Physics of the Earth and Planetary Interiors 170 (2008) 89–94

Contents lists available at ScienceDirect

Physics of the Earth and Planetary Interiors

journa l homepage: www.e lsev ier .com/ locate /pepi

Bulldozing the core–mantle boundary: Localized seismic scatterersbeneath the Caribbean Sea

Meghan S. Miller ∗, Fenglin NiuDepartment of Earth Science, Rice University, 6100 Main Street, MS 126, Houston, TX 77005, USA

a r t i c l e i n f o

Article history:Received 2 November 2007Received in revised form 27 May 2008Accepted 24 July 2008

Keywords:Core–mantle boundary

a b s t r a c t

Images of the scattering field near the core–mantle boundary (CMB) beneath the Caribbean were obtainedfrom migration processing of PKIKP precursors recorded by a broadband seismic array in the southernCaribbean from earthquakes in the Western Pacific. Strong seismic scatterers were found to be concen-trated near the CMB north of Colombia and are surrounded by high velocity anomalies associated withthe subducted Farallon plate. The relative location between the imaged scatterers and the surroundinghigh velocity anomalies leads us to speculate that the observed seismic scatterers are remnants of the

PKIKP precursorsSeismic scatterersFarallon slab

subducted slab. As the Farallon plate penetrated into the lower mantle, segregation allowed its basalticcrust to descend more quickly and to pool in front of the major portion of the subducting slab. The pondingoceanic crustal material could be further collected (or bulldozed) by the residual motion of the subducting

regio

1

lchccGtbaeSArra

pE(et

lgpu1opH2sua

2

(PsL

0d

plate, forming a localized

. Introduction

The nature of the core–mantle boundary (CMB) region, the D′′

ayer, has been a focus of research for decades because it is cru-ial in understanding the Earth’s dynamic structure. Many studiesave found that the D′′ region is extremely heterogeneous and hasomplicated seismic structures that require involvements of bothhemical and thermal processes (e.g., Garnero et al., 1998; Lay andarnero, 2004). One representative and well-studied area is Cen-

ral America and the Caribbean where the D′′ region is featuredy a well-defined seismic discontinuity underlain by anisotropicnd higher velocity materials (e.g., Garnero and Lay, 2003; Hungt al., 2005; Hutko et al., 2006). Prior studies have primarily usedwaves from earthquakes in South America recorded at North

merican stations. Here we present scattering images of the D′′

egion beneath the Caribbean using precursors to the PKIKP phaseecorded by a broadband seismic array in northern South Americand the Caribbean from earthquakes in the Western Pacific.

Scattering of P waves in the lower mantle can be detected asrecursors to the compressional waves that traverse through the

arth’s inner core at the distance range of 130–143◦ (PKIKP, Fig. 1)e.g., Cleary and Haddon, 1972; Haddon and Cleary, 1974; Husebyet al., 1976; Hedlin et al., 1997; Hedlin and Shearer, 2000). Arrivalimes of the precursors and their amplitude are directly related to

∗ Corresponding author. Fax: +1 713 348 5214.E-mail address: meghan.s.miller@rice.edu (M.S. Miller).

VloTocga

031-9201/$ – see front matter © 2008 Elsevier B.V. All rights reserved.oi:10.1016/j.pepi.2008.07.044

n with strong seismic scatterers.© 2008 Elsevier B.V. All rights reserved.

ocation, geometry, density and magnitude of the seismic hetero-eneities in the lower mantle. Previous studies found that PKIKPrecursors can generally be explained by a random scattering fieldniformly distributed in the bottom 200 km (Haddon and Cleary,974), and perhaps in a broader depth range (Hedlin et al., 1997)f the lower mantle. There are a few exceptions that are inter-reted as isolated scatterers (Vidale and Hedlin, 1998; Wen andelmberger, 1998; Thomas et al., 1999; Wen, 2000; Niu and Wen,001; Cao and Romanowicz, 2007). In most case, these isolatedcatterers are found to be related to partial melt. Here we seek tonderstand the features and causes of PKIKP precursors recordedt the BOLIVAR/GEONINOS array.

. PKIKP precursors

We selected 192 seismograms with high signal-to-noise ratiosSNR) from a total of 19 earthquakes that occurred in the Westernacific during 2003–2005 recorded by an array of 82 broadbandeismometers deployed under the U.S. BOLIVAR (Broadband Ocean-and Investigations of Venezuela and the Antilles arc Region) andenezuelan GEODINOS (Geodinámica Reciente del Límite Norte de

a Placa Sudamericana—Recent Geodynamics of the Northern Limitf the South American Plate) projects (Levander et al., 2006) (Fig. 1).

he data were selected based on the signal-to-noise ratio (SNR ≥ 5)f the PKIKP phase. 118 seismograms from 14 events exhibitedlear and strong precursors to the PKIKP phase while 74 seismo-rams from 5 earthquakes showed very weak or no precursor atll (Table 1). The observed PKIKP precursors have clear onsets and

90 M.S. Miller, F. Niu / Physics of the Earth and Planetary Interiors 170 (2008) 89–94

Fig. 1. (A) Location of 14 events with strong precursors to PKIKP and 5 events withno precursors in the Western Pacific. (B) Distribution of seismic stations (temporaryand permanent) in the BOLIVAR deployment. (C) Ray paths of PKIKP phase andP ◦

aoP

rsiaebmpfiai

Table 1Event list

Date Time (GMT) Latitude(◦)

Longitude(◦)

Depth(km)

Magnitude(Mw)

Events with precursorsFeb-22-2004 23:17 18.505 145.387 197.0 5.7Mar-27-2004 06:19 −6.256 154.760 48.0 5.9Jul-22-2004 09:45 26.489 128.894 20.0 6.1Sep-05-2004 10:07 33.070 136.618 14.0 7.2Sep-06-2004 23:29 33.205 137.227 10.0 6.6Sep-08-2004 14:58 33.140 137.200 21.0 6.1Oct-04-2004 19:20 14.546 146.993 7.0 6.0Oct-15-2004 04:08 24.530 122.694 94.0 6.7Dec-11-2004 01:55 −4.613 149.779 541.0 5.5Jan-16-2005 20:17 10.934 140.842 24.0 6.6Jan-17-2005 10:51 10.986 140.682 12.0 6.1Feb-05-2005 03:34 16.011 145.867 142.0 6.6Apr-19-2005 01:46 29.642 138.891 425.0 5.9Aug-13-2005 07:36 20.130 145.800 48.0 6.0

Events without precursorMar-06-2004 06:26 −5.112 152.588 44.0 5.5May-13-2004 09:58 −3.584 150.730 10.0 6.4

WP

cbtsasawoaAgw

sihriWaolftsm

apsgh

KIKP precursors at distance of 135 . Earthquake hypocenter represented as an Xnd seismic station as an inverted triangle. The shaded regions indicate locationsf seismic scatterers, which would produce seismic waves arriving earlier than theKIKP phase.

elatively large amplitudes. One example of seismograms with atrong and isolated precursor is shown in Fig. 2A. These are record-ngs from the deep focus event that occurred at the Izu-Boninrc on April 19, 2005. In comparison, traces for the Oct-08-2004vent near the Solomon Islands are shown in Fig. 2B. Althougheing at roughly the same epicentral distance range, these seis-

ograms show no significant precursory arrivals before the PKIKP

hase. Here the waveform data were preprocessed in two steps. Werst removed the instrument response from the broadband datand then convolved them with the WWSSN (WorldWide Standard-zed Seismograph Network) short-period instrument response. The

isati

Oct-08-2004 08:27 −10.951 162.161 36.0 6.8Nov-02-2004 08:45 28.702 143.214 10.0 5.7Nov-11-2004 17:34 −11.128 162.208 10.0 6.7

WSSN short-period data appeared to possess the best SNR for theKIKP phase and its precursors.

Generally, amplitudes of PKIKP precursors increase with epi-entral distance. The observed difference in precursor amplitudeetween the two groups of earthquakes, however, does not seemo result from a systematic difference in epicentral distance. Fig. 3hows the amplitude ratios between the precursor and PKIKP asfunction of epicentral distance. For the group with clear precur-

ors, we used the time window between the onsets of the precursornd PKIKP phase to calculate the amplitude ratio, while for theeak precursor group, the predicted arrival time window based

n homogeneous random scattering models was employed. Themplitude ratios are further averaged over a 1◦ binning distance.s shown in Fig. 3, the amplitude ratios for the weak/no precursorroup remain very low through all the distance range of 130–138◦,hile those of the strong precursor group are significantly higher.

In general, scatterers located at both the source and receiverides can produce PKIKP precursors (Fig. 1C). Due to the wide rangen location of the 14 events in the Western Pacific (Fig. 1A and C),ere we argue that the observed precursors are most likely due toeceiver-side scattering. The earthquakes are distributed roughlyn two broad regions: the western Pacific and southwest Pacific.

e found mixed observations from both regions, making it hard torgue for the source-side contribution of our observation. On thether hand, the core exits of the PKIKP rays of the two groups areocalized into two separate groups (Fig. 4a). The exit point locationsrom the events with strong precursors appear to be constrained tohe central Caribbean region, while the exit points with no precur-ors are limited laterally to northwest South America and southernost Central America.As shown in Fig. 1C, source- and receiver-side scattered waves

re received in different incident angle ranges. Thus it is in princi-le possible to overcome the ambiguity in source- and receiver-sidecattering with array analysis techniques. Fig. 1C shows that wavesenerated from the source-side scattering (P to PKPbc or P to PKPab)ave larger incident angles than those from receiver-side scatter-

ng (PKPbc to P or PKPab to P). We used the beam-forming or

lowness-azimuth stacking technique (Aki and Richards, 1980; Niund Kawakatsu, 1997; Kaneshima and Helffrich, 1998) to measurehe slowness and back azimuth of the isolated precursor shownn Fig. 2A. In a beam-forming analysis, all the seismograms are

M.S. Miller, F. Niu / Physics of the Earth and Planetary Interiors 170 (2008) 89–94 91

Fig. 2. (A) An example of PKIKP precursors and PKIKP/PKiKP phases observed at northernpicked PKIKP (t = 0). Squares indicate the first arrival of the PKIKP precursors estimated frto the PKIKP phase.

Fan

nlTnvapsso

3

a

dH2c(iasfttieoaeui

coainfb

ha

ig. 3. Averaged amplitude ratio between the precursor and the PKIKP is plotteds a function of epicentral distances. Squares and circles indicate the strong- ando/weak-precursor group, respectively.

th-root stacked (Muirhead, 1968) after a time correction calcu-ated from the assumed slowness and back azimuth before stacking.he best slowness and back azimuth were determined when theth-root stacking amplitude reaches to a maximum (Fig. 5). Wearied slowness from 0.5 to 5 s/deg at increments of 0.05 s/deg,nd searched the back azimuth deviation from the great circleath within a range of ±20◦ at increments of 2◦. The measuredlownesses of precursors are less than that of PKIKP (∼1.92 s/deg),uggesting that receiver-side PKPbc to P scattering is the most likelyrigin of the observed PKIKP precursors.

. Mapping the seismic scatterers

PKIKP precursor onsets were manually picked and were found torrive approximately 3–6 s after the earliest precursory arrivals pre-

essbh

South America and the southeastern Caribbean. Each trace is aligned to manuallyom a random scattering model. (B) An example of events that show no precursors

icted by the random scattering models (Haddon and Cleary, 1974;usebye et al., 1976). We adopted a migration technique (Wen,000) to calculate the scatterer probability and hit count for dis-retized regions at the core–mantle boundary and six other depths50, 100, 150, 200, 250, and 300 km above the CMB). This methods based on the hypothesis that PKIKP precursor energy arrivingt a certain time can be attributed to scatterers along isochronalhells in both the source and receiver sides. As described above,rom beam-forming analysis, we were able to constrain the scat-ering mode as the receiver-side PKPbc to P scattering. We usedhe 1D iasp91 model (Kennett and Engdahl, 1991) to compute thesochrones at each assumed scattering depth. To account for 3Dffects in the migration moveouts, we manually picked the PKIKPnsets by tracking the relative polarity and moveout of the PKIKPnd PKiKP phases recorded from deep focus and intermediate deptharthquakes. Residual times calculated from these events were thensed as the 3D travel time correction for earthquakes that occurred

n the same region.The scatterer probability (Fig. 6A–G) at one grid cell (0.1◦ by 0.1◦

ell) is computed from the ratio of the two numbers: number ofbservations that can be explained by scattering at the grid within0.5 s window and the total number of seismic observations used

n this study. In addition, the hit count at one grid is simply the totalumber of PKIKP precursors sampling the grid cell (Fig. 6H). There-

ore, the grid cells with a large hit count and high probability cane interpreted as the origins of the PKIKP precursor onset energy.

We only used the onset energy to locate the seismic scatterersere since later arrivals could be the result of scattering of the firstrrival by other heterogeneities at shallower depths, and using the

nergy following the onsets requires a complete knowledge of theource-time function (which is unknown). The observed precur-ors are best explained by seismic scattering near the core–mantleoundary north of Colombia (Fig. 6A–G). This region has both theighest probability and significant hit count. It also does not overlap

92 M.S. Miller, F. Niu / Physics of the Earth and Planetary Interiors 170 (2008) 89–94

Fig. 4. (A) Map of the Caribbean region showing the calculated points of ray pathsexiting the core and entering the mantle. Blue circles indicate the events with pre-cursors, and red diamonds indicate the events with no evident precursor. (B–D)Tomographic images of the lowermost mantle from Grand (2002) are shown withthe core exits. Only regions with a 1.6% shear velocity perturbation and greaterabt

wcdiWt

Fig. 5. Normalized energy of the stacked PKIKP (A) and the precursor shown inF4a

4

mohiffmbta

gssuraeli(

re shown. Red lines (and ages) indicate the estimated Farallon plate boundariesetween 100 and 46 Ma (Pindell et al., 2005). (For interpretation of the referenceso color in this figure legend, the reader is referred to the web version of the article.)

ith the area in which the exit points of the events with no pre-

ursors are concentrated. The area with high scattering probabilityecreases steadily for shallower depths (Fig. 6A–G), and becomes

nsignificant at scattering depths ≥300 km above the CMB (Fig. 6G).e thus concluded that most of the scatterers are confined within

he D′′ region.

wtvro

ig. 2B (B) is plotted as a function of back azimuth and slowness. A time window ofs is used in calculating energy. Note that precursors are received in steep incidentngles, indicating that they are generated by receiver-side scattering.

. Results and discussion

The cause of heterogeneities above the core–mantle boundaryay be due to the intrinsic heterogeneity within the subducted

ceanic lithosphere (Hirose et al., 1999). For example, basaltic crustas a different chemistry, density and melting temperature from

ts underlying peridotite layer. Hirose et al. (1999) suggests that theormer basaltic crust becomes negatively buoyant when it trans-orms into a perovskite lithology, allowing it to sink into the deep

antle. At the base of the mantle the former basaltic crust wouldecome partially molten, resulting in very low velocity anomalieshat have been speculated to generate the PKIKP precursors (Vidalend Hedlin, 1998; Wen and Helmberger, 1998; Wen, 2000).

It is likely that these small-scale chemical and thermal hetero-eneities above the CMB are the sources of our observed seismiccatterers. The large variation observed in the precursor amplitudetrongly indicates that these small-scale heterogeneities are notniformly distributed in the study region. We found a good cor-elation between the migrated location of the seismic scatterersnd the calculated exit points (Figs. 4 and 6). A comparison of thexit point locations with a shear velocity tomographic model of theowermost mantle (Grand, 2002) shows that there are high veloc-ty anomalies at both the east and west of the mapped exit pointsFig. 4B–D). The exit points corresponding to the ray paths of eventsithout a precursor lie to the south of the major high velocity fea-

ures. The lateral extent of the scatterers is bounded by the highelocity anomalies on the east and west and by the non-scatteringegion to the south. The high velocity body in the western sidebserved in the shear wave model (Grand, 2002) has been inter-

M.S. Miller, F. Niu / Physics of the Earth and Planetary Interiors 170 (2008) 89–94 93

F C), 153 adily3 , withl

pi

cF

ig. 6. Maps of the “scatterer probability” at depths: 0 km (A), 50 km (B), 100 km (00 km about CMB. Note that the area with high probability (bright yellow) drops ste00 km above the CMB. Probability and hit count are color coded from pink to blue

egend, the reader is referred to the web version of the article.)

reted as the subducted Farallon plate. This high velocity anomalys also shown in the P-wave tomographic images (Ren et al., 2007).

Estimated positions of the Farallon plate boundary at the Ameri-as (Pindell et al., 2005) are plotted with the tomographic image inig. 4B–D. This paleogeographic reconstruction demonstrates the

cgmit

0 (D), 200 km (E), 250 km (F), and 300 km (G) above the CMB. Hit count at depthwith decreasing of scattering depth and become negligible at depth shallower thanblue being the highest. (For interpretation of the references to color in this figure

onvergence of the Farallon plate between 100 and 46 Ma and itseneral relationship with the high velocity anomalies in the lowerantle associated with the ancient subducted Farallon plate. As

t penetrated into the lower mantle and collided with the CMB,he subducting Farallon slab appears to bend along the boundary

94 M.S. Miller, F. Niu / Physics of the Earth and

FTb

abdscta

tiacd(fpll2attaiw

ter2bppvo

A

mbHcTa

R

A

C

C

G

G

G

H

H

H

H

H

H

H

K

K

L

L

L

M

N

N

P

R

S

T

T

V

ig. 7. Cartoon of the subducting Farallon slab and the features relevant to the study.he oceanic crust may have subducted separately and pooled on the core–mantleoundary in front of the subducted oceanic lithosphere (not to scale).

nd continue moving in its east–northeast direction. The formerasaltic crust on the upper portion of the slab may have sub-ucted more quickly and pooled in front of the major portion of theubducting Farallon plate. The descended oceanic crustal materialould have been localized (and bulldozed) by the residual motion ofhe subducting plate and therefore become concentrated into therea shown in Figs. 4 and 6.

Lee and Chen (2007) showed with a simple force balance equa-ion that the subducted oceanic lithosphere can become segregatednto two parts, the eclogitized crust and the lithospheric mantle,nd then journey to different resting grounds. We believe that theombined scattering and tomographic images provide arguable evi-ence for this scenario. According to the calculation of Lee and Chen2007), the eclogitized oceanic crust slips at a velocity slightly dif-erent from the subduction of the Farallon plate. As the result, itools at the base of mantle “ahead” of the large volume oceanic

ithospheric materials which are imaged as fast velocity anoma-ies by seismic tomography (Fig. 7) (e.g., Grand, 2002; Hung et al.,005; Ren et al., 2007). Other models using triplicated S waveformsnd migration methods have found anomalous areas to the east ofhe mapped PKP precursors (Fig. 4), which have suggested to behe presence of folded slabs in the lowermost mantle (Hutko etl., 2006; Sun et al., 2006; van der Hilst et al., 2007) or fast veloc-ty anomalies (Cao and Romanowicz, 2007), which is in agreement

ith our interpretation.While the segregated oceanic crust seems to be a viable explana-

ion for the observed seismic scatterers, other interpretations stillxist. The intensity of seismic scattering observed here is compa-able to those observed in the CMB west of Mexico (Niu and Wen,001) and requires a very large P-wave velocity perturbation (≥6%,ased on the forward modeling of Niu and Wen, 2001), suggestingossible involvement of melts. Tan et al. (2002) found that plumesreferentially develop on the edge of slabs. Plume related lowerelocity anomalies thus could be an alternative explanation for thebserved seismic scatterers.

cknowledgments

We thank the BOLIVAR team, Alan Levander, the PASSCAL Instru-ent Center, and FUNVISIS for providing the data. The work

enefits from discussions with Cin-Ty Lee, Daoyuan Sun, and Donelmberger. We also thank the two anonymous reviewers for theirritical comments which improved the manuscript significantly.he BOLIVAR project is funded by the National Science Foundationnd Miller is funded by an NSERC postdoctoral fellowship.

W

W

Planetary Interiors 170 (2008) 89–94

eferences

ki, K., Richards, P.G., 1980. Quantitative Seismology: Theory and Methods. W.H.Freeman and Co., San Francisco, 932 pp.

ao, A., Romanowicz, B., 2007. Locating scatterers in the mantle using array analysisof PKIKP precursors from an earthquake doublet. Earth Planet. Sci. Lett. 255,22–31.

leary, J.R., Haddon, R.A., 1972. Seismic wave scattering near the core–mantle bound-ary: a new interpretation of precursors to PKP. Nature 240, 549–551.

arnero, E.J., Lay, T., 2003. D′′ shear velocity heterogeneity, anisotropy, and disconti-nuity structure beneath the Caribbean and Central America. Phys. Earth Planet.Inter. 140, 219–242.

arnero, E.J., Revenaugh, J.S., Williams, Q., Lay, T., Kellogg, L.H., 1998. Ultralow veloc-ity zone at the core–mantle boundary. In: Gurnis, M., Wysession, M.E., Knittle,E., Buffett, B.A. (Eds.), The Core–Mantle Boundary Region. American GeophysicalUnion, Washington, DC, pp. 319–334.

rand, S.P., 2002. Mantle shear-wave tomography and the fate of subducted slabs.Philos. Trans. R. Soc. Lond. A 360, 2475–2491.

addon, R.A., Cleary, J.R., 1974. Evidence for scattering of seismic PKP waves nearthe core–mantle boundary. Phys. Earth Planet. Inter. 8, 211–234.

edlin, M.A.H., Shearer, P.M., Earle, P.S., 1997. Seismic evidence for small-scale het-erogeneity throughout the Earth’s mantle. Nature 387, 145–150.

edlin, M.A.H., Shearer, P.M., 2000. An analysis of large scale variations in small-scalemantle heterogeneity using GSN recordings of precursors to PKP. J. Geophys. Res.105, 1870–1882.

irose, K., Fei, Y., Ma, Y., Mao, H.-K., 1999. The fate of subducted basaltic crust in theEarth’s lower mantle. Nature 397, 53–56.

ung, S.-H., Garnero, E.J., Chiao, L.-Y., Kuo, B.-Y., Lay, T., 2005. Finite frequency tomog-raphy of D′′ shear velocity heterogeneity beneath the Caribbean. J. Geophys. Res.110, doi:10.1029/2004JB003373.

usebye, E.S., King, D.W., Haddon, R.A., 1976. Precursors to PKIKP and seismic wavescattering near the mantle–core boundary. J. Geophys. Res. 81, 170–182.

utko, A.R., Lay, T., Garnero, E.J., Ravenaugh, J., 2006. Seismic detection of folded,subducted lithosphere at the core–mantle boundary. Nature 441, 333–336.

aneshima, S., Helffrich, G., 1998. Detection of lower mantle scatterers northeast ofthe Mariana subduction zone using short-period array data. J. Geophys. Res. 103,4825–4838.

ennett, B.L.N., Engdahl, E.R., 1991. Travel times for global earthquake location andphase identification. Geophys. J. Int. 105, 429–465.

ay, T., Garnero, E.J., 2004. Core–mantle boundary structures and processes. In:Sparks, R.S.J., Hawkesworth, C.J. (Eds.), The State of the Planet: Frontiers andChallenges in Geophysics. Geophysical Monograph Series, 150, IUGG, vol. 19.American Geophysical Union, Washington, DC, pp. 25–41.

ee, C.-T.A., Chen, W.-P., 2007. Possible density segregation of subducted oceaniclithosphere along a weak serpentinite layer and implications for compositionalstratification of the Earth’s mantle. Earth Planet. Sci. Lett. 255, 357–366.

evander, A., Schmitz, M., Lallemant, H.G., Zelt, C.A., Sawyer, D.S., Magnani, M.B.,Mann, P., Christeson, G.L., Wright, J.E., Pavlis, G.L., Pindell, J., 2006. Evolution ofthe southern Caribbean Plate boundary. EOS Trans. 87, 97.

uirhead, K.J., 1968. Eliminating false alarms when detecting seismic events auto-matically. Nature 217, 533–534.

iu, F., Kawakatsu, H., 1997. Depth variation of the mid-mantle seismic discontinuity.Geophys. Res. Lett. 24, 429–432.

iu, F., Wen, L., 2001. Strong seismic scatterers near the core–mantle boundary westof Mexico. Geophys. Res. Lett. 28, 3557–3560.

indell, J., Kennan, L., Maresch, W.V., Stanek, K.-P., Draper, G., Higgs, R., 2005.Plate-kinematics and crustal dynamics of circum–Caribbean arc–continentinteractions: tectonic controls on basin development in Proto-Caribbean mar-gins. In: Lallemant, H.G., Sisson, V.B. (Eds.), Caribbean–South American PlateInteractions, Venezuela. Geological Society of America, Boulder, CO, pp. 7–52.

en, Y., Stutzmann, E., van der Hilst, R., Besse, J., 2007. Understanding seismicheterogeneities in the lower mantle beneath the Americas from seis-mic tomography and plate tectonic history. J. Geophys. Res. 112, B01302,doi:10.1029/2005JB004154.

un, D., Song, T.R.A., Helmberger, D.V., 2006. Complexity of D′′ in the presence of slab-debris and phase changes. Geophys. Res. Lett. 33, doi:10.1029/2005GL025384.

an, E., Gurnis, M., Han, L., 2002. Slabs in the lower mantle and their modula-tion of plume formation. Geochemistry, Geophysics, Geosystems 3(11), 1067,doi:10.1029/2001GC000238.

homas, C., Weber, M., Wicks, C.W., Scherbaum, F., 1999. Small scatterers in thelower mantle observed at German broadband arrays. J. Geophys. Res. 104,15073–15088.

idale, J.E., Hedlin, M.A.H., 1998. Evidence for partial melt at the core–mantle bound-

ary north of Tonga from the strong scattering of seismic waves. Nature 391,682–684.

en, L., 2000. Intense seismic scattering near the Earth’s core–mantle boundarybeneath the Comoros hotspot. Geophys. Res. Lett. 27, 3627–3630.

en, L., Helmberger, D.V., 1998. Ultra-low velocity zones near the core–mantleboundary from broadband PKIKP precursors. Science 279, 1701–1703.