Professor Rhodri Davies

I study how dynamical processes within Earth's interior drive change at its surface. I combine high-performance geodynamic simulation, inverse modelling and observations from geophysics, geology and geochemistry to investigate mantle convection, plate motion, volcanism, dynamic topography, melting and chemical evolution.

A central goal of my research is to reconstruct the thermo-chemical structure and flow history of Earth's mantle through time. Through G-ADOPT, my group is developing adjoint and data-informed methods that allow observations to constrain physically consistent models of Earth evolution.

This work connects fundamental geodynamics with major societal challenges, including Antarctic ice-sheet evolution and sea-level change, and the deep-time geodynamic environments that influence mineral systems. I welcome enquiries from students and collaborators interested in inverse geodynamics, computational Earth science and the links between Earth's interior and surface.

Research overview

My current research programme is organised around six connected themes:

• Reconstructing Earth evolution through inverse geodynamics, using adjoint methods, high-performance finite-element modelling and observations to recover mantle structure and flow through space and time.
• Understanding mantle flow and dynamic topography, and how deep and shallow mantle processes shape topography, sedimentary basins, coastlines and ice-sheet stability.
• Revealing the dynamical mechanisms underpinning volcanism and melting, from mantle plumes and intraplate volcanism to mid-ocean-ridge melting and geochemical variability.
• Quantifying plate motion, subduction and lithosphere-mantle interaction, including how slabs, plumes and lithospheric structure alter the force balance governing tectonic change.
• Modelling solid Earth controls on ice sheets and sea level, through glacial isostatic adjustment, mantle rheology and coupled ice-Earth-ocean interactions.
• Identifying geodynamic environments that influence mineral systems, by integrating mantle and lithosphere dynamics, plate reconstructions, melting, geochemistry and data-driven methods.

Research tools

My current software development centres on G-ADOPT, a community platform for forward and inverse geodynamic modelling developed at RSES. Built on Firedrake, it combines automated finite-element methods, adjoint differentiation and high-performance optimisation to make complex, data-informed models more accessible, scalable and reproducible.

I was also a long-standing developer and user of Fluidity, whose adaptive, anisotropic, unstructured meshes enabled efficient simulation of strongly localised features such as slabs, plume conduits and plate interfaces. That work provided an important foundation for my present focus on automated and adjoint-capable geodynamics.

Research highlights

These selected publications trace the development of my research from adaptive modelling and plume and subduction dynamics to data-informed geodynamics, surface processes, sea level and mantle melting.

Scott et al. (2026), Geoscientific Model Development. Automated forward and adjoint modelling of viscoelastic deformation of the solid Earth. This work extends G-ADOPT to viscoelastic solid-Earth deformation and automated adjoint inversion, establishing a scalable route to three-dimensional glacial isostatic adjustment, palaeoclimate and sea-level data assimilation.

Joint inversion for ice load and mantle viscosity. Results after 2, 10 and 100 optimisation iterations progressively recover the target state. Figure 13 of Scott et al. (2026), CC BY 4.0.

Pilia et al. (2025), Earth and Planetary Science Letters. Ghost plumes hidden beneath Earth's continents. Tomography, transition-zone structure, residual topography and plate reconstructions reveal the amagmatic Dani plume beneath eastern Oman, showing that thick continental lithosphere can conceal mantle plumes from the volcanic record.

Tomographic images of the Dani plume

Station geometry, seismic ray paths and P- and S-wave sections reveal a vertically continuous low-velocity anomaly beneath eastern Oman; black lines mark the 410 and 660 km discontinuities. Figure 1 of Pilia et al. (2025), CC BY-NC-ND 4.0.

Sternai et al. (2025), Tectonics. Raising the roof of the world: Intra-crustal Asian mantle supports the Himalayan-Tibetan orogen. Fully coupled petrological and thermo-mechanical models suggest that Indian crust underplates Asian lithosphere, leaving strong mantle lithosphere between Indian and Asian crust to help support the Tibetan Plateau.

Reference model evolution for the Himalayan–Tibetan orogen

Reference-model evolution from ocean-continent subduction to mature collision and exhumation, showing progressive crustal doubling, interlayered mantle lithosphere and uplift. Figure 4 of Sternai et al. (2025), CC BY 4.0.

Ghelichkhan et al. (2024), Geoscientific Model Development. Automatic adjoint-based inversion schemes for geodynamics: reconstructing the evolution of Earth's mantle in space and time. This paper introduces G-ADOPT and shows how automated adjoints can recover earlier mantle states from later observations, moving geodynamics from forward prediction towards quantitative reconstruction.

Reference and reconstructed mantle-convection histories. The inversion recovers the principal temperature, viscosity and surface-stress structures from incomplete information. Figure 11 of Ghelichkhan et al. (2024), CC BY 4.0.

Duvernay et al. (2024), Geochemistry, Geophysics, Geosystems. Coupled geodynamical-geochemical perspectives on the generation and composition of mid-ocean-ridge basalts. Coupling ridge flow, peridotite melting, trace-element partitioning and magma-chamber processing reproduces normal MORB compositions, while a persistent strontium deficit points to recycled oceanic crust in the mantle source.

Mid-ocean-ridge dynamics and melting diagnostics

Reference mid-ocean-ridge flow and melting, with diagnostics showing sensitivity to mantle temperature, spreading rate and water content. Figure 3 of Duvernay et al. (2024), CC BY-NC-ND 4.0.

Davies et al. (2022), Geoscientific Model Development. Towards automatic finite-element methods for geodynamics via Firedrake. This study establishes the computational foundation for G-ADOPT by showing that high-level mathematical descriptions can generate accurate, efficient and extensible geodynamic solvers.

Present-day thermal structure in a global Firedrake mantle-convection model driven by 230 million years of plate-motion history. Red and blue isosurfaces show upwelling plumes and downwelling slabs. Figure 11 of Davies et al. (2022), CC BY 4.0.

Davies et al. (2019), Nature Geoscience. Earth's multi-scale topographic response to global mantle flow. A statistically robust spectrum of oceanic residual topography reconciles observations and models only when lithospheric structure is included, demonstrating that both deep and shallow mantle processes shape Earth's surface.

Predicted versus inferred global topography

Predicted and inferred global topography, with power spectra comparing observational constraints against mantle-flow models with and without shallow lithospheric structure. Figure 1 of Davies et al. (2019).

Jones et al. (2017), Nature. The concurrent emergence and causes of double volcanic hotspot tracks on the Pacific plate. The simultaneous appearance of paired volcanic tracks at Hawaii, Samoa, Society, Foundation and Marquesas is linked to a recent change in Pacific plate motion and a tilted, compositionally heterogeneous plume-melting system.

Recent Hawaiian volcanism along the Loa and Kea tracks, with a schematic of the tilted plume, moving Pacific plate and separated melt-source regions. Figures 1 and 2 of Jones et al. (2017).

Davies et al. (2015), Nature. Lithospheric controls on magma composition along Earth's longest continental hotspot track. This paper identifies the 2,000 km Cosgrove track across eastern Australia and shows that lithospheric thickness exerts a first-order control on the volume and chemistry of plume-derived magma.

Eastern Australian Cenozoic volcanic centres and lithospheric thickness along the Cosgrove track, constraining sub-continental plume-melting depths. Figure 1 of Davies et al. (2015).

Garel et al. (2014), Geochemistry, Geophysics, Geosystems. Interaction of subducted slabs with the mantle transition-zone: A regime diagram from 2-D thermo-mechanical models with a mobile trench and an overriding plate. Systematic models show how slab and overriding-plate strength govern trench migration, slab sinking and the diverse ways slabs penetrate, fold or flatten in the transition zone.

Adaptive-mesh Fluidity simulation of subduction, showing temperature, viscosity, deformation mechanism and mesh evolution as the slab interacts with the mantle transition zone. Garel et al. (2014).

Davies et al. (2012), Earth and Planetary Science Letters. Reconciling dynamic and seismic models of Earth's lower mantle: The dominant role of thermal heterogeneity. Thermodynamic conversion and resolution filtering allow direct comparison of convection models with tomography, showing that large coherent dense piles are not required to explain the main deep-mantle seismic observations.

Observed shear-wave structure beneath Africa compared with purely thermal and thermo-chemical mantle models before and after tomographic resolution filtering. Figure 4 of Davies et al. (2012).

Davies et al. (2011), Geochemistry, Geophysics, Geosystems. Fluidity: A fully unstructured anisotropic adaptive mesh computational modeling framework for geodynamics. Fluidity introduced dynamically adapting meshes that concentrate resolution on evolving boundary layers and flow structures, enabling accurate geodynamic simulations with substantially fewer computational degrees of freedom.

Anisotropic and isotropic adapted meshes in a convection simulation, illustrating how directional refinement resolves thermal boundary layers efficiently. Figure 9 of Davies et al. (2011).

Davies & Davies (2009), Earth and Planetary Science Letters. Thermally-driven mantle plumes reconcile multiple hot-spot observations. High-vigour global convection models generate long-lived, mobile, short-lived, pulsing, merging and splitting plumes, quantitatively reconciling the diversity of hotspot lifespans, productivity and motion.

Evolution of a high-Rayleigh-number global convection model, showing diverse plume behaviours through time. Figure 1 of Davies & Davies (2009).

Review articles

Review articles are important because they test individual results against the wider evidence and make assumptions, disagreements and knowledge gaps explicit. They are also time-stamped syntheses: each captures my overall view of a field at the time of publication and identifies the observations, methods and conceptual advances that I considered most likely to move it forward.

Davies et al. (2023), book chapter. Observations and Models of Dynamic Topography: Current Status and Future Directions. This review explains how observational estimates and mantle-flow predictions can now be reconciled when both deep-mantle flow and shallow lithospheric structure are included. It identifies adjoint inversion and tighter integration of geological, geophysical and geochemical observations as the most promising path to reconstructing dynamic topography through time.

Observed and predicted long-wavelength dynamic topography

Observed long-wavelength residual topography (left) and dynamic topography predicted by a mantle-flow model that incorporates shallow structure (right). Figure 5 of Davies et al. (2023).

Davies et al. (2015), book chapter. Thermally Dominated Deep Mantle LLSVPs: A Review. This review evaluates the principal seismic arguments for chemically dense deep-mantle piles. It concludes that most observations are equally consistent with predominantly thermal structure, while acknowledging geochemically important heterogeneity whose large-scale dynamical effect may be secondary to temperature.

Tomographic and modelled deep-mantle structure beneath Africa

Shear-wave tomography beneath Africa compared with purely thermal and thermo-chemical mantle models before and after applying the tomographic resolution operator. Figure 14.7 of Davies et al. (2015).