D″ observations in the Pacific from PLUME ocean bottom seismometer recordings

Abstract

The seismic investigation of the lowermost mantle is in many places hampered by the lack of suitable source–receiver combinations that sample the D^′′ region and have to meet the requirements of a suitable epicentral distance range. The low velocity regions beneath the central Pacific and Atlantic Oceans in particular have been sampled in fewer places than circum Pacific regions. In this study, we use data from two recent ocean bottom seismometer (OBS) deployments for the Plume-Lithosphere Undersea Mantle Experiment (PLUME) around Hawaii to increase the coverage of the lower mantle with reflected P waves. Through stacking of the data we achieve significant reduction in noise levels. The most favourable epicentral distances to detect D^′′ reflections are around 70–79°. Most of our source–receiver combinations have distances less than that, thereby limiting the number of candidate observed reflections. Nevertheless, using array methods, we are able to test approximately 70 events for arrivals with slowness values and arrival times that would be consistent with a top-side reflection off a hypothetical D^′′ structure (PdP wave). Modelling these data with a 1-D reflectivity method, we identify a few places of detectable PdP waves, for which the velocity contrast in P- and S-wave velocity across the D^′′ reflector have to be relatively large (around 3–5 per cent increase and decrease, respectively) compared to other regions (e.g. beneath the Caribbean or Eurasia where the contrast is closer to 1–2 per cent). For larger distance ranges, smaller velocity contrasts are sufficient to cause observable reflections. This study shows that, despite the possible dominance of microseisms on OBS records, it is possible to use relatively short-period waves, with dominant periods as short as 3–7 s. Our findings suggest that, with future such deployments, OBS deployments will help to extend D^′′ studies to previously unmapped regions.

Composition of the mantle, Body waves, Pacific Ocean

INTRODUCTION

In recent years, the lowermost mantle (named as the D^′′ region by Bullen 1949) has been subject to investigation in many locations around the globe, and many diverse structures have been identified on many length scales. For example, observations of one or more discontinuities (e.g. Lay & Helmberger 1983; Weber 1993; Thomas et al.2004a,b; Lay et al.2006; for a review see Wysession et al.1998; Lay 2007), seismic anisotropy (Kendall & Silver 1998; Lay et al.1998; Maupin et al.2005; Wookey & Kendall 2008) and ultralow velocity zones (e.g. Garnero et al.1998; Thorne et al.2013) have been made in this region close to the core–mantle boundary (CMB).

Several causes for the observed structures in D^′′ have been proposed. These include deep subduction (Kendall & Silver 1998; Thomas et al.2004a; Hutko et al.2006; Lay 2007), plume origins (Williams & Garnero 1996), thermochemical layering (e.g. Lay et al.2004; Lay 2007), and also a phase transition from perovskite to post-perovskite (Murakami et al.2004; Oganov & Ono 2004; Tsuchiya et al.2004). While many properties of the D^′′ region are consistent with a phase transition in MgSiO₃ (e.g. Murakami et al.2004; Wookey et al.2005; Thomas et al.2011), there are also some problems when realistic mineralogy is used: for example, a phase transition in pyrolitic material would occur over a large depth interval of several 100 km (Catalli et al.2009) thereby rendering it invisible for seismic detection due to a large gradient, or the phase transition might occur at pressures found in the core and not in the mantle above (Grocholski et al.2012).

Since the seismic signals from reflections off structures in D^′′ are usually weak arrivals, especially for P waves (e.g. Kito et al.2007; Hutko et al.2008), it is often necessary to enhance them through stacking methods (e.g. Rost & Thomas 2002; Schweitzer et al.2002). Suitable seismic receiver arrays are deployed in only a few regions worldwide thereby limiting the amount of suitable recordings even further in global data sets. Furthermore the distance range where reflections from structures in the D^′′ region are best observed is narrow due to the reflection coefficient for a positive or negative velocity jump (e.g. Weber 1993; Cobden & Thomas 2013) and the best epicentral distance range lies between 70° and 79° for P waves, although observations have also been made at shorter epicentral distances (e.g. Chaloner et al.2009; Rost & Thomas 2010).

Seismic investigations of the presence (or absence) and nature of a seismic reflector at the top of the D^′′ region helps distinguish between different hypotheses. However, such studies require comprehensive data sets with many crossing ray paths. Unfortunately the global coverage of such data is grossly incomplete, and currently available source–receiver combinations illuminate mostly regions where tomographic images show anomalously high seismic velocities (e.g. Masters et al.2000; Grand 2002; Ritsema et al.2011), and most regions examined so far are covered by only very few crossing ray paths (e.g. Thomas et al.2002; Wookey & Kendall 2008; Thomas et al.2011). Some observations of the D^′′ reflector have been made in low velocity regions in some regions in the Pacific (e.g. Yamada & Nakanishi 1998; Kito et al.2004; Avants et al.2006; Lay et al.2006; Kawai & Geller 2010; Takeuchi & Obara 2010). But the results are more diverse than in high-velocity regions around the Pacific, and the African low-velocity anomaly (e.g. S40RTS, Ritsema et al.2011) has rarely been sampled with reflected waves from a D^′′ discontinuity. It would therefore be useful to find source–receiver combinations that allow investigating the deep mantle in new regions or that add crossing ray paths.

As far as station coverage is concerned, the deployment of ocean bottom seismometers (OBSs) potentially provides new source–receiver combinations but, owing to high noise levels in the microseism band between 1 and 10 s period, passive seismic OBS data have so far been used primarily in long-period studies, including surface wave analyses (e.g. Forsyth et al.1998; Endrun et al.2008; Laske et al.2011), SKS splitting analyses (e.g. Wolfe & Solomon 1998) or receiver functions (Kawakatsu et al. 2009). In this work, we present a body wave array analysis of OBS recordings from two deployments for the Hawaiian Plume-Lithosphere Undersea Mantle Experiment (PLUME; Laske et al.2009) in order to test for the possibility of observing, and the presence of, D^′′ reflections in the data.

DATA

We use earthquakes located around the Pacific Ocean that were recorded at PLUME OBS stations. This experiment was undertaken in two phases. Two OBS groups, from the Woods Hole Oceanographic Institution (WHOI) and the Scripps Institution of Oceanography (SIO) in La Jolla, contributed instruments. During the first phase (from 2005 January to 2006 January), 35 OBSs occupied sites in an area covering roughly 300 000 km² (grey triangles in Fig. 1). All but 3 OBSs were recovered and recorded data. During the second phase (2006 April–2007 June), an additional 34 sites were distributed over an area of roughly 1.8 million km² (black triangles in Fig. 1). From these, 26 OBS were recovered where 3 did not record seismometer data, leaving 23 recording OBSs. All OBSs were equipped with a broadband seismometer (Güralp CMG-3T or Nanometrics Trillium T-240) or a wideband seismometer (Nanometerics Trillium T-40 for some phase 1 sites occupied with SIO OBSs). PLUME also provided land based stations on the Hawaiian Islands. These were deployed and maintained by co-principal investigators from the Department of Terrestrial Magnetism (DTM) at the Carnegie Institution of Washington. PLUME had a total of 10 temporary broadband installations featuring a Wielandt-Streckeisen STS-2 seismometer. These stations operated at various times between 2004 August and 2007 December. In addition, data are available from permanent and semi-permanent global seismic network observatories: Kipapa on Oahu (KIP), Pohakuloa on the Island of Hawaii (POHA) and Haleakala on Maui (MAUI). Continuous data streams were collected at various sampling rates: 40 Hz for WHOI OBSs, 31.25 Hz for SIO OBSs, and 20 Hz for PLUME land stations. We filtered the data using several Butterworth bandpass filters (see below) and down-sampled to 20 Hz. The data were then analysed using array stacking techniques (see e.g. Rost & Thomas 2002, 2009).

Figure 1.

Location of the sources (stars), the receivers (triangles) and possible D^′′ reflection points (circles and crosses). Circles indicate places where a PdP signal has been detected (colour and size show the confidence, compare to legend), crosses show places where no clear signal has been detected due to low signal-to-noise ratio (snr). Small white circles show no PdP signal despite good data quality. The tomography model is from Kárason & van der Hilst (2001). The inset shows the wave paths for PdP, PcP and P for an event with 70° epicentral distance. Here, the first PLUME OBS deployment, PLUME-1, is indicated with grey triangles, PLUME-2 with black triangles. Numbers indicate event numbers from Table 1.

Ideally, the earthquake sources should be in an epicentral-distance range of about 70–79° for the best visibility of reflections off the top of D^′′ (e.g. Weber 1993). Additionally, the earthquakes should not be too shallow and large to avoid reverberations and depth phases (pP, sP) that follow the P wave and may potentially mask the small-amplitude D^′′ reflections (PdP, Fig. 1). Since this distance range of 70–80° yielded only a very small number of earthquakes in case of the PLUME data set, we extended our distances requirements to lower epicentral distances. We search the National Earthquake Information Centre (NEIC) catalogue with a distance restriction of 40–80° from Hawaii, source depths of greater than 50 km and body wave magnitudes (m_b) greater than 5.8. 117 events fulfil the criteria for the time of the PLUME deployments. For each event, we perform a 4th-root vespagram analysis (Davies et al.1971; Muirhead & Datt 1976; Rost & Thomas 2002; Schweitzer et al.2002) for the unfiltered broad-band data, and with two different bandpass filters (10–100 s, 3–25 s), and use only those events further for which clear P arrivals are observed in one of these three cases.

OBSERVATIONS

Thirty-four events have a signal-to-noise ratio in either the unfiltered or filtered data for which further analysis was possible (Table 1 and Fig. 1). Of those events, we produce vespagrams with several different bandpass filters and search for possible arrivals between the P wave and the PcP arrival (Fig. 1 inset). Most events have to be low-pass filtered with fairly long corner periods (longer than 8 s) but some events show the arrivals at shorter periods and even at around 1 Hz (e.g. pass bands 3–15 s, 1–10 s,). Some events show clear P and PcP waves already in unfiltered data. We show some examples of events with visible PdP in Figs 2 and 3. Vespagrams were computed with the Nth root staking technique (e.g. Muirhead & Datt 1976) where N = 4, since we find that these are better for picking slowness values compared with linear stacks. For comparison, we also show a linear vespagram in Fig. 4 together with the filtered seismogram section of one event.

Figure 2.

Examples of 4th root vespagrams of 5 PdP observations in vespagrams with P, PdP and PcP marked. Event 12 shows a possible PdP signal, which we classify as such due to its lower than expected slowness. The filters used for each event are specified in each panel. Event parameters are given in Table 1.

Figure 3.

Data for event #31 that shows a PdP arrival (a) together with reflectivity synthetics for this event (b–e). Models are constructed for velocity contrasts of +2, +3, −5 per cent in P- and S-wave velocity and one case for a −5 per cent vp jump and +5 per cent vs jump (e). The reflection coefficient for PdP is shown in (f).The reflection coefficient (for contrasts of ±3 per cent shows larger amplitudes for cases with opposite velocity jumps for P- and S-wave velocities in the distance range considered here.

Figure 4.

Seismogram (a), 4th root vespagram (b) and linear vespagram (c) for event #20 filtered with 14–70 s (compare with Fig. 2c). The slowness resolution for long-period filtered data is poor in the linear vespagram compared with the 4th root vespagram. Arrivals are marked in the bottom vespagram.

Table 1.

Events used in this study. Source parameters are from the National Earthquake Information Centre (NEIC). A ‘y’ in the last column indicates visible and clear PdP arrivals while ‘y?’ denotes the events where the arrival could possibly be a reflection off D^′′. Numbers in the last column indicate detected depth to reflector; * indicates events with good signal-to-noise ratio in the vespagram but no visible PdP arrival.

#	Date	Time	Latitude	Longitude	Depth	mb	Dist.	PdP
	(yyyy/mm/dd)	(hh:mm:ss)	(°)	(°)	(km)		(°)
1	2007/05/06	21:11:53	− 19.4681	− 179.3288	678.6	6.5	40–51
2	2007/04/25	13:34:16	− 14.3155	166.853	67.8	6.4	45–55
3	2007/04/21	07:12:48	− 3.5537	151.3511	407.8	6.1	53–61
4	2007/03/09	03:22:42	43.2267	133.5104	441.4	6.0	57–68	*
5	2007/03/08	05:03:32	29.956	140.2179	140.1	6.1	52–65
6	2007/01/31	03:15:55	− 29.9015	− 177.9179	60.5	6.5	48–60	y 2500
7	2007/01/30	21:37:50	21.1695	144.7656	51.7	6.6	50–62	*
8	2006/10/24	03:03:51	4.9176	125.4541	55.6	5.9	70–83
9	2006/10/18	10:45:35	− 15.095	167.2536	138.2	6.4	45–56
10	2006/10/03	18:03:15	− 18.9296	169.022	169.2	6.3	47–58	y?
11	2006/08/24	21:50:37	51.1751	157.398	54.2	6.5	42–53	*
12	2006/08/20	03:01:05	49.8631	156.3607	50.3	6.0	42–54	y? 2600
13	2006/08/11	14:30:41	18.5391	− 101.0697	60.4	6.0	47–57	y 2450
14	2006/08/07	22:18:55	− 15.8406	167.8174	148.5	6.8	45–55
15	2006/06/22	10:53:12	45.4156	149.3132	106.1	5.9	46–58
16	2006/06/09	05:58:28	− 17.5942	− 178.7259	565.7	6.1	38–48
17	2006/05/22	13:08:02	54.3324	158.4141	190.1	6.2	42–55
18	2006/05/16	10:39:23	− 31.8438	− 179.1549	156.0	7.4	50–62	*
19	2006/01/02	22:13:40	− 19.9742	− 178.1096	584.1	7.2	43–48	*
20	2005/12/08	09:01:26	− 5.4556	146.9877	222.0	6.1	60–65	y 2600
21	2005/11/22	15:11:32	− 5.1553	145.2361	87.8	6.2	60–66
22	2005/11/21	15:36:29	30.9826	130.1217	145.8	6.2	63–71
23	2005/10/15	15:51:07	25.2965	123.4352	200.4	6.4	69–78	y 2600
24	2005/09/21	02:25:06	43.8826	146.2334	102.2	6.0	49–57
25	2005/08/24	10:15:33	38.494	142.9719	58.5	6.1	52–59
26	2005/08/06	09:56:11	− 19.7173	− 175.8137	205.5	6.0	42–46
27	2005/07/02	02:16:47	11.184	− 86.3832	78.6	6.6	64–70	y 2550
28	2005/06/15	10:13:57	− 4.5957	153.1877	76.7	6.2	53–59
29	2005/04/11	17:08:53	− 21.9338	170.5515	73.4	6.7	51–56
30	2005/03/31	01:39:12	− 18.6258	175.9268	66.9	5.9	45–50	y? 2650
31	2005/03/17	13:37:36	15.2256	− 91.4327	197.4	6.2	58–65	y 2480
32	2005/03/02	10:42:09	− 6.5657	129.8801	197.6	7.1	76–81	y? 2700
33	2005/02/08	14:48:21	− 14.3176	167.2603	214.9	6.7	48–52	y 2650
34	2005/02/05	03:34:25	16.0033	146.0387	148.0	6.6	53–59

The wave that results from a reflection off a structure about 300 km above the CMB should arrive with a slowness and traveltime between those of a P and a PcP wave (e.g. Weber 1993). This wave that we call PdP is composed of the reflection and, where possible, the diving wave (sometimes called PdP and PDP, Weber 1993). Since most events in our data set have epicentral distances of less than 70°, and as low as 40°, the seismic arrival of a reflection off a structure about 300 km above the CMB will arrive close to PcP in both slowness and traveltime. The amplitude of the PdP wave for those shorter distances will be very small since the reflection coefficient for a P wave at a velocity increase or decrease (Fig. 3f) is low for distances less than 70°. For positive P-velocity contrasts, the reflection coefficient rises steeply from epicentral distance greater than 70° to a value of 1 whereas it is below 0.05 at smaller take-off angles. Interestingly, the reflection coefficient is larger at shorter distances for opposite behaviour of P and S wave, that is, P wave increases and S wave decreases and vice versa. A case of decreasing P and S velocities would practically not produce any visible reflection unless the velocity decrease was very large.

In our data set, we find seven events with a signal that arrives with a slowness, traveltime and backazimuth as expected for a PdP arrival and can clearly be identified as a D^′′ reflection (i.e. good signal-to-noise ratio, no interference with other phases, clear separation of slowness). Four other events have energy arriving at the correct time and slowness but for those events, the PdP reflection is partially superimposed by other phases and cannot as clearly be identified as the best seven cases. The events with clearly discernible PdP arrivals and one event with a possible PdP arrival are shown in Figs 2 and 3. Event #20 from Table 1 (Fig. 2e) has a small signal consistent with the reflection coefficient for the distance range of 42–54°. The signal arrives at the correct traveltime for PdP time, but we would classify this as possible detection due to the lower-than-expected slowness, although the signal is above the noise level and the expected amplitude is similar to the observed one. The distance range of event #23 (Fig. 2a) is at an epicentral distance were one expects a large reflection coefficients and therefore a clear reflected wave (Fig. 3f), and a fairly strong PdP wave is visible in the data.

We attempt to model the PdP arrivals with synthetic seismograms using a 1-D reflectivity method (Müller 1985). Velocity jumps of less than 2 per cent (both positive and negative P- and S-wave velocity jumps) have a barely visible effect in our seismograms for several of the shorter distance ranges of 40–60°. For velocity contrasts equal and above 3 per cent, we find that small-amplitude arrivals are visible in the synthetics for short distances (Figs 3c–e). Here, due to interference with large amplitude arrivals with P slowness, we find that the PdP arrivals are apparently shifted to slightly lower slowness values in the vespagrams. A change in density also affects the reflection coefficient but for the distances used here, the main effect is due to the change in velocity (e.g. Lay & Garnero 2007).

For event #23, a smaller velocity contrast is sufficient, and modelling indicates that a 1 per cent increase in P-wave velocity can produce a signal comparable to the one found in the data. In contrast, the waveforms of event #20, also at a larger distance range of 60–65°, require a P-velocity contrast of around 2 per cent to explain the observation. In addition to the events with clear PdP and possible PdP, we find some events without any signal between P and PcP (marked with white circles in Fig. 1 and stars in Table 1). These events have very good signal-to-noise ratio but generally very short epicentral distances (43–62°). For these events, very large velocity contrasts would be needed to produce visible reflections. Fig. 5 shows vespagrams of some of the events without a visible PdP reflection.

Figure 5.

Examples of 4th root vespagrams for three events without a visible PdP reflection. Events #7, 18, 19. All events are filtered with a bandpass filter from 11 to 50 s. The PdP prediction of slowness and traveltime for a 300 km thick D^′′ reflector are indicated by the black arrows. Note that in event #18 the large arrival with PcP slowness is pPcP.

The determination of the sign of the velocity jump across the D^′′ reflector could help to distinguish between different causes for the observed structures (e.g. Ohta et al.2008; Thomas et al.2011; Cobden & Thomas 2013). Some of the waveforms of the PdP arrivals in the data would suggest a positive P-velocity contrast for the D^′′ reflector, that is, the waveform of PdP is the same as P and PcP, (Figs 2b, d and f) and one reflection could be due to a negative P-velocity jump (Fig. 2a). Comparing the synthetic data for a case with a +5 per cent velocity jump in P and S waves and a −5 per cent velocity jump in Fig. 3, we find that the apparent polarity of the PdP does not change despite the opposite sign of the velocity contrast, likely due to small amplitudes and interferences with larger amplitude arrivals. In contrast, it has been shown that for larger epicentral distances the polarity of the PdP wave can be used to estimate the direction of the velocity jump, i.e. positive or negative (e.g. Kito et al.2007; Hutko et al.2008; Cobden & Thomas 2013).

We have chosen to display the vespagrams using contouring of amplitude rather than a stacked trace for each slowness stack. The reason for this is that here the focussing of amplitude is easier to observe, hence slowness estimation is easier. It is, however, in some cases misleading to use contour lines. Fig. 6 shows an example of the same vespagram with contour lines (a) and stacked traces for 4th root (b) and a linear stack (c). In the contour vespagram, the PcP arrival shows only one, not very well defined, arrival whereas in the vespagram consisting of stacked traces, the PcP arrival is clearly the same waveform as P. As these vespagrams are produced using synthetic data, it is clear that using both vespagrams together helps to pick slowness as well as polarities of the waves. Since the model used for generating the synthetics has a positive velocity contrast, one would expect a PdP arrival with the same waveform and polarity as PcP and P. However, due to some larger arrivals with P-slowness values (multiples), the signal of the weak PdP arrival is altered by the stacking process and interference and it could be perceived as a negative polarity arrival in the contour vespagram. Fig. 6(b), however, shows a small upswing before the larger downswing. This again leads us to being more careful with picking polarities in our data in the presence of strong, interfering phases such as depth phases or multiples. Since the Nth root stacking technique alters the waveform, it is advisable to use the waveforms from the linear stack (Figs 6c or 4c) for any further polarity or waveform analyses.

Figure 6.

Comparison of a 4th root contour vespagram (a) (synthetics for event #31) and its corresponding vespagram displaying the stacked traces for each slowness (b). Waveforms are clearer in the lower vespagram, whereas slowness picking is easier in the top figure. (c) A linear stack of the same event.

We estimate depths of the reflectors using the arrival times in the vespagrams. The depth estimates range from 2450 to 2650 km depth. The error for estimating depth in this study is around 50 km due to the difficulty in determining the polarity (and therefore choice at which part of the wavelet to pick) as well as the long periods used here. The depths of the reflectors are given in Table 1 for events where we could detect a PdP arrival.

DISCUSSION

The source–receiver combinations that we studied for the PLUME waveforms yielded seven observations of a clear D^′′ reflection. In four additional events, PdP could not be identified unambiguously. Reasons for this are the presence of additional seismic phases in the time window where PdP is expected (such as multiples and depth phases). Low slowness resolution for the longer-period filters and possible superposition of the waves due to stacking techniques make the identification difficult. The seven events we classify as containing PdP observations have distance ranges from 48° to 78°. Most of these events can be found in data from the first PLUME deployment for which the smaller aperture and station spacing likely allowed more effective stacking. The phase-1 OBSs were typically 75 km apart, while the station spacing for phase 2 was on the order of 200 km. One possible strategy for enhancing imaging fidelity for phase-2 events could be to split the network into smaller subarrays. But since the interstation spacing is larger, the stacking will produce poorer results than the first deployment. We do, however, find one case of a PdP arrival and two possible PdP arrivals in the second deployment (Fig. 2f).

To understand the influence of station aperture we conducted a test using the source mechanism of one event and computed synthetic data for PLUME-1, PLUME-2 and a reduced size PLUME-2, where stations outside the ranges of PLUME-1 were deleted. The results are shown in Figs 7(a)–(c). PLUME-1 produced the best resolution whereas the full size PLUME-2 array, likely due to a larger azimuth range, causes interference of the small-amplitude PdP arrival with a P multiple (Fig. 7b). If PLUME-2 aperture is reduced to PLUME-1 size, the smaller number of stations causes a poorer slowness resolution. In order to investigate the deep Earth with future OBS deployments, the aperture and interstation spacing should be considered, to provide optimal resolution.

Figure 7.

Resolution test for event #23 (synthetics) for the different PLUME deployments using 4th root vespagrams. Top: PLUME-1, middle PLUME-2 and bottom, PLUME-2 stations only that are within the aperture of the PLUME-1 deployment.

Some events provide waveforms of very good quality in the filtered and sometimes unfiltered data set, but no D^′′ reflection is obvious (e.g. Fig. 5). These events generally have short distance ranges and the velocity contrast has to be large (above 2 per cent and even higher for the shorter distances) to produce a visible signal in synthetics (Fig. 3). Reported velocity contrasts across D^′′ are of the order of 1–3 per cent globally (see Wysession et al.1998 for a review) but larger values of up to ±5 per cent in S-wave speed in the Pacific have been found by Bréger & Romanowicz (1998) and 4 per cent were reported by Yamada & Nakanishi (1996). If the reflector were produced by a phase transition of perovskite to post-perovskite (e.g. Murakami et al.2004; Oganov & Ono 2004), the P-velocity contrast would be very small (e.g. Wookey et al.2005), and a small S-velocity contrast has been reported (Murakami et al.2007) which would likely not be detectable at these short distance ranges.

Further complications to detect reflection off D^′′ structures come from the nature of the velocity gradient across the reflector since this will influence the amplitude of the reflected wave (Lay 2008). A gradient with large depth-extent across D^′′, as for example produced by a phase transition (e.g. Catalli et al.2009; Grocholski et al.2012) will further decrease the amplitude of the PdP wave. In our synthetic data, a sharp discontinuity has been assumed since this would produce the largest possible amplitude and the best-case scenario but the amplitude could also point to a larger contrast with a gradient zone. If additionally topography exists on the reflector, the D^′′ reflection could be focused or defocused, again leading to a large variability of the amplitude of the seismic wave (e.g. Thomas & Weber 1997). In addition, anisotropy also has an influence on the reflection coefficient (e.g. Thomas et al.2011) as well as the polarity of the reflected wave and may add to the number of unidentified or absent D^′′ reflections.

Interestingly, while remembering that we need to be cautious with polarities, in some cases the velocity contrast that causes the reflections is associated with a velocity increase (Fig. 8), even though those reflection points occur in parts of the large low velocity region beneath the Pacific Ocean (e.g. Ritsema et al.2011). One would therefore expect a negative velocity contrast for this region although we should also point out that whole-mantle seismic tomography that images the LLSVPs likely misses some of the fine-structure that may exist within them. Since models with negative P- and S-velocity contrasts produce smaller reflection coefficients than positive velocity contrasts, it is possible that some of the events could have reflections with very small amplitude and therefore remain undetectable. Only in cases where a negative P-velocity jump is coupled with a positive S-velocity jump or vice versa would we expect to see a reflection (see Figs 3b–f). Data that agree with positive P-velocity jumps in a low-velocity region have been observed before and have been explained as possible reflection off MORB (e.g. Cobden & Thomas 2013). In the Pacific, Lay et al. (2006) find several reflectors in the D^′′ region near the edge of the LLSVP, both positive and negative and Ohta et al. (2008) explain these reflections as occurring at the transition to MORB material and a post-perovskite transition in pyrolite. Since our reflections have very small amplitudes, we would still exercise caution when interpreting polarities (compare Fig. 4) since stacking in the presence of larger, interfering phases could produce erroneous waveforms in the stacks.

Figure 8.

4th root vespagrams of data (top panel) and synthetic data (bottom panel) for event #13 from Table 1. The D^′′ reflection PdP as seen in the data has been modelled with a positive reflector at 2450 km depth and a positive P-and S-velocity jump of 5 per cent.

Some of our reflection points are near regions where D^′′ has been identified before. For example, Kito & Krüger (2001) find two reflectors with negative velocity contrast near one of our PdP observations in the western Pacific Ocean. Their source–receiver paths (Fiji to Japan) are almost perpendicular to our path (Papua New Guinea to Hawaii, events #32 and #20) and could offer the possibility of testing for anisotropy in D^′′ using P waves (see e.g. Thomas et al.2011) since our data suggest a velocity increase. In contrast to Kito & Krüger (2001) and Kito et al. (2004), Shibutani et al. (1993), Yamada & Nakanishi (1996, 1998) and Takeuchi & Obara (2010), using similar paths, find positive velocity contrasts in P and S velocity in the Southwest Pacific, so some data may need revisiting for mutual, internal consistency.

The timing of the PdP arrival with respect to the P-wave arrival suggests a reflector around 300 km above the CMB (or 2600 km depth) for several of our events. One event shows an arrival from a discontinuity slightly deeper at 250 km above the CMB (event #33) and three have arrivals from shallower reflectors (events #6, #27 and #31). The event that samples a region east of the points sampled by Kito et al. (2004), Takeuchi & Obara (2010) and Yamada & Nakanishi (1996, 1998) indicates a reflector height of 300 km above the CMB while the previous studies find depth ranges of 250–380 km above the CMB (Kito et al.2004), 200–340 km (Takeuchi & Obara 2010) and 170–270 km above the CMB (Yamada & Nakanishi 1996, 1998). An event with a possible D^′′ reflector (event #32) further west and closer to their imaged region indicates a height of 200 km above the CMB, consistent with those previous results.

Several of the places where we find a D^′′ reflector with PLUME data have not been illuminated before with waveform studies. When we compare to the ISC bulletin data of Weber & Körnig (1990, 1992) we find agreement between two of our observations (events #23 and #12) and their predictions in the Northwest Pacific, although our point lies a few degrees further southeast. There are, however, a few studies that find S-wave reflectors near our D^′′ reflector observations (see Wysession et al.1998) although their compilation figure seems to indicate a large scatter between previously reported observations and non-observations of a D^′′ reflector. We find three reflections near the Coast of Central America as indicated in Fig. 1 with the vespagrams shown in Figs 2(b), 3(a) and 8(a). These points show traveltimes consistent with a shallow reflector (between 2450 and 2550 km depth) with possibly a positive velocity contrast of around 4–5 per cent (Fig. 8). This region lies north of the points imaged by several groups (see overview in Kito et al.2007) and is close to the region imaged by van der Hilst et al. (2007) although their reflector depths are slightly lower than the ones we find, but we are several degrees further west. It should be noted that the slowness of PdP is slightly different when comparing synthetics and data for this event although the traveltime agrees in the synthetic data and the observed data. Extensive modelling with variations of velocities and densities above and below the D^′′ reflector or dipping reflectors would be needed to perfectly match the slowness.

Our observations of teleseismic shorter-period PdP waves show that it is possible to use data from OBS experiments for body wave analyses for deep Earth structures. OBS data often exhibit multiples from reflections in the water column (water reverberations) as well as the sediment layer beneath. With water depths between 4500 and 5800 m around Hawaii, the first water multiple is expected to arrive 6–8 s after the P arrival. They can be identified more or less clearly in high-pass filtered PLUME data (see figs S1 and S2 in Wolfe et al.2011). Wolfe et al. (2011) pointed out that these have strongly variable amplitudes. We can identify some water multiples in the data (see vespagrams in Figs 2, 3 and 5—events 12 and 18) as these have slowness values comparable to P waves. The PdP and PcP waves are well separated in slowness and can therefore still be recognized. Effects from sediments will likely coincide with the tail end of the P wave and slightly alter these for the long frequencies used in this study. The sediment layer around Hawaii is typically 250 m thick though it can be 1 km thick in the moat close to the islands. To our knowledge, there is no evidence of significant heterogeneity so the waveforms should be affected equally at all stations we consider here. With more data from new OBS experiments becoming publicly available at the data management centre of the Incorporated Research Institutions for Seismology (IRIS), we expect to expand this study in coming years. However, it will be important to keep the aperture and station spacing of the array in mind for utilizing such short-period waves. New source–receiver combinations to test different hypotheses for lower mantle structures are possible and could be used in the future to distinguish between mineral phase changes, anisotropy, or chemical causes for D^′′ reflectors. We should also mention that the microseism peak is unusually broad near Hawaii so that analyses of other OBS deployment may reach to shorter periods than what we achieve for PLUME.

We have so far not used S waves to study the D^′′ region since we have yet to correct our seismogram database for the misalignment of the horizontal components with geographic north. SKS arrives in a similar time range as the D^′′ reflection SdS for distances around 70°–78°, with a slowness close to ScS. The SdS reflections should be readily observable for positive velocity contrasts, but one would have to be certain of the rotation of the seismometers into radial and transverse component to distinguish between SKS waves and SdS waves.

CONCLUSION

We analysed events recorded by the two PLUME OBS deployments around Hawaii for reflections of P waves at a structure in the lower most mantle, the D^′′ region. The epicentral distance ranges for the events used here is shorter than for many studies of D^′′, and the reflection coefficient is expected to be very small. Nevertheless, we found several events with a PdP arrival, and the timing of several of those is consistent with depths of a reflector at around 300 km above the CMB, but the depth varies between regions. The polarity of the waves appears to be mostly positive (same as PcP and P) and suggests a positive velocity contrast but for these small signals caution should be exercised with using polarities. Modelling suggests that the velocity contrast needs to be large in some of the regions where we see reflections at short epicentral distances. Some events with good signal-to-noise ratio show no reflection which could either mean a small velocity contrast (below 2–3 per cent) or an absence of PdP. We also map previously uncharted regions with our new data set, and in other regions providing the possibility of using crossing paths. This study shows, for the first time, that OBS data can be used to sift teleseismic body waves for reflections off D^′′ discontinuities even though OBS records are said to be affected greatly by ocean noise. OBS data are therefore invaluable to increase our coverage of illuminating regions in the deep Earth.

We would like to thank reviewer Alex Hutko and editor Frank Krüger for their comments which improved the clarity of the manuscript. The PLUME deployment used instruments from the NSF-funded OBS Instrument Pool (OBSIP). Data from the two OBS deployments are distributed through IRIS. The deployments were financed through NSF grant OCE00–02470. GL was supported by grant EAR12–15636. Data were analysed with Seismic Handler (Stammler 1993) and maps were produced with GMT (Wessel & Smith 1995).