dynamical links between small- and large-scale mantle...
TRANSCRIPT
This is a repository copy of Dynamical links between small- and large-scale mantle heterogeneity: seismological evidence.
White Rose Research Online URL for this paper:http://eprints.whiterose.ac.uk/123701/
Version: Accepted Version
Article:
Frost, DA, Garnero, EJ and Rost, S orcid.org/0000-0003-0218-247X (2018) Dynamical links between small- and large-scale mantle heterogeneity: seismological evidence. Earth and Planetary Science Letters, 482. pp. 135-146. ISSN 0012-821X
https://doi.org/10.1016/j.epsl.2017.10.058
(c) 2017, Elsevier B.V. This manuscript version is made available under the CC BY-NC-ND4.0 license https://creativecommons.org/licenses/by-nc-nd/4.0/
[email protected]://eprints.whiterose.ac.uk/
Reuse
Items deposited in White Rose Research Online are protected by copyright, with all rights reserved unless indicated otherwise. They may be downloaded and/or printed for private study, or other acts as permitted by national copyright laws. The publisher or other rights holders may allow further reproduction and re-use of the full text version. This is indicated by the licence information on the White Rose Research Online record for the item.
Takedown
If you consider content in White Rose Research Online to be in breach of UK law, please notify us by emailing [email protected] including the URL of the record and the reason for the withdrawal request.
1
Title: Dynamical links between small- and large-scale mantle heterogeneity:1seismologicalevidence23
Authors:DanielA.Frost1*,EdwardJ.Garnero2,andSebastianRost34
5
Affiliations:6
1Earth&PlanetaryScience,UniversityofCalifornia,Berkeley,California,USA7
8
2SchoolofEarthandSpaceExploration,ArizonaStateUniversity,Tempe,Arizona,9
USA10
11
3InstituteofGeophysicsandTectonics,SchoolofEarthandEnvironment,University12
ofLeeds,Leeds,UK13
14
*Correspondenceto:[email protected]
Abstract1617We identify PKP�PKP scattered waves (also known as P′�P′) from earthquakes18
recordedatsmall-apertureseismicarraysatdistances less than65°. P′�P′energy19
travelsasaPKPwavethroughthecore,upintothemantle,thenscattersbackdown20
throughthecoretothereceiverasasecondPKP.P′�P′wavesareuniqueinthatthey21
allowscatteringheterogeneitiesthroughoutthemantletobeimaged.Weusearray-22
processing methods to amplify low amplitude, coherent scattered energy signals23
and resolve their incoming direction. We deterministically map scattering24
heterogeneity locations from the core-mantle boundary to the surface.Weuse an25
2
extensive dataset with sensitivity to a large volume of the mantle and a location26
methodallowingustoresolveandmapmoreheterogeneitiesthanhavepreviously27
been possible, representing a significant increase in our understanding of small-28
scalestructurewithin themantle.Ourresultsdemonstrate that thedistributionof29
scattering heterogeneities varies both radially and laterally. Scattering is most30
abundant in the uppermost and lowermost mantle, and a minimum in the mid-31
mantle,resemblingtheradialdistributionoftomographicallyderivedwhole-mantle32
velocity heterogeneity. We investigate the spatial correlation of scattering33
heterogeneities with large-scale tomographic velocities, lateral velocity gradients,34
the locationsofdeep-seatedhotspotsandsubductedslabs. Inthe lowermost150035
kmofthemantle,small-scaleheterogeneitiescorrelatewithregionsoflowseismic36
velocity,highlateralseismicgradient,andproximitytohotspots.Intheupper100037
km of the mantle there is no significant correlation between scattering38
heterogeneity location and subducted slabs. Between 600 and 900 km depth,39
scatteringheterogeneitiesaremorecommonintheregionsmostremotefromslabs,40
andclosetohotspots.Scatteringheterogeneitiesshowanaffinity forregionsclose41
to slabs within the upper 200 km of the mantle. The similarity between the42
distribution of large-scale and small-scale mantle structures suggests a dynamic43
connection across scales, whereby mantle heterogeneities of all sizes may be44
directedinsimilarwaysbylarge-scaleconvectivecurrents.45
Keywords:seismology;deepEarth;scattering;mantlestructure;mantledynamics;46seismicarrays47481.Introduction49
3
The high frequency (~1 Hz) seismic wavefield provides evidence of50
kilometre scale structure within the Earth [Cleary and Haddon, 1972]. Seismic51
energy that is not explained by wave propagation in smoothly varying velocity52
modelsof theEarthhasbeenattributedtoreflectionsandscattering fromsharply53
contrasting volumetric heterogeneities and roughness on interfaces [Chang and54
Cleary,1981].Theinteractionofthewavefieldwithdiscrete,small-scalevariations55
inelasticpropertiesand/ordensitycandivertseismicenergyontonewpaths,often56
generatingprecursorsorpostcursors(coda)tothemainseismicphasesthattravel57
inthegreatcircleplane.Thesizeofthescatterersthatcanbeimagedisdependent58
upon the wavelength that is analysed; for the teleseismic high-frequency P-59
wavefieldabove1Hztheyaretypicallyontheorderof1to10km.60
Global imaging of Earth’s small-scale heterogeneities is difficult due to the61
uneven distribution of earthquake sources and seismic receivers, and the low62
amplitudeof the scatteredsignals involved. Scattering canbe studiedusing single63
stations,butwiththisapproachthelocationofthescatteringheterogeneitycanbe64
ambiguous[Wen,2000].Alternatively,seismicarrays,i.e.,3ormorecloselylocated65
sensors,canresolvetheincomingdirectionofscatteredwaves,thusitispossibleto66
deterministicallylocateheterogeneities[Thomasetal.,1999;RostandEarle,2010;67
Frost et al., 2013]. In the last few decades a number of studies have started to68
unravelthedistributionofsmall-scaleheterogeneitiesofEarth’smantle.Hedlinetal.69
[1997], and later Mancinelli and Shearer [2013, 2016] studied the depth70
distribution of heterogeneity within the mantle through analysis of PKP pre- and71
postcursorsrecordedatsinglestations.UsingastochasticRayleigh-Bornscattering72
4
approach, Mancinelli and Shearer [2013, 2016] developed a global model of73
scattering heterogeneity containing 0.1% root-mean-square velocity variations in74
thedeepest1200kmofthemantlewithheterogeneityscalesizesrangingfrom2to75
30km.76
Thisworkiscomplementedbystudiesthatdeterministicallymapsmall-scale77
scattering heterogeneity within the upper and lower mantle. These studies have78
noted lateral variations in heterogeneity distribution, as well as variations in79
amplitudesofscatteredwaves.ScatteredP-to-P(P�P,wherethe“�“representsthe80
locationofscattering)andP-to-S(P�S)wavesaresensitivetoheterogeneitiesinthe81
upperhalfofEarth’smantle;theyhavebeenusedtomapscatteringheterogeneityin82
regionsinfluencedbyrecentsubduction[KaneshimaandHelffrich,1998;Bentham83
andRost,2014].Scatteringinthelowermostmantlehasalsobeenobservedtovary84
laterally [Waszek et al., 2015]. Strong scattering has been observed in regions85
beneathmantlehotspots [Wen,2000],nearsmall, regionalultra-lowvelocityzone86
(ULVZ)structures[YaoandWen,2014],beneathsubductionzones[MillerandNiu,87
2008], and near the edges of LLSVPs [Frost et al., 2013]. A near-global study of88
PK�KP–aPKPwavethatisback-scatteredinthelowermantleontoasecondPKP89
path – suggests a spatial correlation between scattering and LLSVP edges in the90
lowermost300kmofthemantle[RostandEarle,2010;Frostetal.,2017].91
The volume of the mantle that can be investigated for scattering92
heterogeneityiscontrolledbythespecificsoftheseismicprobe.PK�KPcanbeused93
toinvestigatethelowermantleclosetotheCMB[ChangandCleary,1981;Rostand94
Earle, 2010; Frost et al., 2017]. ThedirectwavePKPPKP (also calledP′P′) results95
5
from a PKP wave (P′) reflecting from the underside of the surface, back into the96
Earth as a secondPKPwave, along the great-circlepath (GCP).Thisphase canbe97
precededbyscatteredenergycalledPKP�PKP(P′�P′),causedbyback-scatteringof98
PKPatanydepthinthemantle[Rostetal.,2015].LikePK�KP,P′�P′hasanunusual99
scatteringgeometry(Fig.1)andcanscatter from locationsoff theGCP,and theP′100
segmentsneednotbesymmetric toeachother.P′�P′ is thecontinuationofPK�KP101
towardsthesurface,thusthisphaseisabletosamplethewholemantlefromCMBto102
crust(Fig.2).Weextendourearlierworkandinvestigatethemantleupwardsfrom103
theCMBtothesurfacetodeterministicallymaptheverticalandlateraldistribution104
ofscatteringheterogeneitiesthroughoutthemantle.Incontrasttootherscattering105
probes, the unusual (and versatile) raypath geometry of P′�P′ allows the study of106
previouslyunsampledregionsoftheEarth.107
TheinternalstructureoftheEarthandthenatureofmantleconvectionare108
inherently connected across scales [e.g. Tackley 2015]. The distribution of large-109
scale mantle structure as imaged by seismic tomography has been investigated110
usingthermo-chemicalgeodynamicmodels,whichindicatethatdownwellingofcold,111
dense slabs at subduction zones moves and shapes the hot, convecting piles of112
seismically slow material at the CMB, forming the Large Low Shear Velocity113
Provinces (LLSVPs) [McNamara and Zhong, 2005; Li et al., 2014; Domeier et al,114
2016]. The LLSVPs, if compositionally distinct, may modulate mantle dynamics115
through thermal instabilities that result in mantle plumes that rise up causing116
hotspot volcanism [Thorne et al., 2004; French and Romanowicz, 2015].117
Furthermore, calculations suggest thatmantle plumesmaybe spatially correlated118
6
with the LLSVPs [Thorne et al., 2004; Doubrovine et al., 2016]. Geodynamic119
modelling of thermo-chemical structures in the deep mantle indicates that small-120
scaleheterogeneities(assmallaskilometre-sized)canbepassively transported in121
thelarge-scaleflow[BrandenburgandvanKeken,2007;Lietal.,2014,Mulyukova122
et al., 2015]. Furthermore, geochemical analysis of intraplate volcanism suggests123
thatheterogeneitiessituatedinthedeepEarthmaybetransportedtothesurfaceby124
entrainment in mantle convection [Williams et al., 2015]. Therefore, there is125
compelling evidence that the distribution of small-scale seismic structure in the126
mantleislinkedtothelarge-scalestructures.127
Hereweuseaglobalcollectionofearthquakesrecordedatseismicarraysto128
identifyP′�P′ anddeterministically locate thepositionof the causativevolumetric129
scatteringheterogeneitywithinthemantle.Weinvestigatetherelationshipbetween130
scatteringheterogeneityandotherseismologicallyimagedstructuresinthemantle.131
We use our observations to understand the distribution of small-scale132
heterogeneities throughout the whole of the mantle, and the connection with133
dynamicprocesses.134
7
135Figure1:PKP�PKP(P′�P′)examplepath.(a)AP′�P′pathfromthesource(star)to136
a scattering point in the mantle (circle) and then to the receiver (triangle). P′�P′137
travelsalongtwogreat-circlepaths(solidlines)toandfromthescatteringpoint,off138
thegreat-circlepathbetweenthesourceandreceiver(dashedline).PKPraypaths139
from(b)sourcetoscatterer(PKPab)and(c)scatterertoreceiver(PKPbc).Thetwo140
PKPlegsmaybesymmetricorasymmetric(asinthiscase)andcanscatterfromany141
depthinthemantlefromtheCMBtothesurface.Raysobservedatthesurfacearrive142
fromaspecificdirectionknownastheback-azimuth,ȟ,measuredrelativetoNorth,143
ortherelativeback-azimuthmeasuredfromtheGCP,andfromaverticalincidence144
angle,referredtoastheslowness,u.145
[SINGLEOR1.5COLUMNFIGURE]146
147148149150151
c Scatterer to Receiver
!
b Source to Scatterer
a
8
152Figure2:Travel-timecurvedisplayingP′�P′andotherscatteredphasesinthehigh-153
frequencyseismicwavefield.BlacklinesmarkmajorP-wavephases.Theblueregion154
marks the time and distance region investigated for P′�P′ waves in this study.155
HatchedregionmarkstimeanddistanceregioninvestigatedforPK�KPinFrostetal.,156
[2017]. Grey and pink lines mark the P- and S-waves, respectively, that may157
contaminatetheP′�P′studyregion.OtherP-andS-wavesarenotshownforclarity.158
Differently shaded grey regions denote time and distance regions previously159
investigatedforotherscatteredwaves.AdaptedfromRostetal.,[2015].160
[SINGLECOLUMNFIGURE]161
1622.Data163164
We collect data from 643 earthquakes at any depth with magnitudes M≥6165
recordedatupto12smallandmediumaperturearrayswithin65°ofanyevent(Fig.166
3).Thearrayscontainamixtureofshortperiodandbroadbandinstruments;weuse167
onlythemostcommoninstrumenttypeineacharray.Thesearraysweredesigned168
9
to determine the directivity of short-period P-waves thus are ideally suited for169
analysis of high-frequency scattered waves. The aperture of an array controls its170
directivityresolution,thusweselectonlyarrayswithaperturesof10kmto30kmto171
ensurethatweareabletoresolvewelltheincomingdirectionofwaves.172
Eachevent-arraypairhasaspecificgeographicalvolumeofthemantlefrom173
whichpossibleP′�P′scatteredwavescanbedetected(Fig.3).Thesizeandshapeof174
thesamplingregionatanygivenscatteringdepthisdependentontheevent-array175
distance. Using estimations of the potential scattering volumes combined for all176
source-arraypairs,wedevelopa“potentialsamplingdensitymap”ofourdatasetfor177
differentdepths,whichrepresents theabundanceof scattererswewoulddetect if178
the actual distribution of scattering in the Earth distribution were uniform. The179
potentialscatterersamplingdistributionofthedatasetisuneven,butincontrastto180
otherprobes,thesouthernhemisphereiswellcoveredthroughoutthedepthofthe181
mantle, allowing investigation of the relationship between scattering182
heterogeneities and the South and Central American subduction zones, and the183
AfricanandPacificLLSVPs. 184
10
185Figure 3: Earthquakes (dots) and arrays (triangles) in our dataset and resultant186
potential P′�P′ scattering sampling. The 643 events and up to 12 arrays yielded187
1715 event-array pairs. Global sampling distributions are constructed by188
summationofthepotentialscatteringsamplingforallsource-arraypairsat:(a)the189
surface(0kmdepth);(b)transitionzone(600kmdepth);(c)mid-mantle(1200km190
depth);and(d)theCore-MantleBoundary(2889kmdepth).Samplingisdensestin191
themid-mantle andmost geographically extensive in the lowermostmantle. Grey192
wedge in (a) displays an example of the potential scattering regions for a single193
event-arraypair.194
[2COLUMNFIGURE]195
196
1973.Methods198199
We investigate energy associatedwithmantle scattering in a timewindow200
from the firstpossible arrival of P′�P′ at~1700safter theearthquakeorigin time201
(for a surface focus event) corresponding to scattering at theCMB, up to the first202
Z=0 kma Z=600 kmb
d Z=2889 kmc Z=1200 km
0 325 650 975 1300Potential sampling
11
possiblearrivalofthedirectwaveP′�P′dfat~2400s,whichistheearliestP′�P′GCP203
phase, reflecting from the underside of the surface on the antipodal side (Fig. 2).204
P′�P′ scattering related to interactions with small-scale mantle heterogeneity is205
feasibleforanytimeanddistanceinthiswindow(blueshadedregioninFig.2).206
Wede-trend the data and discard any discontinuous traces i.e. gaps in the207
recording.Theremainingtracesarefilteredwitha2ndorderbandpassbetween0.5208
and2Hztoenhancethefrequenciesmostassociatedwithsmall-scalescatteringin209
paststudiesthatinvestigatedfrequencycontent[Mancinellietal.,2016;Frostetal.,210
2017]. Wesearchforscatteredsignalswithinthewavefielddatausingfk-analysis211
(frequency-wavenumber), which performs a grid-search over incoming directions212
to maximise coherence (the similarity of two or more signals in the frequency213
domain)ofthesignalstackedacrossthearray,calculatedinthefrequencydomain214
[Capon, et al., 1967]. We search over slownesses from 0 to 8 sec/deg and back-215
azimuthsbetween -180° to180° relative to theGCP.By selecting signalswith the216
highestcoherencewedeterminethebestfittingslownessvector(acombinationof217
theback-azimuth,θ,andthehorizontalslowness,u)oftheincomingsignals inthe218
scattering search time-window (1700s to 2400s after earthquake origin). To219
improve the resolution of the slowness vector of incoming signals, as well as to220
furtherincreasetheprominenceofsignalsabovethenoise,weapplytheF-statistic221
to the fk-analysis (Fig.4) [Blandford,1974;Selby,2011].TheF-statisticcalculates222
theratiooftheamplitudeofthestackedsignaltothesumofthedifferencesbetween223
the stack and each trace used to form the stack. The F-trace has the effect of224
penalising stacks that differ from individual input traces i.e. signals that are225
12
incoherentacrossthearray.Thus,thebestfittingslownessandback-azimuthfrom226
the grid-search are those that produce the most representative stack of the227
individual array traces.Byperforming these calculations in the frequencydomain228
weincreaseefficiencybyreducingthenumberoftransformationsrequiredbetween229
thetimeandfrequencydomains.However,thefkapproachreturnsasinglevalueof230
coherencefromeachslownessvectoraveragedacrossthewholetimewindow,thus231
collapsingthetimeaxis.CombiningtheF-statisticwithtraditionalfk-analysisresults232
in much-improved slowness vector resolution, even for the small-aperture arrays233
usedhere[Frostetal.,2013].Thustheoriginofthescatteredenergycanbemore234
preciselyestimated.235
We measure the slowness and back-azimuth of the most coherent signals236
receivedatthearrayinconsecutive50s longtimewindows(Fig.4).Thiswindow237
length gives depth resolution comparable to that obtained in global tomography238
models, and is sufficient to identify broad-scale trends in scattering distribution,239
both laterally and with depth. We assume the arrival time of a signal to be the240
middleofthe50stimewindow,andgiventhatscatteringofP′�P′fromarangeof241
depthscanarriveatthearraywithsimilartravel-times,each50stimewindowthat242
weinvestigateissensitivetoscatteringfroma50to200kmthicknessofthemantle.243
The thickness of the scattering region that each time window is sensitive to244
decreases with scattering depth hence, at shallower depths, there is overlap in245
depth sensitivity between windows – adjacent time windows can contain energy246
scatteredfromthesamedepth(albeitfromdifferentlocations).247
13
Mantle scattered P′�P′ waves are expected to arrive with slownesses248
between2.1and4.4s/deg.TherangeofdirectionsfromwhichP′�P′wavescanbe249
observed is dependent upon the event-array distance and scattering depth. Array250
analysespermitrecognitionandomissionofcontaminatingwavesbydetermination251
oftheincomingdirectionofenergy,comparedwiththedirectionspossibleforP′�P′.252
We compute the expected arrival times for possible contaminating waves: direct253
phases, depthphases, andmultiple reflectionsof bothP- andS-waves.Wedonot254
calculate multiples reflecting off upper mantle discontinuities (i.e. a downgoing255
wave reflecting off the 660 km discontinuity, then reflecting back down from the256
410 km discontinuity). Contaminating waves would likely be detected along the257
GCP (we take both minor and major arc arrivals into account). In contrast, P′�P′258
scatteredenergymostcommonlyarrivesofftheGCP,allowingclearidentificationof259
thescatteredarrivals.However,atshortevent-arraydistancesitispossibleforP′�P′260
toarrivealongtheGCP;thesesituationscanbepredictedandextracareistakento261
exclude contaminatingphases.As there are fewphases that can arrivewithin the262
P′�P′window (Fig 2),wewould expect few timewindows to be contaminatedby263
other seismic phases. Nonetheless, we discard any time window where we both264
observeasignalwithin20degreesoftheGCP(inmajororminorarcdirections)and265
anyknownseismicwaveispredictedtoarriveinthesametimewindowandalong266
the same backazimuth (i.e. minor or major great-circle path) (e.g Fig. 4d). Of all267
identified signals, only 2% match the time and direction predicted for known268
seismicphases,andthusarediscarded.269
14
Thewavefieldmayalsobecontaminatedbyforeshocksoraftershockstothe270
analysedevents,thusweexcludefromfurtheranalysisanyscatteredsignalswhere271
any magnitude ≥6 earthquake occurs within two hours of the origin time of the272
studied earthquake (11% of identified scattered signals). As a further test we273
removeany scattered signal that couldbe contaminatedbyamagnitude≥5event274
butfindnosystematicdifferenceinthedistributionofscatteringheterogeneity.Our275
focusoncorewavearrivalswithslownessesfrom2.1to4.4s/deghelpstoexclude276
contaminationfromsmaller,closerevents,whichhavehigherslownessesassociated277
with more horizontal incoming energy (and the discarding of GCP signals further278
minimizes energy from small local events contributing to data we analyse).279
Therefore,wearecertainthatourdataselectionpreventsanycontaminationofthe280
results by local and regional events. Lithospheric scattering directly beneath the281
arraymayredirecthighslownesscontaminatingenergytolowerslownessestypical282
ofmantlescatteringthatweconsiderhere.However,thedirectcontaminatingwave283
would arrive in the same time window as the lithospheric scattered energy, and284
would likely be more coherent with an obviously inappropriate slowness. This285
allows a straightforward identification (and removal) of energy scattered from286
lithosphericstructure.287
Aftercontaminatedtimewindowshavebeenremoved,scatteredsignalsare288
identified. We pick time windows containing energy prominently above the289
backgroundnoise level in f-k spaceandconsistentwith thedirectivity criteria for290
P′�P′ scattering (e.g. Fig. 4b). We identify the slowness and back-azimuth of the291
scattered signal, and select the time at the middle of the 50 s long window as292
15
scattered travel time; therefore, we only identify one scattered signal per 50 s293
window.IfmultipleP′�P′signalsareobservedinthesametimewindowweretain294
the signal with the highest coherence, as this will be the best spatially resolved.295
Multiplewavesarrivingat a similar time, either scatteredordirect,may interfere296
causing the apparent arrival direction of energy at the array to be incorrect. The297
apparentsignalwould likelyappearblurredacrossdirections, thusweonlyselect298
signalswithtightlyresolvedslownessandback-azimuth(withinthecapabilitiesof299
thearray).300
Theback-azimuth,slowness,andtimeinformationforeachscatteredsignal301
are used to calculate a scattering location in the mantle. The back-azimuth of a302
signal indicates thehorizontal direction alongwhich thewave travelledwhile the303
slownessdefinesadiscretepathfora1DEarthmodel,andthetravel-timerelatesto304
the scattering depth (Fig. S1). Thus there is a trade-off between the distance and305
depth of a scattered path, hence we attempt to fit both slowness and travel-time306
simultaneouslywithagridsearch.Weraytracebackwardsfromthearrayalongthe307
observedback-azimuthtoarangeofpossiblescatteringdepthsanddistances,and308
then ray-trace from these scattering locations to the source. Possible scattering309
locationsarespacedevery0.01° indistancebetween theminimumandmaximum310
possiblepath lengthsofPKPalongtheresolvedback-azimuthand50kmindepth311
from theCMB to the surface.Wemodel the scattering locationbyminimising the312
misfit between the calculated slowness and time for each potential scattering313
location and the observed values. Mapped scattering heterogeneity locations are314
discarded if traced rays to the solution location do not well fit the observed315
16
slownessandtime:ifthesquaredslownessmisfit(observedminuspredicted)plus316
twicethesquaredtimemisfitisgreaterthan10,i.e.aweightingfactorof2isused317
fortraveltimemisfitandthereforewefavourfittingscatteringlocationswithsmall318
travel-time misfits. The misfit value selected fits signals within the slowness319
resolution limit of the arrays. Overall, of the original 4319 identified scattered320
signals, we discard signals contaminated by other events (11% of the original321
population),otherphases(2%),andpoorlyfitsignals(44%),leaving1876mapped322
scatteringheterogeneities.323
Duetotheuncertaintyintravel-time(fromusingthemiddleofthe50stime324
window)andtheuncertaintyinslowness(duetotheabilityofthearraystoresolve325
theincomingdirection)wedeterminethedimensionsoftheregionthatcontainsthe326
heterogeneity based on these limitations. We calculate scattering locations for327
signals arriving at the start and end of the 50s time window, and with slowness328
variationof±0.3s/degrelative to thatmeasuredat thearray(estimated fromthe329
slowness spacingof thegrid-search).Thisdefinesa regionaround thebest fitting330
heterogeneitylocationthatis,onaverage,±100kmlaterallyandvertically.Formid-331
mantle scattering at high slowness values (~1000-1800 km depth), the error332
regionscanoccasionallygrowtovaluesaslargeas±800kmlaterallyand±500km333
vertically but this larger misfit is only relevant for around 5% of the solution334
scatteringlocations,thusthemajorityofthescatteringheterogeneitiesidentifiedin335
ourdatasetarelocatedtowithin±100kmverticallyandlaterally.336
Sub-surface structurebeneath themajorityof thearraysused in this study337
hasbeendemonstratedtohaveaninsignificanteffectontheresolvedslownessand338
17
back-azimuth (Bondar et al., 1999). Nevertheless, removing scattering339
heterogeneitiesobservedatChiangMaiarray,whichismostaffectedbysub-surface340
structure, dominantly reduces scattering in the upper 200 km of the mantle and341
causes no significant change in our conclusions on the relationship with lower342
mantlestructure.343
344
345
346Figure4:Arraydataareshownfor(aandb)amagnitude6.5event,24kmdepth,347
52°awayfromYellowknifearray,and(candd)amagnitude7.8event,0kmdepth,348
37°awayfromWarramungaarray.(a)ThetimewindowforP′�P′scattering(1700-349
2450secforthisevent,blueregioninFig.2).Thepredictedtimeofthedirectphase,350
P′P′df,isshownbytheverticalline,markingtheendofthescatteringwindowused351
here. Data are filtered between 0.5 and 2.0 Hz. The grey shaded time window352
18
corresponds to information shown in (b). (b) f-k processing of the 50 sec time353
windowshowngreyin(a),displayedintermsofback-azimuth(ȟ,azimuthalaxis)354
andslowness(u,radialaxisoutwardsfrom0to8s/degwithringsmarking2to6355
s/deg). Back-azimuth is measured relative to the great-circle path (vertical blue356
line).Thewhitestarshowsthemaximumcoherenceinthef-kanalysis,arrivingwith357
relative back azimuth = -106° (blue dashed line). The 90% coherence contour is358
roughly ±10° wide in back-azimuth and ±0.5 s/deg in slowness around the359
maximum. The green regions show the range of possible slownesses and back-360
azimuths for P′�P′ waves scattering at this distance and the median depth of361
scatteringforthistimewindow(fromtheshapeofthepotentialscatteringregions,362
greyregions inFig.3).(c) f-kprocessingofatimewindowshowingnoclearP′�P′363
waves.(d)f-kprocessingofatimewindowthatislikelycontaminatedbythedirect364
phasePKKKP (predicted slownessandback-azimuthmarkedbypurplediamond).365
Timewindows(c)and(d)arenotpickedforfurtherprocessing.366
[2COLUMNFIGURE]367
3683693704.Results371372
Themappedlocationsofscatteringheterogeneitiesareunevenlydistributed373
in themantle,both laterallyandwithdepth.This isnotunexpectedgiventhat the374
potentialsamplingcapacityofourdatasetalsovariesinlocationanddepth(Fig.3).375
Wedividethenumberofmappedscatterersbythepotentialsamplingdensity(Fig.376
3) in order to compare relative scattering density for different regions. This377
normalised scattering population shows that heterogeneities are distributed378
throughout the mantle, but more abundant scattering heterogeneity is present in379
theuppermostandlowermostmantle(Fig.5).Theradialscattererdistributionalso380
showsasmallincreaseinscatteringheterogeneitybetween600and900kmdepth,381
19
justbelowthetransitionzone,andaminimuminthemid-mantlebetween1400to382
1800kmdepth.383
We find that the radial abundance of small-scale scattering heterogeneity384
matchestheRMSamplitudeoflarge-scaletomographicvelocities(Fig.5):scattering385
ismostcommonand theRMSvariationof tomographicvelocities ishighest in the386
uppermost and lowermost mantle. This correlation holds roughly for all387
tomographicmodels(Fig.S2).388
389390
391
0
200
400
600
800
1000
1200
1400
1600
1800
2000
2200
2400
2600
2800
0.0 0.2 0.4 0.6 0.8 1.0
0 1 2 3 4
dVs RMS (SEMUCB-WM1)
Depth
(km
)
Normalised
scattering/sampling
20
Figure 5: Normalised scattering heterogeneity density with depth (number of392
scatterers divided by number of samples in each 100 km thick layer) for the393
completedataset(black,lowerx-axis)andRMSoftheshearvelocityperturbations394
fromtheglobaltomographicmodelSEMUCB-WM1(greyanddashed,upperx-axis)395
[French and Romanowicz, 2014]. The depth distribution of small-scale scattering396
heterogeneity roughly correlates with the RMS of long-wavelength dVs397
perturbations.Bothlinesarescaledtofitthesameaxis.398
[SINGLECOLUMNFIGURE]399
400401
Weinvestigatepossiblespatialcorrelationbetweentheresolvedscattering402
heterogeneities and large-scale mantle features, which may be interpreted as403
proxiesfordynamicprocesses,asinFrostetal.,[2017].Wecomparethelocationof404
scatteringheterogeneitytogeographicalregionsbeneathhotspots,subductedslabs,405
regionsofhighandlowtomographicvelocities,andregionsofhighandlowlateral406
tomographic velocity gradients. The high/low velocities and gradients from407
tomographicmodelslikelyrelatetothelocationsofLLSVPsandsubductedslabsin408
the mantle. The spatial locations of scattering heterogeneities are shown,409
normalisedbysampling,intheSupplementaryMaterial,whiletheabsolutelatitude,410
longitude, and depth information for each scattering heterogeneity is shown in411
SupplementaryTable1412
413
414
4.1Relationshipbetweenscatteringheterogeneitiesandmantlestructure415
We compare the distribution of scattering heterogeneity with S-wave416
tomographicmodels,bothbecausetheyarethebasisforthedefinitionoftheLarge417
21
LowShearVelocityProvinces,andalsoshowconsistencybetweenmodels[Garnero418
etal.,2016].Weuseseveral tomographicmodels:GyPSuM[Simmonsetal.,2010],419
SEMUCB-WM1[FrenchandRomanowicz,2014],S40RTS[Ritsemaetal.,2011],and420
TX2011 [Grand,2002].Additional comparisonswithP-wavemodels are shown in421
thesupplementarymaterial(Figs.S7-9).Wecalculatelateralvelocitygradientsfrom422
tomographic models, revealing abrupt changes in mantle structure, which thus423
serveasaproxy forboundariesof theLLSVPs [Thorneetal.,2004,Garneroetal.,424
2016].Wecalculategradientsoveradistanceof10°astheresultinggradientswell425
replicatethemarginsoftheLLSVPsfoundinforwardmodellingstudies[Garneroet426
al.,2016andreferencestherein].427
We use hotspots from the study of Courtillot et al. [2003]. French and428
Romanowicz [2015] analysed the tomographic model SEMUCB-WM1 [French and429
Romanowicz, 2014] and characterised hotspots based on associated tomographic430
velocityanomalies.Weusethe20hotspotsthatwerelabelledaseither“primary”or431
“clear”meaningthatthehotspotoverliesacolumnoflowvelocitiesfromtheCMBto432
1000kmdepthwithdVslessthan-1.5%orlessthan-0.5%,respectively.433
We use slab locations from the Regionalized Upper Mantle (RUM) model,434
which locates slabs at depth using intra-slab seismicity [Gudmundsson and435
Sambridge,1998].Whencomparingwithscatteringheterogeneitylocationsweuse436
slablocationsatthesurface(zerodepth).Slabsmoveonlyasmallamountlaterally437
astheysubduct(ӊ5°relativetotheplateboundaryatthesurface[Steinbergeretal.,438
2012]),whichisunlikelytostronglyinfluenceourcorrelations.439
22
To account for differences in themagnitude, range, andpattern of velocity440
anomalies and velocity anomaly gradients between tomographic models, and441
differencesinthenumberoflocationsinofmapsofhotspotandslablocations,we442
convertmapsoftomographicvelocitiestomapsofpercentagecumulativeareaona443
spheresortedbydecreasingvelocityanomaly(fromfasttoslow).Forexample,the444
20%area corresponds to the area of the 20%highest tomography velocities of a445
givendepthshell(Fig.6).Weonlyconsidertheregionsofthetomographicmodels446
thatmatchtheregionssampledbytheP′�P′datasetateachdepth.Inthisway,the447
highest and lowest velocities in several tomographic models with inherently448
differing amplitudes of velocity variation can be directly compared. We establish449
geographical area percentages associated with the locations of hotspots and450
subductedslabsbycomputingthecumulativeareasurroundingthefeatureswithin451
specific distances from them (within the area sampled by the P′�P′ dataset). For452
example,thefirst20%areaforslabscorrespondstotheregionclosesttoslabsthat453
addsupto20%oftheEarth’ssurfacearea;converselythelast20%areaindicates454
thatamountofsurfaceareafurthestfromslabs.455
To estimate correlations between the abundance of small-scale scattering456
and subducted slabs, hotspots, tomographic velocities and gradients, we compare457
the location of these features to the distribution of scattering heterogeneity. For458
each 100 km depth shell, we count scattering heterogeneities in each 20% area459
division from the feature of interest. To account for the variability in sampling460
coverage of our dataset (Fig. 3), we count our estimation of potential scatterers461
(afforded by our event-array distributions) in the same 20% area regions, and462
23
calculate the ratio of the number of observed-to-potential heterogeneities. This463
allowsus to construct amapof normalised scatteringprevalence, thus effectively464
removingthebiasofourunevensampling.465
ThefirstsetofcomparisonsisdisplayedinFig.7asacumulativehistogram466
asafunctionofdepth.Intheupper200kmofthemantle,scatteringheterogeneities467
are most common in regions of high velocity (Fig. 7a), which is evident from the468
horizontalwidthofthelightanddarkblueshadingbeinggreaterthanthewidthof469
the light and dark red shading over the same depth range. In the lower mantle,470
especially in the deepest 500 km or so, the opposite is true: scattering471
heterogeneitiesaremoreabundantinlowvelocityregions(asevidentbywiderred472
shading). Regions of the lowermost mantle with high seismic velocities show473
virtuallynocorrelationwith scatteringheterogeneities.Scattering is slightlymore474
commoninregionsofhighseismicvelocitybetween600and900kmdepth.475
In the deepest 200 km of the mantle, scattering heterogeneities are more476
common inregionsofhigh lateralseismicvelocitygradients(Fig.7b: thewidthof477
theblackanddarkblueshadingissignificantlygreaterthanthelightgreencolors).478
In the lowest ~1000 km of the mantle, scattering heterogeneities are in greater479
abundance in the20%areaaroundhotspots than inanyotherbin; there isalsoa480
slight increase in mapped heterogeneities beneath hotspots in the mid-mantle481
between600-900kmdepth(seethewideredcolorsshading,Fig.7c).Ourmapped482
scattering heterogeneities show little correlation with regions surrounding the483
surface location of slabs, except in areas furthest from slabs in the 600-900 km484
depth range (indicated by the wide orange-yellow shading, Fig. 7d). In the upper485
24
200kmofthemantle,scatteringstronglycorrelateswithhighseismicvelocitiesand486
proximitytoslabs(Figs.7aand7d,blueandyellowshading,respectively),which,at487
theseshallowdepthsismostcloselyrelatedtothelocationofcontinents.Whilethe488
precise locations of the heterogeneities is different, the heterogeneities resolved489
withP′�P′showaverysimilardistributioninthelowermost300kmofthemantle490
tothoseheterogeneitiesresolvedwithPK�KPinanearlierstudy[Frostetal.,2017].491
Totesttherobustnessofthesecorrelationswedeterminehowlikelytheyare492
tohavebeenproducedby chance.We rotate the tomographicmodels (of velocity493
and lateral gradient), and hotspot and slab locations by a random angle about a494
randomlylocatedpoleofrotation.Wethenrecomputethecorrelationsbetweenthe495
rotated geographical features and the distribution of the unrotated scattering496
heterogeneities. The random rotation is repeated 200 times for each tomography497
model,aswellasforthehotspotandslablocations,tocalculatetherangeofpossible498
correlations.Themeanandstandarddeviationoftherangeofcorrelationsateach499
depthiscomputed,assumingGaussianstatistics.Wecomparethiswiththeoriginal,500
unrotateddatainFig.8,andconsideranycorrelationtobestatisticallysignificantif501
thecorrelationvaluebetweenscatterersandregionsof theunrotatedphenomena502
plots outside one standard deviation from the mean correlation of the rotated503
phenomena (demonstrating that at least 84% of the random correlations are a504
lower value). When we do not assume a distribution and instead calculate the505
proportionofsamplesaboveandbelowonestandarddeviationofthedata,wefind506
very similar patterns of significant observations. Using this metric, we define the507
followingcorrelationsassignificantandunlikelytheproductofchance:508
25
(1) Anincreasedcorrelationwithscatterersinregionsoflowvelocityatdepths509
greaterthan1800km(solidredlineintheleftpanelofFig.8a)510
(2) An increased abundance of scattering heterogeneities in regions of high511
velocity gradient in thedeepest fewhundredkmof themantle, aswell as512
between1600-2000kmdepth(solidredlineintherightpanelofFig.8b).513
(3) Anincreasedabundanceofheterogeneitiesclosetosurfacehotspotlocations514
atdepthsgreaterthan2100kmdepth(solidredlineinleftpanelofFig.8c).515
(4) Adecreasedabundanceofheterogeneitiesfarfromsurfacehotspotlocations516
atalmostalldepthsgreaterthan800kmdepth(solidredlineinrightpanel517
ofFig.8c).518
There is no significant correlation seen between scatterer locations and slab519
locations, except an increase in correlation between heterogeneities and large520
distancesfromslabsbetween600and900kmdepth,whichmatchesthedepth521
rangeoftheincreasedcorrelationwithlowvelocitygradients(solidredlinesin522
leftpanelofFig.8bandrightpanelof8d),andanincreasedcorrelationbetween523
heterogeneities and large distances from slabs throughout much of the lower524
mantle(whichiswhatoneexpectsifcorrelationsarestrongforlowvelocities).525
526
4.2Dependenceuponchoiceofmodel527When comparing small-scale scattering locations with tomographically528
derived high or low velocities, the results may depend upon the choice of the529
tomography model. In our previous analysis, we compared the distribution of530
scattering heterogeneities to tomography model SEMUCB-WM1 [French and531
Romanowicz,2014].Wefurtherexploretherelationshipbetweenourmappedfine-532
26
scale scattering heterogeneities with large-scale structures in other tomography533
models:GyPSuM,S40RTS,andTX2011(Figs.S4-6andS7-9forP-wavemodels).We534
find small differences in precise depths and magnitudes of correlations with535
differentmodels,butthecorrelationbetweenscatteringandlowvelocitiesatdepths536
below1600km,andwithhighvelocitiesatdepthsof200kmandshallowerandthe537
robustnessofthesecorrelationsareconsistentbetweenmodels.538
Totestthedependenceofcorrelationonthepatternofhotspots,inaddition539
to comparing with rotated hotspot locations, we create a population of randomly540
locatedhotspots,equalinnumbertotheprimaryandclearhotspotsfromCourtillot541
et al., [2003] and French and Romanowicz [2015]. We find that a synthetic542
population generates no preferential spatial correlation with the scattering543
heterogeneities (Fig. S10). Furthermore, when the population of random hotspot544
locations isrotatedto test therobustnessof thecorrelation, thecorrelationof the545
randompopulationveryoftenfallswellwithintheonestandarddeviationrangeof546
the rotated data (Fig. S12). This implies that the observed correlation between547
hotspot locations and scattering heterogeneities in the lowermantle is caused by548
thespecificdistributionofhotspots.549
Wetesttheinfluenceofourdecisiontouseonlythesurfaceslablocationsof550
the RUM model. We calculate the spatial correlation between scattering551
heterogeneitiesandslablocationsasdescribedabove,butuseslablocationsatthe552
depthof theheterogeneity.Whenconsideringscatteringheterogeneitiesatdepths553
greaterthanthatwhichtheslabismappedtoweusethelocationoftheslabatthe554
last mapped depth and project this position vertically down to the CMB. This555
27
method of vertical extrapolation likely still misrepresents the locations of slabs:556
someamountoflateralmovementatgreaterdepthsisevidentintomographicand557
geodynamicmodelsbut is typicallyontheorderofa fewdegrees[e.g.Frenchand558
Romanowicz,2014andSteinbergeretal.,2012].Nonetheless,wefindnosignificant559
difference in the correlations between using the surface slab location and slab560
locationswithdepth(Fig.7andFigs.S11andS12).561
562
563Figure 6: Tomography and distance from subduction zones and hotspots by564
percentage area calculated for sampling at the CMB. (a) The magnitude of the565
velocity anomalies in SEMUCB-WM1 [French and Romanowicz, 2014] in the area566
sampledbyourdatasetattheCMBdisplayedbydecreasinganomaly(fromfastblue567
28
areastoslowredareas)inregionsoccupying20%oftheareaoftheCMB.(b)The568
magnitudeofthelateralvelocitygradientdecreasingfromhightolowin20%area569
regions. (c) Distance from hotspots (connected to plumes identified as either570
primary or clear in the analysis of French and Romanowicz [2015]). (d) Distance571
fromslabs(atzerodepthsliceinRUM[GudmundssonandSambridge,1998].Black572
linemarkstheextentofthesampledarea(asinFigure3d).573
[2COLUMNFIGURE]574
575
576
577Figure 7: Scattering abundance (bar width) with depth compared with the578
distribution of large-scale heterogeneity throughout the mantle (colour scale).579
Scattering abundance is calculated cumulatively across all areas, is divided by580
0
200
400
600
800
1000
1200
1400
1600
1800
2000
2200
2400
2600
2800
Depth
(km
)
0 0.25 0.5 0.75 1.0
0 20 40 60 80 100Dist. from HIGH dVs by % area
a dVs by area
0 0.25 0.5 0.75 1.0
0 20 40 60 80 100Dist from HIGH D(dVs) by % area
b D(dVs) by area
Dist. from SLABS by % area 0 20 40 60 80 100
0 0.25 0.5 0.75 1.0
d Dist. to Slabs at z=0
Dist. from HOTSPOTS by % area 0 20 40 60 80 100
0 0.25 0.5 0.75 1.0
c Dist. to Hotspots
Normalised cumulative
scattering/sampling
Normalised cumulative
scattering/sampling
Normalised cumulative
scattering/sampling
Normalised cumulative
scattering/sampling
29
sampling, and is normalised to unity, representing the maximum scattering581
abundance at any depth. (a) Scattering heterogeneity and tomographic velocity582
anomalies (from SEMUCB-WM1 [French and Romanowicz, 2014]) sorted from583
highest (blue) to lowest (red)measuredasa functionofsurfacearea in20%area584
bins.(b)Scatteringheterogeneityandlateraltomographicvelocitygradientsorted585
fromhighest (dark blue) to lowest (light green). (c) Scattering heterogeneity and586
distance from hotspots from low to high (red and yellow, respectively). (d)587
Scatteringheterogeneityanddistancefromslabsfromlowtohigh(redandyellow,588
respectively). Scattering heterogeneity in the lower mantle shows an affinity for589
both low seismic velocities and hotspots. Black lines encapsulate the highest and590
lowest40%arearegions.591
[2COLUMNFIGURE]592
593
30
594Figure 8: Scattering abundance with depth, divided by sampling, showing the595
unrotated model (red line) compared with rotated models (grey). The unrotated596
0
200
400
600
800
1000
1200
1400
1600
1800
2000
2200
2400
2600
2800
Lowest 20% dist. area Highest 20% dist. area
0 0.25 0.5 0.75 1.00 0.25 0.5 0.75 1.0
0
200
400
600
800
1000
1200
1400
1600
1800
2000
2200
2400
2600
2800
Lowest 20% dVs area Highest 20% dVs area
0 0.25 0.5 0.75 1.00 0.25 0.5 0.75 1.0
0
200
400
600
800
1000
1200
1400
1600
1800
2000
2200
2400
2600
2800
0 0.25 0.5 0.75 1.00 0.25 0.5 0.75 1.0
Lowest 20% D(dVs) area Highest 20% D(dVs) area
0
200
400
600
800
1000
1200
1400
1600
1800
2000
2200
2400
2600
2800
Lowest 20% dist. area Highest 20% dist. area
0 0.25 0.5 0.75 1.00 0.25 0.5 0.75 1.0
d
ba
c
dVs heterogeneity gradient
Hotspot locations Subduction zone locations
dVs heterogeneity
Depth
(km
)D
epth
(km
)
Depth
(km
)D
epth
(km
)
Normalised
scattering/sampling
Normalised
scattering/sampling
Normalised
scattering/sampling
Normalised
scattering/sampling
Normalised
scattering/sampling
Normalised
scattering/sampling
Normalised
scattering/sampling
Normalised
scattering/sampling
31
model(redline)isdashedwhenwithinonestandarddeviation(darkgrayshading)597
of themeanof the spatial correlations (blackdiamonds)with the rotatedmodels,598
andsolidwhenoutsidethis level.Thelightershadedregionmarkstherangeofall599
correlationswiththerandomlyrotatedphenomena.Comparisonsareshownfor:(a)600
tomographically derived velocity heterogeneity from SEMUCB-WM1 for the 20%601
area corresponding to the lowest (left panel) and highest velocities (right panel).602
Correlation between increased scattering abundance and low velocities appears603
robust in the deepest mantle, and correlation to high velocities is robust in the604
shallowest200kmof themantle, aswell asaround1200kmdepth.(b)As in(a)605
except correlations are between observed scattering and rotated shear velocity606
gradients in model SEMUCB-WM1. Correlations are most significant for the607
strongest gradients (right panel) at the base of the mantle. (c) As in (a) except608
correlations are between scatterers and distance to rotated hotspot regions.609
Correlationsaremostsignificant inthedeepestmantle incloseproximitytobeing610
beneath hotspots (left panel). (d) as in (c) except correlations are between611
scatterersanddistancetorotatedslabregions.Ourrandomrotationtestshowsno612
significantcorrelationbetweenscatterersandproximitytoslabs.613
[2COLUMNFIGURE]614
615
32
6165.Discussion617618
In this study we mapped scattering heterogeneities and explored their619
geographical relationship to tomographic velocities and gradients, as well as620
hotspotsandslabs. Ourresultsmaybe interpreted in termsof thedistributionof621
mantleheterogeneity,whichwewilldiscusshere.622
6235.1Possibleoriginsofscatteringheterogeneityinthemantle624625
We observe scattering from small-scale heterogeneity throughout the626
mantle,with increasedheterogeneityat thetopandbottomofthemantle.Seismic627
waves can be scattered by volumetric heterogeneity with sharp impedance628
contrasts,when theheterogeneity has aminimumscale length comparable to the629
wavelengthoftheincidentwave.Ourmethodisnotcapableofresolvingtheprecise630
partitioning of the incident wavefield into scattered versus transmitted energy,631
sincewedonothaveaconsistentreferencephasetocomparetotheamplitudeof632
the scattered wave. Thus we are unable to constrain the properties of the633
heterogeneities (e.g. impedance contrast). Nonetheless, the frequencies of waves634
that we study (between 0.5 and 2.0 Hz) imply that observed scattering635
heterogeneitieshaveaminimumscalelengthofonetotensofkm.636
Avarietyofstructurescouldscattertheenergyobservedinourdata.Wecan637
usethedistributionandsizesofscatteringheterogeneitiestoaddressthefeasibility638
ofpossiblecauses.Materialundergoingphasechangessuchasfrombridgmaniteto639
post-perovskite(pPv)inthelowermantle(orthebacktransformation)[Murakami640
etal.,2004;OganovandOno,2004],aswellastransitionsofolivinetowadsleyiteto641
33
ringwooditetoperovskitethroughtheuppermantletransitionzonecouldprovide642
an impedance contrast with the ambient mantle. The bridgmanite to pPv phase643
transitionispredictedtooccurinthedeepestfew100kmofthemantle,andonlyin644
relatively cold regions of the mantle for a standard pyrolitic composition, thus645
would not be appropriate to explain scattering at all depths and locations, unless646
mineralogical alterations are considered [Lay et al. 2006]. Thephase transition is647
controlledby temperature, composition, andpressure.Whilepressure is assumed648
hydrostatic, localchanges incomposition,perhapsbycontaminationof themantle649
by subducted mid-ocean ridge basalt (MORB), may influence the pPv transition650
[Grocholski et al., 2012], possibly causing the transition to occur locally in the651
vicinity of the MORB contamination. Metastability of phase transitions due to652
chemical heterogeneity [Catalli et al., 2009] could allow transformed minerals to653
persist outside of their expected stability range. High thermal conductivity in the654
lowermantle[Stackhouseetal.,2015]renderssmall-scaletemperaturechangesan655
unlikely cause of spatially limited occurrence of the pPv transition. While many656
morphologies and scale lengths of pPv regions can be envisioned that could657
contributetowavefieldscatteringobservedhere,thedetailsofsuchprocessesare658
notconstrained.However,pPvshouldnotbestable in theuppermantle,andthus659
cannot explain observed scattering there. Nonetheless, pPv remains a viable660
contributortowavefieldscatteringinthedeepestmantle.661
The subduction process continuously introduces compositional662
heterogeneityintothemantle.Scatteringhaspreviouslybeenmappedintheupper663
mantle and lower mantle in the proximity of subduction zones [Kaneshima and664
34
Helffrich, 1998; Rost and Earle, 2010; Miller and Niu, 2008; Bentham and Rost,665
2014].Wedonotobservearobustpreferenceofscatteringheterogeneityinupper666
mantle regions of subduction over other regions. While we do observe slightly667
increasedscatteringinregionsassociatedwithsubductionataround600to900km668
depth (Fig. 7a), this does not appear to be statistically significant (Fig. 8a, right669
panel). Nonetheless, the increased concentration of scattering heterogeneity670
between600and900kmdepthshowsrobustspatialcorrelationwithregionsaway671
fromsubductionzonesandareasoflowamplitudelateralvelocitygradient(Figs.7672
and7). Insome tomographicmodelssubductingslabsareobserved to flattenata673
similardepth,between~800-1200kmdepth[e.g.FrenchandRomanowicz,2015].674
Oceanic crust may be responsible for scattering throughout the mantle.675
Subductedoceaniccrustmayremainunmixedduetoslowchemicaldiffusionrates676
[Olson et al., 1984] and is only homogenised into the mantle through mechanical677
stirring. If the observed scattering heterogeneities are oceanic crust then the678
dispersalofheterogeneitiesthroughoutthemantlemustbefasterthanstirringand679
removal of heterogeneities since scattering heterogeneity is also observed in680
regionsthathavenotbeeninfluencedbysubductionforalongtime.681
The iron spin transition affects the velocity and density of iron-bearing682
mantlematerials[Linetal.,2005].Recently,thishasbeenobservedtooccurovera683
60GPapressurerange(~600to2000kmdepth)[HolmstromandStixrude,2015]684
and thus would likely not generate discrete heterogeneities capable of causing685
scattering.686
35
Products of chemical reactions between core and mantle materials are687
predictedtohavephysicalpropertiesincontrastwiththeambientmantle[Knittle688
andJeanloz,1989]thusmaybecapableofcausingseismicscattering.Experiments689
demonstratethatsuchmantlematerialenrichedinironwouldlikelybedenserthan690
the ambient mantle [Wicks et al., 2010]. An interesting possibility is the691
developmentofareactionproductlayerthatwouldinhibitfurtherinteractionwith692
the core; for this case, products are likely to be constrained to a very limited693
thicknessclosetotheCMB,ontheorderofafewmeterstokilometers[Kandaand694
Stevenson, 2006]. However, flow in the deep mantle could generate thicker695
accumulationsofreactionproducts[Maoetal.,2006],whichcouldscatterwaves.In696
addition,ULVZsarecommonly imagedtohavevastlyreducedseismicvelocitiesof697
up to -10%dVp and -30%dVs, and increased density of +10-20% relative to the698
surroundingmantle[e.g.,McNamaraetal.,2010].Partialmeltofmantlematerialhas699
been proposed as an explanation of ULVZs [Williams and Garnero, 1996]. Partial700
meltmaybedenserthanthesolidstate[OhtaniandMaeda,2001]aswellashaving701
stronglyreducedseismicvelocities.WhileULVZsandCMBreactionproductscould702
explain deeper scattering heterogeneities, simulations have suggested that dense703
materialmayalsobeentrainedupto200kmabovetheCMB,dependentondensity,704
viscosity, and vigor of mantle flow [Bower et al., 2011]. CMB topography or705
roughnessmightcausescattering[ChangandCleary,1981;Mancinellietal.,2016],706
butthiscouldnotexplainheterogeneitieswemapupoffoftheCMBthroughoutthe707
mantle.LLSVPsmaybecompositionallydistinct fromthesurroundingmantle [e.g.708
Garnero et al., 2016], anddynamical flowmodels predict that the LLSVPmaterial709
36
willbegraduallyentrainedintomantleflowonsmall lengthscales[Lietal.,2014;710
Williams et al., 2015; Mulyukova et al., 2015]. Thus, depending on the LLSVP711
properties and entrained heterogeneity scale, this process might give rise to712
scattering. Geodynamic models also predict that surrounding ambient mantle713
material can be downward entrained into the LLSVPs, thus offering an origin of714
scatteringwithinLLSVPregions.715
7165.2Distributionofscatteringheterogeneity717718
Thedistributionof small-scale volumetric heterogeneities is likely strongly719
dependentonthedynamicpropertiesandprocesseswithintheEarth.Innumerical720
simulationsofmantledynamicssmall-scaleheterogeneity,particularlythatderived721
fromsubductedoceaniccrust,tendstobeconcentratedinregionsofupwellingfrom722
the lower mantle around plumes and downwelling from the surface around723
subductionzones(Fig.1ofLietal.,[2014]).Thesamefocusingbeneathupwellings724
isexpectedforbasalheterogeneities[McNamaraetal.,2010](e.g.,compositionally725
distinct ULVZ material, CMB reaction products, and entrained LLSVP material).726
Furthermore, large-scale mantle heterogeneity may influence radial small-scale727
heterogeneity distribution by modifying the convective flows in which the728
heterogeneitiescouldbeentrained[Lietal.,2014].729
As wavelength at some fixed frequency is a function of the local velocity,730
whichchangeswithdepth, and thewavelengthof scattering structure that canbe731
resolvedisdependentontheincidentfrequency,itfollowsthatinbandlimiteddata,732
theresolvablescatteringwavelengthchangeswithdepth.Wefilteralldatabetween733
0.5 and 2.0 Hz, therefore, we resolve scattering heterogeneity with wavelengths734
37
between about 7-28 km at the CMB, decreasing to about 3-12 km at the surface.735
Stirring of initially larger-scale heterogeneity is suggested to lead to a cascade of736
heterogeneity sizes, increasing in abundance with decreasing scale [Olson et al.,737
1984].Apreviousstudyofthescaleofscatteringheterogeneitiesinthelowermost738
mantle found the most common scale-length to be 4-7 km, but other scales were739
also present [Frost et al., 2017]. Despite the limited frequency range used in this740
study, we are likely imaging heterogeneity of a similar size (around 7 km)741
throughoutthemantle.742
Thesimilaritybetweenscatteringheterogeneityabundanceandtomographic743
amplitude (Fig. 5) may arise from processes relating to convection and chemical744
differentiationthat likelygeneratestrong lateralvelocityvariationsoncontinental745
scalesandsmallerthroughstirringanddiffusion.Lowermantleanomaliesmanifest746
atarangeofspatialscales(LLSVPs,ULVZs,D′′,CMBreactionproducts),andstirring747
andentrainmentmayfurtherdecreasetheirsize[Olsonetal.,1984,Lietal.,2014],748
leadingbothtohigh-amplitudelarge-scalevelocityanomaliesandabundantsmall-749
scalescattering.Uppermantleheterogeneityrelatedtosubduction,magmatism,and750
convectiveprocessesarealsolikelytooccuracrossscales.Inadditiontoincreased751
scatteringatthetopandbottomofthemantle,wealsoobserveaslightbutmarked752
increase in scattering abundance from 600-900 km depth, independent of the753
tomographic velocity structure,whichmay relate to slab subductionprocesses or754
large-scaleverticalviscositychanges[Rudolphetal.,2015].755
7566.Conclusion757758
38
Through analysis of the high-frequency seismic wavefield we map the759
distributionofsmall-scaleseismicheterogeneity,ontheorderof~1-10kminsize,760
throughout Earth’s mantle. We deterministically locate vastly more scattering761
heterogeneities than has been done previously, significantly improving our762
understanding of small-scale mantle structure. The spatial distribution and scale-763
lengthofthisscatteringheterogeneitysuggestsitmaybetheproductofseveralon-764
goingprocessesinthemantle.Theseincludeoceaniccrustdisseminatedthroughout765
the mantle, entrainment of basal heterogeneities such as ULVZ material or core-766
mantle reaction product, and compositionally distinct LLSVP material swept into767
mantle flow. Subducted MORB may suitably explain all scattering observations768
withoutscatteringcontributions fromothersources.However,wecannotruleout769
that scattering is caused by a mixture of heterogeneities with different origins in770
different regions and depths. While small-scale heterogeneity appears present in771
much of the mantle, we find increased scattering heterogeneity within the772
uppermost200kmofthemantleandthelowermost300kmofthemantle,similar773
to heterogeneity amplitudes seen in tomography models. We find no statistically774
significant correlationbetweenscatteringandsubductingslabs in theupper1000775
kmof themantle. In the lowermantle (fromaround1500kmdepthdown to the776
CMB), scattering is most common in regions related to the LLSVPs and close to777
deeply sourcedmantlehotspots.Meanwhile, scattering is rare in regions far from778
deeply sourced mantle hotspots. This suggests that large-scale convective lower779
mantlestructuresmayentrainandconcentratesmall-scaleheterogeneityinregions780
ofupwelling,downwelling,andstagnantflow.781
39
782Acknowledgements:ThisworkwassupportedNSFgrantEAR1401270andNERC783
grantsNE/K006290/1andNE/H022473/1.CTBTInternationalMonitoringSystem784
datausedherewereacquiredwhiletheauthorwasundertakingastudentship785
sponsoredbytheUKNationalDataCentreatAWEBlacknest.Thepaperbenefited786
fromdiscussionswithBarbaraRomanowicz,andimprovedfromhelpfulcomments787
oftworeviewers.788
789
790
791
References:792Bentham,H.L.M.&Rost,S.ScatteringbeneathWesternPacificsubductionzones:793evidenceforoceaniccrustinthemidmantle.GeophysicalJournalInternational197,7941627–1641(2014).795796Blandford,R.R.AnautomaticeventdetectorattheTontoForestseismic797observatory.Geophysics39,633–643(1974).798799Bower,D.J.,Wicks,J.K.,Gurnis,M.&Jackson,J.M.Ageodynamicandmineralphysics800modelofasolid-stateultralow-velocityzone.EarthPlanet.Sci.Lett.303,193–202801(2011).802803Bondar,I.,North,R.G.,Beall,G.,1999.Teleseismicslowness-azimuthstationcorrec-804tionsfortheInternationalMonitoringSystemseismicnetwork.Bull.Seism.Soc.805Am.89,989–1003.806807Brandenburg,J.P.&vanKeken,P.E.Deepstorageofoceaniccrustinavigorously808convectingmantle.JournalofGeophysicalResearch112,1–15(2007).809810Capon,J.,Greenfield,R.&Kolker,R.Multidimensionalmaximum-likelihood811processingofalargeapertureseismicarray.Proc.IEEE55,192–211(1967).812813Catalli,K.,Shim,S.-H.&Prakapenka,V.ThicknessandClapeyronslopeofthepost-814perovskiteboundary.Nature462,782–5(2009).815816Chang,A.C.&Cleary,J.R.ScatteredPKKP:Futherevidenceforscatteringatarough817core-mantleboundary.Phys.Earthplanet.Int.24,15–29(1981).818
40
819Cleary,J.&Haddon,R.A.W.Seismicwavescatteringnearthecore-mantle820boundary:anewinterpretationofprecursorstoPKP.Nature240,549–551(1972).821822Courtillot,V.,Davaille,A.,Besse,J.&Stock,J.Threedistincttypesofhotspotsinthe823Earth’smantle.EarthPlanet.Sci.Lett.205,295–308(2003).824825Domeier,M.,Doubrovine,P.V.,Torsvik,T.H.,Spakman,W.&Bull,A.L.Global826correlationoflowermantlestructureandpastsubduction.Geophys.Res.Lett.8274945–4953(2016).828829Doubrovine,P.V.,Steinberger,B.&Torsvik,T.H.Afailuretoreject:Testingthe830correlationbetweenlargeigenousprovincesanddeepmantlestructureswithEDF831statistics.Geochem.Geophys.Geosyst.1130–1163(2016).832833French,S.W.&Romanowicz,B.A.Whole-mantleradiallyanisotropicshear-834velocitystructurefromspectral-elementwaveformtomography.Geophys.J.Int.199,8351303–1327(2014).836837French,S.W.&Romanowicz,B.A.BroadplumesrootedatthebaseoftheEarth’s838mantlebeneathmajorhotspots.Nature525,95–99(2015).839840Frost,D.A.,Rost,S.,Selby,N.D.&Stuart,G.W.Detectionofatallridgeatthecore-841mantleboundaryfromscatteredPKPenergy.Geophys.J.Int.195,558–574(2013).842843Frost,D.A.,Rost,S.,Garnero,E.J.&Li,M.SeismicevidenceforEarth’scrustydeep844mantle.EarthandPlanetaryScienceLetters470,54–63(2017).845846Garnero,E.J.,McNamara,A.K.&Shim,S.-h.Continent-sizedanomalouszoneswith847lowseismicvelocityatthebaseofEarth’smantle.NatureGeosci.1–9(2016).848849Grand,S.P.Mantleshear-wavetomographyandthefateofsubductedslabs.Phil.850Trans.R.Soc.LondA360,2475–2491(2002).851852Grocholski,B.,Catalli,K.,Shim,S.-H.&Prakapenka,V.Mineralogicaleffectsonthe853detectabilityofthepostperovskiteboundary.ProceedingsoftheNationalAcademy854ofSciences109,2275–2279(2012).855856Gudmundsson,O.&Sambridge,M.Aregionalizeduppermantle(RUM)seismic857model.J.Geophys.Res.103,7121–7136(1998).858859Hedlin,M.A.H.,Shearer,P.M.&Earle,P.S.Seismicevidenceforsmall-scale860heterogeneitythroughouttheEarth’smantle.Nature387,145–150(1997).861862Holmström,E.&Stixrude,L.SpinCrossoverinFerropericlasefromFirst-Principles863MolecularDynamics.Phys.Rev.Lett.117202,1–5(2015).864
41
865Kanda,R.V.S.&Stevenson,D.J.Suctionmechanismforironentrainmentintothe866lowermantle.Geophys.Res.Lett.33,4–7(2006).867868Kaneshima,S.&Helffrich,G.Detectionoflowermantlescatterersnortheastofthe869Mariannasubductionzoneusingshort-periodarraydata.J.Geophys.Res.103,8704825–4838(1998).871872Knittle,E.&Jeanloz,R.Simulatingthecore-mantleboundary:Anexperimentalstudy873ofhighpressurereactionsbetweensilicatesandliquidiron.Geophys.Res.Lett.16,874609–612(1989).875876Lay,T.,Hernlund,J.,Garnero,E.J.&Thorne,M.S.Apost-perovskitelensandD”heat877fluxbeneaththecentralPacific.Science314,1272–6(2006).878879Li,M.,McNamara,A.K.&Garnero,E.J.Chemicalcomplexityofhotspotscausedby880cyclingoceaniccrustthroughmantlereservoirs.NatureGeosci.7,366–370(2014).881882Lin,J-f.,Struzhkin,V.V.,Jacobsen,S.D.,Hu,M.Y.,Chow,P.,Kung,J.,Liu,H.,Mao,H-k.883&Hemley,R.J..SpintransitionofironinmagnesiowustiteintheEarth’slower884mantle.Nature436,377–380(2005).885886Mancinelli,N.J.&Shearer,P.M.Reconcilingdiscrepanciesamongestimatesofsmall-887scalemantleheterogeneityfromPKPprecursors.Geophys.J.Int.195,1721–1729888(2013).889890Mancinelli,N.,Shearer,P.&Thomas,C.Onthefrequencydependenceandspatial891coherenceofPKPprecursoramplitudes.J.Geophys.Res.121,1873–1889(2016).892893Mao,W.L.etal.Iron-richpost-perovskiteandtheoriginofultralow-velocityzones.894Science312,564–5(2006).895896McNamara,A.K.&Zhong,S.ThermochemicalstructuresbeneathAfricaandthe897PacificOcean.Nature437,1136–9(2005).898899McNamara,A.K.,Garnero,E.J.&Rost,S.Trackingdeepmantlereservoirswithultra-900lowvelocityzones.EarthPlanet.Sci.Lett.299,1–9(2010).901902Miller,M.&Niu,F.Bulldozingthecoremantleboundary:Localizedseismic903scatterersbeneaththeCaribbeanSea.Phys.Earthplanet.Int.170,89–94(2008).904905Mulyukova,E.,Steinberger,B.,Dabrowski,M.&Sobolev,S.V.SurvivalofLLSVPsfor906BillionsofYearsinaVigorouslyConvectingMantle:ReplenishmentandDestruction907ofChemicalAnomaly.J.Geophys.3824–3847(2015).908909
42
Murakami,M.,Hirose,K.,Kawamura,K.,Sata,N.&Ohishi,Y.Post-perovskitephase910transitioninMgSiO3.Science304,855–858(2004).911912Oganov,A.R.&Ono,S.Theoreticalandexperimentalevidenceforapost-perovskite913phaseofMgSiO3inEarth’sD”layer.Nature430,445–448(2004).914915Ohtani,E.&Maeda,M.Densityofbasalticmeltathighpressureandstabilityofthe916meltatthebaseofthelowermantle.EarthPlanet.Sci.Lett.193,69–75(2001).917918Olson,P.,Yuen,D.A.&Balsiger,D.Mixingofpassiveheterogeneitiesbymantle919convection.J.geophys.Res.89,425–436(1984).920921Ritsema,J.,Deuss,A.F.,vanHeijst,H.J.&Woodhouse,J.H.S40RTS:adegree-40922shear-velocitymodelforthemantlefromnewRayleighwavedispersion,teleseismic923traveltimeandnormal-modesplittingfunctionmeasurements.Geophys.J.Int.184,9241223–1236(2011).925926Rost,S.&Earle,P.S.Identifyingregionsofstrongscatteringatthecoremantle927boundaryfromanalysisofPKKPprecursorenergy.EarthPlanet.Sci.Lett.297,616–928626(2010).929930Rost,S.,Earle,P.S.,Shearer,P.M.,Frost,D.A.&Selby,N.D.Seismicdetectionsof931small-scaleheterogeneitiesinthedeepEarth.InKhan,A.&Deschamps,F.(eds.)The932Earth’sHeterogeneousMantle,chap.12,367–390(SpringerInternational933Publishing,2015).934935Rudolph, M. L., Lekic, V. & Lithgow-Bertelloni, C. Viscosity jump in Earth’s mid-936mantle.Science350,1349–1352(2015).937938Selby,N.D.ImprovedTeleseismicSignalDetectionatSmall-ApertureArrays.Bull.939seism.Soc.Am.101,1563–1575(2011).940941Simmons,N.A.,Forte,A.M.&Grand,S.P.Jointseismic,geodynamicandmineral942physicalconstraintsonthree-dimensionalmantleheterogeneity:Implicationsfor943therelativeimportanceofthermalversuscompositionalheterogeneity.Geophysical944JournalInternational177,1284–1304(2009).945946Stackhouse,S.,Stixrude,L.&Karki,B.B.First-principlescalculationsofthelattice947thermalconductivityofthelowermantle.EarthPlanet.Sci.Lett.,427,11-17(2015).948949Steinberger, B., Torsvik, T. H. & Becker, T. W. Subduction to the lower mantle a950comparisonbetweengeodynamicandtomographicmodels.SolidEarth3,415-432951(2012).952953Tackley,P.J.MantleGeochemicalGeodynamics.InTreatiseonGeophysics,chap.9547.12,521–585(ElsevierB.V.,2015),2ndedn.edn.955
43
956Thomas,C.,Weber,M.,Wicks,C.W.&Scherbaum,F.Smallscatterersinthelower957mantleobservedatGermanbroadbandarrays.J.Geophys.Res.104,15073–15088958(1999).959960Thorne,M.,Garnero,E.J.&Grand,S.P.Geographiccorrelationbetweenhotspots961anddeepmantlelateralshear-wavevelocitygradients.Phys.EarthPlanet.Int.146,96247–63(2004).963964Waszek,L.,Thomas,C.&Deuss,A.PKPprecursors:Implicationsforglobal965scatterers.Geophys.Res.Lett.(2015).966967Wen,L.IntenseseismicscatteringneartheEarth’score-mantleboundarybeneath968theComoroshotspot.Geophys.Res.Lett.27,3627–3630(2000).969970Wicks,J.K.,Jackson,J.M.&Sturhahn,W.Verylowsoundvelocitiesiniron-rich971(Mg,Fe)O:Implicationsforthecore-mantleboundaryregion.Geophys.Res.Lett.37,9721–5(2010).973974Williams,C.D.,Li,M.,Mcnamara,A.K.,Garnero,E.J.&Soest,M.C.V.Episodic975entrainmentofdeepprimordialmantlematerialintooceanislandbasalts.Nat.976Commun.6,1–7(2015).977978Williams,Q.&Garnero,E.J.SeismicEvidenceforPartialMeltattheBaseofEarth’s979Mantle.Science273,1528–1530(1996).980981Yao,J.&Wen,L.Seismicstructureandultra-lowvelocityzonesatthebaseofthe982Earth’smantlebeneathSoutheastAsia.PhysicsoftheEarthandPlanetaryInteriors983233,103–111(2014).984985986987