The deep secrets of hotspot volcanoes
he improvements in geological seismic imaging techniques over the last 10 years have given us robust images of the deep mantle. But can they be used to settle a 40 year old debate on the origin of hotspot volcanism? Professor Barbara Romanowicz tells us more
hundred years ago this year, Alfred Wegener laid out the theory of continental drift in his paradigm-shifting book The Origin of Continents and Oceans. It took 50 years for this to lead to the now widely accepted theory of plate tectonics, which provides a powerful framework to explain many geological and geophysical observations, such as, for example, the concentration of seismicity and volcanism along plate boundaries. Yet, the simple assumption that plates are rigid and consequently deformation is concentrated along their boundaries fails to account for the presence of mid-plate, so called “hotspot”, volcanism.
These volcanoes – a textbook example is Hawaii – often form linear chains whose age increases regularly in the direction of plate motion. In 1963, Tuzo Wilson had already proposed that such a “hotspot” volcanic island chain is the surface expression of a deep-seated plume, a narrow conduit bringing hot and buoyant material to the surface from a boundary layer somewhere in the earth’s deep mantle. These plumes would be fixed with respect to the moving plates above them. Then Jason Morgan suggested in 1972 that mantle plumes might originate at the earth’s core-mantle boundary, 2900km down, tapping a primordial reservoir of material shielded from direct sampling by convection within the mantle. This could explain the difference in trace element compositions of hotspot and mid-ocean ridge volcanoes – landmarks of plate boundaries where new crust is being formed.
The origin of hotspot volcanism has generated vigorous debate, with another camp arguing that they simply represent melting processes related to stress distribution within the lithospheric (the solid outer section of Earth, including the crust and upper part of the mantle) plates themselves. To settle this debate, imaging of the deep structure at high resolution is necessary in order to detect the thin plume stems – if they exist. Indeed, in a simple model of convection in the viscous mantle, one would expect these stems to be no wider than 100-200km, if buoyancy is driven only by temperature differences (known as Rayleigh Benard convection). This represents a daunting challenge for seismic tomography, which has to rely on the very non-uniform illumination of the earth’s interior provided by seismic waves propagating through it, because the only sources of these waves powerful enough are natural earthquakes distributed mostly along plate boundaries, and sensors are sparsely distributed, primarily on continents. Quite a different situation from medical imaging of the brain, where sources and sensors can be placed regularly over the surface of the body.
The debate was revived when, in 2004, Raphaella Montelli, then a graduate student working with Guust Nolet at Princeton University, produced a global image of the earth’s mantle, where elongated zones of lower than average seismic velocities (a proxy for high temperature) could be traced from many hotspots around the world almost all the way to the core-mantle boundary.
These were exciting results, but skeptics remained, because these conduits were not quite continuous – requiring a stretch of the imagination to connect the dots. More importantly, some questions remained as to the biases inherent in the imaging technique, which was based on the interpretation of propagation travel times of only the first arriving waves from each earthquake. This method is prone to artifacts, as images can be smeared along seismic wave ray paths in regions, such as ocean basins, where
illumination is poor due to the lack of cross-paths for thousands of miles around the recording site. That is, an elongated structure reaching deep into the mantle right beneath an isolated hotspot island could be mistaken for a plume, when in fact it was artificially stretched due to lopsided illumination.
Progress in our understanding of earth’s mantle dynamics and its evolution requires combined efforts in several earth sciences disciplines
With improvements in imaging techniques perfected over the last 10 years, we now have, for the first time, much clearer, robust images of deep mantle plumes¹. The results are intriguing! First, the detected plume conduits all have roots in patches of strongly reduced shear velocity at the core-mantle boundary, and extend vertically for almost 2000km – i.e. across 2/3 of the mantle. Second, they are much broader than expected (more than 500km in diameter, even after correcting for imperfect sharpness of the images). Third, they are found in the vicinity of many catalogued hotspots, but only those that are located above the two landmark antipodal and equatorial zones of lower than average shear velocity, present in the lowest mantle, one beneath the Pacific and the other beneath Africa². There are indications that these giant structures, sometimes called “superplumes” are not only hot – but also of a somewhat different mineralogical composition than the surrounding mantle, and that perhaps they have been in a stable position for at least 200 Million years and possibly much longer, acting as anchors to mantle circulation. But the nature and role in mantle dynamics of these prominent structures is not yet fully understood.
The fact that the detected plumes are found within their footprint, and are broader by a factor of two or more than expected, and therefore likely quite old (geologically speaking), are strong indications that the simple model of purely thermally driven convection in the mantle needs to be revisited. There is already ample evidence from numerical and experimental convection modeling that wide, dome-shaped, stable plumes are formed in the case of a fluid heated from below, with a chemically distinct, slightly denser layer at its base. The new images will help improve such models.
Another intriguing feature of these plume images is the change in character of the plume conduits after rising through 2000km of the mantle (i.e. around 1000km depth from the surface). Instead of continuing straight up, most of them are deflected horizontally for some distance and, while some re-emerge right under the corresponding hotspot and can be followed as somewhat narrower features meandering through the upper mantle, we lose track of others – an indication that they may be too thin to be detected tomographically at present. This suggests that there is a rheological boundary around 1000km depth, with significantly lower viscosity, allowing more vigorous convection, above this depth.
It has been known for a long time that the upper mantle is less viscous than the lower mantle, but in general, the limit between the two domains is assumed to be at a depth of ~660km, which corresponds to a known mineralogical phase transition, where the high pressure version of olivine – the major silicate mineral in the upper mantle ((Mg,Fe)SiO4) – is decomposed into bridgmanite ((Mg,Fe)SiO3) and magnesio-wustite ((Mg,Fe)O). What happens at around 1000km must be significant, since in some regions, subducted slabs of cold lithosphere appear to extend horizontally for several thousands of kilometers, rather than plunging straight into the lower mantle – not only around 660km depth where it might be expected, but sometimes at 1000km, for example in the Tonga-Fiji region or in South America. There is no evidence for a mineralogical phase change at that depth, so a change in the deformation mechanism of lower mantle minerals may be at play.
As usual, progress in our understanding of earth’s mantle dynamics and its evolution requires combined efforts in several earth sciences disciplines: seismological images of present day structure serve as reference for geodynamical modelling that increasingly includes information on the nature and deformation behaviour of mantle minerals. The sharpness of present seismological images can still be improved: it is possible that other types of plumes, thinner and more similar to those expected from thermal convection modelling, also exist in the deep mantle, but are beyond the present image resolution. Also, sharper images are needed to better understand how plumes interact at large scale with the tectonic plates that move above them. For example, do the regularly spaced features elongated in the direction of plate motion observed in ocean basins beneath the lithosphere correspond to spreading over time of plume material brought from the deep, or do they represent secondary scale convection set up by the motion of the rigid lithospheric plates over the soft asthenosphere?
This brings us to how the sharpness of seismic images of the mantle has been improved and can be further refined in the future: the first arriving waves used until recently in most of deep mantle tomography represent only a very small fraction of the information contained in records of seismic waves propagating through the earth from distant earthquakes. So, there was hope for improvement, if one could only compensate for the poor distribution of sources and sensors by exploiting information from waves bouncing multiple times within the mantle, reflecting on the surface and the core-mantle boundary, propagating along the surface (surface waves) or diffracting along the core-mantle boundary. In other words, if one could use entire time series (in our jargon “full waveforms”) rather than only the first arriving, isolated bursts of energy. While work on waveform tomography started in our group more than 20 years ago, such an approach has only recently been fully exploitable owing to the implementation of accurate and relatively fast numerical computations of the seismic wavefield across the earth’s interior, and predicted for earth models with arbitrary strength and size of elastic heterogeneities.
The method of choice, currently, is the so-called “spectral element method”, and its main drawback is that it requires access to HPC facilities in order to perform the many iterations needed in solving the non-linear imaging problem in a reasonable time. The computational time also increases as the cube of the highest wave frequency. The final construction of our current model itself required over 3 Million of CPU hours on the National Energy Research Scientific Computing Center’s supercomputers. Sharpening the images will require an extension to higher frequency and zoom in regions with optimal illumination by dense arrays of observing seismic stations, including on the ocean floor.