Abstract
One-dimensional seismic reference models for the mantle, such as PREM and AK135, are widely applied throughout the earth sciences, from 3-D tomography through to mineral physics and geodynamical simulations. In such cases the models are typically assumed to represent the "average" themochemical structure of the mantle. Using recent mineral physics compilations to construct synthetic velocity profiles for different thermochemical models, and comparing them with a range of 1-D seismic data types, we tried to decifer what such a structure might be. However it soon became clear that, given the current mineral physics constraints, it is very difficult to fit all the data with a smooth temperature profile and uniform chemical composition. The 1-D average thermochemical structure appears not only to vary with depth, but the cause of such variations is most easily explained by 3-D chemical heterogeneity. Furthermore, there are distinct differences between the upper and lower mantle. In the upper mantle, seismic data indicate an enrichment in basaltic component between 250 and 350 km, and an enrichment in harzburgitic component immediately below 660 km, as well as chemical changes within the transition zone. In the lower mantle, 1-D seismic profiles are much smoother, and the differences between seismic data and a near-adiabatic, chemically homogeneous mantle are smaller. However with increasing depth it is increasingly difficult to fit the velocity gradients with a homogeneous structure, and there are indications of superadiabatic, basalt-enriched material contributing to the 1-D average. Such material could correspond to the large, low shear velocity provinces (LLSVPs) observed in seismic tomography. Our results emphasise that one-dimensional seismic reference models should not be regarded as some background structure relative to which there are faster and slower regions, but rather a weighted average of the 3-D structure which can be biased by certain thermochemical components. Future seismic investigations should focus on the physical interpretation of three-dimensional seismic data sets, and will be aided by an increasing accuracy and breadth of the mineral physics data compilations.