2021-Mantle plumes and their role in Earth processes.pdf

REVIEWS ­­ Mantle plumes and their role in Earth processes Anthony A. P. Koppers 1 ✉, Thorsten W. Becker 2, Matthew G. Jackson...

REVIEWS ­­ Mantle plumes and their role in Earth processes Anthony A. P. Koppers 1 ✉, Thorsten W. Becker 2, Matthew G. Jackson 3 , Kevin Konrad 1,4, R. Dietmar Müller 5, Barbara Romanowicz6,7,8, Bernhard Steinberger 9,10 and Joanne M. Whittaker 11 Abstract | The existence of mantle plumes was first proposed in the 1970s to explain intra-plate, hotspot volcanism, yet owing to difficulties in resolving mantle upwellings with geophysical images and discrepancies in interpretations of geochemical and geochronological data, the origin, dynamics and composition of plumes and their links to plate tectonics are still contested. In this Review, we discuss progress in seismic imaging, mantle flow modelling, plate tectonic reconstructions and geochemical analyses that have led to a more detailed understanding of mantle plumes. Observations suggest plumes could be both thermal and chemical in nature, can attain complex and broad shapes, and that more than 18 plumes might be rooted in regions of the lowermost mantle. The case for a deep mantle origin is strengthened by the geochemistry of hotspot volcanoes that provide evidence for entrainment of deeply recycled subducted components, primordial m­an­tle domains and, potentially, materials from Earth’s core. Deep mantle plumes often appear deflected by large-scale mantle flow, resulting in hotspot motions required to resolve past tectonic plate motions. Future research requires improvements in resolution of seismic tomography to better visualize deep mantle plume structures at smaller than 100-km scales. Concerted multi-proxy geochemical and dating efforts are also needed to better resolve spatiotemporal and chemical evolutions of long-lived mantle plumes. Hotspot Narrow mantle upwellings, or plumes, are an integral the relationships between various geophysical and geo- Region of intra-plate volcanism part of Earth’s convection system, yet many controver- chemical observations, and the dynamics and shapes of that forms an age-progressive sies have surrounded mantle plumes since the idea was plumes are still hotly contested, leaving numerous open volcanic chain (like Hawaii) or first invoked1. Where the plate tectonic revolution in the questions in the understanding of Earth’s interior. a region of excessive volcanism 1960s provided an explanation for volcanism at plate To date, it is understood that mantle plumes could along plate boundaries (like Iceland), with temperatures boundaries, it did not provide a model for the occur- start as deep as the core–mantle boundary at roughly exceeding that of ambient rence of volcanism in the middle of plates. However, in 2,900-km depth8–15. These plumes are also thought to asthenospheric mantle by 1963, it was the northwest-oriented lineament of vol- have temperatures in excess of ~100–300 °C compared ~100 °C or more. canic islands forming the Hawaiian archipelago that with the ambient mantle16–21 and require tens of millions Large igneous provinces could be explained by a rigid tectonic plate moving of years to ascend to the base of Earth’s lithosphere, (LIPs). Extremely large over a hotspot in Earth’s asthenosphere, allowing for where they form hotspots (Fig. 1a). Computer model- accumulations of intrusive and intra-plate magma production2. Almost a decade later, in ling and laboratory experiments, however, have shown extrusive volcanic rocks (areal 1971, it was hypothesized that such a hotspot could form how entrainment of chemical heterogeneity (Fig. 1b) extents >0.1 million km2 with above a vertical, narrow, hot mantle plume rising from may change key characteristics of plume conduits, for volumes >0.1 million km3) deposited over a short the deep mantle1. In these initial models, mantle plumes example, by inclusion of recycled subducted ocean geological period of time, often were defined as narrow thermal upwellings1 with a wide crust22–26. As a result, most plumes are now considered in less than 1 million years. ~1,000-km plume head, followed by a thinner ~100-km thermochemical plumes rather than purely thermal tail3,4. The deep mantle sources of plumes were later con- plumes, and are thought to have more intricate struc- nected to the recycling of old oceanic lithosphere (and tures and broader shapes. At asthenospheric depths, the sediments) conveyed to Earth’s deep interior via subduc- ascending plumes can exceed the pressure-dependent ✉e-mail: anthony.koppers@ tion over the course of hundreds of millions to billions melting temperature for mantle rocks, leading to par- oregonstate.edu of years5–7, thus, completing the convection cycle. How tial melting, resulting in widespread magmatism that https://doi.org/10.1038/ many mantle plumes exist, at what depth in the mantle creates large igneous provinces (LIPs) and age-progressive s43017-021-00168-6 they originate, their longevity and petrological make-up, volcanic hotspot trails27,28. Intra-plate hotpot volcanism, 382 | June 2021 | volume 2 www.nature.com/natrevearthenviron 0123456789();:  Reviews Key points However, the dynamics of mantle plumes and their relation to mantle convection and the formation of hot- Thermochemical mantle plumes are an integral part of Earth’s dynamic interior. spots in the shallow asthenosphere are still debated. For More than 18 mantle plumes appear to originate from the deepest regions example, it is unclear whether hotspots are formed by in Earth’s mantle. broad upwellings of thousand(s)-kilometre scale trig- Mantle plumes influence surface processes, including continental break-up and mass gered by passive return flow from subducting slabs55,56 extinctions. or by narrower upwellings of hundred(s)-kilometre The location of two large low-seismic-velocity provinces in the lowermost mantle can scale that are actively rising from a thermal boundary be associated with most present-day hotspots and ancient large igneous provinces. layer in the form of plume-like conduits25,57–59. Whether Individual plume motions are required to resolve changes in absolute plate tectonic all plumes form either somewhere near the core–mantle motions. boundary or in the upper mantle above the 660-km seismic Mantle plumes are geochemically heterogeneous, incorporating deeply recycled discontinuity in relation to subduction-driven convec- subducted components, primordial mantle domains and, potentially, materials from tion or the ponding of lower mantle plumes is also still Earth’s core. debated28,60–64. These differences are important in deci- phering to what extent there is a separate convectional Large low-shear-velocity therefore, provides a unique means of sampling deep regime in the upper mantle, besides whole-mantle provinces mantle plumes, providing key insights into the convec- convection. (LLSVPs). A large region tive scales and physical conditions of Earth’s interior Furthermore, the observation that some near-surface (several thousand kilometres and the chemical evolution of the mantle over billions processes in the Earth’s lithosphere and upper mantle in size) of low shear velocity (a few percent lower than of years of Earth history7,16,29,30. (such as propagating lithospheric cracks, diffuse exten- average) in the lowermost few Studies of hotspot volcanoes and mantle plumes also sion in the interior of tectonic plates and small-scale hundreds of kilometres of the provide an understanding of the broader interplay with sublithospheric convection) can result in similar style mantle. There are two of them, deep mantle superstructures, mantle convection, plate intra-plate volcanic features22,65 led some to question the beneath Africa and the Pacific. tectonics and surface processes. For example, it has existence of deep mantle plumes at all55,56,66,67. However, Hadean been hypothesized that large-scale mantle flow driven advances in seismic tomography have now allowed deep Geologic eon in Earth history, by slabs, many of which are subducted to the deep mantle plumes to be imaged with more confidence. starting with Earth’s accretion lower mantle and piled up into slab graveyards11,31–35, Several global-scale mantle tomography studies have about 4.567 billion years ago will trigger plumes, characteristically above large low- shown the existence of broader continuous low-velocity and ending 4 billion years shear-velocity provinces (LLSVPs), where seismic veloc- seismic anomalies extending from the deep mantle that ago at the beginning of the Archaean. ities are several percent lower than in surrounding can be interpreted as mantle plumes68–72. In particular, regions36–42, or focused along their edges11,13,43,44. These the first whole-mantle conduit has now been imaged73 Core–mantle boundary processes are all part of global elemental cycles governed that is sufficiently narrow to be considered a purely Boundary between the liquid by mantle convection and, therefore, mantle plumes thermal plume, as originally proposed1. metal outer core and the mostly solid lower mantle possibly sample of a variety of global-scale geochemi- As well as seismic imaging and geodynamic simula- at ~2,891-km depth. cal reservoirs, including ancient mantle domains from tions, advances in high-precision geochemical techniques Hadean times and a variety of deeply recycled crustal have caused a revolution in understanding the complex 660-km Seismic reservoirs5–7,45–47. Plumes impinging on Earth’s litho- formation history of mantle plumes. Most notably, discontinuity Seismic discontinuity sphere might also initiate rifts in continents, aiding in plume-fed hotspots exhibit geochemical hetero­geneity, corresponding to a solid–solid continental break-up and the formation of new ocean indicating that both ancient domains and recycled crust phase transition in the mantle basins48–51. Ultimately, mantle plumes are able to produce reservoirs are feeding upwelling plumes, providing at ~660-km depth. extensive volcanism on Earth’s surface, causing major insight into mantle dynamics5,7,74,75. In some cases, these perturbations in its environment and in ocean chem- heterogeneities are expressed on Earth’s surface and sus- istry, leading to climate change and possibly triggering tained in bilaterally zoned ‘striped’ volcanic chains, as mass extinctions52–54. seen in Hawaii, Galápagos, Marquesas and other volcanic islands chains76–82. In addition, extinct short-lived isotope systems (such as 129I-129Xe, 182Hf-182W and 146Sm-142Nd) Author addresses indicate the presence of early-formed isotopic anom- 1 College of Earth, Ocean, and Atmospheric Sciences, Oregon State University, Corvallis, alies in the source regions of mantle plumes that have OR, USA. remained unchanged since they formed during Earth’s 2 Institute for Geophysics and Department of Geological Sciences, Jackson School of accretion from ~4.567 billion years ago until the end of Geosciences, The University of Texas at Austin, Austin, TX, USA. the Hadean83–85. These isotope systems also reveal that 3 Department of Earth Science, University of California, Santa Barbara, Santa Barbara, mantle plumes might even tap into Earth’s core86,87, which CA, USA. was previously considered to be an isolated reservoir not 4 Department of Geoscience, University of Nevada, Las Vegas, Las Vegas, NV, USA. interacting with the mantle. 5 EarthByte Group, School of Geosciences, The University of Sydney, Sydney, NSW, Australia. In this Review, we discuss progress in imaging the 6 Department of Earth and Planetary Science, University of California, Berkeley, lower mantle, in understanding whether low-velocity Berkeley, CA, USA. anomalies are seismic signatures of active upwellings 7 Collège de France, Paris, France. 8 Institut de Physique du Globe de Paris, Paris, France. and in modelling the shapes and behaviours of plumes 9 Helmholtz Centre Potsdam, GFZ German Research Centre for Geosciences, Potsdam, depending on composition and rheology. We confer on Germany. how the geochemistry of oceanic island basalts increases 10 Centre for Earth Evolution and Dynamics (CEED), University of Oslo, Oslo, Norway. knowledge of deep mantle reservoirs and plume forma- 11 Institute for Marine and Antarctic Studies, University of Tasmania, Hobart, TAS, Australia. tion at the core–mantle boundary. We also synthesize nATuRe RevIeWS | EArTH & EnVIronmenT volume 2 | June 2021 | 383 0123456789();: Reviews a Mid-ocean ridge Subduction zone Hotspot Hotspot tracks and large igneous provinces Subduction zone b Entrained subducted material 660 km Ascending Upwelling Downwelling chemically zoned 1,000 km mantle plume Ambient lower LLSVP ULVZ ? mantle ULVZ LLSVP LLSVP 2,900 km Core ? Core–mantle ULVZ Core boundary Fig. 1 | Dynamic nature of Earth’s interior. a | Schematic cross section through Earth’s interior, depicting the key components of plume generation and upwelling near, above and along the edges of a large low-shear-velocity province (LLSVP) and near the core–mantle boundary. These LLSVPs might contain localized ultra-low velocity zones (ULVZs) and along their edges subducted material may pile up over hundreds of millions of years. b | Schematic cross section of a plume root showing the entrainment of subducted materials, LLSVP and ULVZ components, and possibly core materials at the edge of a LLSVP and centred above an ULVZ (that might be a unique deep mantle locality containing partial melt). palaeomagnetic and age information in seamount chemical plumes originating near the core–mantle trails in the context of informing past plate and plume boundary. Both the thermal and compositional expla- motions, including how and if true polar wander might nations were quite speculative at the time; however, they happen, given the observed overall mantle structure. We are still actively pursued. The large-scale correlation of discuss how continental break-up and the formation of seismic structure in the deep mantle with anomalies new ocean basins can be affected when mantle plumes in Earth’s geoid and subduction zone configuration36,89 impinge on the base of Earth’s lithosphere. In addition, was established soon thereafter. The correlation with we provide a catalogue of more than 50 identified hot- the distribution of hotspots and superswells in oceanic spots (Supplementary Table 1), of which 18 are consid- lithosphere was also noticed early on90–96 and, today, we ered most likely to have a deep mantle plume origin. understand these show a strong correspondence with Lastly, we provide perspectives on future integrative LLSVP locations (Fig. 2a). research themes that can help improve understanding Since then, scores of seismic studies have confirmed True polar wander of mantle plumes and their role in the interconnected the presence of LLSVPs and interpreted the seismically Reorientation of the entire Earth system and processes. fast areas elsewhere as the remnants of the downgoing solid Earth with respect to the spin axis. slabs from present and past subduction zones31–33,97 with Deep mantle superstructures a time-depth progressive correlation demonstrated for Eclogite Based on today’s global seismic tomography models, the last 130 million years35. Tomographic models show An ultra-high pressure two large-scale features with anomalously low seismic considerable agreement on the extent of the LLSVPs metamorphic rock forming at velocities can be recognized in the deepest parts of the and the shape of their borders on large scales, when depths greater than 35 km and typically associated with the mantle (Fig. 2). Although their general outlines and loca- filtered up to 2,500-km wavelengths98–100. The LLSVP transformation of basalt during tions underneath the Pacific and below Africa have been structures are likely hot; however, the anti-correlation oceanic plate subduction. known for decades, the composition, fine-scale structure of bulk-sound and shear velocities within them40,42,101,102, and origin of LLSVPs are still debated, as well as their as well as the sharpness of their borders103–105, suggest Geoid Equipotential surface of the role in the generation of deep mantle plumes. that they might include a denser component of different Earth’s gravity field that most composition. In spite of these observations, some stud- closely coincides with mean LLSVP structure and origin. The first tomographic ies question the necessity of the denser component106,107 sea level. images of the Earth’s lower mantle revealed the presence and imply that LLSVPs could be purely thermal features. of a very long wavelength structure at the base of the If, indeed, LLSVPs are denser than the surrounding Superswells Regions of elevated mantle that is anti-correlated with the gravity field88. The lower mantle108,109, this would help resist entrainment topography too large to be first possible explanation proposed was a dynamic inter- of LLSVP components in mantle convection and pre- attributed to a single hotspot. pretation encompassing thermal anomalies and core– vent mixing with the overlying mantle. This could help Two prominent superswells mantle boundary deflections owing to mantle-wide facilitate the survival of LLSVPs over hundreds of mil- overlie the corresponding large low-shear-velocity convection, and the second involved lateral variations in lions of years34 and, perhaps, throughout most of Earth provinces in the Pacific composition because of the presence of denser eclogite- history110,111. However, resolving the internal density and Africa. like material in regions of past subduction and/or structure in LLSVPs is challenging112–114 and different 384 | June 2021 | volume 2 www.nature.com/natrevearthenviron 0123456789();:  Reviews a SEMUCB Connected conduit length b Yellowstone Degrees along profile Yellowstone 0.0 0.2 0.4 0.6 0.8 1.0 10° 15° 20° 5° 25° SSW 0° 30° NNE 0 500 1,000 Depth (km) 1,500 2,000 2,500 –2 –1 0 1 2 −2 −1 0 1 2 δvs (%) at 2,850 km % Velocity perturbation c Hawaii Pitcairn Guadalupe SEMUCB-WM1 SEMUCB-WM1 Hawaii Pitcairn Guadalupe S40RTS S40RTS Hawaii Pitcairn Guadalupe PRI-S05 PRI-S05 500 1,000 1,500 2,000 2,500 –2.0 +2.0 Depth (km) dlnVs (%) nATuRe RevIeWS | EArTH & EnVIronmenT volume 2 | June 2021 | 385 0123456789();:  Reviews ◀ Fig. 2 | Tomographic imaging of mantle plumes and LLSVPs. a | Global map linking Mantle plume characteristics surface hotspots (circles) with depth-projected source locations (diamonds) of the In this section, we discuss what geodynamic modelling mantle plumes based on the methods of ref.9, using the plume catalogue of ref.195 and predicts about the make-up and behaviour of plume tomographic model SEMUCB-WM1 (refs72,327). The blue to red background colouring heads and tails during plume generation and ascent, depicts the amplitude of the seismic shear velocity anomaly δvs at 2,850-km depth and we review what seismic imaging reveals about deep in SEMUCB-WM1. Two large low-shear-velocity provinces (LLSVPs; red background colours) are located beneath Africa and the mid-Pacific. Most plumes originate above mantle plumes. these LLSVPs with a smaller group (Bowie, Cobb, Guadalupe and Socorro in the eastern Pacific) forming away from these lower mantle anomalous regions. The upper colour Models of plume generation and ascent. Localized hot bar represents the normalized length of the mantle plume conduits as identified in upwelling plumes are expected in any terrestrial-type tomography, where unity (light colours) mean that 100% of the plume length is mapped planet mantle, where convection operates through over the full mantle depth and zero (dark colours) mean that none of the plume length bottom heating in a temperature-dependent viscosity is mapped. b | Whole-mantle depth cross section (from the equator about 6° west of fluid of uniform composition or where a domain with Baja California up to 30° N in Saskatchewan, Canada) showing the Yellowstone mantle concentrated heat production can sustain a thermal plume structure (red-orange colours) based on shear wave tomography models using boundary layer25,26,59,139,140. Differences in temperature the dense USArray seismic network73. c | Collage of mantle plume tomography images and/or composition will cause variations in density that for Hawaii and Pitcairn that compare three seismic shear wave tomography models developed over the last 15 years: PRI-S05 from ref.68, S40RTS from ref.41 and can result in thermochemical instabilities near bound- SEMUCB-WM1 from ref.327. Small globes show the position of the cross section ary layers as the beginning of an upwelling and, when it lines. Panel b adapted from ref.73, Springer Nature. persists, a rising mantle plume25. Typical thermal mantle plumes in Earth are expected to have a broad plume head, up to roughly thousand inferences on their estimated density115,116 might be kilometres in diameter, followed by a narrow plume tail, because of thermal effects counteracting intrinsic den- not wider than a couple of hundred kilometres23,28,57,141,142. sity contrasts to a different extent and in different parts These plumes can persist over long geological times, of the LLSVPs117. The distinct composition and observed but only if the thermal boundary layer from which they chemical heterogeneity of the two LLSVPs could result arise is also maintained for hundreds of millions of years. from a reservoir of primitive material at the base of the Indeed, for plumes to keep rising through the entire man- mantle, as suggested by the relationships between high tle, a lower limit of ~500–1,000 kg s−1 of anomalous mass 3 He/4He hotspots and plumes originating near the core– flux (volume flux times density anomaly) is required to mantle boundary87,118–121, from the accumulation of sub- sustain the plume tails143. Anomalous mass flux can ducted crust transformed into eclogite34 or from some be estimated from the size of hotspot swells21,57,90,144,145, depth-dependent stratification of both111. However, the interaction of plumes with migrating spreading whether the LLSVPs are compact, continuous piles of ridges146,147, the geometry of V-shaped ridges148 or based compositionally distinct material11,43,122 where mantle on seismic tomography149. However, owing to different plumes are generated across their tops and along their amounts of buoyant material available in the thermal edges or whether LLSVPs represent bundles of closely boundary layer, the anomalous mass flux is expected spaced thermochemical plumes64,123 separated by a ring of to vary between plumes and can change with time for Ultra-low velocity zones downwellings that are constrained by the geometry individual plumes144,145. (ULVZs). Smaller regions (hundreds of kilometres in size) of tectonic plates at the surface124,125 remains a subject of Even though rising plumes from near LLSVPs are of very low shear velocity debate37,117,123,126. expected to be buoyant owing to their hotter tempera- (exceeding 10% lower than tures, they are also likely to entrain chemically distinct average) in the lowermost Mantle plume nurseries. Also debated is whether or not materials from the LLSVPs that can be denser than the few tens of kilometres of the mantle. There are perhaps plumes — specifically, those that can be correlated to surrounding mantle115 and, thus, their positive buoy- a dozen or more of them. ancient LIPs — originate primarily along the edges or ancies might be substantially reduced34,150. The anom- anywhere above LLSVPs8–15. LLSVP edges are defined alous density of LLSVPs 115,116 and the nature of the Thermal mantle plumes by sharp vertical and horizontal gradients in seismic entrainment of other materials during ascent of these Columnar upwellings of hot shear-wave velocities103,105,127. Many plumes and recon- thermochemical mantle plumes 151,152 are still debated. material from the lowermost mantle to the base of the structed LIP locations tend to occur close to vertically Because of time-dependent and variable amounts of lithosphere. above LLSVPs (Fig. 2a). Hence, LLSVPs are proposed entrainment, highly complex plume behaviour and by some to act as plume generation zones or nurseries shapes (Fig. 3) can result that substantially differ from Hotspot swells for all major deep-mantle-sourced hotspots and LIPs the classical head-and-tail structure153–155. For exam- Regions of elevated topography (several hundred that have been active and remained approximately in ple, the negative buoyancy of material entrained from kilometres wide) caused place over at least the last 200 million years110,128–130. LLSVPs in plume heads can cause material to sink back by buoyant mantle plume However, because the African LLSVP is rather narrow into the ascending plume156,157, leading to broader plume material rising beneath and (Fig. 2a), a concentration of plumes near the edges is conduits that are a few hundred kilometres wider than being dragged along with ambiguous. Interestingly, the roots of at least some of typical thermal plumes. a moving plate. the LLSVP-rooted plumes do seem to contain unusu- How long it takes for a plume head to traverse the Thermochemical mantle ally large ultra-low velocity zones (ULVZs). For example, mantle after forming at the core–mantle boundary is plumes mantle anomalies below Hawaii131, Samoa132, Iceland133 difficult to estimate. It primarily depends on its buoy- Columnar upwellings of and Marquesas 134 are indicative of a composition- ancy, arising from a density contrast of ~30 kg m−3 for hot and chemically distinct material from the lowermost ally different, likely denser, component117,135, possibly thermal plumes, but which could be much less for thermo­­ mantle to the base of the because of the presence of partial melt136 or at least iron chemical plumes, and the average viscosity of the sur- lithosphere. enrichment137,138 in these ULVZs. rounding mantle. Widely discrepant estimates exist for 386 | June 2021 | volume 2 www.nature.com/natrevearthenviron 0123456789();:  Reviews a Thermal plume head Axisymmetric Thermochemical plumes Partially failed thermal plume thermochemical plume b Rising plume c Rising plume Subducted slab LLSVP-like thermochemical LLSVP-like structure thermochemical structure Excess temperature +400 °K –400 °K Fig. 3 | Examples of rising thermal and thermochemical plumes. a | Representations of rising thermal plume head models: a thermal plume head connected by a stem to the source region27; an axisymmetric thermal plume; three thermochemical plumes150 with colours representing temperature differences (purple/red = hottest; orange = cooler; yellow = coolest; white = chemically different plume material); and, finally, a model of a plume that partially fails when entrained materials (that are too heavy) collapse in the top of the conduit (based on laboratory syrup tank experiments155). b | 3D and time-dependent numerical model with plumes (blue) rising from a large low-shear-velocity province (LLSVP) situated below southern Africa13. c | 3D model showing rising plumes from the LLSVP margins that are substantially hotter than the surrounding lowermost mantle, with large temperature anomalies of about 500 K (ref.77). Panel b adapted with permission from ref.13, Wiley, © 2015. American Geophysical Union. All Rights Reserved. Panel c adapted with permission from ref.77, Proceedings of the National Academy of Sciences. mantle viscosity, as it can be (locally) controlled by vari- per year, such that slabs require between ~150 and 200 ations in temperature and stress that might render global million years to reach the bottom of the mantle13,31,161,162. average viscosity estimates not applicable for plumes158. As plumes rise through the mantle, they are predicted One traversal time estimate can be made because to become increasingly tilted with time as their roots reconstructed LIP eruptions are correlated with LLSVP become shifted towards large-scale upwellings, likely margins and, therefore, LIPs are hypothesized to erupt above the two LLSVPs62,163. Numerical models yield tilts from those margins129,159. However, in order for that (mostly) up to — but sometimes even exceeding — 60° correlation to be maintained during their rise through (refs11,62,164). These strong tilts are partly a consequence the mantle, plume heads must rise up from the lower of plumes being modelled such that they match the mantle rather fast, probably within 30 million years or present-day hotspot positions; the tilts tend to be less less, to avoid large lateral deflections, consistent with in models where plumes dynamically form, but do numerical models13. For smaller plumes with smaller not exactly match actual hotspot positions11,13,77,165. plume heads, such as the Yellowstone plume (Fig. 2b), Eventually, when the plume head reaches the litho- the plume heads are predicted to rise more slowly, taking sphere, a LIP can form over short amounts of geological 80 million years or longer160. Despite these long ascent time, likely in less than a million years108. Long-lived times, plume heads rise considerably faster than slabs hotspot tracks can form after that and persist in some sink, which are estimated to sink at speeds of 1–2 cm cases for 80 to 120 million years60,166,167 or even longer79. nATuRe RevIeWS | EArTH & EnVIronmenT volume 2 | June 2021 | 387 0123456789();: Reviews Several hotspot tracks are associated with flood basalts they appear to be rooted near the core–mantle bound- at one end (such as Tristan, Réunion, Kerguelen) and, ary and extend into upper mantle depths in the vicinity thus, might correspond to typical thermal plumes with of hotspots69,71,72 (Fig. 2c). Numerical models show that a tail following a plume head28,168. thermochemical plume conduits are wider than a purely Although the typical lifetime of a deep mantle plume thermal plume156, in order to attain sufficient buoyancy. is likely long, on the order of ~100 million years, some Hence, plume conduits larger than ~500 km in diameter of the older Pacific oceanic plateaux (for example, Hess as observed in tomographic models indicate that they are and Shatsky Rise) were volcanically active only in the likely thermochemical rather than purely thermal72,189, Cretaceous and do not correspond to any currently and most of these broad conduits are observed over the active hotspot169,170. Evidence from the bathymetric LLSVPs (Fig. 2a). Anomalies in the travel times of seismic record in the Pacific Ocean also indicates that many hot- core waves, recorded by the dense USArray seismic net- spot trails are characterized by shorter seamount trails, work in North America, now also reveal a low-velocity apparently only active for up to 30 million years171,172. plume-like anomaly in the lower mantle beneath the These surface expressions, however, do not have to Yellowstone hotspot73 (Fig. 2b) that is probably unrelated mean that the mantle sources are short-lived per se but, to the Pacific LLSVP. rather, that instabilities can develop in rising mantle These seismically imaged, low-velocity plume-like plumes, for example, once conduits are tilted more than structures are nearly vertical in much of the lower 60° from the vertical, they might break apart163,173. Other mantle, but some are strongly tilted above a depth of studies suggest that plume heads might cut themselves 1,000 km (ref.72), which is different from geodynamically off from a supply of hot mantle material153 or detach modelled conduits that tend to be tilted throughout the from their plume conduit174. Plumes might also become entire mantle62,177. Such a strong tilt above ~1,000 km internally unstable and collapse (Fig. 3a) because of their is also found for the Yellowstone plume73, which, so far, is insufficient buoyancy155 or else switch on or off if rising the only plume where the predicted tilted conduit shape plumes are pulsating or boudinaging150,154. can be matched in detail with a tomographic conduit All possible explanations for such intermittent plume image160. The difference in tilt from geodynamic mod- behaviours and plumes without heads are still being elling and seismic observations can provide important mapped out and debated. Some hotspot tracks end at constraints on mantle rheology. It might indicate that subduction zones and it has been proposed that the the overall large-scale mantle flow at depths greater than respective LIP might have been accreted to the overriding ~1,000 km could be substantially slower than currently plate, for example, for the Hawaii plume in Kamchatka modelled, with horizontal flow speeds not exceeding a and the Okhotsk Sea175. An alternative hypothesis is that few millimetres per year. the Hawaiian flood basalts might have been subducted In addition, global seismic tomography indicates that into the upper mantle below Kamchatka176. both downgoing slabs190 and broad upwelling plume conduits69,72 appear to be deflected horizontally around Imaging mantle plumes. Within the framework of purely a depth of 1,000 km. This deflection is deeper than the thermal convection, thermal mantle plume conduits are known deflection at 660-km depth — where there is a expected to be only 100–200 km in diameter in the upper seismic discontinuity that most likely corresponds to mantle and no more than 400 km in the lower half of a phase transition from less dense ringwoodite crystal the mantle, which has a higher viscosity by about two structures to denser lower mantle bridgmanite and other orders of magnitude177. Given these predicted small high-pressure oxides — and is accompanied by a decor- plume conduit diameters, plume detection via seismic relation between the longest wavelength seismic struc- tomography is especially difficult, because of limited tures in the extended transition zone (400–1,000 km) resolution owing to a lack of earthquake sources and and the deeper mantle191. Similar observations were receiver stations in the ocean basins58,178–180. In addition, already visible in lower-resolution seismic mantle plume signals might be hiding in the small-amplitude models and indicated by seismic data suggesting a ver- coda of shear and compressional seismic waves, so that tical decorrelation at ~800-km depth192. The 1,000-km the full waveforms must be analysed in their entirety to boundary, thus, may require a change of material prop- better resolve plume conduits, as illustrated by synthetic erties not coinciding with the 660-km discontinuity and experiments181. Such new analytical approaches, there- could be explained by an increase in mantle viscosity fore, are required to resolve the smaller wave speed var- somewhat deeper than traditionally considered, given iations that could be interpreted as plume-like conduits the rather non-unique constraints provided by geoid and in seismic tomography. post-glacial rebound data193,194. In earlier studies, teleseismic travel time tomo­graphy In summary, there is still progress to be made in has found plume-like conduits of at least 400 km in resolving ~100-km-scale seismic heterogeneities in the diameter182–184. For example, the Iceland plume at first lower mantle relevant to the scale of deep mantle plumes. could only be seen in the upper mantle with an upwell­ However, modern-day global tomography models are ing broader than expected185,186, and imaging of the starting to resolve the presence of deep mantle plume Hawaii plume originally resulted in the detection of conduits below hotspots. a broad, inclined upwelling, disappearing from view below 1,500-km mantle depths68,187,188. However, in Deep mantle plume catalogue. We compiled a deep higher-resolution models, broader (>400 km) plume-like mantle plume catalogue (Supplementary Table 1) based conduits are found beneath many major hotspots, and on a listing of 57 potential hotspots195 that was then 388 | June 2021 | volume 2 www.nature.com/natrevearthenviron 0123456789();:  Reviews Ocean island basalts individually scored by the authors of this paper in order a short hotspot track does not exclude a plume, because A volcanic rock type typically to generate an expert opinion poll average, with the aim the plume could have arrived in the shallow mantle of basaltic composition, of providing a database of hotspots ordered by strongest only over the last few tens of million years197 or the hot- forming in ocean basins away to weakest evidence for a deep mantle plume origin. In spot could have exhibited only intermittent volcanic from plate tectonic boundaries and associated with intra-plate their scoring, each of the authors used their own set of activity over the last ~140 million years171. Finally, hot- hotspot volcanoes and mantle criteria, but everyone could score plumes of deep origin spots that are so closely spaced that they correspond plumes. as probably (1 point), perhaps (0.5 points), not (0 points) to separate upwellings in the upper mantle, but likely or unknown (no score). We added up all points and one (wider) upwelling in the deep mantle, are consid- divided these by the number of scores given for each ered as one hotspot. Examples are Tristan-Gough and plume. The full catalogue (Supplementary Table 1) and Bouvet-Meteor/Shona. a map (Supplementary Figure 1) lists only 18 plumes above the threshold of 0.8 points that are most likely Geochemical heterogeneity to have a deep mantle origin. These high scores in our Although hotspot trails and LIPs have complex histories, plume rankings suggest strong evidence for a deep man- variations in their chemical compositions provide clues tle plume beneath a hotspot; however, the reader should about the number and nature of distinct domains that be careful when interpreting lower scores to mean that reside in Earth’s interior and that are sampled by mantle we exclude the possibility of a plume. Because we have plumes. These clues lead to intriguing debates on where not yet observed a plume beneath a hotspot does not and at what length scales and timescales those geochem- mean that such a plume does not exist, owing to imper- ical domains are manifested in the deep mantle, and how fect seismic datasets and models or missing ages and/or geochemically and geographically zoned hotspot trails geochemistry data. might reflect heterogeneity within mantle plumes. One concern is that Yellowstone-type plumes might exist beneath hotspots with lower scores in the table, in Enduring ancient signatures. Extinct short-lived isotope particular, for the African continent, as we understand systems provide insights into the processes happening in now that such plumes are ‘thin plumes’ that are observed the earliest approximately 50 (for 182W) to ~500 million only when datasets from high-density seismic arrays are years (for 142Nd) of Earth’s history85,120,198,199. Because available73. Another concern is that high 3He/4He can be radioactive decay in these isotope systems is rapid, with indicative of a deep mantle conduit sourcing a hotspot, half-lives from 8.9 to 103 million years, the radioactive but the lack of high 3He/4He does not exclude the pres- parent isotopes were only present in the early stages ence of a plume, as many plume-fed hotspot volcanoes of the Earth’s formation, so resolvable changes in the only exhibit low helium isotope ratios196. In addition, radiogenic isotopic ratios of 129Xe/130Xe, 182W/184W and 142 Nd/144Nd are restricted to the Hadean, which ended 10 4 billion years ago. The presence of such ancient iso- topic anomalies in mantle plume source regions, there- 5 fore, suggests that some early-formed mantle reservoirs are still present in the Earth’s interior, despite extensive 0 convective mixing83,84,87,119. For example, the 129I-129Xe system shows a marked difference in 129Xe/130Xe between –5 Earth’s mantle and atmosphere29 indicating that hetero­ geneous 129Xe/130Xe signatures are preserved in the –10 mantle since the Hadean83. Datasets for the 182Hf-182W μ182W and 146Sm- 142Nd systems confirm the presence of –15 Hadean-generated signatures in the modern mantle, Samoa Galápagos with resolvable 182W (Fig. 4) and 142Nd anomalies in –20 plume-fed ocean island basalts84,120,200, but the discovery of Hawaii Iceland anomalous 182W in plume-head-generated flood basalts86 –25 Réunion remains controversial201. Heard Several studies found that most ocean island basalts –30 Core- Juan Fernández influenced? Other hotspots that host negative 182W anomalies also exhibit high 3 He/4He lavas87,120,121 that, in turn, could be interpreted –35 0.702 0.703 0.704 0.705 0.706 0.707 to sample deep primordial mantle signatures202 pres- 87 Sr/86Sr ent in only the hottest and most buoyant (deep) man- tle plumes196. It is hypothesized that these ocean island Fig. 4 | Isotope systematics in global plume volcanics. Negative μ182W anomalies basalt 182W anomalies reflect a contribution from appear in some plume-generated ocean island basalt volcanoes, with no relation to Earth’s core, which has preserved a very low μ182W value long-lived 87Sr/86Sr signatures, but a relation between μ182W and the presence of recycled (the deviation of 182W/184W from the terrestrial stand- oceanic and continental material does emerge from the data239. One interpretation of ard in ppm). This extreme core value developed because these observations is that some deeply sourced mantle plumes with strongly negative μ182W anomalies have inherited a W-isotopic signature of Earth’s core86,87, but other tungsten is a moderately siderophile element that, dur- interpretations exist, as discussed in the main text. The grey band represents the ing core formation, became enriched in the Earth’s core 2σ reproducibility (±5 ppm) on the terrestrial tungsten isotope standard, which, by relative to the short-lived, lithophile radioactive parent definition, has μ182W = 0. Data and errors shown are 2σ measurement errors summarized (182Hf) that remained in the mantle86,87. It is possible that in refs87,239. We note that the estimated μ182W is as low as −220 in Earth’s core328. this core material is partitioned back into the mantle at nATuRe RevIeWS | EArTH & EnVIronmenT volume 2 | June 2021 | 389 0123456789();: Reviews the base of mantle plumes, aided by silicate melting87 though the high 3He/4He Yellowstone hotspot231 that is and possibly in ULVZs at the core–mantle boundary positioned outside of the LLSVP regions73,160 suggests (Fig. 1b). However, the origin of the low μ182W values as that primordial domains might also exist outside of observed in the mantle sources of ocean island basalts the LLSVPs195. However, to date, 129Xe, 142Nd and 182W can be attributed to late accretion203 or early Hadean sil- short-lived isotope datasets in ocean island basalts are icate Earth differentiation204, and it will be important to still statistically too small (Fig. 4) to conclusively link further evaluate which models are consistent with the the Hadean isotopic anomalies to the LLSVPs. Much positive μ182W and 3He/4He correlation in hotspot lavas87. larger Sr–Nd–Pb isotope datasets confirm that hot- spots with geochemically enriched (low) 143Nd/144Nd Mantle plume heterogeneities inferred from hotspots. are geographically linked to both of the LLSVPs82,232,233. In addition to sampling ancient reservoirs in the deep These enriched mantle domains are concentrated Earth, mantle plumes erupt lavas that exhibit geochem- in the Southern Hemisphere234 and appear to exist at ical signatures associated with several different ancient the largest (hemispheric) length scale in the mantle46, subducted crustal materials. These surficial geochemi- although geochemical heterogeneities also exist at much cal signatures in mantle plume source regions, therefore, shorter centimetre to kilometre length scales235 or even provide first-order evidence that Earth is recycling its sub-centimetre scales in peridotite samples236. Some lithospheric plates in subduction zones at a global scale have suggested that only the African LLSVP hosts these and is resupplying the deep source regions of mantle enriched (low) 143Nd/144Nd mantle signatures237, but such plumes with various crustal materials205–210. Radiogenic arguments hinge on the selection of plumes195 and more isotopes (such as 87Sr/86Sr, 143Nd/144Nd, 206Pb/204Pb and work is needed to better identify reservoir geometries 176 Hf/177Hf) provide insight into which, how many and in and dynamics. Global hotspots with higher radiogenic what way different recycled materials are involved in the Pb isotopes (with increased 206Pb/204Pb) seemingly are chemical dynamics of mantle plume formation5–7,45,47,74,75, not linked to the LLSVPs75 and, thus, leave the location of and show that many of the global hotspot systems have this scarce mantle domain uncertain. Finally, it has been two, three or more distinct components in their mantle argued that neither recycled oceanic crust nor continen- plume source regions76,79,82,211–213. tal crust are specifically associated with the LLSVPs119 Oceanic and continental crust, and associated sed- and that hotspots with low 143Nd/144Nd signatures and iments, have long been thought to contribute to much hotspots with radiogenic Pb isotopic signatures appear to of the mantle’s compositional variability sampled by be spatially decoupled in Earth’s mantle195. This decou- plumes5–7,214–217. Recycled marine carbonate218, recy- pling is unexpected, given that both mantle domains are cled metasomatized oceanic or continental mantle likely to have formed via subduction and recycling as lithosphere219–222 and processes occurring at the core– part of the plate tectonic cycle. mantle boundary223 are among the proposed mecha- ULVZs are argued to be compositionally distinct nisms for the generation of compositional variability in from their surrounding mantle and from LLSVPs the mantle and, consequently, in providing the building when they are hosted within those dense layers238. blocks of (deep) mantle plumes. However, the hypoth- Thus far, only a few plume conduits, including Hawaii, esis that ancient subducted oceanic crust is returned to Samoa, Iceland and Marquesas, can be associated with the surface in upwelling plumes was challenged because ULVZs131–134. However, it is not yet possible to evaluate subducted crust can appear too dense, and, instead, it whether ULVZs sample geochemical mantle reservoirs has been postulated that upper mantle peridotite is the that are different from LLSVPs. Until the ULVZs have predominant lithology melting beneath plume-fed been conclusively mapped out in seismic images, it will hotspots110. However, other studies show that recycled not be possible to determine whether ULVZ-related mafic (eclogite and pyroxenite) lithologies (derived from plumes have a different composition than plumes that subducted oceanic crust) are, ultimately, needed in gen- are not related to ULVZs, and this currently limits erating the major element compositions of ocean island what can be inferred about the isotope geochemistry basalts224–226. Recycling of materials that were once at the of the ULVZs. Ultimately, the evolution of these geo- Earth’s surface and from as early as the Archaean207,227,228 chemical domains is linked to plate tectonic processes, therefore appear to be required to explain the geochemical including recycling of oceanic and continental crust variability in plume-fed hotspot lavas. and transport by plumes, and these processes control the composition, location, size, amount of chemical Primordial and recycled reservoirs inferred from overprinting239 and longevity of geochemical reservoirs plume-fed hotspots. Mechanisms for the long-term in the Earth’s interior. preservation of the Hadean geochemical anomalies are imperfectly understood229, but storage in dense, viscous Chemical plume structures. Surface expressions of domains of the deep mantle can isolate Hadean-formed plume-fed volcanism in the oceanic realm are highly reservoirs from convective motions of the mantle146,150,230. varied. A notable complexity is the formation of The two LLSVP regions at the core–mantle boundary are double-track volcanic hotspot trails that exhibit a bilat- Peridotite attractive locations in this regard83,118 and plume-fed hot- eral geochemical zoning or striping over millions of The dominant mantle rock spots may provide important clues. For example, hotspots years of plume history. These kinds of surface expres- type mostly made from silicate minerals olivine and pyroxene, with the highest (most primordial) 3He/4He ratios appear sions can be governed by the make-up of the plumes rich in Mg and Fe, with less to be positioned over the LLSVPs118,119. This appears con- themselves76–82 and by their interactions with the over- than 45% silica. sistent with an ancient source for these seismic features, riding tectonic plates and any changes in plate motion 390 | June 2021 | volume 2 www.nature.com/natrevearthenviron 0123456789();:  Reviews and direction240,241. The presence of these geochemically in the plume conduit and melting rates, and differences in resolved dual volcanic trends in ocean island systems was the refractory nature of various elements to explain geo- first noticed for the Hawaiian Islands242 and follow the graphic variations in isotopic compositions251,256 or the subparallel Loa and Kea volcanic tracks243 (Fig. 5). Since generation of small-scale sublithospheric cells22. Melting then, these double-track trends have been observed in of peridotite and pyroxenite lithologies at differential many other hotspot tracks, including Cape Verde, Easter, depths in a tilted plume conduit241 could then lead to Galápagos, Marquesas, Samoa, Society, Tristan-Gough the emergence of double-track volcanism, but this is and Rurutu79,81,212,244–250. dependent on a change in the direction of the overriding There are currently several models on the origin of the tectonic plate. chemical structure of mantle plumes and the role that this When plume tilt and plate motion are aligned — for structure plays in generating the observed double-track example, as hypothesized for the Northwest Hawaiian hotspot trends. The primary models (Fig. 5) invoke dif- Ridge — then both melt sources overlie each other, and ferent mantle components in an unzoned heterogeneous no zonation is observed. When a change in plate motion plume22,241,251, a concentrically zoned plume241,252–255 or a would occur — such as the proposed shift in Pacific Plate bilaterally zoned plume76–79,81,82,246. The unzoned hetero­ motion ~6 million years ago — the lithospheric motion geneous plume model calls for plumes with different could become faster than plume tilt can adjust for, result- mantle components that are evenly distributed or contain ing in a geographic separation of deeply sourced peri­ no notable internal organization241,251,256. These models dotite melts (Kea composition) and shallower secondary typically rely on variations in both the temperature profile pyroxenite melts formed from eclogitic melt reactions a East Kea N Nīhoa Middle 100 km West ? Bank West Bilaterally Nīhoa zoned model Kauaʻi Loa 7–8 Ma Niʻihau W. Kaʻena Central Kea East Koʻolau Kauaʻi Kauaʻi 2 Ma 5 Ma Waiʻanae Makapuʻu ? E. Molokaʻi Loa W. Molokaʻi W. Maui Lānaʻi Haleakalā Kahoʻolawe Hāna Ridge Concentrically zoned model Kohala Māhukona Mauna Kea b 0.98 Hualālai Kīlauea 0.97 Mauna Loa 0 Ma Lōʻihi Pb*/206Pb* 0.96 0.95 LOA trend 208 KEA trend 0.94 0.93 0.92 800 700 600 500 400 300 200 100 0 Approximate distance to Kīlauea (km) Fig. 5 | Isotopic zonation of the Hawaiian plume. Double-track volcanism present among the Hawaiian Islands as evident in Pb-isotopic ratios for lava flows as a function of distance from the active underwater Lō‘ihi volcano264. a | The location of the Loa (blue) and Kea (red) isotopic trends are superimposed on a map of the Hawaiian Islands with its major active and inactive volcanic centres (triangles). b | The box-and-whisker plots of 208Pb*/206Pb* provide a direct measure of the difference in 208Pb/204Pb and 206Pb/204Pb as measured in the Hawaiian lavas relative to the composition of chondritic meteorites that formed our Solar System ~4.6 billion years ago. The 208Pb*/206Pb* ratios, therefore, provide insight into the chemical heterogeneity within the Hawaii plume, which, for this isotopic ratio, is the time-integrated function of the Loa and Kea plume source differences in Th/U. Ma, million years ago. Figure adapted with permission from ref.264, Wiley. © 2019. American Geophysical Union. All Rights Reserved. nATuRe RevIeWS | EArTH & EnVIronmenT volume 2 | June 2021 | 391 0123456789();: Reviews (Loa composition). This model infers that double tracks thickness and the lithology that is melting — can alter only appear during periods of plate motion change, and, the resulting lava compositions. In addition, changes in therefore, does not explain the complex zonation pat- plate motion can strongly influence how different com- terns observed along, for example, the Tristan-Gough ponents in the plume are sampled. Therefore, multiple track on the African Plate79. petrological, chemical and physical parameters need to The concentrically zoned plume model argues that be considered when further refining and testing models plumes concentrate the hottest and densest materials of plume structure. in their centres during ascent from the core–mantle boundary241. Concentric zonation is mostly seen in the Mantle plumes and plate tectonics distribution of primitive helium isotopic signatures253–255. Understanding if and why certain plumes move, in The origin of this zonation has been hypothesized to which directions, in unison or not, and how fast is an result from higher rates of melting in the plume core rela- ongoing debate. In this section, the evidence that plumes tive to the surrounding asthenosphere253,254 or preferential can move independently and at speeds typically less than concentration of dense, primitive material near the core of plate tectonic movements are described. We also discuss the plume255,257. Alternatively, preferential distribution how plumes provide insights into possible reorienta- of helium into carbonate melts could result in more prim- tions of the entire Earth relative to its spin axis, during itive, higher 3He/4He-bearing melts ascending faster (and events of true polar wander, and how they are consid- more vertically) than solid plume material, and, thus, ered powerful initial mechanisms in the global plate resulting in a geographically zoned pattern of distinct tectonic cycle, for example, in breaking up continents helium signatures on the surface, which are not reflective or initiating subduction. of the underlying chemical structure of the plume258. The bilaterally zoned plume model postulates that Distinguishing plume motions. The presumed stability the geochemical zonation observed among most ocean of thermal mantle plumes initially allowed scientists to island chains originate from a mantle plume structure use the shapes and age progressions of seamount trails that is divided into two distinct chemical domains. The to derive directions and speeds of past plate motions, and, plume could contain a bilaterally continuous structure80, with those models in hand, to chart out the positions of two zones consisting of vertically continuous filaments tectonic plates back through geological time1,2,159,169,265–269. with some spacing in between76, or a partly ordered In these models, fixed mantle plumes were presumed to structure with some mixing between the zones251. As persist over more than 100 million years. It would be some mantle plumes appear to be rooted on the bound- expected that all seamount trails forming on a particu- ary between the LLSVPs (and/or ULVZs) and the geo- lar tectonic plate would record the same history of plate chemically depleted ambient lower mantle (Fig. 2), it is motions around the same Euler poles. It would also be possible that zoned plumes sample distinct materials expected that, when plate motion changes occurred, from these LLSVP ‘edge’ domains in the lowermost the timing of the bends (or turns) that would form in mantle80,81,259,260. each seamount trail would be contemporaneous (Box 1). In the case of Hawaii, the southern Loa component In other words, in a fixed mantle plume scenario, the would represent incorporation of more enriched LLSVP geometries between different seamount trails would be materials, whereas the northern Kea component would fixed and their chronologies identical. represent ambient (more primordial) lower mantle Palaeomagnetic inclination data from seamount materials82 that are common to most Pacific plumes261. trails indicate that plumes are, in fact, not stationary However, to explain the general (but incomplete262) with respect to the spin axis166,270,271. Distance compar- absence of a Loa component in the Hawaiian plume isons between coeval seamount trails show that plumes prior to ~5 Ma, it is required that the LLSVP-derived are moving away from or closing in on each other60,272, Loa component only became entrained in the plume and the 40Ar/39Ar dating of the Louisville and Rurutu conduit intermittently between ~47 and 6.5 Ma, and hotspot tracks on the Pacific Plate show that the most then more consistently from 6.5 Ma to the present day263. acute parts in their bends (not as pronounced or Alternatively, this shift from Kea-only to dual Loa-Kea clearly visible as for Hawaii) appear to occur around components could also have resulted from a change in 47 Ma and thus about 3 million years earlier than the plate motion around 6 Ma, as the plate traversed over Hawaiian-Emperor bend 60,248,273,274 that exhibits its a bilaterally zoned plume264. Similar bilateral trends are strongest bending around 50 Ma. For these Cretaceous seen within several other Pacific hotspots81,246,259, allow- and early Cenozoic times, analysis of related palaeomag- ing us to link surficial geochemical signatures to the netic data and age dates concluded that the Hawaiian lowermost mantle geophysical domains (such as LLSVPs hotspot moved at 48 ± 8 mm per year between 63 and and ULVZs). 52 Ma, whereas other Pacific hotspots have moved The bilateral zoned plume appears to be the cur- much more slowly275. Similarly, based on age progres- rently favoured model; however, a combination of sions along tracks and changing distances between different models is also feasible. For instance, plumes tracks, a total relative motion of 53 ± 21 mm per year Euler poles can contain bilateral zonation of recycled and ambient between Hawaii and Louisville and of 57 ± 27 mm per Poles describing the rotation mantle components, whereas dense primordial com- year between Hawaii and Rurutu has been found, point- of a plate on a sphere. Any rigid plate motion can ponents preferentially concentrate in the plume core. ing to a likely large individual Hawaiian hotspot motion be described as rotation With all plume models, the depth and degree of melt- from 60 to 48 Ma, compared with more limited plume around an Euler pole. ing — which relate to plume temperature, lithospheric motions for Louisville and Rurutu60. Such observations 392 | June 2021 | volume 2 www.nature.com/natrevearthenviron 0123456789();:  Reviews Box 1 | Scenarios for Hawaiian–Emperor bend formation The prominent 120° bend in the Hawaiian–Emperor seamount trail (HEB; see figure) was first interpreted as representative of a major change in Pacific Plate motion267 occurring ~47 million years ago (Ma)274. However, an absence of geological evidence for a change in Pacific Plate motion at that time, and recognition that hotspots are mobile, led to the proposal that the HEB might represent Hawaiian hotspot motion, which came to an abrupt stop at the time of the bend329. This hypothesis was tested through the analyses of magnetic palaeolatitudes in scientific ocean drilling cores recovered from Emperor seamounts, which support a notable southward hotspot motion between ~80 and 47 Ma (refs270,275,330). Comparison of the relative distances of Pacific hotspots using age progressions along three tracks (Hawaii, Louisville and Rurutu) show the Hawaiian hotspot moving southward, with a maximum ~600–700 km of motion occurring between 62 and 47 Ma that is insufficient to explain the entire, more than 2,000 km, length of the Emperor chain60. There remains an active debate on the relative contribution of hotspot drift and plate motion changes to the shape of the 120° bend, with scenarios ranging from the HEB being entirely caused by a ~60° change in plate motion331,332 to being mostly caused by hotspot drift271,275,276. In this debate, however, it is becoming apparent that at least one-third of the length of the Emperor chain can only be explained by a southward Hawaii plume motion in the beginning of the Cenozoic. The three plate reconstructions shown in the figure all use the Antarctic Plate circuit. Two of them (refs280,333) use relative motions in Zealandia as additional constraints, whereas (ref.334) does not. The mismatch between the predicted 47-Ma hotspot locations and the actual location of the bend (Daikakuji Seamount) is consistent with ~1 cm per year south-east hotspot motion (short orange arrows) from 47–0 Ma that is obtained by numerical models276,332. The mismatch between the predicted 61-Ma hotspot locations and Suiko Guyot can be amended by adding another second rapid southward hotspot motion (long maroon arrow) that is consistent with age progressions along the hotspot tracks60 and numerical modelling276. Seton et al. (2012) Tetley et al. (2019) Suiko 61 Ma Matthews et al. (2016) 600–700 km of hotspot motion 62–47 Ma Nintoku 56 Ma –100 0 100 Gravity anomaly (mGal) Koko 49 Ma Daikakuji 47 Ma Hotspot motion 47–0 Ma Hawaii 0 10 20 30 40 50 60 70 Age (Ma) are not unexpected in a convecting mantle, as computer kinematics-based modelling approach obtains hotpot models show that rising mantle plumes will be advected motion rates278 that are highly variable between 2.5 mm in Earth’s overall mantle circulation regime62,163,276,277 and per year (Afar) and 49.4 mm per year (Caroline), with cause the locations of hotspots on the Earth’s surface to Hawaii moving at a relatively low speed of 11.6 mm per wander over geological time. year since 5 Ma. This same study also shows that the For the most recent 5 million years in Earth’s history, Pacific hotspots move at speeds between 10 and 50 mm the rate of motion of major hotspots can be computed per year, whereas Atlantic Ocean and Indian Ocean hot- differently using a maximum likelihood optimization spots would move more slowly, below 20 mm per year. that incorporates present-day plate motion models278 Yet another alternative kinematics-based approach was and which is compared with the azimuths and age developed280 to determine the motions of major hot- progressions of hotspot tracks 279. This alternative spots for the last 80 million years by estimating misfits nATuRe RevIeWS | EArTH & EnVIronmenT volume 2 | June 2021 | 393