Plume Conduit Research Articles

Paleolatitudes of the Kerguelen hotspot: new paleomagnetic results and dynamic modeling

The Kerguelen Plateau, a Large Igneous Province in the southern Indian Ocean, was formed as a product of the Kerguelen hotspot in several eruptive phases during the last 120 Myr. We obtained new paleolatitudes for the central and northern Kerguelen Plateau from paleomagnetic investigations on basalts, which were drilled during ODP Leg 183 to the Kerguelen Plateau–Broken Ridge. The paleolatitudes coincide with paleolatitudes from previous investigations at the Kerguelen Plateau and Ninetyeast Ridge (the track of the Kerguelen hotspot) and indicate a difference between paleolatitudes and present position at 49°S of the Kerguelen hotspot. We show that true polar wander, the global motion between the mantle and the rotation axis, cannot explain this difference in latitudes. We present numerical model results of plume conduit motion in a large-scale mantle flow and the resulting surface hotspot motion. A large number of models all predict southward motion between 3° and 10° for the Kerguelen hotspot during the last 100 Myr, which is consistent with our paleomagnetic results.

Seismic evidence for a tilted mantle plume and north–south mantle flow beneath Iceland

Shear waves converted from compressional waves at mantle discontinuities near 410- and 660-km depth recorded by two broadband seismic experiments in Iceland reveal that the center of an area of anomalously thin mantle transition zone lies at least 100 km south of the upper-mantle low-velocity anomaly imaged tomographically beneath the hotspot. This offset is evidence for a tilted plume conduit in the upper mantle, the result of either northward flow of the Icelandic asthenosphere or southward flow of the upper part of the lower mantle in a no-net-rotation reference frame.

Seismic evidence for a deeply rooted low-velocity anomaly in the upper mantle beneath the northeastern Afro/Arabian continent

We present seismic results that support the presence of a small, low shear velocity anomaly deeply rooted in the upper mantle transition zone beneath southern Arabia and the Red Sea. The low shear velocity anomaly persists down to the 660 km discontinuity. It is found from the waveform inversion of 2741 Rayleigh wave seismograms taking into account several higher modes. We use records from the permanent IRIS and GEOSCOPE stations completed with data collected after various field deployments of portable stations in the Horn of Africa (INSU experiment), Tanzania, Saudi Arabia and Tibet (PASSCAL experiments). The complete dataset provides a dense ray coverage of the Afro/Arabian continent and allows shear-wave heterogeneities to be resolved with wavelengths of a few hundred kilometers. To achieve a good vertical resolution in the whole upper mantle, we analyze up to the fourth Rayleigh mode in the period range 50–80 s, in addition to the fundamental Rayleigh mode in the period range 50–160 s. We discuss whether the pattern of upper mantle shear velocity anomaly could be related to local causes or to one or several plume conduits in the region. Our lateral resolution may intuitively not be sufficient to resolve a narrow plume conduit at transition zone depths. However, we show that when a dense coverage is available, a narrow low-velocity anomaly will affect the path-average measurements for a large number of individual seismograms crossing the anomaly. In this case, the low-velocity perturbation is mapped in the tomographic model, even though smoothed by the lateral resolution. We conclude that our observation is difficult to attribute to a shallow origin or to reconcile with a single narrow plume conduit in the region. It can be explained either by several close narrow plume tails or by a broad region of upwelling.

An optical method for measuring temperature in laboratory models of mantle plumes

SUMMARY We present a method for measuring radial temperature profiles in laboratory thermal plumes using the deflection of a laser beam that passes through the fluid. Plumes are created by injecting hot corn syrup into a column of cold syrup at a well-defined rate. Every second a new radial temperature profile can be taken, which makes the method suitable for monitoring time-dependent phenomena. We compare the thermal structure of stationary plume conduits and of propagating solitary waves with numerical results obtained with a 2-D axisymmetric convection code. The agreement is excellent and shows that accurate high-resolution temperature profiles can be obtained without perturbing the flow.

Mapping the Hawaiian plume conduit with converted seismic waves

The volcanic edifice of the Hawaiian islands and seamounts, as well as the surrounding area of shallow sea floor known as the Hawaiian swell, are believed to result from the passage of the oceanic lithosphere over a mantle hotspot. Although geochemical and gravity observations indicate the existence of a mantle thermal plume beneath Hawaii, no direct seismic evidence for such a plume in the upper mantle has yet been found. Here we present an analysis of compressional-to-shear (P-to-S) converted seismic phases, recorded on seismograph stations on the Hawaiian islands, that indicate a zone of very low shear-wave velocity (< 4 km s(-1)) starting at 130-140 km depth beneath the central part of the island of Hawaii and extending deeper into the upper mantle. We also find that the upper-mantle transition zone (410-660 km depth) appears to be thinned by up to 40-50 km to the south-southwest of the island of Hawaii. We interpret these observations as localized effects of the Hawaiian plume conduit in the asthenosphere and mantle transition zone with excess temperature of approximately 300 degrees C. Large variations in the transition-zone thickness suggest a lower-mantle origin of the Hawaiian plume similar to the Iceland plume, but our results indicate a 100 degrees C higher temperature for the Hawaiian plume.

On the fate of mantle plumes at density interfaces

The motion of plumes through compositionally layered systems is investigated by laboratory experiments in order to understand the ascent modes of deep mantle plumes. Experimental plumes entrain the ambient heavier fluid and the average density of the plume head ρ P( t) increases with time. The degree of entrainment depends on the viscosity ratio of the matrix fluid μ matrix to the buoyant fluid μ buoyant, and the mass flux. Two types of plume penetration through the density interface are possible. The first, which can be characterized as a ‘pass-through mode’ (PTM), exhibits relatively little entrainment. Here the whole plume head passes directly through the interface because ρ P( t) is smaller than the density of the upper layer ρ u. In contrast, in the second type of interaction (named the ‘rebirth mode’, RBM), a large entrainment effect takes place. The plume is now greatly disturbed as it passes through the interface between the two fluid layers and entrained material remains at the density interface since ρ P( t)> ρ u. In addition, a new diapir emerges from the plume conduit and rises through the upper layer. The difference between these two types of interaction is reflected in the spatial and the temporal characteristics of geochemical data obtained from mantle plume products such as Ontong Java Plateau (OJP).

Mapping out the conduit of the Iceland mantle plume with helium isotopes

In order to investigate the lateral dimensions and the shallow dispersion of the Iceland mantle plume we have measured the 3He/ 4He ratios in closely spaced primitive (Mg#: 59–70) basaltic subglacial pillow lavas erupted along the Northern Rift Zone (NRZ) of Iceland. This detailed sampling has revealed for the first time a clear gradient in 3He/ 4He ratios along the NRZ, increasing from 8.0 times atmospheric ( R/ R a) in the northernmost samples to 20.4 R/ R a in central Iceland. In combination with previously published 3He/ 4He data, the new results from subglacial basaltic glasses from the neovolcanic rift zones of Iceland reveal a ‘plateau’ of high values ( 3He/ 4He ∼20 R/ R a), which is approximately 100 km in diameter and located over the southeastern part of the island. The high 3He/ 4He plateau is well-defined and coincides with a regional gravity minimum and a seismic low velocity anomaly within the upper mantle beneath Iceland. We suggest that this zone of elevated 3He/ 4He ratios (>20 R/ R a) outlines the width of the Iceland plume conduit at depth. 3He/ 4He ratios are well correlated with Nb/Th and Nb/Rb, but not well correlated with other isotope ratios (as indicated by Pb isotopes). The correlations between helium isotopes, the geophysical variations, and some extremely incompatible elements, such as Rb and Th, and the lack of correlation between He and Pb isotopes, are most likely due to preferential concentration of He into volatile-rich incipient melts, and the subsequent lateral dispersion of the plume at shallower levels.

Blob instability in rotating compositional convection

The onset and structure of compositional convection in a rotating system are investigated experimentally. A vertically oriented cylindrical annulus filled with NH4Cl‐H2O solution is cooled from the bottom and can be rotated about its axis at rates ranging up to 10.5 rad s−1, corresponding to Ekman numbers down to 7.5 × 10−6. The Coriolis force has a strong effect on the structure of plumes above the mush‐liquid interface. Helical motion of the conduit, which is weakly developed in the non‐rotating case, is amplified by Coriolis forces that twist the plume conduits to lie nearly horizontally. This results in secondary plumes (or blobs) that rise from the sub‐horizontal primary plume conduits. This new instability could be an efficient mechanism for producing small‐scale flow in the form of buoyant blobs that ascend through the polar regions of the outer core.

Seismic evidence that the source of the Iceland hotspot lies at the core–mantle boundary

Although Morgan proposed in 1971 that hotspots such as Iceland were the result of hot, rising mantle plumes, it is still debated whether plumes originate from a thermal boundary just above the core-mantle boundary or at the base of the upper mantle. Although seismic evidence of plumes in the upper mantle is accumulating, narrow plume conduits in the deep mantle have yet to be detected. Details of plume formation in the lower mantle have therefore remained largely unconstrained. Here, however, we present seismic evidence for the presence of a localized patch of material with ultra-low seismic wave speed, located at the core-mantle boundary beneath the Iceland hotspot, and propose that this zone represents the hot, partially molten source region of the Iceland mantle plume. Through the modelling of seismic waveforms, we constrain the seismic velocity structure at this patch of the core-mantle boundary using a numerical-analytical interfacing code5 designed to reproduce the complex interference of shear-wave phases transmitted through, and refracted at, the boundary. Although this structure is difficult to constrain precisely, our preferred model consists of a dome which is 250 km wide, 40 km high and contains P- and S-wave velocity (wave-speed) reductions of 10% and 30%, respectively.

Plume‐ridge interaction in the Easter‐Salas y Gomez seamount chain‐Easter Microplate system: Pb isotope evidence

Lead isotope ratios are reported on fresh basalt glasses from 28 seamounts located along the western end of the Easter‐Salas y Gomez seamount chain (ESC), southeast Pacific for the purpose of testing the mantle plume source‐migrating ridge sink (MPS‐MRS) model for the origin of this extensive intraplate neovolcanic seamount chain. Most of these basalts west of 103°W are less than 1.5 m.y. old. Results reveal a 1100‐km‐long westward gradient decreasing in radiogenic Pb from Salas y Gomez to the Easter microplate (EMP). The results confirm the previously established four‐point average trend based on data from Salas y Gomez, Easter Island, the east rift, and the west rift of the Easter microplate [Hanan and Schilling, 1989]. Tight linear arrays in Pb isotope space of the combined EMP and ESC data sets indicate binary mixing between the radiogenic isotope‐rich Salas y Gomez plume and the depleted asthenosphere source is the dominant factor in controlling Pb isotopic variation in the region. Mapping of the plume component mass fraction reveals not only a westward dilution of the plume component from Salas y Gomez to the EMP but also dilution with depleted material across the ESC on both flanks. The mapping is consistent with the MPS‐MRS model which calls for a subhorizontal, sublithospheric plume channel connecting the mantle plume conduit near Salas y Gomez with the East Pacific Rise, where plume flow is confined along a thermal groove. Mixing by entrainment of surrounding depleted material takes place progressively downstream. A numerical thermal lithosphere cooling model reveals a rheological boundary layer (i.e., the base of the lithosphere) shoaling by 1.3° to 3.5° from Salas y Gomez to about 100 km east of the EMP. The slope magnitude is consistent with those suggested by laboratory and numerical fluid dynamic models for hot buoyant plume‐ridge channeled flow to develop and be maintained.

Detection of mantle plumes in the lower mantle by diffraction tomography: theory

We investigate the properties of long-period seismic waves scattered by idealized plume conduits in the lower mantle. We build a schematic, yet realistic, model of the seismic velocity and density anomalies caused by a thermal plume. We devise a method to construct realistic seismograms of P-waves scattered by this anomaly. The results show that the scattered wave takes the form of an Airy phase, which arrives after the direct P-wave, at a time that corresponds to the shortest time it takes for the P-wave to travel from the source to the plume, and from there to the station. The predicted amplitude at a long period ( T≃20 s) is in the range of 1–5% of the amplitude of the direct wave. We explore the variation of the amplitude as a function of the geometrical parameters, and show that it is well explained by considering the contribution of the column of the plume in a T/4 Fresnel zone around the fastest scattered ray. We compare the amplitudes and waveforms obtained in the Rayleigh and Mie approximations, and find that, for realistic geometries, wider plumes yield a larger signal. Nevertheless, the predicted amplitudes are too small to yield a detectable signal. In the Born approximation, the image reconstruction of nearly vertical features reduces to a 2D linear inversion. We present a method, based on LSQR, to produce such images from the global set of long-period seismograms. In a companion paper [Ying Ji, H.-C. Nataf, Earth Planet. Sci. Lett., this issue], this method is applied to real data that sample the lower mantle beneath Hawaii.

Advection of plumes in mantle flow: implications for hotspot motion, mantle viscosity and plume distribution

Summary Because of their slow relative motion, hotspots, mainly in the Pacific, are often used as a reference frame for defining plate motions. A coherent motion of all Pacific hotspots relative to the deep mantle may, however, bias the hotspot reference frame. Numerical results on the advection of plumes, which are thought to cause the hotspots on the Earth’s surface, in a large-scale mantle flow field are therefore presented. Bringing the results into agreement with observations also leads to conclusions regarding the viscosity structure of the Earth’s mantle, as well as the sources and distribution of plumes. The abrupt change in direction of the Hawaiian–Emperor chain implies an upper-mantle viscosity under the Pacific of ∼ 1.5 × 1020 Pa s or less. Slow relative motion of hotspots requires high lower-mantle viscosity, unless hotspots are located at large-scale stationary upwellings that are currently unresolved by seismic tomography. For our preferred model, we obtain coherent motion of Pacific hotspots in a reference frame of no net rotation, as well as coherent motion relative to African hotspots, caused by return flow antiparallel to plate motion. Advection and regional differences in life expectancy can to a large part explain the distribution of plumes in relation to ridges, subduction zones (present and past) and seismic anomalies. Plume conduits are substantially tilted in the lower mantle. The surface motion of hotspots is often smaller than the advection rate of plume conduits in the lowermost mantle.

Melt migration and mantle chromatography, 2: a time-series Os isotope study of Mauna Loa volcano, Hawaii

We have determined the major element, trace element, and Os isotopic compositions of a stratigraphic suite of tholeiitic basalts spanning >30,000 years of the eruptive history of Mauna Loa volcano. Good correlations are observed between Os isotopes and the isotopes of Sr, Nd, Pb and He. In addition, the isotopes correlate with fractionation-corrected major element abundances within this single volcano, and provide a record of increased melting of mafic material with time at Mauna Loa. Chromatographic element fractionation during melt transport is shown to be negligible based on the good correlations of the isotopes of the compatible element Os with the other incompatible element tracers. The temporal variation at Mauna Loa is best described by the translation of the volcano over a Hawaiian plume which is radially zoned in composition. The radial zonation is a predicted consequence of thermal entrainment during flow in a mantle plume conduit.

The fluid dynamics of plume-ridge and plume-plate interactions: An experimental investigation

A number of hotspots occur at or near mid-ocean ridges (e.g., Iceland) causing elevated bathymetry, gravity anomalies, and geochemical enrichment for hundreds of kilometers along otherwise normal ridge axes. This is commonly presumed to be the result of hot plume material from the deep mantle flowing outward from a narrow plume conduit. Here we explore the dynmics of plume-ridge and plume-plate interaction using laboratory experiments. We model chemically buoyant plumes, as opposed to thermally buoyant plumes, so we ignore the effects of thermal diffusion while concentrating on the strictly fluid dynamical aspects of the interaction. We find that the spreading of a continuously fed plume beneath a fixed plate is described by a simple expression derived from a balance between plume buoyancy and viscous resistance to spreading. For plumes spreading beneath ridges we find two remarkably distinct phases of flow: At first the plume spreads approximately as it would beneath a fixed plate, then a rapid transition occurs to a steady-state condition in which the plume width along the ridge (‘waist’) is constant. This width is found to be proportional to a natural length scale that increases with the square root of the plume volumetric flux and the ratio of the mantle/plume viscosities, and decreases with the square root of the plate spreading velocity. Although our experimental conditions, with uniform mantle viscosity and no thermal diffusion, are perhaps oversimplified compared to the real Earth, the results scale reasonably well to observation. Our model suggests that sufficient time has elapsed (∼ 60 Myr) for the Iceland plume to reach steady state in its interaction with the Mid-Atlantic Ridge. Using independent estimates for the plume volumetric flux at the Iceland hotspot, and assuming a viscosity ratio of one, the model predicts a minimum waist width of about 850 km, in close agreement with the observed value of about 920 km. Data from other hotspots (Jan Mayen, Azores, Galapagos) are also consistent with the experimental results for average mantle/plume viscosity ratios of 1–3. Using estimated plume fluxes with a simple conservation of mass equation yields reasonable estimates of plume thickness ranging from 70 to 180 km for these hotspots. Thus the experimental model appears to be reasonably consistent with observation, given that thermal diffusion has been disregarded.

Fluid dynamic and geochemical aspects of entrainment in mantle plumes

A similarity solution has been developed for vertical, steady‐state mantle plume conduits by considering the boundary layer flow emanating from a point source of heat in an axisymmetric geometry. This model includes the effects of temperature and shear stress on viscosity, and incorporates depth‐dependent viscosity and thermal expansivity. Plumes with variable viscosity have upward velocities of 0.30–100 m/yr and radii of 30–250 km, depending on temperature, rheology and buoyancy flux. These results demonstrate the small lateral scale of plume features relative to the resolution of most large‐scale convection models and seismological studies. All of the plumes studied showed significant entrainment of ambient mantie surrounding the plume conduit, driven by the radial conduction of heat from the plume. This heat raises the buoyancy and lowers the viscosity of the ambient mantel, thereby entraining it into the conduit flow. For buoyancy fluxes of 0.1–10 Mg/s, similar to the range estimated for plumes in the Earth's mantle, we calculate a range of entrainment of &gt;90% to &lt;5% ambient mantle, correlating negatively with buoyancy flux. Examination of the streamlines of mantle material which is entrained into thermal plumes indicates that most of the entrained fraction originates from approximately the lower half of the layer traversed by the plume, and shows minor entrainment of upper level material. This is especially true for non‐Newtonian and depth dependent rheologies, and for depth‐dependent thermal expansivity. Upwelling of depleted upper mantle, viscously coupled to the plume flow, is proposed as a mechanism for generating post‐shield stage alkalic basalts erupted on oceanic island chains and their associated flexural arches. The existing Sr‐Nd‐Pb isotope data for oceanic basalts indicate the presence of a component which is common to hotspot basalts worldwide, and which is distinct from the upper mantle source of mid‐ocean ridge basalts. This component (termed “FOZO” by Hart et al. (1992)) has moderately depleted Sr and Nd signatures, radiogenic Pb isotopes, and elevated 3He/4He ratios. The high He isotope ratios of FOZO suggest an origin for this component in the lower mantle, and would appear to provide independent evidence to support the fluid dynamic observations for significant entrainment of lower mantle in plumes and exclusion of upper mantle. If the composition of FOZO is representative of the isotopic composition of the lower mantle, then it would appear that this reservoir has been differentiated relative to estimates for the bulk silicate earth (BSE). This may be due either to melting and differentiation at higher levels in the mantle, or to fractionation of high pressure phases from a terrestrial magma ocean.

Intershield geochemical differences among Hawaiian volcanoes: implications for source compositions, melting process and magma ascent paths

As Hawaiian volcanoes develop, their lavas systematically change in composition and isotopic ratios of Sr, Nd and Pb. These trends provide important constraints for understanding plume-related volcanism as a volcano migrates away from the hotspot. There are also geochemical differences between Hawaiian shields. In particular, lavas from adjacent shields such as Kilauea and Mauna Loa on Hawaii and Koolau and Waianae on Oahu have significant differences in abundances of some major and incompatible elements and isotopic ratios of Sr, Nd and Pb. Some incompatible element abundance ratios, such as Zr/Nb and Sr/Nb, are correlated with intershield differences in Sr and Nd isotope ratios, but these isotopic ratios are not correlated with intershield differences in major element composition, or even parent/daughter abundance ratios such as Rb/Sr and Sm/Nd. Moreover, at Kilauea and Mauna Loa the intershield differences have apparently persisted for a relatively long time, perhaps 100 ka. These intershield geochemical differences provide important constraints on plume volcanism. Specifically, (i) each volcano must have distinct magma ascent paths from the region of melt segregation; (ii) the 25-50 km distance between adjacent, but geochemically distinct, shields requires that the sources vary on a similar scale, and that the melt production region is similarly restricted. The absence of correlations between lava compositions and radiogenic isotope ratios provides evidence for significant differences in melting process such as each shield forming by a different mean extent of melting with melt segregation at different mean pressures. Two types of models are consistent with the intershield geochemical differences: (i) a relatively large radius, ca . 40 km, plume conduit with a systematic spatial distribution of geochemical heterogeneities; or (ii) a small radius, less than 20 km, plume conduit composed of geochemically distinct diapirs. Because relatively small radius diapirs of limited vertical extent are too small to create the large Hawaiian shields, a possible alternative is a continuous conduit containing solitary waves which transport geochemically distinct packets of material.

Entrainment and stirring in viscous plumes

Buoyant plumes in very viscous fluids, although laminar, entrain their surroundings as a result of conduction of heat. Once heated, the fluid takes part in the plume motion and is stirred within the plume by internal circulation. Entrainment modifies the behavior of plumes by enlarging and cooling them. Similarity solutions for several cases, including isolated ‘‘thermals’’ having fixed buoyancy, thermals having a constant internal heat generation, the initial stages of a plume formed by a continuous volume flux (a starting plume), and steady plume conduits fed by a constant source flux are reviewed. Both laboratory experiments and computations of particle paths show that stirring in thermals and starting plumes also leads to well-defined shapes for passive tracers, shapes that depend upon the magnitude of an appropriate Rayleigh number. The computations of particle paths are extended to the case of a steady plume conduit ascending through a horizontal shear flow in order to show the internal structure of the plume. Implications of these models for mantle plumes and volcanic hotspots are discussed briefly. It is possible that the structured chemical heterogeneities produced by stirring within plumes may contribute to observed chemical variability of lavas generated by plume activity.

On the dynamics of long-lived plume conduits in the convecting mantle

Long-lived plumes, responsible for mantle hotspots and associated chains of volcanism, will be deflected from the vertical as they penetrate through a mantle which is overturning on much larger length scales in response to surface cooling. Recent experiments have shown that plume inclination leads to entrainment of surrounding material as a result of coupling between conduction of heat and laminar stirring. We develop an analysis for entrainment and its effects on the structure and dynamics of continuing, steady conduits in a mantle undergoing horizontal shear. It is found that the stirring of surrounding mantle with plume source material can result in major modifications of the diameter, temperature and mass flux in plumes. This model is a departure from the axisymmetric vertical pipe models of plumes and predicts that the temperature in a defected plume will decrease with height, thus reducing the plume temperature in the upper mantle while allowing a sizeable temperature difference at the source. The cooling of plume conduits, after they are deflected by plate motion, can lead to a lower maximum temperature and lower MgO contents of basalts along some linear hotspot tracks compared with picrites from their associated flood basalt provinces. The effects of entrainment depend primarily on the buoyancy (or heat) flux carried by the plume, with greater cooling for smaller fluxes. Strong continuing plumes such as Hawaii, Easter Island, Macdonald, Tahiti and Reunion are predicted to produce picrites having MgO contents comparable to the picrites found in continental flood basalt provinces. Weak plumes such as Crozet, Bouvet and Juan de Fuca will give only cooler melts. Through entrainment long-lived plumes can sample the whole depth of the mantle from the source to the surface, and stirring within the plume leads to compositional zonation which may contribute to compositional heterogeneity in hotspot melts. We also predict that plumes originating at the core-mantle boundary and having buoyancy fluxes less than about 3 × 10 3 N s −1 will not give surface hotspots, in good agreement with the lower limit of hotspot buoyancy fluxes inferred from seafloor topography.

Mantle plumes

The current state of our understanding of mantle plumes and their effect on other features within and on the Earth is reviewed The physical conditions necessary for the existence of plumes, i.e., narrow vertically oriented cylinders of reduced viscosity and density, are discussed and it is argued that these conditions occur within the mantle. Heat is believed to provide the reduction in viscosity, via thermally activated creep, and the reduction in density, via normal thermal expansion, although metasomatic plumes could occur. The heat for the plumes probably originates with the core. Theoretical and experimental models of plume structure and of plume initiation are reviewed, with special attention paid to the possibility of solitary waves propagating up the plume conduit. The suggestion that impacts of extraterrestrial bodies could initiate plumes is critically investigated and found to be implausible. Possible ways that plumes can interact with general mantle convection are reviewed.

Thermal entrainment by deflected mantle plumes

HOTSPOTS are thought to result from deep mantle plumes, perhaps representing an ancient geochemical source isolated from the uppermost mantle. As they rise through the mantle, thermal plumes entrain surrounding material by thermal conduction, and small plumes or narrow plume conduits may consist of only a small fraction of original source material. Here we present new experimental results which show this effect dramatically for plume conduits that are deflected horizontally by shear flow in the mantle beneath lithospheric plates. Thermally entrained material can become concentrated at the centre of a plume (or surface hotspot trace) with original source material in the periphery. Also, the diapiric instability observed in chemical plume conduits is suppressed by thermal entrainment. Varying degrees of geochemical enrichment observed at hotspots such as the Galapagos Islands may result from thermal entrainment, implying that mantle plumes are driven by thermal (rather than compositional) buoyancy.