Seismic Observation of a New ULVZ Beneath the Southern Pacific

We present new observations of core‐diffracted shear waves which contain anomalous waveforms sampling the lowermost mantle beneath the southern Pacific region. Data in two distinct geometries, one from New Zealand to North America and the other from the Fiji and Solomon Islands to South America, show evidence of postcursor phases. The postcursor delays and move‐outs imply that they are caused by an ultra‐low velocity zone (ULVZ). Beamforming analyses of the observed diffracted postcursors show a strong backazimuthal deviation, suggesting this new ULVZ is likely to have a cylindrical shape similar to broad ULVZs sampled by shear diffracted waves elsewhere. Full‐waveform modeling suggests that the postcursors seen in North America might be due to the previously modeled ULVZ located to the west of the Galápagos, while those seen in South America are due to a previously unknown ULVZ beneath the Southern Pacific. We cannot fit observations in both geometries by a single ULVZ. For the new location, we propose one cylindrical ULVZ model with a radius of 400 km and a shear wave velocity decrease of 20% centered at geographical coordinates (−33.6, 130) close to the Pitcairn hotspot. Despite some uncertainty in the west‐east direction, this new ULVZ observation likely provides another example to support the hypothesis that ULVZs exist at the base of mantle plumes where primordial signatures are observed in the ocean island basalts.


Introduction
Ultra-low velocity zones (ULVZs) are anomalous structures on the core-mantle boundary-the discontinuous boundary between the rocky mantle and the fluid outer core-which have been increasingly discovered over recent decades (see Garnero et al. (1998) for an early summary of work in the 1990s).ULVZ structures are the most extreme seismic heterogeneity observed in the lower mantle, and the existence of ULVZs potentially influences the degree of thermal coupling (Glane & Buffett, 2018;Olson, 2003) and chemical exchange (Brandon & Walker, 2005;Fu et al., 2023;Knittle & Jeanloz, 1991) across the core-mantle boundary.
Tens of studies detect ULVZs and report large variations in size and shape (see compilation by Yu and Garnero (2018)).This may be due to, on the one hand, natural variation in ULVZ structures; and on the other hand, variation in the body wave phases studied, lack of suitable data coverage, and limitations of the forward modeling techniques, which are often simplified to two or one-dimensional heterogeneity.The typical seismic features related to ULVZs include a pronounced reduced shear wave velocity, often between 10% and 30% compared to surroundings.Some studies suggest ULVZs have ridge-like features (Garnero & McNamara, 2008;Ni & Helmberger, 2001;Rost et al., 2006), while cylindrical shapes are particularly successful in explaining the out-ofplane energy of observed S diff postcursors (Cottaar & Romanowicz, 2012;Cottaar et al., 2022;Z. Li et al., 2022;Yuan & Romanowicz, 2017).The lateral extent of cylindrical structures found to date are on the order of 600 to 1,000 km in diameter, which we refer to as "mega-ULVZs" following the suggested name for the broad ULVZ mapped using SPdKS phase beneath Samoa (Thorne et al., 2013).
The underlying cause of ULVZs remains unclear.Partial melt provides a plausible explanation for the drastic reduction in shear wave velocity (Q.Williams et al., 1998).However, a pure partial melt origin does not explain the localized presence of ULVZs by itself, without introducing chemical heterogeneity or varying topography (Rost et al., 2005).Moreover, modeling of the partial melt dynamics has been shown challenging as the dense melt is likely to sink out to form a global melt layer (e.g., Dannberg et al., 2021).If this scenario exists, the layer must be very thin (Russell et al., 2022).A chemically distinct component, such as iron enrichment, has been suggested for ULVZ based on mineral experiments (Dobrosavljevic et al., 2019;Wicks et al., 2010Wicks et al., , 2017)).The iron enrichment is also supported by the large aspect ratio observed for several ULVZs, which implies its composition is significantly denser than the surrounding mantle (Bower et al., 2011).
New observations from the geochemical community may provide more insight into the connection between surface hotspots and ULVZs.Geochemical samples at Earth's surface support the existence of deep mantle chemical heterogeneity (Hofmann, 1997).Efforts have long been made to track these deep mantle reservoirs.The noble gas and tungsten isotopes collected at numerous plume-fed surface hotspot volcanism suggest a primordial origin for a mantle reservoir (Jackson et al., 2017;Mukhopadhyay, 2012;Mundl et al., 2017;Mundl-Petermeier et al., 2020), which could be linked to ULVZs with a primordial origin at the base of the plumes.An increasing body of evidence is strengthening the correlation between mapped mega-ULVZs and the most anomalous isotopic signatures (Cottaar et al., 2022).The anomalous helium and tungsten isotopes have also been observed at many hotspots beneath the southern part of the Pacific Ocean (Jackson et al., 2017;Mundl-Petermeier et al., 2020), where no mega-ULVZs have been previously mapped (Hansen et al., 2023;Thorne et al., 2021;Yu & Garnero, 2018).
In this work, we examine shear wave diffracted data that has a sensitivity to the lower mantle structure beneath the southern Pacific region.We found tens of events showing evident anomalous postcursor waveforms after the ScS/ S diff phase and analyzed five high quality ones to confirm the direction the postcursors are coming from.Using waveform modeling, we show that three events point to a new ULVZ structure situated to the south of the Pitcairn hotspot, while the two other events might be due to the previously modeled ULVZ near Galápagos.

Methods
We use seismic shear waves that diffract along the core-mantle boundary (S diff , Figure 1) to sample potential ULVZs beneath the southern part of the Pacific Ocean.S diff is typically observed at stations of epicentral distances larger than ∼98°, although the exact value depends on the event depth and the velocity structure, but only varies slightly.In the diffracted region, S diff waves propagate parallel to the core-mantle boundary (Figure 1) and have a strong sensitivity to near-boundary layered structures, such as ULVZs.When the diffracted waves encounter an extremely slow velocity patch, some energy gets trapped within, propagates slower and refracts.This behavior is comparable to surface wave energy getting trapped in a slow sediment zone.The delayed S diff energy appears as a postcursor phase, arriving with a noticeable delay after the main S diff phase and showing move out with time as a function of azimuth.The robust observation of the postcursor energy mainly relies on the dense coverage of seismic stations.In previous Sdiff studies, most observations of postcursors have been made on the dense deployment of the Transportable Array (IRIS Transportable Array, 2003), for example, Cottaar and Romanowicz (2012), Cottaar et al. (2022), Z. Li et al. (2022), and Yuan and Romanowicz (2017).In our current data set, there are some geometries at sub-diffracted distances in the 90°-98°range, which are technically ScS phases.We find that at these distances, ScS energy becomes similarly trapped within the ULVZ, and produces postcursors that look and behave the same as S diff postcursors.Throughout the paper we will also refer to these observations as "S diff postcursors." In our analyses, we use transverse component seismograms which record the horizontal energy (SH) of diffracted waves.We only focus on SH energy as SV energy attenuates much more along the core-mantle boundary (Doornbos & Mondt, 1980;Komatitsch et al., 2010;Teng & Richards, 1968).We plot the seismograms as a function of azimuth as this enables clearer identification of the move-out of the postcursor phase (Cottaar & Romanowicz, 2012).

Analysis of Core-Diffracted Shear Waves
The exact location of the ULVZ can be challenging to pinpoint when employing seismic diffraction observation (Cottaar & Romanowicz, 2012;Thorne et al., 2021).Any point along a core-diffracted path could potentially contribute to the waveform complexity we observe at a surface station.If a ULVZ is sampled from a single azimuth there are strong trade-offs between size, velocity reduction, and location along the path.To determine the location, we require the same ULVZ to be sampled from multiple directions, which in turn depends on available earthquake-station geometries.
We searched all ScS/S diff data with a distance range from 90°to 140°that sample the core-mantle boundary beneath the southern Pacific for earthquakes above moment magnitude 6.0 and across all depths between 2000 and 2021.
Data are plotted as a function of azimuth and assessed by eye for potential postcursors, judging the signal-to-noise ratio of individual events and potential interference with depth phases.We find a total of 28 events containing some evidence of postcursors (Table S1 in Supporting Information S1).We select five earthquakes, which represent two different earthquake-station geometries and all have a moment magnitude above 6.3.For these events we remove individual seismograms with a poor signal-to-noise ratio (judged by eye).Data in this final data set ranges in distance from 90°to 135°.The data is introduced in detail in Section 3.
With this data set, we apply two approaches to determine the location of a potential ULVZ: 1. We beamform the postcursor energy across different groups of stations to find its direction of propagation.
These can be propagated back to the core-mantle boundary, assuming the energy results from the boundary of the ULVZ.See Section 2.2. 2. We forward model synthetic data with a simplified cylindrical ULVZ in different locations and evaluate how the waveform of resulting postcursors in synthetics fit the real data.See Section 2.3.

Beamforming Analysis
The lensing effect of the S diff postcursor raypath was first discovered in the study of the Hawaiian ULVZ (Cottaar & Romanowicz, 2012).The receiver arrays indicate the incoming S diff postcursor signals have a gradual deviation of the backazimuth as a function of azimuth.In this study, our beamforming procedures closely follow those applied to data for the Hawaiian ULVZ (Z.Li et al., 2022).We establish our array groups based on each station and its nearest 10 stations, and then only stack if the maximum aperture of an array is between 1°and 4°distance, in order to provide robust beamforming quality.
For each array, we apply phase-weighted stacking to obtain clean and robust beamforming results (Schimmel & Paulssen, 1997).The phase coherence proves to be a good weighting factor for S diff waves in the stacking procedure (Z.Li et al., 2022).The phase-weighted beamforming stack, B, is formulated as: where s i represents the time series of ith station of an array, x i is the distance vector for ith station with respect to the reference station, u is the trial slowness vector, and Φ k denotes the instantaneous phase obtained from the Hilbert transform of the original data series s i (t).The weighting factor ν is set to 2 to avoid further data distortion.The absolute slowness is fixed at 8.323 s/°, and we note that slight variations in this value does not significantly affect the beamforming results of S diff or its postcursor (Z.Li et al., 2022).The local peak values of the local maxima are picked using the package scikit-image (a toolbox of Image processing in Python (van der Walt et al., 2014)).Significant peaks within expected time ranges are picked as the main arrival and postcursor.We present one source of uncertainty by showing the standard deviation of the station azimuths in each array.A more broadly spaced subarray means we also expect a broader range of backazimuths across the array, while our stacking assumes a plane wave.

Forward Modeling
The direct simulation of a global wavefield for 3D models down to the period required to produce postcursors (10 s period) comes at a significant computational expense.We thus apply the Coupled Spectral Element Method (CSEM, Capdeville, Chaljub, Vilotte, & Montagner, 2003) to compute waveform synthetics for different ULVZ models.This method couples the spectral element method for a 3D heterogeneous model in the lowermost 300 km of the mantle with a normal mode solution for a 1D layered global Earth model in the core and the rest of the mantle.It proves to be efficient and accurate for full waveform solutions for core-diffracted phases (Capdeville, To, & Romanowicz, 2003).We use Preliminary Reference Earth Model (Dziewonski & Anderson, 1981) as the background global Earth model for the 1D mode solution and SEMUCB (French & Romanowicz, 2014) as the background 3D lower mantle model for the spectral element region.The trial ULVZ models assume a cylindrical shape that is constrained by only a few parameters and has been shown effective in explaining S diff postcursors (Cottaar & Romanowicz, 2012;Cottaar et al., 2022;Yuan & Romanowicz, 2017).
In this study, we use forward modeling to mainly analyze the potential locations of a ULVZ.The overall data availability and the eventual uncertainty in location do not allow us to rigorously explore trade-offs between the other characteristic parameters.Empirical experience in the postcursor modeling suggests 300 and 400 km are good starting parameters (Cottaar & Romanowicz, 2012;Z. Li et al., 2022).Constraining the height of the ULVZ also depends on the resolution of the vertical mesh and the corresponding frequency content.Therefore, for most of our trial models, we assume the ULVZ has a radius of 300 or 400 km, height of 20 km, and velocity reduction of 20%.Our trial locations are based on the beamforming results and by analyzing which azimuths observe minimally delayed postcursors, which we assume have minimal refraction and propagated through the center of the ULVZ.

Data
Five high quality events (see Figure 2 and Table 1) are chosen, which include two different geometries across the southern Pacific.Events 1 and 2 occurred beneath New Zealand and are sampled in North America.Events 3 and 4 occurred beneath the Fiji Islands and are sampled across South America.Event 5 has a similar geometry, but the event occurred further west beneath the Solomon Islands.Data with postcursors for five further events in the same geometry toward South America are shown in Figure S16 in Supporting Information S1.Data from the South American trench toward stations across Australasia and from the Central America trench toward Antarctica were also considered, but have poor signal-to-noise ratios.The waveforms for the five main events are shown in Figure 3, and individual earthquake-station geometries are shown in Figures S2-S6 in Supporting Information S1.

Event 2
The second event (Figure 3b) of magnitude 6.3 occurred on 07 December 2012 in North Island, New Zealand, which is close to, but slightly further north compared to the epicenter of the first event.The events also occurred in the same year, and thus have similar coverage beneath the Southern Pacific Ocean.This event equally has a fortunate radiation pattern (Figure S3 in Supporting Information S1) resulting in a high signal-to-noise ratio.The main S diff arrival is clear, and its depth phase is also identifiable roughly 80 s after the main S diff arrival.The postcursors appear from 60°to 80°azimuth and delay with time from 30 to 60 s comparable to the first event.

Event 3
The third event (Figure 3c) occurred near the Fiji Islands on 06 September 2018 and sampled the Southern Pacific through stations in South America, mainly at ScS distances.This is a much larger event of moment magnitude 7.9 and at a depth of 686.6 km.The event radiates significant SH energy from 90°to 135°(Figure S4 in Supporting Information S1).The waveforms of the main phase suggest a complicated source time function.The postcursor has a move-out in time at azimuth from 110°to 120°and from 125°to 140°, a minimum in travel time delay occurs between 120°and 125°.

Event 4
The fourth event (Figure 3d) occurred at 554.7 km depth beneath the Fiji Islands on 18 November 2018 with a 6.8 moment magnitude.Its geometry is comparable to Event 3, but its magnitude is smaller and the radiation pattern is less favorable (Figure S5 in Supporting Information S1).The postcursors are comparable, but less strong than for Event 3. We observe an arc of postcursor after the main arrivals.The minimum of the postcursor traveltime lies between 120°and 125°azimuth.

Event 5
The fifth event (Figure 3e) occurred on 23 April 2011 beneath the Solomon Islands.This is a large event (magnitude 6.8) of intermediate depth (76.9 km).The depth phase (sS diff ) of this event is strong and arrives roughly 40 s later than the main S diff phase.The same postcursors appear after the main phase and the depth phase, around 110°-120°in azimuth, and in both cases delayed by 20-40 s.The postcursors have a roughly hyperbolic shape in their delay.This event is located further west than Event 3 and Event 4 and thus the S diff has longer paths that sample the core-mantle boundary further to the west (Figure 2).

Beamforming Results
Examples of individual beamforming stacks are included in Supporting Information S1 (Figures S7-S10 in Supporting Information S1).The times and backazimuths for the main phase and postcursors picked from the phase-weighted stacks for Events 1, 2, 3, and 5 are plotted in Figure 4.For Event 4, we could not clearly and unambiguously identify the postcursor arrival from the beamforming stacks.
For Event 3 the postcursor signal shows a clear arrival time move-out from 20 to 50 s with the relative backazimuth deviation also showing a trend from 6°to 14°(Figure 4c).The pattern shows that the minimally delayed postcursor has little backazimuth deviation, and increase in the delay time correlates with increase in backazimuth deviation.This pattern is comparable to what is seen in beamforming results for the Hawaiian ULVZ (Cottaar & Romanowicz, 2012).This suggests this newly observed ULVZ could also be close to cylindrical in shape.
The beamforming results of Events 1 and 2 are more scattered (Figures 4a and 4b), but there are slight trends as a function of azimuth.We notice that the values of the backazimuthal deviations for Events 1 and 2 are much larger (up to 30°-40°) which correlate with the observed delay times also being longer.This is likely due to strong offplane refraction away from the ULVZ center.
Based on the beamforming results of postcursor backazimuth, we plot the back-projected raypaths from the receiver arrays of all the events together in Figure 5 to better infer the ULVZ location.Despite the scattered results, the overall evaluation of the back-projected rays suggests a ULVZ could exist beneath the southern Pacific region and might lie close to the Macdonald and Pitcairn hotspots.In particular, the back-projected raypaths in Events 2 and 3 demonstrate some lensing effect that could be caused by an approximately cylindrical ULVZ.The back-projected energy for Events 1 and 2 suggest that these could also come from the previously modeled Galápagos ULVZ (Cottaar et al., 2022), indicated as location G.

SEMUCB Reference Model
For reference, we compute waveforms for the 3D mantle tomographic model SEMUCB (French & Romanowicz, 2014) without any ULVZ.The resulting waveforms are shown in Figure 6 for events 1 & 2 and in Figure 8 for events 3-5.SEMUCB synthetics account for any variations due to the source pattern and 3D long-wavelength

Previously Detected ULVZs
Based on the beamforming result of events 1 & 2, we find that the postcursor signals could come from the recently detected Galápagos ULVZ.To avoid the potential misinterpretation of the postcursor signal from the existing ULVZs elsewhere, we make synthetics implementing the previously detected ULVZs.We include the major ULVZ structures in the Pacific including Samoa (Thorne et al., 2013), Galápagos (Cottaar et al., 2022) and Hawaii (J.Li et al., 2022;Z. Li et al., 2022).Details of the model setting are shown in Table S2 in Supporting Information S1.Note, that we have not included the suggested ULVZ near Marquesas since its location was not modeled (Kim et al., 2020).
Figure 6 shows the synthetics for this multi-ULVZ model for events 1 & 2 as these show noticeable postcursors.
Combining this observation with the beamforming results (Figure 5), we infer that events 1 & 2 are likely affected by the Galápagos ULVZ.The fit is not perfect, and this data might provide new constraints on any asymmetry of the Galápagos ULVZ, but this is beyond the scope of this study.Events 3-5 remain unaffected by the multi-ULVZ model (waveforms shown in Figure S11 in Supporting Information S1).Thus, we focus our detailed modeling in the following section on events 3-5.

Modeling the Location of the ULVZ
For each event, we assess the azimuth range for which the postcursors are strong, and we estimate at which azimuth the postcursors are minimally delayed.The assumption of the cylindrical ULVZ model means that this azimuth corresponds to the center of ULVZ.We backproject these azimuths and assess potential ULVZ location by their interception.We also account for the appropriate distance where the S diff paths are expected to have sensitivity to the lowermost mantle (dashed lines in Figure 7).
Based on these estimates, we perform a trial modeling for tens of ULVZs (Figure 7).Details of the modeling parameters are included in Table S3 in Supporting Information S1.After visually evaluating the waveform fit for each ULVZ model, we propose one best cylindrical ULVZ model with a 400 km radius and a 20% shear wave velocity decrease centered at geographical coordinates ( 33.6, 130).The waveforms for this model for events 3-5 are presented in Figure 8.However, we do note that the ULVZ can be shifted in the west-east direction and produce similar results.To illustrate this, we show waveforms for five models (ULVZ 40-44, Table S3 in Supporting Information S1) shifted sequentially in the west-east direction (Figure S12 in Supporting Information S1), which show only slight variations in the postcursors.Results for ULVZ 44 do present a potential constraint for the east boundary, as we see a strong reduction in the energy of postcursors at out-of-plane azimuths here (Figure S20 in Supporting Information S1).Based on these models, we provide an estimated area of uncertainty shown in Figure 10.Given the quality of the data, and the uncertainty in the location of the ULVZ, it is difficult to draw conclusions from any systematic search on the size and the velocity reductions of the ULVZ, and the trade-offs between these parameters.From our modeling, we mainly conclude that a significant ULVZ, comparable to those observed with Sdiff in other places, needs to be present.
While we excluded events 1 & 2 based on potential postcursors from the Galápagos ULVZ, we do test if these can also contain signatures from this new ULVZ.In Figures S14-S18 in Supporting Information S1, we show synthetic data for all events for our preferred model (presented as model C).For events 1 & 2, this shows postcursor energy at the very largest azimuths, with a move-out that does not fit the observations.Given the sparsity and quality of data at these azimuths, it is difficult to say that this postcursor is not actually present in the data.We present waveforms for two further models shifted to the west (model A) and to the northwest (model B, Figure S14 in Supporting Information S1), to test if some location can reasonably fit all five events at once.Particularly model A, produces postcursors that could be the ones seen in events 1&2.However, the postcursors this model produces for events 3-5 are shifted away from the azimuths where they are observed.We conclude that events 1&2 are either due to the Galápagos ULVZ, or are affected by a ULVZ further north that is not sampled by events 3-5.

Discussion
Our main results are based on the postcursors observed in South America which give evidence of a previously unknown ULVZ beneath the Southern Pacific.Although the location of the ULVZ has some uncertainty in the west-east direction, it is relatively close to the Pitcairn hotspot, which lies at the northern edge of the proposed model.We tentatively name it "Pitcairn ULVZ," although we note that the Macdonald and Easter hotspots lie to the west and east of the modeled ULVZ, respectively, within the region of uncertainty.

Structure of a Potential Pitcairn ULVZ
We modeled the Pitcairn ULVZ using a simplified cylindrical model.The strongest evidence that the Pitcairn ULVZ is roughly cylindrical in shape comes from the results of beamforming event 3. The backazimuths observed show a similar lensing effect in the S diff postcursor data as seen for data sampling the Hawaiian ULVZ (Cottaar & Romanowicz, 2012).This implies some similarity in the 3D structure between these mega-ULVZs, which means they might have comparable composition and origin and are shaped similarly by their dynamic surroundings (Bower et al., 2011;M. Li et al., 2017).
Besides the location of the Pitcairn ULVZ, other parameters in describing the cylindrical ULVZ include the height, radius, and velocity reduction.Unfortunately, the trade-offs between these are quite strong.For instance, a larger ULVZ with a weaker velocity reduction result in a similar result in travel time delay compared to a smaller ULVZ with a stronger velocity reduction.Additionally, the best-fitting size or velocity reduction of the ULVZ will depend on its location along the diffracted path.The seismic observations for this ULVZ are limited to sampling the structure in one direction, and the station coverage is sparse and the data relatively noisy.These factors make it difficult to constrain ULVZ parameters and their uncertainties.Here we have limited ourselves to modeling with parameters similar to other mega-ULVZs.

Potential Primordial Repositories
While there is considerable uncertainty on the location of the new ULVZ, there is some correlation with the geographic location of the surface hotspots in the southern Pacific (see Figure 10).Besides the Pitcairn hotspot, the ULVZ could also underlie the more westerly hotspot Macdonald or the more easterly hotspot Easter.These hotspots at the southern edge of the Pacific LLSVP are convincingly rooted by deep mantle plumes, as imaged by recent whole mantle tomographic models (e.g., SEMUCB, French and Romanowicz (2014); GLAD-M25, Lei et al. (2020)).Slices through a SEMUCB demonstrate the possible plume conduits connecting the coremantle boundary region and the surface (Figure 9).These hotspots show anomalous primordial signatures of 3 He/ 4 He values (Jackson et al., 2017) and Macdonald and Pitcairn also show deviations in tungsten isotopes ( Mundl-Petermeier et al., 2020).The strongest deviations observed are not as anomalous as other locations linked to the presence of mega-ULVZ, that is, Hawaii, Iceland, Samoa, and Galápagos (Cottaar et al., 2022), although this difference could also be due to other localities being more extensively studied.The anomalously high 3 He/ 4 He values at the surface hotspots suggest a primordial source that has been chemically distinct from the convective mantle for at least 4.5 Ga (Jackson et al., 2017).The siderophile tungsten isotope data further suggest the source not only preserves the primordial composition, but should contain metal as well.To what degree these signatures could be sourced purely from the LLSVP or ULVZs and what role they each play remains up for debate.Equally uncertain is if all hotspots are underlain by a ULVZ or mega-ULVZ (Cottaar et al., 2022;Yu & Garnero, 2018).We note that our full data set covers a lot of the southern Pacific and suggests there is not a mega-ULVZ associated with each of the many hotspots in this region.

Conclusions
We present analyses of postcursors to core-diffracted shear waves sampling the southern part of the Pacific.The observed postcursors, which can be grouped in two different earthquake-station geometries, cannot be explained by a single ULVZ.We suggest the postcursors sampled in North America result from the Galápagos ULVZ, while the postcursors observed in Southern America suggest the presence of a previously unknown ULVZ roughly to the south of the Pitcairn hotspot.Its location remains uncertain in the west-east direction as this is the direction of all the paths that sample this ULVZ.The beamforming analyses of the postcursors show a strong backazimuthal deviation, suggesting strong focusing and the ULVZ is likely to have a cylindrical structure similar to the Hawaiian mega-ULVZ.The ULVZ may represent a potential root to a mantle plume that feeds into one or multiple nearby hotspots.Nearby hotspots have anomalous isotope signatures that could represent sampling of ULVZ material.

Data Availability Statement
The seismic waveform data used for processing and plotting in the study are available at IRIS (Incorporated Research Institutions for Seismology) data center via The IRIS DMC Web Services (https://service.iris.edu/irisws/) under open access.Waveform processing was conducted using the Python library ObsPy (Beyreuther et al., 2010).

Figure 1 .
Figure 1.(a) Raypaths of core-diffracted shear waves, S diff , that are sensitive to the core-mantle boundary region.S diff paths (blue) with epicentral distance from 100°to 130°are computed using the Preliminary Reference Earth Model velocity model (PREM, Dziewonski & Anderson, 1981).Part of the core-mantle boundary is highlighted in red at the distance where S diff raypaths are within 100 km of the core-mantle boundary, indicating where the wave senses the near-boundary structure when accounting for its finitefrequency nature.(b) Schematic horizontal wavefront of S diff postcursors (blue) caused by cylindrical ultra-low velocity zone (red).

Figure 3 .
Figure 3. Tangential displacement waveform observation of (a) Event 1, (b) Event 2, (c) Event 3, (d) Event 4, and (e) Event 5. Waveforms are sorted by source-station 1°a zimuth bins and aligned by the predicted S diff travel times.The average of each azimuth bin is plotted behind with thicker gray lines.The colored patches highlight the approximate moveout of the delayed S diff postcursors analyzed in this study (colors correlate with Figure 2).
. The computed waveforms do not contain any postcursor energy.To create postcursors, we have to include stronger anomalies than long-wavelength variations in SEMUCB.Overall, we only observe small variations in the main arrival as a function of azimuths.There are some travel time variations which can be attributed to the paths going through the Pacific large low-shear-velocity provinces (LLSVP) structure.

Figure 4 .
Figure 4. Beamforming results for (a) Event 1, (b) Event 2, (c) Event 3, and (d) Event 5.For each, the left subplot shows the travel times of the energy peak in the beamforming stack for the main phase (blue) and the postcursor (red) as a function of station azimuth relative to the predicted travel time for S diff for Preliminary Reference Earth Model (Dziewonski & Anderson, 1981).The right subplot shows the observed backazimuth deviations with respect to the epicenter backazimuth in the beamforming stacking.The error bars show the standard deviation for the spread in azimuths within each subarray.

Figure 5 .
Figure5.Back-projected paths of the S diff postcursor using the observed backazimuth deviations for the S diff postcursors (Figure5).The source locations, receivers and back-projected rays of Event 1 (blue), Event 2 (orange), Event 3 (green), and Event 5 (purple) are plotted on the background tomography at 2,791 km depth SEMUCB_WM1(French & Romanowicz, 2014).The dashed line starts where the back-projected rays encounter the bottom 100 km of the lowermost mantle.Our proposed ultra-low velocity zone (ULVZ) model (labeled "C"), and Galapagos ULVZ (labeled "G") are shown in black circles.

Figure 6 .
Figure 6.Waveform demonstration of the effects from background tomography structure and previously detected ultra-low velocity zones (ULVZs) (event 1 & 2, top and bottom).From left to right, the columns show the waveforms for the real data, for the SEMUCB tomography model, and for the multi-ULVZ model.We identify postcursor signals (highlighted in color) in the waveforms for the multi-ULVZ model, resembling the postcursor signals in the real data, which are caused by the Galápagos ULVZ.

Figure 7 .
Figure 7. Geographical location of the proposed best ultra-low velocity zone (ULVZ) model (C, black atop) and other trial models (transparent gray behind).The proposed ULVZ model has the best waveform fit to the real data in events 3-5.The colored dashed line indicates the azimuth of the minimum delayed postcursor.

Figure 8 .
Figure 8. Waveform comparisons between real and synthetic for events 3-5.The first column shows real data, the middle columns represent synthetics for tomography model SEMUCB, and the right column shows the waveform of our proposed ultra-low velocity zone (ULVZ) model.From top to bottom the rows represent events 3-5.Plots show the tangential displacement waveform displayed in 1°azimuth bins.The waveforms are bandpass filtered between 10 and 20 s.The time axis is aligned by the predicted S diff travel times.The horizontal yellow dashed lines in the synthetics indicate the azimuth that corresponds to the center of the cylindrical ULVZ model.We only show the azimuth range where the postcursor is present.The moveout of the postcusors is captured by our proposed model.

Figure 9 .
Figure 9. SEMUCB tomography cross-section slicing seismic structure beneath Pitcairn and Macdonald hotspots.(a) Crosssection lines and global distribution of recognized hotspots and ultra-low velocity zones (ULVZs) (updated compilation by Yu and Garnero (2018)).(b) Cross-section 1 slicing the Pitcairn hotspot.(c) Cross-section 2 slicing the Macdonald hotspot.

Table 1
Events Analyzed in This Study Note.Further events assessed are shown in TableS1in Supporting Information S1.Source information is obtained from the global CMT catalog (http://www.globalcmt.org/).