of similarities and differences between recent seismic tomography models of the Earth's mantle in an attempt to determine a benchmark for geodynamic interpretation. After a spherical harmonic expansion, we find the spectral power and radial correlation of each tomographic model as a function of depth and harmonic degree. We then calculate the correlation, at the same depths and degrees, between all possible pairs of models, to identify stable and model-dependent features (the former being usually of longer spatial wavelength than the latter). We can therefore evaluate the degree of robust structure that seismologists have mapped so far and proceed to calculate ad hoc mean reference models. Tomographic models are furthermore compared with two geodynamic subduction models that are based on plate motion reconstructions. We find systematically low intermediate-wavelength correlation between tomography and convective reconstruction models and suggest that the inadequate treatment of the details of slab advection is responsible. However, we confirm the presence of stable, slab-like fast anomalies in the mid-mantle whose geographic pattern naturally associates them with subduction. This finding -in addition to our analysis of heterogeneity spectra and the absence of strong minima in the radial correlation functions besides the one at $\sim 700$ km- supports the idea of whole mantle convection with slab penetration through the 660-km phase transition, possibly accompanied by a reorganization of flow.