We present a detailed study of the advective contribution to the radial flow of heat in the mantle as deduced using a compressible internal loading theory in which the flow is assumed to be driven by the density heterogeneities implied by recent global seismic tomographic models. We calculate the radial flow velocity response of a viscous mantle and find that for reasonable values of the parameters which enter the theory, the heat flow correlation integral delivers the correct area-integrated value of the heat flow observed at Earth's surface. This result is unlikely to be significantly affected by the low-degree truncation of the tomographic models we employ as in both the upper mantle and lower part of the lower mantle, heat is primarily transported by the degree 2 components of the flow. We propose that the radial profile of heat advection is a particularly useful diagnostic with which to ''prospect'' for the existence of thermal boundary layers in the mantle. For split mantle tomographic models, we find a sharp drop in advected heat at a depth of 670 km. We investigate in detail the rheological consequences of a circulation that is layered at this depth by the action of the endothermic phase transformation of spinel to a mixture of perovskite and magnesiow¿stite. On physical grounds this is expected to lead to the development of a dipolar viscosity structure centered on the internal thermal boundary layer. We investigate the impact that such a structure has on the predicted aspherical geoid.

A sequence of forward modeling calculations of this geophysical observable demonstrates that a viscosity profile which includes a dipolar structure centered at 670 km depth and a significant increase of viscosity below mantle depths of 1200--1500 km optimally reconciles the long-wavelength GEM-T2 observations. The increase in viscosity in the lower part of the lower mantle is also required by the heat flow data whereas the introduction of the dipolar structure in the viscosity profile allows the upper mantle value of the viscosity which is close to that inferred by Haskell of 1021 Pa s to be continued to a depth of 1200--1500 km in accordance with the requirements of recent postglacial rebound inferences. In the context of a whole mantle model of the circulation, we are also able to accommodate the constraints imposed by the data by introducing only a low-viscosity notch at 670 km depth rather than the dipolar structure. The viscosity models derived herein therefore provide a fully self-consistent reconciliation of these distinctly different geodynamic data and would appear to resolve a previously unresolved conflict between them. ¿ American Geophysical Union 1995