Slabs in the lower mantle - results of dynamic modelling compared with tomographic images and the geoid

The recent increase in resolution of tomographic images of the Earth's interior has enabled us to see slabs in the lower mantle. On the other hand, the distribution of slabs can be inferred from subduction history, using a dynamical model of mantle flow. Comparison of tomographic images with model results can help to distinguish between alternative models of subduction history, tomographic models, and mantle flow models and thus improve our understanding of the Earth. Here, a simple dynamic model of subduction is used and some results are presented: Amounts and locations of subduction are inferred from published models of plate motion and plate boundaries for the past 120 Ma; the latter have been interpolated on 2 Ma time intervals. Mantle flow driven by the density anomalies corresponding to subducted slabs is calculated with the method of Hager and O'Connell for a viscosity structure that features an increase in viscosity from 4 . 10(20) Pa below the lithosphere to 3 . 10(22) Pa in the lower part of the lower mantle. Slabs are advected in the now field. In the model, slabs sink on average about 1700 km through the mantle during 100 Ma; however, the calculated average horizontal motion of slabs throughout most of the mantle is only about 600-700 km; for slabs that have sunk to the lowermost mantle it is about 1000 km. This is in accordance with the previously observed good correlation between subduction locations and lowermost mantle heterogeneities. Slabs sink fastest in regions with abundant subduction and little trench migration (e.g., below East Asia). In the lowermost mantle, predicted slab locations tend to lie in areas with high seismic velocities, however, there are large areas of fast seismic velocities where no slabs are predicted, and which therefore presumably correspond to earlier subduction. In some regions (e.g., below Japan), our model approximately reproduces the observed shape of the slab. In other regions, where the agreement is poor, the results can help to further constrain models of subduction history and associated mantle flow. Additionally, the predicted geoid is computed and compared with the actual geoid. Predicted and observed geoid have similar magnitude and a number of features in common, however, an optimization of the fit is not attempted here, since, in order to do this, also the hot upwellings from the lower boundary layer (presumably in the form of plumes) would have to be included. (C) 2000 Elsevier Science B.V. All rights reserved.