E idence o a ULVZ nea Vanua u om Sdi pos cu so s1
Ca l Ma in1∗, Lobke Ha msma1, James A kins2, A wen Deuss1, Sanne Co aa 2
2
1Depa men o Geosciences, U ech Uni e si y, 3584 CB, Ne he lands3
2Bulla d Labo a o ies, Depa men o Ea h Sciences, Uni e si y o Camb idge, CB3 0EZ,4
UK5
*Co esponding au ho ([email p o ec ed])6
Key poin s7
•We p esen new e idence o an ul a-low eloci y zone (ULVZ) on he CMB o he sou heas 8
o Vanua u using Sdi pos cu so s (Sdi +) om ea hquakes in he Sou h Paci ic Rise egion9
owa ds seismic a ays in Eas Asia.10
•The da a a e explained by a quasi-cylind ical ULVZ loca ed a 172.2±0.9°E and 22.9±1.1°S11
wi h heigh 20 ±5 km, adius 240 ±50 km, and shea wa e eloci y educ ion 30 ±5%.12
•Combined wi h p e ious s udies using ScP and SPdKS in he a ea, his is a b oad-scale egion13
wi h a pa chwo k o ULVZs compa able in a ie y o Hawaii.14
•The Vanua u ULVZ lies wi hin he Paci ic la ge low eloci y p o ince, and migh po en ially15
be a oo o a man le plume ha de lec ed owa ds he eas e n Aus alian ho spo s a ound16
he Tonga slab.17
1
1 Abs ac 18
Thin anomalous s uc u es known as ul a-low eloci y zones (ULVZs) ha e been ound on he19
co e-man le bounda y (CMB) and ha e ex eme eloci y educ ions. These ea u es a e de ec ed20
due o hei e ec on seismic wa es ha a el h ough hem, ypically p oducing p ecu so s o 21
pos cu so s. In his s udy we use pos cu so s o shea co e-di ac ed wa es (Sdi +) ha sample22
he CMB nea Vanua u o de ec and cha ac e ise he p ope ies o a ULVZ. We iden i ied a o al23
o 19 ea hquakes o igina ing om he Sou h Paci ic Rise egion de ec ed by s a ions ac oss Eas 24
Asia – pa icula ly Japan – showing Sdi + signals. O hese e en s, six wi h he highes quali y25
Sdi + signals a e included in a Bayesian in e sion o a el imes using he 2D Wa e on T acke 26
we p e iously de eloped. A subse o e en s was selec ed o u he analysis by modelling using27
3D ull wa e o m syn he ics o a ange o pa ame e s. The compa ison o he eal da a wi h he28
syn he ic wa e o ms sugges s ha a ULVZ is loca ed o he sou heas o Vanua u a 172.2±0.9°E29
and 22.9±1.1°S and i s b oad-scale s uc u e can be app oxima ed as a cylinde wi h a heigh o 30
20 ±5 km, adius 240 ±50 km, and shea wa e eloci y educ ion o 30 ±5%. These pa ame e s31
a e compa able o o he ULVZs p e iously de ec ed and modelled wi h Sdi and Sdi +. The e32
a e app eciable unce ain ies in he loca ion along he NW-SE di ec ion due o he dis ibu ion o 33
ea hquakes and seismic a ays, as well as ade-o s be ween he heigh , size and eloci y educ ion34
o he ULVZ. O he s udies using SPdKS, ScP and PcP ha e epo ed de ec ions o ULVZs in he35
p oxima e egion, some o which a e consis en wi h he well- i ing pa ame e space o he ULVZ36
in his s udy. The Vanua u ULVZ lies wi hin he sou hwes edge o he Paci ic la ge low eloci y37
p o ince. The e is po en ially a man le plume oo ed by his ULVZ ha has di e ed owa ds38
he ho spo s on he eas e n Aus alian pla e a ound he Tonga slab, al hough mos omog aphic39
models do no show a con inuous plume he e.40
2
2 In oduc ion41
The lowe mos man le abo e he co e-man le bounda y (CMB) con ains s ong la e ally he e o-42
geneous seismic s uc u es no obse ed in he es o he lowe man le [e.g. Ri sema e al., 2011;43
F ench and Romanowicz, 2014]. A ound 30% o he CMB is co e ed by wo la ge egions wi h44
ela i ely low shea wa e eloci ies, called la ge low eloci y p o inces (LLVPs), which a e b oadly45
loca ed benea h he Paci ic and A ica [e.g. Co aa and Lekic, 2016; Ga ne o e al., 2016]. The46
LLVPs ha e been seen in omog aphic models o o e 40 yea s [Dziewonski and Ande son, 1981],47
wi h hei de ailed s uc u e inc easing wi h ime [e.g. Th as a son e al., 2024; Cui e al., 2024].48
Whe he hey a e he mal plume clus e s o he mochemical s uc u es ela ed o supe plumes o 49
s able piles is s ill a subjec o deba e [e.g. Ga ne o e al., 2016; Lau e al., 2017; Koelemeije e al.,50
2017; Tala e a-Soza e al., 2025], al hough a po en ial consensus image is eme ging o a hin highe 51
densi y s able base wi h bundles o he mochemical plumes abo e [Richa ds e al., 2023; Da aille52
and Romanowicz, 2020].53
Besides LLVPs, smalle and hinne s uc u es ha e been iden i ied on he CMB. These s uc-54
u es – known as ul a-low eloci y zones (ULVZs) – ha e much highe shea wa e eloci y educ-55
ions, on he o de o 10–50% [e.g. Yu and Ga ne o, 2018; Li, Leng, Jenkins and Co aa , 2022].56
ULVZs ha e been de ec ed since he mid-1990s using a ious seismic phases ha in e ac wi h he57
CMB [Williams e al., 1998] and a e epo ed o ha e a b oad a ia ion in size ( om a ew o 100 km58
in heigh and a ew o 1000 km in la e al ex en ) and eloci y educ ion [e.g. summa y by Yu and59
Ga ne o, 2018]. Na u ally, he e a e s ong ade-o s be ween heigh , size and eloci y educ ion.60
ULVZs de ec ed by bounce poin phases (ScS, ScP, PcP) o by sca e ing (PKP) gene ally lack61
cons ain s on hei la e al ex en o shape, whe eas hose de ec ed by di ac ed phases (Sdi ,62
SPdKS) sample la ge po ions o he CMB bu he e a e ambigui ies in whe e along hose pa hs63
anomalies o igina e. In cases whe e he e is c ossing da a and/o b oad co e age o an a ea, 3D64
models o hese s uc u es ha e been p oposed. Fo example, syn he ic modelling o ec angula o 65
amo phous shapes o he Samoa mega-ULVZ ha e been e ec i e in ep oducing obse ed SPdKS66
da a [Tho ne e al., 2013; Jensen e al., 2013; K ie e al., 2021]. Fo Sdi wa e o ms, cylind ical67
shapes a e ypically assumed as hey ep oduce he obse ed ou -o -plane pos cu so signals; in-68
cluding ULVZs ound nea Hawaii [Co aa and Romanowicz, 2012; Li, Leng, Jenkins and Co aa ,69
3
2022], Iceland [Yuan and Romanowicz, 2017], Galapagos [Co aa e al., 2022], Pi cai n [Li e al.,70
2024], S Helena [Da ison e al., 2024] and in he cen al Paci ic [Kim e al., 2020; Wol and Long,71
2023; Ma in e al., 2024]. Wi h good da a co e age o he Hawaii ULVZ, an elonga ed o ellip ical72
cylinde has been cons ained [Li, Sun and Bowe , 2022; Ma in e al., 2023b]. These b oades UL-73
VZs o 500–1000 km o mo e in la e al ex en a e some imes e e ed o as ‘mega-ULVZs’ [Tho ne74
e al., 2013].75
Since ULVZs a e ela i ely hin s uc u es ha lie on he CMB, s udies a e ypically ocussed76
on a a ge egion, ely on dense seismic a ays, and mainly use o wa d modelling o in e p e 77
he seismic obse a ions. The i s s udies o in e o ULVZ mo phology in es iga ed laye ed78
s uc u es in he adial di ec ion using Bayesian me hods o explain p e- and pos cu so s in ScP79
wa e o ms [Pachhai e al., 2015, 2023]. Fo Sdi pos cu so s (Sdi +), Ma in e al. [2023a] de-80
eloped a Bayesian me hod mapping he s uc u e in he 2D ho izon al plane, using a wa e on 81
acke as he o wa d model [Hause e al., 2008]. These me hods illus a e he ange o models82
ha can i he da a and educe he explo a ion equi ed h ough o wa d modelling.83
Majo ques ions emain as o wha ULVZs a e made o and how hey o m. A chemically84
dis inc composi ion ha is en iched in i on- ich e ope iclase could explain he obse ed eloci y85
educ ions [e.g. Wicks e al., 2010; Dob osa lje ic e al., 2019; Esdaille and Chen, 2024]. Geody-86
namic modelling sugges s ULVZs ha e a composi ion wi h a majo inc ease in densi y, implying87
ha ULVZs ha e a solid-s a e o igin [Bowe e al., 2011]. Chemical anomalies ha e been sugges ed88
o be a emnan o a basal magma ocean [Lab osse e al., 2007] o sedimen s om he ou e co e89
[Fu e al., 2023]. The p esence o pa ial mel has also been sugges ed o explain high dVp/dVs90
a ios [e.g. Williams e al., 1998]. Mel ing o subduc ed slab ma e ial has been p oposed o explain91
ULVZs loca ed in subduc ed slab egions [e.g. Fes in e al., 2024; Li, 2023]. Howe e , i is unlikely92
ha he mel will emain s able wi hin he ULVZ o e long imescales as i will pool on he CMB,93
esul ing in a global mel laye [e.g. He nlund and Tackley, 2007; Dannbe g e al., 2021], unless he e94
is su icien in e nal s i ing [He nlund and Jellinek, 2010]. Such a global mel laye has no ye 95
been de ec ed unambiguously due o he seismic esolu ion limi , bu a global laye o up o se e al96
kilome es may be possible [e.g. Russell e al., 2022; Hansen e al., 2023]. Longe pe iod no mal97
modes a e also consis en wi h a hin global laye [Russell e al., 2023] bu al e na i ely his could98
be due o a long wa eleng h componen in he dis ibu ion o piled ULVZ ma e ial [Koelemeije 99
4
e al., 2012].100
Besides hei na u e, hei ela ionship wi h su ounding low emains unknown. Some s udies101
ha e sugges ed a ela ionship be ween mega-ULVZs and ho spo s a he su ace and po en ial102
man le plumes connec ing hem [Tho ne e al., 2013; Yuan and Romanowicz, 2017; Kim e al.,103
2020; Co aa and Romanowicz, 2012], which is consis en wi h geophysical models [e.g. Jellinek104
and Manga, 2002]. ULVZs migh he e o e ep esen a ese oi o explain anomalous signa u es105
o 3He/4He and µ183W [Mundl-Pe e meie e al., 2020; Co aa e al., 2022]. Howe e , anomalous106
geochemical obse a ion a e no always obse ed a plumes nea ULVZs [He e e al., 2023; Da ison107
e al., 2024], no can all mega-ULVZs be ela ed o a ho spo o man le plume [Ma in e al., 2024].108
The la e is pa icula ly ue o he la ge ange o smalle ULVZs ha a e obse ed globally [e.g. Yu109
and Ga ne o, 2018; Hansen e al., 2023]. The la ges obse ed ULVZs a e also p edominan ly loca ed110
a he bounda ies o he LLVPs, which is consis en wi h geodynamical modelling ha sugges s111
ha dense ULVZs ha o m ou side he LLVPs a e swep o he bounda ies and accumula e he e112
[Liu e al., 2024].113
In his s udy, we ocus on he egion o he CMB benea h he Sou hwes Paci ic, nea Vanua u114
and Fiji. Posi i e de ec ions and null obse a ions o ULVZs ha e been epo ed he e om mul iple115
seismic p obes [Yu and Ga ne o, 2018] (summa ised in Figu e 1). Due o he dis ibu ion o 116
ea hquakes caused by he subduc ion o he Paci ic pla e unde he Aus alian pla e wi h espec 117
o he Alice Sp ings and Wa amunga seismic a ays in Aus alia, he e a e nume ous s udies118
which ha e in es iga ed he s uc u e o ULVZs in his egion using ScP wa e o ms [e.g. Ros 119
and Re enaugh, 2001; Ideha a e al., 2007; B own e al., 2015; Pachhai e al., 2015]. Using he120
same e en s, o he s udies ha e in es iga ed he possibili y o ULVZs he e om anomalous SPdKS121
wa e o ms [e.g. Tho ne and Ga ne o, 2004; Jensen e al., 2013; Tho ne e al., 2021]. The wide ange122
o epo ed pa ame e s and loca ions o a possible ULVZ indica e ha subs an ial unce ain y123
emains ega ding he ine-scaled s uc u es on he CMB in he New Caledonia/Vanua u egion.124
He e, we p esen new obse a ions o Sdi pos cu so s (Sdi +) which sample he CMB and125
p o ide e idence o a ULVZ in he egion benea h Vanua u and Fiji, om ea hquakes in he126
Paci ic Rise egion owa ds s a ions ac oss Eas Asia. We hus e e o he ULVZ causing hese127
Sdi + signals as he ‘Vanua u’ ULVZ. We in e o ULVZ loca ion and pa ame e s om Sdi +128
a el imes wi h a 2D wa e on acke o six high quali y e en s. F om he ensemble o models129
5

Figu e 1. (a) P obabili y map o he p esence o ULVZs based on he 1% mis i map by Tho ne e al. [2021].
No e ha he e is no dis inc ion in his s udy be ween low p obabili y and poo da a co e age. The dashed
black line is he Samoa ULVZ om Tho ne e al. [2021] and K ie e al. [2021]. The black dashdo ed line
gi es he p oposed ULVZ om Jensen e al. [2013]. The dashed blue line ep esen s he con ex hull o ou
well- i ing pa ame e space o he ULVZ ound in his s udy. The solid blue line gi es he p e e ed model
o he ULVZ in his s udy, wi h a adius o 240 km and loca ion 172.2°E and 22.9°S. (b) Zoom-in o he
black box in (a) wi h loca ions o bounce poin s on he CMB gi en by: ci cles [Ros and Re enaugh, 2001,
2003; Ros e al., 2005, 2006, 2010], s a s [Ideha a e al., 2007], squa es [Thomas e al., 2009], diamonds
[Pachhai e al., 2015], pen agons [B own e al., 2015], iangles [Hansen e al., 2020], and c osses [Pachhai
e al., 2022]. The lines a e ay pa hs o SPdKS wa e o ms on he CMB om Tho ne and Ga ne o [2004].
Thei colou s indica e he p esence ( ed) and absence (black) o anomalous wa e o ms; complex wa e o ms
(o ange) a e no in e p e ed. The colou ed dashdo ed lines a e p oposed ULVZs in he egion su ounding
Vanua u om mul iple p e ious s udies [Thomas e al., 1999; Tho ne e al., 2013; Pachhai e al., 2022, 2023].
6
which i he da a we compu e 3D ull wa e o m syn he ic models o se e al o he highes quali y130
e en s o es cons ain s on he hickness, la e al ex en , and shea wa e eloci y educ ion o 131
he ULVZ. Finally, we go on o compa e ou model wi h o he s epo ed in his egion, which a e132
la gely consis en .133
3 Da a and Me hods134
3.1 Sdi pos cu so s (Sdi +)135
Shea wa es di ac along and p opaga e pa allel o he CMB and can be obse ed beyond epicen al136
dis ances o app oxima ely 100°(Sdi , Figu e 2a), al hough his a ies depending on he assumed137
1D eloci y model and e en dep h. Ene gy o he Sdi wa e ge s apped and p opaga es as138
a guided wa e wi hin he ULVZ, lagging behind ene gy o he Sdi wa es p opaga ing abo e o 139
a ound he ULVZ. Due o he eloci y con as a he bounda ies o he ULVZ, he guided wa e is140
(a) (b)
Figu e 2. (a) P opaga ion o ScP, ScS, SPdKS, S, and Sdi seismic phases om an ea hquake (s a ) o
ecei e s ( iangles) h ough a c oss-sec ion o he Ea h. ScP and ScS a e co e- e lec ed phase ha sample
bounce poin s on he CMB. SPdKS and Sdi a e phases ha ha e di ac ing legs along he CMB, wi h
di ac ion ay pa hs ha a enua e wi h dis ance. (b) Wa e on p opaga ion o Sdi om an ea hquake
(s a ), calcula ed using he 2D wa e on acke [Hause e al., 2008; Ma in e al., 2023a]. The wa e on
p opaga es h ough ime (sequen ial g ey lines), p oducing pos cu so signals as i passes ac oss a ULVZ
(pink ci cle). A ecei e ( iangle) de ec s he di ec a i al (Sdi , g een line) and an addi ional delayed
a i al due o e ac ion om he ULVZ (Sdi +, blue line) a a di e en angle o he di ec backazimu h.
7
also e ac ed upon en y and exi o he ULVZ. As a esul , he p opaga ion o he wa e on is141
dis u bed and c ea es pos cu so s o he main Sdi phase (Figu e 2b), he e e e ed o as Sdi +.142
The ime delay o he Sdi + compa ed o he di ec Sdi phase is caused by bo h he p opaga ion143
wi hin he ULVZ and he longe pa h aken due o e ac ion when o -axis. Sdi + signals caused144
by a cylind ical ULVZ ha e an app oxima ely hype bolic a el ime mo e-ou , wi h he minimum145
a i al ime a he on-axis azimu h ela i e o he ULVZ midpoin om he sou ce. De ec ion o 146
Sdi + equi es a la ge and dense seismic a ay. In pa icula , mos ecen s udies o ULVZs using147
Sdi + ha e only been possible due o he dense co e age o seismic s a ions in No h Ame ica [e.g.148
Co aa and Romanowicz, 2012; Yuan and Romanowicz, 2017; Co aa e al., 2022; Wol and Long,149
2023; Ma in e al., 2024], while some Sdi + signals ha e been obse ed by ne wo ks h oughou 150
Eu ope [Da ison e al., 2024], cen al Asia [Yuan and Romanowicz, 2017], and Sou h Ame ica [Li151
e al., 2024]. He e, we use a new geome y om he Sou h Paci ic Rise o s a ions in Eas Asia.152
In his s udy, Sdi speci ically e e s o SHdi , obse ed on he ans e se componen , which153
is obse able a epicen al dis ances up o 150° o su icien ly la ge ea hquakes. SVdi , on he154
o he hand, is s ongly a enua ed in he di ac ed po ion due o leaking in o he co e, esul ing155
in a la ge decay in ampli ude as a unc ion o epicen al dis ance [Koma i sch e al., 2010] and is156
no s udied u he he e.157
3.2 Da a selec ion158
We sea ch o e idence o Sdi + o e en s in he Sou h Paci ic Rise egion using he dense seis-159
mome e dis ibu ion in Eas Asia. We inspec ed all ea hquakes loca ed nea he Sou h Paci ic160
Rise egion a any dep h om 1995 o 2024 and wi h a momen magni ude la ge han 5.5. Da a161
a e ob ained o a dis ance ange o 90–150°, al hough mos da a in his geome y is obse ed162
Da e Lon (°) La (°) Dep h (km) Mag ∆ (s) Region
12003/08/28 -115.20 -49.92 15 6.2 -5.0 Sou he n Eas Paci ic Rise
22005/05/12 -138.91 -57.57 12 6.5 -3.0 Paci ic-An a c ic Ridge
32001/09/02 -136.75 -54.31 15 6.3 0.0 Paci ic-An a c ic Ridge
4a 2007/03/31 -123.61 -55.94 12 6.2 -9.5 Sou he n Eas Paci ic Rise
4b 2016/08/18 -123.63 -55.97 13.44 6.0 -9.5 Sou he n Eas Paci ic Rise
52015/05/19 -132.39 -54.53 14.91 6.6 -1.5 Paci ic-An a c ic Ridge
Table 1. Lis o selec ed e en s wi h high quali y Sdi + signals. Sou ce pa ame e s a e om he Global
CMT P ojec [Eks ¨om e al., 2012]. The ull lis o e en s can be ound in Table S1.
8
Figu e 3. Map o e en loca ions (s a s), ecei e s ( iangles) and he Sdi ay pa hs sensi i e o he CMB
(below 2800 km dep h). The ea hquakes and ay pa hs a e colou ed pe e en : 1–g een, 2–pu ple, 3– ed,
4a–blue, 4b–black and 5–o ange (Table 1); ligh g ey s a s a e e en s iden i ied bu no analysed (Table S1).
Ho spo loca ions a e shown as o ange ci cles. The dashed blue line ep esen s he con ex hull o ensemble o
models o ULVZs which i he Sdi + a i al ime da a in his s udy. The solid blue line gi es he p e e ed
model o he ULVZ in his s udy, wi h a adius o 240 km and loca ion 172.2°E and 22.9°S. The models o
he Samoa ULVZ a e gi en as black dashed lines: A–i egula shaped [Tho ne e al., 2021] and B– ec angula
shaped [K ie e al., 2021]. The black dashdo ed line shows he in e ed C–diamond shaped ULVZ om
Jensen e al. [2013]. The backg ound displays he LLVP o e map a 2800 km dep h [Co aa and Lekic,
2016]. The solid ed line and g eyscale ma ke s deno e he c oss-sec ion ansec in Figu e 8.
a 115–130°. The seismog ams we e downloaded om ne wo ks a ailable h ough he Ea hScope163
Conso ium Da a Managemen Cen e and he Na ional Resea ch Ins i u e o Ea h Science and164
Disas e P e en ion (he ea e NIED) F-ne , a b oadband seismog aph ne wo k consis ing o mo e165
han 70 s a ions ac oss Japan [Okada e al., 2004]. A e downloading, he da a we e esampled o166
10 Hz, componen s checked o missing da a, and o a ed in o he ZRT ( e ical, adial, angen ial)167
o ien a ion since SHdi is p ima ily isible on he angen ial componen .168
A o al o 78 ea hquakes loca ed in he Sou h Paci ic Rise egion we e analysed. Mos o he169
ea hquakes had a low signal- o-noise a io (SNR) due o hei low momen magni ude. The e we e170
9
Howe e , he e is signi ican a ia ion in he ela i e ampli udes be ween Sdi and Sdi +, wi h301
small bu s ong anomalies esul ing in small Sdi + ampli udes, and la ge bu weak anomalies302
p oducing la ge Sdi + ampli udes.303
We also compu e syn he ics o models away om he ade-o cu e o demons a e he unsui -304
abili y o hese models and illus a e unce ain ies. A numbe o es ed models a e la ge o mo e305
s ongly educed (Models B1, B3 & C1; ‘abo e’ he cu e in Figu e S7), esul ing in p onounced306
Sdi + ha a e mo e delayed and ha e longe pe iods han obse ed in he eal da a (Figu es S13307
& S14). Fo models ha a e smalle o mo e weakly educed (Models B2, B4 & C2; ‘below’ he308
cu e in Figu e S7), he Sdi + a e compa able in ampli ude when il e ed be ween 10–20 s bu a e309
oo weak o be obse ed in 20–30 s. As expec ed, he a el ime delays o he Sdi + a e educed.310
O e all, he e a e app eciable ade-o s o he size and eloci y educ ion wi h he hickness o he311
ULVZ, which is e lec ed in he unce ain ies in ou inal p e e ed model.312
Figu e 6. Displacemen syn he ics o PREM wi h a ULVZ wi h heigh 20 km, adius 240 km, and shea
eloci y educ ion o 30% (Model A3, Table S2) il e ed be ween pe iods o 10–20 s o he six e en s lis ed
in Table 1. The same Sdi + highligh s om Figu e 4 a e o e laid, and he colou s co espond o he e en
loca ion s a s in Figu e 3.
16

4.3 P e e ed model313
The consis en p esence o Sdi + a i als in he obse ed da a is s ong e idence o he exis ence314
o a ULVZ benea h Vanua u on he CMB. In e sion o he Sdi + a i al imes wi h he 2DWT315
sugges s a ULVZ loca ed a 172.9±0.9°E and 22.9±1.1°S. Fu he modelling wi h 3D ull wa e o m316
syn he ics shows he da a a e consis en wi h a cylind ical ULVZ wi h adius 240 ±50 km, heigh 317
20 ±5 km, and shea wa e eloci y educ ion 30 ±5%. Syn he ics o all o he e en s used in his318
s udy a e shown in Figu e 6.319
Whils he ULVZ was modelled using a simpli ied cylind ical shape, which e ec i ely ep oduces320
he hype bolic mo e-ou o Sdi +, he limi ed azimu hal co e age o he egion o in e es makes i 321
di icul o jus i y o wha ex en his app oxima ion is easonable. As Sdi a els long dis ances322
along he CMB, anomalous s uc u es anywhe e along hose ay pa hs may cause in e e ence and323
p oduce addi ional wa e o ms. The a ailable da a se samples he ULVZ along a single azimu h324
(NW–SE), meaning he loca ion is poo ly cons ained along he ay pa hs in he NW–SE di ec ion325
bu well cons ained in he NE–SW di ec ion (Figu e 5a).326
5 Discussion327
5.1 Compa ison o nea by ea lie s udies328
The Sou hwes Paci ic has been in ensi ely in es iga ed, wi h p e ious s udies inding e idence o 329
he p esence and absence o ULVZs using ScP and SPdKS (summa ised in Figu e 1). A p obabili y330
map de i ed om anomalous SPdKS wa e o ms sugges s he likelihood o ULVZs o he wes ,331
no h, and pa icula ly o he eas o ou p e e ed model loca ion [Tho ne e al., 2021], whe e332
locally a ge ed s udies ha e mapped he Samoa ULVZ [Tho ne e al., 2013; K ie e al., 2021]. To333
he wes , he o ui ous sou ce- ecei e dis ibu ion has led o a ple ho a o s udies using ScP o334
p obe benea h he Co al Sea [e.g. Ideha a e al., 2007; Ros e al., 2010; Pachhai e al., 2022], as335
well as a localised s udy using SPdKS [Jensen e al., 2013]. Since he analyses o hese da a assume336
an in-plane o igin o p e- o pos cu so s in he wa e o ms, i is possible ha ou -o -plane ene gy337
om he ULVZ iden i ied in his s udy could cause hese signals [Pachhai e al., 2024] and we no e338
ha he Vanua u ULVZ is loca ed in a gap o in-plane da a co e age o bo h phases.339
17
We es i ou da a would be able o obse e he o he ULVZs p e iously p oposed in he340
egion by compu ing ull wa e o m syn he ics o E en s 1 and 4a & 4b. Fo he wo published341
Samoa ULVZ models – an i egula shaped ULVZ [Tho ne e al., 2021] and a ec angula shaped342
ULVZ [K ie e al., 2021] – we use a hickness o 26 km and a shea wa e eloci y educ ion o 343
20% as implemen ed by K ie e al. [2021]. To he wes , we es a ULVZ in he egion o highes 344
p obabili y (pu ple con ou in Figu e 1), which is compa able in size o ou p e e ed model. Fo 345
his we use pa ame e s om ou p e e ed model wi h adius 240 km, heigh 20 km and shea 346
eloci y educ ion o 30% eloca ed o he midpoin o he local high p obabili y egion, 166°E and347
24.5°S. The majo i y o p e ious ScP s udies in his egion he e o e all wi hin he bounda ies o 348
ou ial model, as well as wi hin he uppe limi o unce ain y in adius o ou p e e ed model349
(Figu e S15). All ULVZ models es ed p oduced Sdi + signals (Figu es S16 & 7 o E en s 1 and 4a350
& 4b, espec i ely). Fo he models o ULVZs o he eas , Sdi + wa e o ms a e o se signi ican ly351
o he la ge azimu hs, which means ha he Sdi + in he obse ed da a canno be caused by he352
Samoa ULVZ (Figu e 7d-e). Howe e , his does no exclude he possibili y ha hese con ibu e o353
he wa e o m complexi ies obse ed, no would lack o signals necessa ily con adic he p esence354
o he Samoa ULVZ (Figu e S16g), as we ha e p e iously no ed ha he isibili y o ULVZs migh 355
a y by di ec ion [Ma in e al., 2023b]. Fo he model o he wes , he wa e o ms look – pe haps356
unsu p isingly – ai ly simila , as we used he p ope ies o ou p e e ed model, bu hey a e o se 357
o smalle azimu hs. While he o se in azimu h is only sligh o E en 1 (Figu e S16 ), i is qui e358
app eciable o E en 4 (Figu e 7 ). This demons a es ha he loca ion o ou p e e ed model is359
well cons ained by he combina ion o all six e en s, each wi h sligh ly di e en co e age, a leas 360
in he SW-NE di ec ion.361
While a mega-ULVZ model u he o he wes migh no be expec ed in ou da a, a ange o 362
ScP s udies in his egion ha e obse a ions ha all wi hin he con ex hull o ou model space363
(Figu e 1). These obse a ions om bounce poin s udies, howe e , sugges hinne and pa chie 364
ULVZ co e age han we in e he e [e.g. Ideha a e al., 2007; Ros e al., 2010; Pachhai e al., 2022],365
which could be he na owe edge o a la ge , hicke s uc u e [Jensen e al., 2013]. Gi en he low366
sensi i i y o Sdi and Sdi + wa e o ms o small-scale s uc u es a ela i ely long pe iod, his367
s udy canno esol e he complex mo phology o his egional s uc u e, bu suppo s he p esence368
o a la ge, hick quasi-cylind ical s uc u e (∼500 km ac oss, ∼20 km deep). The possibili y o in-369
18
Figu e 7. Wa e o m da a and syn he ics o he Vanua u ULVZ in his s udy and o o he models o
p oxima e ULVZs. (a) Obse ed and (b- ) syn he ic displacemen wa e o ms o E en s 4a & 4b. Syn he ics
o PREM wi h (b) no ULVZ, (c) he p e e ed ULVZ in his s udy, (d) he i egula ly shaped Samoa ULVZ
[Tho ne e al., 2021], (e) he ec angula -shaped Samoa ULVZ [K ie e al., 2021], and ( ) a ULVZ wi h
pa ame e s o ou p e e ed model loca ed a he p oxima e p obabili y peak o Tho ne e al. [2021] (Figu e
S15). The Sdi + signals a e highligh ed whe e isible. Da a a e il e ed be ween 10–20 s pe iod.
plane and ou -o -plane pos cu so s caused by a mega-ULVZ should be conside ed in u u e s udies370
a emp ing o econcile obse a ions o ScP wi h obse a ions o Sdi and SPdKS. O e all, his371
egion appea s o ha e many complexi ies obse ed by di e en seismic phases, simila o he372
pa chie a eas ound a ound he mega-ULVZ nea Hawaii [Jenkins e al., 2021].373
Fu u e s udies using Sdi + and SPdKS could also examine he possible in e nal laye ing o 374
ULVZs by using sho e pe iod pos cu so s. Iden i ying Sdi + a highe equencies is challenging375
as hey a e o en obscu ed by noise, al hough Li, Leng, Jenkins and Co aa [2022] ha e been able376
o iden i y he in e nal laye ing o he Hawaii ULVZ using his me hod.377
5.2 Compa ison o o he mega-ULVZs and ela ion o LLVPs378
The Vanua u ULVZ has compa able pa ame e s o – bu is somewha smalle in la e al ex en han –379
o he modelled mega-ULVZs which ha e been obse ed using Sdi + [e.g. Co aa and Romanowicz,380
19
Figu e 8. C oss-sec ions o ecen whole man le omog aphic models (a) SEMUCB-WM1 [F ench and
Romanowicz, 2014], (b) GLAD-M35 [Cui e al., 2024], and (c) REVEAL [Th as a son e al., 2024] h ough
he ansec on Figu e 3, which passes h ough he Vanua u and Samoa ULVZs and nea he Lo d Howe
and Samoa ho spo s. Veloci y de ia ions a e wi h espec o he (whole Ea h) adial a e age o each omo-
g aphic model. Nea by ho spo s a e p ojec ed on o he c oss-sec ion: EA = Eas Aus alia, T = Tasman id,
LH = Lo d Howe, and S = Samoa. ULVZ hickness (solid black lines) is exagge a ed o isualisa ion.
2012; Da ison e al., 2024]. Since he e en s sampling his ULVZ a e ypically shallow and low381
magni ude, i is di icul o di e en ia e be ween weak Sdi + signal and noise a sho e pe iods.382
The e is dis inc Sdi + ene gy be ween 10–20 s pe iod and i is isible up o 30 s, sugges ing a383
compa able equency con en and hickness o ha o he Hawaii ULVZ [20 km in heigh , Ma in384
e al., 2023b] and hicke han ha o Iceland [15 km, Yuan and Romanowicz, 2017] o he mid-385
Paci ic ULVZ [10 km, Ma in e al., 2024].386
A he sou hwes edge o he Paci ic LLVP, he Vanua u ULVZ is he eigh h mega-ULVZ mod-387
elled in 3D. All o hese – so a – a e loca ed a o nea he LLVP bounda ies, ein o cing a po en ial388
co ela ion be ween he la ges ULVZs and he bounda ies o he LLVPs. P e ious mega-ULVZs389
ha e also been in e p e ed as whole man le plume oo s due o hei geochemical signa u es [Co -390
aa e al., 2022]. The Vanua u ULVZ is almos equidis an o he Samoa and Lo d Howe ho spo s,391
wi h oughly ∼15–20° o ei he . Howe e , he Samoa ULVZ lies close o ha ho spo and has392
p e iously been sugges ed as i s po en ial oo [e.g. Tho ne e al., 2013; K ie e al., 2021].393
The Lo d Howe seamoun chain lies on he eas e n Aus alian pla e. This seamoun chain,394
combined wi h he Tasman id and Eas e n Aus alia ho spo , acks u he wes and has no395
obse able opog aphic swell, implying ha he hea lux in he a ea is e y mino [Hogga d e al.,396
2020]. Howe e , he Lo d Howe seamoun chain has simila geochemical signa u es o o he Paci ic397
ocean island basal s, wi h S -Nd iso ope analysis sugges ing a deep man le sou ce [Roge s e al.,398
2023]. These signals a e also seen on he Tasman id seamoun chain sugges ing a simila sou ce, bu 399
3He/4He measu emen s o con i m a p imo dial ese oi a e lacking in he egion [Roge s e al.,400
2023].401
20
We show c oss-sec ions o h ee omog aphic models [F ench and Romanowicz, 2014; Cui e al.,402
2024; Th as a son e al., 2024] be ween he Lo d Howe and Samoa ho spo s, ac oss he loca ion403
o he Vanua u ULVZ (Figu e 8). The e a e di e ences be ween he c oss-sec ions, bu all h ee404
models show he Tonga slab, he Paci ic LLVP, and a b oad-scale po en ial plume benea h Samoa,405
unde lain by he Samoa ULVZ. Ano he , olde slab has been sugges ed in his egion a a dep h406
o ∼1100 km om a ossil subduc ion zone [Schella e al., 2009; Van de Mee e al., 2018]. The407
Sou h Loyal y Basin slab is di icul o dis inguish om he Tonga slab in he omog aphic models408
we show due o esolu ion. Towa ds Lo d Howe, one model in pa icula , REVEAL, sugges s a409
con inuous low eloci y anomaly om he egion o he Vanua u ULVZ o he su ace. While he410
o he wo models do no show a con inuous plume he e, all models show some sugges ion o ho 411
upwelling ma e ial de lec ed a ound he Sou h Loyal y Basin slab and down-dipping Tonga slab412
owa ds he sou hwes . Po en ially, spli ing o his plume in o hinne plumele s owa ds he uppe 413
man le makes i ha d o image hese. Al e na i ely, he plume may ha e been ecen ly cu o by414
he slab and could be e ea ing sou hwa ds, which is consis en wi h an age-p og essi e educ ion415
in magma ic lux obse ed o he Lo d Howe seamoun chain [Se on e al., 2019] and he lack o 416
p esen -day opog aphic swell [Hogga d e al., 2020].417
The image o wo nea by b oad-scale plumes, oo ed by ULVZs, aligns wi h he ‘bundle o 418
plumes’ iew p oposed as an explana ion o LLVPs by Da aille and Romanowicz [2020]. Since419
he connec ion o he ULVZ o he man le plume and a po en ial ho spo is enuous, we s ick o420
naming i ‘Vanua u ULVZ’ and no he ‘Lo d Howe ULVZ’. The Vanua u and Samoa ULVZs hus421
ep esen wo la ge scale s uc u es a a dis ance o app oxima ely 20°(∼1200 km a he CMB).422
Fu u e geodynamical models could es i hese ea u es can emain s able, o would be expec ed423
o e en ually me ge.424
21

6 Conclusions425
Sdi + p oduced by 19 ea hquakes loca ed in he Sou h Paci ic Rise egion and de ec ed by s a ions426
ac oss Eas Asia p o ide e idence o he Vanua u ULVZ. A combina ion o in e se modelling o 427
a el imes o Sdi + and o wa d modelling o Sdi + wa e o ms o six o he highes quali y428
e en s places cons ain s on he ULVZ pa ame e s. Since all o he ea hquakes a e co-loca ed, he429
posi ion o he ULVZ is well cons ained in he SW-NE o ien a ion bu poo ly cons ained along430
he NW-SE di ec ion. The p e e ed model o he ULVZ is a cylinde wi h a heigh o 20 ±5 km, a431
adius o 240 ±50 km, and a shea wa e eloci y educ ion o 30 ±5% loca ed o he sou heas o 432
Vanua u a 172.2±0.9°E and 22.9±1.1°S. The e a e s ong ade-o s be ween he heigh , eloci y433
educ ion and size o he ULVZ.434
Al hough he egion su ounding he p oposed loca ion o he Vanua u ULVZ has been in en-435
si ely s udied using ScP and SPdKS wa e o ms, ou s udy o Sdi + add ess a gap o da a co e age.436
Full wa e o m syn he ics o nea by p e iously p oposed models do no explain ou obse a ions.437
Join modelling o SPdKS and Sdi o his egion may p o ide c ucial addi ional cons ain s on438
he dVp/dVs a io, and po en ially densi y, o he ULVZ and he e o e di e en ia e be ween o igin439
hypo heses.440
Syn hesising ou obse a ions wi h p e ious s udies sugges s he CMB he e is a complex land-441
scape comp ising a ‘mega-ULVZ’ and smalle -scale ULVZs, which is wi hin he sou hwes edge o 442
he Paci ic LLVP and ela i ely close o he Samoa mega-ULVZ. Recen omog aphic models hin a 443
he possibili y ha he Vanua u ULVZ lies a he oo o a plume ha is de lec ed o he sou hwes 444
a ound he Tonga slab. Howe e , omog aphic models do no ag ee whe he his plume eeds in o445
he Lo d Howe and neighbou ing ho spo s o i i is now a waning plume.446
22
Acknowledgemen s447
CM and AD we e unded by a Vici awa d (g an numbe 016.160.310/526) om he Du ch Re-448
sea ch Council (NWO). CM and SC ecei ed unding om he Eu opean Resea ch Council (ERC)449
unde he Eu opean Union’s Ho izon 2020 esea ch and inno a ion p og amme (g an ag eemen 450
No. 804071 -ZoomDeep). SC and JA ecei ed unding om he Na u al En i onmen Resea ch451
Council (NE/V018213/1 and NE/S007164/1, espec i ely). Some o he wo k in his s udy was452
pe o med using esou ces p o ided by he Camb idge Se ice o Da a D i en Disco e y (CSD3)453
ope a ed by he Uni e si y o Camb idge Resea ch Compu ing Se ice (www.csd3.cam.ac.uk). We454
also acknowledge he clus e acili ies, Eeji in U ech and Gauss in Camb idge, and hank he455
suppo s a who main ain hem. We would like o hank S ua Russell, Flo ian Mille , Douwe an456
Hinsbe gen, and he U ech seismology g oup o help ul discussions; and Neil Ma jo am, Lukas457
an de Wiel and Theo an Zessen o echnical suppo . We hank he edi o , Ve non Co mie ,458
and wo anonymous e iewe s o hei hough ul commen s and sugges ions which imp o ed he459
manusc ip . Finally, we would like o hank NIED – and e e yone in ol ed wi h he deploymen 460
and main enance o he seismic ne wo k in as uc u e – o making such a ich da a se eely461
a ailable.462
Da a and code a ailabili y463
The acili ies o IRIS Da a Se ices (www.i is.edu), and speci ically he IRIS Da a Managemen 464
Cen e , we e used o access o wa e o ms and ela ed me ada a. IRIS Da a Se ices a e unded465
h ough he Seismological Facili ies o he Ad ancemen o Geoscience (SAGE) Awa d o he Na-466
ional Science Founda ion unde Coope a i e Suppo Ag eemen EAR-1851048. Da a om he F-467
ne and Hi-ne ne wo ks we e made eely a ailable by Na ional Resea ch Ins i u e o Ea h Science468
and Disas e Resilience [Okada e al., 2004]. Da a om NIED was downloaded using Hine Py [Tian,469
2024]. Ea hquake pa ame e s we e used om he Global CMT P ojec (www.globalcm .o g).470
Codes used will be made a ailable by CM upon eques .471
23
Compe ing in e es s472
Au ho s decla e ha hey ha e no compe ing in e es s.473
24
Re e ences474
Bodin, T. and Samb idge, M. [2009], ‘Seismic omog aphy wi h he e e sible jump algo i hm’,475
Geophysical Jou nal In e na ional 178(3), 1411–1436.476
Bowe , D. J., Wicks, J. K., Gu nis, M. and Jackson, J. M. [2011], ‘A geodynamic and mine al physics477
model o a solid-s a e ul alow- eloci y zone’, Ea h and Plane a y Science Le e s 303(3-4), 193–478
202.479
B own, S. P., Tho ne, M. S., Miyagi, L. and Ros , S. [2015], ‘A composi ional o igin o ul alow-480
eloci y zones’, Geophysical Resea ch Le e s 42(4), 1039–1045.481
Co aa , S. and Lekic, V. [2016], ‘Mo phology o seismically slow lowe -man le s uc u es’, Geophys-482
ical Supplemen s o he Mon hly No ices o he Royal As onomical Socie y 207(2), 1122–1136.483
Co aa , S., Ma in, C., Li, Z. and Pa ai, R. [2022], ‘The oo o he Gal´apagos man le plume on484
he co e-man le bounda y’, Seismica 1(1).485
Co aa , S. and Romanowicz, B. [2012], ‘An unusally la ge ULVZ a he base o he man le nea 486
Hawaii’, Ea h and Plane a y Science Le e s 355, 213–222.487
C o well, H. P., Owens, T. J., Ri sema, J. e al. [1999], ‘The TauP Toolki : Flexible seismic488
a el- ime and ay-pa h u ili ies’, Seismological Resea ch Le e s 70, 154–160.489
Cui, C., Lei, W., Liu, Q., Pe e , D., Bozda˘g, E., T omp, J., Hill, J., Podho szki, N. and Pugmi e, D.490
[2024], ‘GLAD-M35: a join P and S global omog aphic model wi h unce ain y quan i ica ion’,491
Geophysical Jou nal In e na ional 239(1), 478–502.492
Dannbe g, J., Myhill, R., Gassm¨olle , R. and Co aa , S. [2021], ‘The mo phology, e olu ion and493
seismic isibili y o pa ial mel a he co e–man le bounda y: implica ions o ULVZs’, Geophys-494
ical Jou nal In e na ional 227(2), 1028–1059.495
Da aille, A. and Romanowicz, B. [2020], ‘De la ing he LLSVPs: Bundles o man le he mochemical496
plumes a he han hick s agnan “piles”’, Tec onics 39(10), e2020TC006265. e2020TC006265497
10.1029/2020TC006265.498
URL: h ps://agupubs.onlinelib a y.wiley.com/doi/abs/10.1029/2020TC006265499
25
Th as a son, S., an He waa den, D.-P., Noe, S., Jose Schille , C. and Fich ne , A. [2024], ‘RE-654
VEAL: A global ull-wa e o m in e sion model’, Bulle in o he Seismological Socie y o Ame ica655
114(3), 1392–1406.656
Tian, D. [2024], ‘Hine Py: A Py hon package o accessing and p ocessing NIED Hi-ne seismic657
da a’, Jou nal o Open Sou ce So wa e 9(98), 6840.658
Van de Mee , D. G., Van Hinsbe gen, D. J. and Spakman, W. [2018], ‘A las o he unde wo ld:659
Slab emnan s in he man le, hei sinking his o y, and a new ou look on lowe man le iscosi y’,660
Tec onophysics 723, 309–448.661
an D iel, M., K ische , L., S ¨ahle , S. C., Hosseini, K. and Nissen-Meye , T. [2015], ‘Ins aseis:662
ins an global seismog ams based on a b oadband wa e o m da abase’, Solid Ea h 6(2), 701–663
717.664
Wicks, J., Jackson, J. and S u hahn, W. [2010], ‘Ve y low sound eloci ies in i on- ich (Mg, Fe) O:665
Implica ions o he co e-man le bounda y egion’, Geophysical Resea ch Le e s 37(15).666
Williams, Q., Re enaugh, J. and Ga ne o, E. [1998], ‘A co ela ion be ween ul a-low basal eloci-667
ies in he man le and ho spo s’, Science 281(5376), 546–549.668
Wol , J. and Long, M. D. [2023], ‘Lowe mos man le s uc u e benea h he cen al Paci ic669
Ocean: Ul alow eloci y zones and seismic aniso opy’, Geochemis y, Geophysics, Geosys ems670
24(6), e2022GC010853.671
Yu, S. and Ga ne o, E. J. [2018], ‘Ul alow eloci y zone loca ions: a global assessmen ’, Geochem-672
is y, Geophysics, Geosys ems 19(2), 396–414.673
Yuan, K. and Romanowicz, B. [2017], ‘Seismic e idence o pa ial mel ing a he oo o majo 674
ho spo plumes’, Science 357(6349), 393–397.675
32

E idence o a ULVZ nea Vanua u om Sdi pos cu so s1
Ca l Ma in1∗, Lobke Ha msma1, James A kins2, A wen Deuss1, Sanne Co aa 2
2
1Depa men o Geosciences, U ech Uni e si y, 3584 CB, Ne he lands3
2Bulla d Labo a o ies, Depa men o Ea h Sciences, Uni e si y o Camb idge, CB3 0EZ,4
UK5
*Co esponding au ho ([email p o ec ed])6
SUPPLEMENTARY7
In his supplemen a y a e:8
•Sec ion S1: Ea hquake da a used in his s udy9
•Sec ion S2: Wa e o ms o syn he ics o di e en pa ame e s o demons a e ade-o s10
•Sec ion S3: Wa e o m modelling o ULVZs in ea lie s udies o compa ison11
1
S1 Ea hquake da a12
S1.1 Ea hquake lis 13
In o al we iden i ied 19 ea hquakes in he Sou h Paci ic Rise egion owa ds seismic a ays Eas 14
Asia – pa icula ly he Japanese F-ne [Okada e al., 2004] – which showed Sdi pos cu so e idence15
o a ULVZ (Table S1).16
Da e Lon. [°E] La . [°N] Dep h [km] Mag. Loca ion
2001/08/06 -123.05 -55.67 15 Mw 6.7 Sou he n Eas Paci ic Rise
32001/09/02 -136.75 -54.31 15 Mw 6.3 Paci ic-An a c ic Ridge
12003/08/28 -115.20 -49.92 15 Mw 6.2 Sou he n Eas Paci ic Rise
2004/01/29 -114.85 -50.03 15 Mw 6.1 Sou he n Eas Paci ic Rise
22005/05/12 -138.91 -57.57 12 Mw 6.5 Paci ic-An a c ic Ridge
2006/10/10 -122.41 -56.18 12 Mw 6.0 Sou he n Eas Paci ic Rise
4a 2007/03/31 -123.61 -55.94 12 Mw 6.2 Sou he n Eas Paci ic Rise
2007/11/02 -128.80 -55.40 13.55 Mw 6.2 Paci ic-An a c ic Ridge
2011/11/02 -129.08 -55.34 15 Mw 6.2 Paci ic-An a c ic Ridge
2012/07/18 -128.96 -55.36 20.34 Mw 5.9 Paci ic-An a c ic Ridge
2014/05/12 -115.06 -49.90 13.24 Mw 6.4 Sou he n Eas Paci ic Rise
52015/05/19 -132.39 -54.53 14.91 Mw 6.6 Paci ic-An a c ic Ridge
2015/12/24 -123.11 -56.06 12.13 Mww 6.2 Sou he n Eas Paci ic Rise
2016/02/16 -124.57 -55.74 20.13 Mww 6.1 Sou he n Eas Paci ic Rise
4b 2016/08/18 -123.63 -55.97 13.44 Mw 6.0 Sou he n Eas Paci ic Rise
2017/06/15 -124.44 -55.77 20.03 Mww 5.8 Sou he n Eas Paci ic Rise
2018/11/15 -122.26 -56.23 12 Mww 6.3 Sou he n Eas Paci ic Rise
2024/05/18 -123.14 -56.05 12 Mww 5.9 Sou he n Eas Paci ic Rise
2024/06/09 -133.88 -54.06 12.62 Mww 6.2 Paci ic-An a c ic Ridge
Table S1. Ea hquake da a and pa ame e s o e en s showing Sdi pos cu so s ha sample he CMB
nea Vanua u. This s udy uses he e en s labelled 1–5 (Table 1 in main pape ). E en loca ions a e plo ed
in Figu e 3. Ea hquake pa ame e s and momen enso solu ions a e aken om he Global CMT P ojec
[Eks ¨om e al., 2012].
2
S1.2 E en s used in he s udy17
We ep oduce Figu e 4, showing he wa e o ms o six ea hquakes, E en s 1–5, il e ed be ween18
7–12, 20–30 and 30–40 s pe iod (Figu es S1–S3, espec i ely). We also plo displacemen wa e o ms19
om he Hi-ne sho pe iod seismome e s il e ed be ween 10–20 and 7–12 s pe iod (Figu es S4 &20
S5, espec i ely) [Okada e al., 2004].21
Figu e S1. Same as Figu e 4 bu il e ed be ween 7–12 s pe iod.
3
Figu e S2. Same as Figu e 4 bu il e ed be ween 20–30 s pe iod.
Figu e S3. Same as Figu e 4 bu il e ed be ween 30–40 s pe iod.
4
Figu e S4. Displacemen wa e o ms om he sho pe iod Hi-ne seismome e s o E en s 1–5 il e ed
be ween 10–20 s pe iod. The linea s ack o each 1°azimu hal bin ( hick black lines) o e lay he indi idual
seismog ams ( hin g ey lines). No e he di e en (y axis) azimu hal ange compa ed o b oadband da a.
Figu e S5. Same as Figu e S4 bu il e ed be ween 7–12 s pe iod.
5

S2 T ade-o s in wa e o m modelling22
Figu e S6 shows he syn he ics p edic ed o he sou ce solu ions om he Global CMT P ojec 23
wi h he dep h se o 0 km which bes ma ches he obse ed Sdi wa e o ms. We no e ha he24
da a and syn he ics o E en 2 ha e been lipped o ma ch he wa e o ms o he o he e en s o 25
ease o isual compa ison.26
Figu e S6. Displacemen syn he ics o PREM il e ed be ween pe iods o 10–20 s o he six e en s lis ed
in Table 1. Ea hquake pa ame e s and momen enso solu ions a e aken om he Global CMT P ojec
[Eks ¨om e al., 2012], wi h dep hs se o 0 km.
6
We explo e he ade-o in model pa ame e s now using ull wa e o m syn he ics. The loca ion and27
unce ain ies a e in e ed om he 2DWT in e sion (Figu e 5a), which a e assumed o be ixed o 28
he emainde o his unce ain y assessmen . We ial a numbe o models along he eloci y-size29
ade-o cu e o he 2DWT ensemble o demons a e ha his ade-o is ep oduced by ull30
wa e o m syn he ics as well as some o -cu e pa ame e s (Figu e S7 & Table S2). Addi ionally, we31
cons ain he heigh by compa ing he p esence o absence o pos cu so s in syn he ics wi h ULVZs32
o heigh s 15, 20, 25 and 30 km o di e en il e bands. We also show p edic ed a el imes om33
he 2DWT o he a ious models lis ed in Table S2 (Figu e S8).34
Figu e S7. Duplica e o Figu e 5b, wi h pa ame e
labels o e laid. Ensemble o models ha i he Sd-
i pos cu so a i al imes, showing he ade-o in
eloci y and size. The median model is ma ked wi h
a ed c oss and his og ams o each o he pa ame e s
a e p ojec ed on o he espec i e axes.
Model dVs (%) R (km) H (km)
A1 −40 165 20
A2 −35 200 20
A3 −30 240 20+15,25,30
A4 −25 305 20
A5 −20 400 20
B1 −35 240 20
B2 −30 200 20
B3 −30 305 20
B4 −25 240 20
C1 −35 305 20
C2 −25 200 20
Table S2. Pa ame e s used o es 3D cylind ical
models o he ULVZ. All models a e cen ed a he
median loca ion, 172.16 °E and 22.92 °S. Syn he ics
a e also un o Model A3 wi h heigh s 15, 25, and
30 km o demons a e sensi i i y o he equency
con en o he wa e o ms o model heigh .
7
Figu e S8. P edic ed a el imes om he 2DWT o he a ious models o which we compu e ull
wa e o m syn he ics (Figu e S7 & Table S2). (a) is he same as Figu e 5c. No e ha he 2DWT p edic ions
do no accoun o heigh o he model.
8
Figu e S9. (a) Real da a and syn he ic wa e o ms o (b) PREM wi h no ULVZ and PREM wi h a
cylind ical ULVZ o heigh (c) 15 km, (d) 20 km, (e) 25 km, and ( ) 30 km o E en s 4a & 4b, il e ed
be ween 10–20 s pe iod. All models use dVs o -30% and a adius o 240 km (Model A3, Table S2).
Figu e S10. Same as Figu e S9 bu il e ed be ween 20–30 s pe iod.
9