The shapes of mantle plumes are sensitive to mantle viscosity, density structure, and flow patterns. Increasingly, global tomographic models reveal broad plume conduits in the lower mantle and highly-tilting conduits in the mid and upper mantle. Previous studies mostly relied on 2D slices to analyze plume shapes, but fully investigating the complexity of 3D plume structures requires more effective visualization methods. Here, we use immersive virtual reality (VR) headsets to visualize the full-waveform global tomographic models SEMUCB-WM1 and GLAD-M25 (VS). We develop criteria for the identification of plume conduits based on the relationship between the plume excess temperature and the VS anomaly (∂VS). We are able to trace 20 major plume conduits, measure the offsets of the conduits in azimuth and distance with respect to the hotspots, calculate the tilt angle, and evaluate the ∂VS along all traced conduits. We compare our traced conduits with the conduits predicted by global mantle convection models and vertical conduits. The wavespeed variations along conduits traced from each tomographic model are slower than modeled or vertical conduits, regardless of which tomographic model they are evaluated in. The shapes of traced conduits tend to differ greatly from modeled conduits. Plume ponding and the emergence of secondary plumes, which could result from a combination of different plume compositions, phase transitions, small-scale convection, and variations in viscosity and density of the ambient mantle, can contribute to the complex observed plume shapes.

The long-wavelength geoid is sensitive to Earth’s mantle density structure as well as radial variations in mantle viscosity. We present a suite of inversions for the radial viscosity profile using whole-mantle models that jointly constrain the variations in density, shear- and compressional-wavespeeds using full-spectrum tomography. We use a Bayesian approach to identify a collection of viscosity profiles compatible with the geoid, while enabling uncertainties to be quantified. Depending on tomographic model parameterization and data weighting, it is possible to obtain models with either positive- or negative-buoyancy in the large low shear velocity provinces (LLSVPs). We demonstrate that whole-mantle density models in which density and $V_S$ variations are correlated imply an increase in viscosity below the transition zone, often near ~1000~km. Many solutions also contain a low-viscosity channel below 650~km. Alternatively, models in which density is less-correlated with $V_S$ – which better fit normal mode data – require a reduced viscosity region in the lower mantle. This feature appears in solutions because it reduces the sensitivity of the geoid to buoyancy variations in the lowermost mantle. The variability among the viscosity profiles obtained using different density models is indicative of the strong non-linearities in modeling the geoid and the limited resolving power of the geoid kernels. We demonstrate that linearized analyses of model resolution do not adequately capture the posterior uncertainty on viscosity. Joint and iterative inversions of viscosity, wavespeeds, and density using seismic and geodynamic observations are required to reduce bias from prior assumptions on viscosity variation and scalings between material properties.

The viscosity of the mantle affects every aspect of the thermal and compositional evolution of Earth's interior. Radial variations in viscosity can affect the sinking of slabs, the morphology of plumes, and the rate of convective heat transport and thermal evolution. Below the mantle transition zone, we detect changes in the long-wavelength pattern of lateral heterogeneity in global tomographic models, a peak in the the depth-distribution of seismic scatterers, and changes in the dynamics plumes and slabs, which may be associated with a change in viscosity. We analyze the long-wavelength structures, radial correlation functions, and spectra of four recent global tomographic models and a suite of geodynamic models. We find that the depth-variations of the spectral slope in tomographic models are most consistent with a geodynamic model that contains both a dynamically significant phase transition and a reduced-viscosity region at the top of the lower mantle. We present new inferences of the mantle radial viscosity profile that are consistent with the presence of such a feature.