Research Article

Finite-Frequency Tomography Reveals a Variety of Plumes in the Mantle

See allHide authors and affiliations

Science  16 Jan 2004:
Vol. 303, Issue 5656, pp. 338-343
DOI: 10.1126/science.1092485


We present tomographic evidence for the existence of deep-mantle thermal convection plumes. P-wave velocity images show at least six well-resolved plumes that extend into the lowermost mantle: Ascension, Azores, Canary, Easter, Samoa, and Tahiti. Other less well-resolved plumes, including Hawaii, may also reach the lowermost mantle. We also see several plumes that are mostly confined to the upper mantle, suggesting that convection may be partially separated into two depth regimes. All of the observed plumes have diameters of several hundred kilometers, indicating that plumes convey a substantial fraction of the internal heat escaping from Earth.

Hotspots are characterized by higher temperature, topographic swells, and recent volcanism with isotopic signatures distinct from those that characterize mid-ocean ridge or andesitic basalts (13). The best known example is the Hawaii-Emperor volcanic chain, which may have formed as the Pacific plate moved over a deep magmatic source (310). Narrow thermal up-wellings in the form of plumes are commonly observed in laboratory experiments (11, 12) and numerical simulations (1315), and deep-mantle plumes have been invoked to explain flood basalts, the isotopic signature of ocean island basalts, and the topography of the swells and plateaus that often accompany volcanic hotspots. Although this has led to a coherent (albeit incomplete) theory of much of the geology that characterizes hotspots, undisputed evidence for the existence of lower-mantle plumes in tomographic images of the mantle is lacking. High temperatures reduce the velocity of seismic waves, so that plumes should be evinced as columnar low-velocity anomalies. In the absence of convincing tomographic evidence, it has recently been argued that hotspots could instead be the manifestation of shallow, plate-related stresses that would fracture the lithosphere, causing volcanism to occur along these cracks (1618).

The inversion. A unique feature of our tomographic inversion is the use of finite-frequency sensitivity kernels (19, 20) to account for effects of wavefront healing on the travel times of low-frequency P waves; this enables us to combine long- and short-period data sets. We use a remeasured, expanded set of long-period data and very carefully selected short-period delay times, and we adapt the model parameterization to the lower resolution at depth. Global tomographic models of seismic P-wave velocity have so far relied almost exclusively on classical ray theory. In this infinite-frequency approximation, the travel time of a P wave is only influenced by seismic velocity along an infinitesimally narrow path—the seismic ray. This simplifies the mathematics, but it ignores sensitivity of the travel time to velocity structure off of the ray, within a volume known as the Fresnel zone (21). The cross-ray diameter of the Fresnel zone is on the order of (λL)1/2 for a wave of wavelength λ and ray length L, and can be in excess of 1000 km at the long periods we consider. As a result of diffractive wavefront healing, objects much smaller than the width of the Fresnel zone will not appreciably influence the travel time of the wave (22, 23). For heterogeneities comparable in size to the Fresnel zone, one may account for these wavefront healing effects with the use of a new method of interpretation (19, 20), which we refer to as finite-frequency tomography. A preliminary analysis of low-frequency P-wave arrival times shows that the amplitudes of deep, small-scale velocity heterogeneities are underestimated by 30 to 60% when interpreted by classical ray theory (24).

Mantle plumes are probably narrow, and their images are potentially degraded if we ignore the effects of wavefront healing (25). We present the result of a finite-frequency tomographic inversion based on the travel times of 66,210 P-waves and the differential travel times of 20,147 PP-P and 2382 pP-P waves, with a dominant period of 20 s, combined with 1,427,114 short-period P and 68,911 short-period pP travel times extracted from bulletins (26). The travel times of the long-period phases were measured by cross-correlation with a synthetic pulse (27) and were inverted using three-dimensional finite-frequency sensitivity kernels (19); the short-period travel times were inverted using ray theory. Although part of the long-period data set also served to construct model P16B30 (28), the original seismograms were reanalyzed, and previously undetected clock errors have been corrected (29). The standard deviations of the long-period times were assigned to three accuracy classes and range from 0.5 to 1.15 s (24). The short-period travel times were picked onset times reported to the International Seismological Centre (ISC) and later reinterpreted with smaller uncertainties (30). Of a total of 9.2 million P measurements, we selected only the ∼16% listed with two-decimal precision. The standard deviations of the short-period times are 0.9 s for P and 1.1 s for pP (31); about 45,000 outliers with deviations larger than 3σ were rejected after a first iteration.

We simultaneously inverted for the fractional perturbation in the compressional velocity, δvP/vP, and the perturbations in the earthquake hypocentral parameters (origin time, longitude, latitude, and depth). For technical aspects of the inversion, see (29).

Subducted lithospheric slabs are large features that cause little or no wavefront healing. Our slab images are in good agreement with earlier studies; therefore, we restrict our analysis to the newly imaged plumes.

As in every high-resolution tomographic study, the fact that we have far more data than unknowns does not guarantee that every model parameter is well resolved. Much of our effort has been spent on determining which of the imaged plume features are reliable. Lack of resolution may take different forms: (i) An existing plume may not be imaged with sufficient contrast over its full length, or to its source at large depth, where resolution often decreases; (ii) the regularization may spread a shallow anomaly to larger depth in the image (“leakage”); and (iii) the regularization leads to horizontal “smearing” of an anomaly, resulting in a larger anomaly image with a smaller amplitude than present in Earth.

The resolution tests we have undertaken enable us to determine the minimum radius that a real plume must have for it to be visible in our tomographic images. This minimum plume radius is listed in Table 1. The imaged plume radii should be seen as generous upper limits; because of the limitations to the resolving power combined with the effects of smoothing, the actual plume radii may be much smaller. The depth listed in Table 1 is the deepest level at which the absolute contrast |δvP/vP| exceeds 0.3%. We code these maximum depths in Table 1 with “≥” whenever we conclude from the resolution tests that the absence of a plume at greater depth may be due to a lack of resolution, or with “≤” whenever there is a possibility that the image is generated by leakage to this depth. In cases where we determined that the resolving power is sufficient, we added “∼” to emphasize that the depths in Table 1 are only estimates. Even in the case of well-resolved plumes, the depths are uncertain to several hundred km, particularly in the deepest mantle. This codification of depth values in Table 1 is somewhat incomplete in the sense that resolution may depend on the unknown width of the plume. See (29) for more extensive resolution information.

Table 1.

Summary of the results for the 32 hotspots present in our tomographic images. Depth limits and minimum radius (in lower mantle unless the plume is confined to the upper mantle) have been determined from the resolution analysis (29). Code for plate names: af, African plate; an, Antarctic plate; au, Australian plate; eu, Eurasian plate; nz, Nazca plate; pa, Pacific plate; sa, South American plate.

Label Name Latitude and longitude Plate Depth (km)View inline Radius (km)View inline Remarks
AF Afar 7°N 39°E af ≥1450 200 Could be a deep plume
AR Atlantic Ridge 15° & 25°N 45°W na ∼1900 200 Lack of resolution at the base of the mantle gives rise to horizontal smearing
AS Ascension 8°S 14°W sa ∼2800 100 Robust
AZ Azores 38°N 26°W eu ∼2800 300 Lack of mid-mantle resolution may be responsible for merging with Canary
BV Bouvet 54°S 3°E af ≤1450 400 Connection to African superplume is not well resolved
BW Bowie 53°N 136°W pa ≤650 100 Plume wider than 100 km would be visible beneath 300 km depth
CN Canary 28°N 18°W af ∼2800 400 Robust
CV Cape Verde 15°N 24°W af ≥1900 300 Lack of resolution in the deep mantle
CR Caroline 3°N 167°E pa ≥1000 300 Lack of resolution at the bottom of the mantle
CC Cocos/Keeling 17°S 95°E au ≤1000 200 Nearby plumes strong only in the upper mantle
CK Cook Island 22°S 158°W pa ≥1450 200 Lack of resolution in the lower mantle beneath nearby plumes
CS Coral Sea 15°S 155°E au ∼2800 300 Robust, does not reach the surface, stops at 1450 km depth
CZ Crozet 46°S 50°E an ∼2350 400 Vertical leakage present, but confined above 1450 km depth; same origin as Kerguelen at 2350 km depth; synthetic plumes are separated down to 1000 km depth
ES Easter 27°S 108°W nz ∼2800 400 Robust
EA East Australian 41°S 146°E au ∼650 100 Robust
EF Eifel 50°N 4°E eu ∼650 100 Robust
ET Etna 38°N 15°E eu ≤1000 200 Strong only in the upper mantle
SL East of Solomon 5°S 165°E pa ∼1000 Nearby plumes robustly resolved
GL Galapagos 92°W nz ≤1000 300 Strong only in the upper mantle
HN Hainan 20°N 110°E eu ∼1000 200 Robust
HW Hawaii 19°N 155°W pa ≥2350 300 Not enough constraint on depth
IC Iceland 64°N 17°W na ≤1000 100 Strong only in the upper mantle (see discussion in text)
IO Indian Ocean 35°S 100°E au ≥1900 400 Not well resolved
JF Juan de Fuca/Cobb 46°N 130°W pa ≤1000 100 Strong only in the upper mantle
JZ Juan Fernandez 34°S 81°W nz ≥1450 300 Lack of resolution below 2350 km depth in the nearby plumes
KG Kerguelen 50°S 69°E an ∼2350 400 Robust
LS Louisville 54°S 141°W pa ≥1000 300 Very weak, newly discovered anomalies located at 55°S, 150°W, and 60°S, 120°W
RN Reunion 21°S 56°E af ≥1900 200 Could be a D″ plume
SM Samoa 15°S 168°W pa ∼2800 200 Robust, except between 1000 and 1450 km depth
SJ South of Java 12°S 112°E au ∼2800 300 Does not reach the surface, stops at 1450 km depth
SY Seychelles 5°S 56°W af ∼650 Nearby plumes robustly resolved
TH Tahiti 18°S 148°W pa ∼2800 300 Robust, except between 1000 and 1450 km depth
  • View inline* See text for explanation, including meaning of symbols.

  • View inline Minimum radius of the plume constrained from the resolution tests.

  • Deep plumes. To emphasize features, including deep plumes that are vertically continuous over all or much of the lower mantle, we compute a vertical average of the P velocity anomaly δvP/vP in the lowest part of the mantle (1800 to 2800 km depth) (Fig. 1). The African superplume extends high enough to survive this averaging over 1000 km, and is visible as the large low-velocity anomaly beneath southern Africa. Inspection of the unaveraged tomographic model shows that the superplume extends locally upward to depths of 1500 km, in accordance with previous observations (32, 33). This anomaly is part of a broad low-velocity region that underlies the Atlantic, the African continent, and much of Europe. The broad anomaly in the lower mantle beneath the Pacific Ocean, south of the equator, is different in character. Here there are several large velocity anomalies enhanced by the averaging, which can be identified as individual plumes rising from the “superplume”: beneath the Coral Sea (15°S, 155°E), east of the Solomon Islands (5°S, 165°E), beneath Samoa (15°S, 168°W), and a broad anomaly centered beneath Tahiti (18°S, 148°W).

    Fig. 1.

    Vertical average over the lowermost 1000 km of the mantle of the relative velocity perturbation δvP/vP. The averaging emphasizes features that are continuous with depth. Map has been wrapped around to have complete views of both the Atlantic and the Pacific oceans.

    North of the equator, the lower mantle beneath the Pacific plate exhibits several high-velocity anomalies forming a ring that extends beneath East Asia and the eastern part of the Indian Ocean. In the unaveraged model, fast anomalies with perturbations greater than 1% are observed beneath Tonga and Asia, the latter presumably caused by the subduction of the Tethys slab (3439). Together, these form a ring of high-velocity anomalies around the Pacific, which reaches down to the D″ region. Finally, the polar regions are characterized by lower mantle of opposite velocity anomaly at all longitudes; if we assume that these anomalies mainly reflect temperatures, then the deep mantle is hot beneath the North Pole but cold beneath Antarctica. This suggests that the large J3 coefficient in the geoid [the “pear” shape of Earth (40)] may be caused by this deep-seated bipolar anomaly.

    The Ascension, Azores, Canary, Easter, Samoa, and Tahiti hotspots all have well-resolved deep-rooted origins near the bottom of the mantle (Fig. 2) (figs. S2 to S9). Ascension and St. Helena (fig. S2) merge at about 1000 km depth, a well-resolved confluence that is observed for several other deep plumes. For example, Azores and Canary (fig. S3) are distinct plumes down to 1450 km depth, where they merge together and begin to bend eastward to reach the bottom of the mantle at about 30°N, 10°W. Farther south, but less well resolved, the Cape Verde (fig. S3) plume joins this complex at 1900 km depth. Kerguelen (fig. S7) and Crozet (fig. S4) are seen to originate from a common broad anomaly located north of Crozet at 2350 km depth; the width of this anomaly is affected by horizontal smearing. Beneath the Pacific superswell, the Tahiti, Cook Island, and Samoa plumes (figs. S8 and S9) are closely spaced. The images for the Tahiti and Samoa plumes are robust and show independent features to large depth in the mantle. Cook Island merges with Tahiti at about 1450 km depth, although we cannot rule out that this is an artifact caused by lack of resolution.

    Fig. 2.

    Three-dimensional view of deep plumes present in our tomographic model. Maps are 40° by 40°, appropriately scaled with depth. Note the vertical exaggeration. The depth spacing changes at 1000 km. The color scale is the same as in Fig. 3. Two-letter identifiers show hotspot locations.

    Hawaii (fig. S6) is one of the longest lived plumes and is by far the strongest in flux, as measured by the topographic swell it creates (1, 2); the associated fast anomaly in our model falls somewhat short of expectations. This is most likely due to the dearth of seismic ray paths sampling this structure at midmantle depth, due to the distance of Hawaii from the circum-Pacific seismicity. Seismic rays from the Tonga subduction zone to North American stations are abundant, but these pass southeast of the suspected plume location near the core-mantle boundary. Resolution analysis indicates that only a very wide plume of radius >300 km would be clearly resolved. In a separate inversion using only low-frequency data (which sense a wider region around the ray), the Hawaii plume anomaly extends to 2800 km depth (24).

    Afar, Juan Fernandez (fig. S5), Cook Island (figs. S8 and S9), Caroline, Reunion, and two newly discovered anomalies near Louisville could also be deep plumes (fig. S1). Our resolution analysis shows that narrow plumes beneath these hotspots suffer a loss of resolution at increasing depth, making any deep plumes invisible.

    Shallow plumes. A relatively large number of plumes are clearly imaged only in the upper mantle: Bowie, Eastern Australia, Eifel, Etna, Iceland, Cocos-Keeling, Galapagos, and Juan de Fuca/Cobb (Fig. 3) (figs. S10 to S16). Eastern Australia (fig. S12) and Eifel (fig. S13) show robust velocity anomalies clearly constrained down to 650 km depth. Bowie and Juan de Fuca/Cobb (fig. S10) are connected to form a broad low-velocity anomaly at 300 km. Bowie, which is visible only down to 300 km depth, is not really identifiable as an isolated plume. Juan de Fuca/Cobb reaches 1000 km depth; however, the velocity perturbation at this depth is much weaker than in the upper mantle, and it is unlikely that the source region of the Juan de Fuca/Cobb hotspot is at a depth of 1000 km or greater. An origin shallower than 1000 km is also suggested for Cocos/Keeling (fig. S16), Etna (fig. S13), Galapagos (fig. S14), and Iceland (fig. S15). At about 650 km, Etna shows a connection to a plume-like low-velocity anomaly beneath the Gulf of Suez.

    Fig. 3.

    Three-dimensional views of the shallow plumes and the newly discovered plumes present in our velocity model. Plotting format, including change in depth spacing at 1000 km, is identical to that of Fig. 2.

    In our images, the very strong upper-mantle plume beneath Iceland has almost disappeared at 1000 km depth. Vertical leakage down and below this depth could easily explain the weak lower-mantle anomaly of –0.3%. We suspect that such leakage has led an earlier study to suggest a deep plume (41). It is clear from our image that the strong velocity anomaly observed in the upper mantle beneath Iceland is not generated by a large upwelling from the lower mantle.

    A large drop in the temperature dependence of vP in the transition zone (42) could make plumes a much weaker velocity anomaly in the lower mantle and therefore more difficult to image, leading to apparent source regions near 670 km. However, the fact that we do observe numerous plumes extending to the deep mantle weakens this argument, at least as a phenomenon affecting the stronger upper-mantle plumes.

    Newly discovered plumes. Several new velocity anomalies that we associate with plumes are visible in our model and are not related to well-known hotspots. A Mid-Atlantic ridge plume is identifiable with anomalies at 12°N and 25°N (fig. S1); an axial hotspot at this location has recently been hypothesized (43). A deep plume stops at about 1450 km depth south of Java (Fig. 3) (fig. S16). An anomaly feeding the Southeast Indian ridge (fig. S1) is defined down to 2350 km depth, where it merges with the anomaly located south of Java. The resolution tests indicate that all of these structures are robust. The deep plume beneath the Coral Sea (Fig. 3) (fig. S11) extends upward to 2350 km, with a weak connection to a newly discovered plume beneath East Solomon (5°S, 165°E) (Fig. 3) (fig. S11). The East Solomon plume extends upward to 1000 km, where a weak connection is visible with a plume beneath the Caroline Islands (Fig. 3). The Louisville hotspot, characterized by a well-delineated narrow island chain, has waned in recent geological times. We do not observe a plume beneath either of the disputed present-day locations (1, 44). Instead, we recover two weak low-velocity anomalies situated to the northwest and southeast of the Louisville hotspot (fig. S1), which we list as “Louisville” in Table 1. Finally, we image a hitherto unreported plume north of Reunion, approximately beneath the Seychelles (5°S, 56°E) (fig. S1). This plume extends down to 650 km and is well resolved. We note that Seychelles is a continental fragment left behind from the breakup of Gondwana, and not a conventional oceanic island (45).

    Absent plumes. A number of proposed plumes (1, 2, 9) do not exhibit associated velocity anomalies in our images. Our resolution tests show that even a 0.3% anomaly of only 100 km radius would be visible beneath Yellowstone, indicating that there is not a substantial plume. Guadalupe is not an isolated low-velocity anomaly in our tomographic image, but is rather part of a broad low-velocity region connecting Bowie, Juan de Fuca/Cobb, and what seems to be a signature of the East Pacific Rise (15°N, 115°W), that is, south of Guadalupe. Perhaps because of the absence of a co-sited station, we do not resolve a plume in the upper mantle beneath Macdonald Island.

    Implication for mantle dynamics. Our P-wave velocity images show unambiguous evidence that at least nine hotspots cap a plume originating near the core-mantle boundary. However, at least another eight are associated with plumes whose imaged base lies near the 670-km discontinuity. Although the phase transition at 670 km has been a candidate for a second thermal boundary layer, recent tomographic evidence (34, 35) for whole-mantle convection had made this less likely. Yet if the 670-km discontinuity can temporarily delay penetration, upwelling material may accumulate and give rise to an apparent source region at or just below 670 km (15). The coexistence of at least two preferred depths where our plumes images stop is suggestive of the existence of at least two convecting regions, symbiotically linked in a regime that is more complex than either the whole-mantle or layered convection models. It is, however, also possible that the variety of plume styles we image is simply a manifestation of the expected intermittency and ephemerality of high–Rayleigh number, wholemantle convection.

    Our tomographic images also show plume images that fade at mid-mantle depths. From a geodynamical point of view, steady, nonintermittent plumes that do not originate near 670 km or near the D″ layer are problematic, because no thermal boundary layer has so far been observed in the mid-mantle. The presence of a change in mantle chemistry has been postulated to explain changes in the pattern of subduction (34, 35, 46) and the presence of an anticorrelation between P-wave and S-wave velocity in the lower mantle (47). Anderson (48) advocates a transition at a depth near 1000 km. So far, the presence of such a mid-mantle transition has not been confirmed (49, 50). A geochemical model has been proposed with a thermal boundary layer in the mid-mantle, which is strongly varying in depth and seismically invisible (because compositional changes compensate for the higher temperature, keeping velocities approximately the same) (51). Laboratory experiments show that thin plumes can arise from such a deep layer (11); however, we observe thicker plumes (Table 1). Tackley's model (52) identifies the deep layer with the location of the superplumes, which is not where we observe the fading of the plume images. The simplest explanation for the plume images extending only to mid-mantle depths is that they are really deep-mantle plumes, which are not resolved at deeper levels. Alternatively, we could be observing a large number of hiatuses in plume flux; however, the fact that we see many more plume tails than plume heads around these depths makes this less probable on statistical grounds.

    The plumes must have diameters of several hundred km, or they would not have been resolved even with our improved imaging techniques. Because of the very low efficiency of heat diffusion, the thermal “halo” around the plume is only a fraction of the width (typically 50 km for a plume that has been active for 100 million years). Thus, plumes must be wider than is commonly assumed, and this provides qualitative information about the rheology in the lower mantle. Although a Newtonian viscosity for the upper mantle seems to be required by the modeling of glacial rebound observations, the rheology of the lower mantle is not well constrained by surface observations (53). However, the wide plumes we observe would clearly be favored by a lower mantle with a sluggish convective regime such as would be expected for a high, Newtonian viscosity (54).

    Of course, our images only provide a snapshot of the current mantle, and we are unable to infer the rise velocity of the plumes. Because of this, we are unable to place a firm constraint on the heat flux carried by the plumes; however, an order-of-magnitude calculation indicates that the observed heat flux of about 44 TW at Earth's surface imposes a limit on the rise velocity of the plumes, which is probably below 10 cm/year (55). This loose constraint reinforces our conclusion that convection in the lower mantle is slow, and indicates that the role of plumes in transporting heat from the core to the surface of the Earth is larger than previously suggested (1).

    Another surprising observation is the lack of correlation between the maximum depth of a plume image and the associated 3He/4He anomaly. Of the ocean islands with high values of 3He/4He (3), Easter, Hawaii, Kerguelen, Samoa, and Tahiti are well-resolved deep plumes (Table 1). Afar, Cape Verde, Caroline, Reunion, and Jean Fernandez are potential D″ plumes and also have correspondingly high 3He/4He. But other plumes for which a deep origin can be decisively ruled out (Galapagos, Iceland) have high 3He/4He. Conversely, St. Helena and Canary Islands have a low 3He/4He ratio but have anomalous velocities that extend deeply into the mantle.

    The maximum velocity anomaly of the observed plumes in the lower mantle is reduced by a factor of about 3 with respect to the value found in the upper mantle (which is often in excess of –1.5%). Even though the magnitude of each plume anomaly is affected by the resolution, this diminution of δvP/vP with depth is in accordance with predictions of dvP/dT (42).

    Supporting Online Material

    SOM Text

    Figs. S1 to S21


    References and Notes

    View Abstract

    Stay Connected to Science

    Navigate This Article