Seismic evidence for a mantle plume beneath the Cape Verde hotspot

P- and S-wave tomography of the upper mantle beneath the Cape Verde hotspot is determined using arrival-time data measured precisely from three-component seismograms of 106 distant earthquakes recorded by a local seismic network. Our results show a prominent low-velocity anomaly imaged as a continuous column <100 km wide from the uppermost mantle down to about 500 km beneath Cape Verde, especially below the Fogo active volcano, which erupted in 1995. The low-velocity anomaly may reflect a hot mantle plume feeding the Cape Verde hotspot.


Introduction
The Cape Verde hotspot is located in the African plate, about 2000 km east of the nearest plate boundary (Figure 1). It is composed of a group of late Cenozoic oceanic islands resting on a broad bathymetric swell on mature (>110 Ma) oceanic lithosphere (Mitchell et al. 1983;McNutt 1988;Holm et al. 2008;Müller et al. 2008; Figure 1). This hotspot has a positive surface heat flow (Courtney and White 1986), high geoid anomaly (Crough 1982;Monnereau and Cazenave 1990), and long-term volcanism (e.g. Holm et al. 2008; Figure 1). The last known volcanic eruption occurred at Fogo volcano in 1995.
Such an intra-plate hotspot on an almost stationary plate (Morgan 1983;Pollitz 1991;Gripp and Gordon 2002) is well suited for studying the origin of hotspots without the complexities due to plate motion or plate boundary processes (e.g. Sleep 1990;Ribe et al. 1995;Agrusta et al. 2013;Rychert et al. 2014;Villagómez et al. 2014). Many researchers have investigated the crustal and lithospheric structure under the hotspot using different geophysical methods (e.g. Ali et al. 2003;Lodge and Helffrich 2006;Pim et al. 2008;Wilson et al. 2010;Represas et al. 2012;Vinnik et al. 2012), and proposed several potential mechanisms that are often attributed to a plume origin (Wilson 1963;Morgan 1971). This attribution is supported by many geochemical studies (e.g. Gerlach et al. 1988;Doucelance et al. 2003;Holm et al. 2006Holm et al. , 2008Mata et al. 2010;Mourão et al. 2012a, b), and global tomographic models (e.g. Zhao 2001Zhao , 2007Montelli et al. 2006). However, it is still not clear whether a mantle plume exists or not beneath the Cape Verde hotspot, because of the low resolution of global tomographic models and contradictory receiver-function results on thickness of the mantle transition zone under this hotspot Vinnik et al. 2012). Higher-resolution local tomography under the hotspot may provide useful constraints to solve this debate, such as those for other intra-plate volcanoes (e.g. Gupta et al. 2009;Lei et al. 2009;Wolfe et al. 2009Wolfe et al. , 2011Zhao et al. 2009;Tian and Zhao 2012;Zhao and Tian 2013). However, there is still no local tomography model for the Cape Verde region.
To clarify the origin of the Cape Verde hotspot, in this work we carefully collected many arrival-time data from original seismograms of teleseismic events recorded by a portable seismic network and a Global Seismographic Network (GSN) station installed on the Cape Verde archipelago ( Figure 1). Detailed three-dimensional (3D) P-and S-wave velocity (Vp, Vs) images of the upper mantle under the study area ( Figure 1) were obtained, for the first time, using teleseismic tomography, which shed new light on the deep structure and origin of the Cape Verde hotspot.

Data and method
We used a GSN permanent seismic station (SACV) and seven temporary broadband stations deployed on the Cape Verde archipelago between July 2002 and September 2004 by a collaborative team from institutions in the UK, Cape Verde, Portugal, and Japan (Lodge and Helffrich 2006;Helffrich et al. 2010; Figure 1). Teleseismic waveform data recorded by the eight stations were downloaded from the database of Incorporated Research Institutions for Seismology (IRIS). We selected teleseismic events with epicentral distances of 30-90°and magnitudes greater than M 5.8 ( Figure 2). Hypocentral parameters of the teleseismic events were obtained from the Bulletin of the International Seismological Center (ISC). We measured relative P-and S-wave arrival times accurately from the three-component seismograms filtered in a frequency band of 0.01-0.1 Hz (Figure 3). The measurement accuracy is estimated to be~0.1 s for P-wave and 0.1-0.2 s for S-wave data. Only the seismograms with clear P and S onsets were used. We adopted only those events which were recorded by more than five seismic stations. As a result, we measured a total of 659 P-wave arrivals from 106 teleseismic events and 573 S-wave arrivals from 95 teleseismic events recorded by the eight seismic stations ( Figure 2).
The tomographic method of Zhao et al. (2006Zhao et al. ( , 2012 was used to determine the 3D Vp and Vs models under the study region ( Figure 1). Theoretical travel times were calculated using the iasp91 Earth model (Kennett and Engdahl 1991). Raw travel-time residuals were obtained by subtracting the theoretical travel times from those observed, and relative residuals were obtained for each event by subtracting the event's mean residual from the  (Mitchell et al. 1983;Plesner et al. 2003;Duprat et al. 2007;Holm et al. 2008;Foeken et al. 2009;Madeira et al. 2010), and blue numbers show Moho depth (in km) beneath the individual islands (Lodge and Helffrich 2006). (b) The blue box denotes the present study area, and the white dashed lines denote the plate boundaries (Bird 2003). The topography data are derived from the GEBCO_08 Grid, version 20100927, http://www.gebco.net. The ages of oceanic plates are from Müller et al. (2008). raw residuals. We used the CRUST1.0 model (Laske et al. 2013) to correct the teleseismic relative residuals for the heterogeneous crustal structure in the study area following the approach of Zhao et al. (2006). A 3D grid net was set up in the modelling space, and velocity perturbations from the 1D iasp91 Earth model (Kennett and Engdahl 1991) at every grid node were taken as unknown parameters. The velocity perturbation at any point in the model was computed by linearly interpolating the velocity perturbations at the eight grid nodes surrounding that point. A 3D ray tracing technique was used to compute travel times and ray paths accurately (Zhao et al. 2006). The velocities and depths of crustal layers from the CRUST1.0 model (Laske et al. 2013) and station elevations were taken into account in the 3D ray tracing. The damped least-squares method was used to solve the large and sparse system of observation equations that relate the observed relative residuals to the unknown velocity parameters. Damping and smoothing regularizations were adopted to suppress the dramatic short-scale variations of the velocity anomalies (Zhao et al. 2006. To minimize the effect of the uncertainties of the initial velocity model, the final velocity perturbation at each grid node was calculated from the average of the velocity perturbations at each depth. The root-mean-square (RMS) relative travel-time residual was reduced after tomographic inversion (see Figure S1 in the online supplemental material at http://dx.doi.org/10.1080/ 00206814.2014.930720). Figure 4 shows the distribution of average relative traveltime residuals at the eight stations, which reflects the lateral heterogeneity under the study area. Early arrivals mainly appear at the eastern islands of the Cape Verde archipelago, which have older volcanism (e.g. Holm et al. 2008; Figure 1) and a thicker crust (Lodge and Helffrich 2006; Figure 1). Large delayed arrivals are visible at the western islands, which have a thinner crust and younger volcanic rocks (e.g. Holm et al. 2008; Figure 1), in particular at the Fogo active volcano, which last erupted in 1995, suggesting that a significant low-velocity (low-V) anomaly exists beneath this volcano. Figure 4 also shows the distributions of P-and S-wave ray paths from all the teleseismic events used in this study, indicating that both  (Bird 2003). The number of (c) P-and (d) S-wave arrival times from the teleseismic events is compared to event azimuth. P-and S-rays criss-cross well in the upper mantle under the study area.

Analysis and results
To assess the adequacy of ray coverage and to evaluate the resolution of the tomographic image, we conducted many checkerboard resolution tests (CRTs) following the approach of Zhao et al. (2006Zhao et al. ( , 2012. To perform a CRT, we first assigned positive and negative velocity perturbations of 3% to all 3D grid nodes, then calculated synthetic relative residuals for the checkerboard model with the same numbers of seismic stations, events, and ray paths as those in the real data set, and finally inverted the synthetic data to determine whether the input checkerboard model could be recovered. Figures S4-S6   intervals, which indicate that our tomographic model has a resolution of~100 km in both horizontal and vertical directions. We conducted many tomographic inversions to find the optimal values of the damping and smoothing parameters ( Figure S7). Likewise, the maximum depth of the grid model was also obtained by investigating the trade-off between the RMS relative residual and the maximum model depth ( Figure S8). The optimal tomographic images thus obtained are shown in Figures 5-7. Our results show that the Vp and Vs images are generally similar to each other (Figures 5-7). A prominent low-V anomaly, imaged as a continuous column, is visible under the Cape Verde hotspot, especially beneath the Fogo active volcano from the uppermost mantle down to about 500 km depth (Figure 7). In the optimal 3D Vp and Vs models, the grid interval is 0.5°in the lateral direction and 50 km in depth and the bottom of the grid model is set at 600 km depth. The results with a different grid interval (1°in the lateral direction and 100 km in depth, Figure S12) are similar to the optimal models (Figure 7), indicating that the main features of our results are robust.
We also conducted a restoring resolution test (RRT) (Zhao et al. 2006 to confirm the 3D Vp and Vs images obtained. The RRT input model contains only the major features of the obtained tomographic images, and thus the smearing effects and resolution of the major features can be better known through the test (Zhao et al. 2006. The procedure of the RRT is the same as that of the CRT (Figures S4-S6), except for the input model, which is constructed from the tomographic results (Figure 7). The RRT results ( Figures S13 and S14) show that the main features of velocity anomalies in the obtained tomographic images are well recovered, suggesting that those anomalies are reliable features. In addition, we also conducted many synthetic tests for examining the resolvability of the image along the vertical cross-sections, and the test results (Figures 8 and S15-S19) further confirmed our tomographic images (Figure 7). Lodge and Helffrich (2006) made an analysis of P-to-Sconverted seismic phases with teleseismic waveforms recorded by the same seismic stations used in this study. They found a high-velocity (high-V) layer in the uppermost mantle under the Cape Verde islands. The high-V layer may reflect intrusive bodies at the crust-mantle interface, which is perhaps a ubiquitous feature of hotspot magmatism formed at mature oceanic lithosphere (Richards et al. 2013). Our results (Figures 5a and 6a) show a similar feature except for the low-V anomalies beneath the western islands, which are the most active volcanic and seismic area in the archipelago (Holm et al. 2008;Grevemeyer et al. 2010;Madeira et al. 2010).

Discussion
Receiver functions show a deeper 410 km discontinuity under Cape Verde, suggesting a low-V upper mantle there . Our results (Figure 7) show a columnar low-V anomaly in the upper mantle beneath the Fogo active volcano, being consistent with the receiverfunction result. Our detailed resolution tests indicate that the low-V column is a reliable feature (Figures 8 and S13-  . Distribution of (A) P-and (B) S-wave ray paths (black lines) used in this study. Map views (a, d) and eastwest (b, e) and north-south (c, f) vertical cross-sections. The squares show the seismic stations used. Red and blue shading denotes delayed and early arrivals, respectively. The scale for the relative residuals is shown at the bottom. The red triangles denote the Fogo volcano. Green lines show bathymetric contours, and red dashed lines show the 410 km discontinuity. Note that the depth scale is reduced by~2/3 in the vertical crosssections (b, c, e, and f). S19). This feature is similar to the tomographic images under other hotspots, such as Hawaii (Wolfe et al. 2009(Wolfe et al. , 2011, Iceland (Foulger et al. 2001;Hung et al. 2004;Rickers et al. 2013), the South Pacific (Suetsugu et al. 2009), Afar (Hansen and Nyblade 2013), Yellowstone (Schmandt et al. 2012;Tian and Zhao 2012), the Azores (Yang et al. 2006), Eifel (Ritter et al. 2001;Zhu et al. 2012), Galápagos (Villagómez et al. 2014), Erebus (Gupta et al. 2009), andHainan (Lei et al. 2009;Wei et al. 2012). We believe that the low-V column reflects a hot mantle plume that feeds the Cape Verde hotspot. Wilson et al. (2013) showed that the Cape Verde swell compensation depth is approximately 70 km near the base of the oceanic lithosphere (Schmerr 2012), suggesting the existence of a mantle upwelling that supports the swell (e.g. Courtney and White 1986). Holm et al. (2008) obtained a low magma production rate for this hotspot, which corresponds to a mantle volume flux of 28 m 3 s -1 , much smaller than that for the Iceland (63 m 3 s -1 ) or Hawaii (300 m 3 s -1 ) hotspots (Sleep 1990). The prominent swell is supported by such a weak mantle upwelling, probably due to its quasistationary position between the mantle upwelling and the overlying African plate during a long period, at least more than 20 million year (e.g. Morgan 1983;Pollitz 1991;Gripp and Gordon 2002). We did not attempt to estimate thermal anomalies with the obtained velocity model (Figures 5-7) because the amplitude of velocity anomalies depends to a large extent on the adopted damping and smoothing parameters (e.g. Nataf 2000; Zhao et al. 2006;Nolet et al. 2007;Foulger et al. 2013).
The Cape Verde swell crust is generally thicker under the individual volcanic islands (Lodge and Helffrich 2006; Figure 1), whereas between islands the swell has a normal oceanic crust of~7-8 km thickness (Pim et al. 2008;Wilson et al. 2010). This feature could be related to the episodic swell growth . The swell probably formed in the early Miocene, accompanied by the initial alkaline volcanism significantly acentric with respect to the swell (Lancelot et al. 1978;Ruddiman et al. 1988;Holm et al. 2008). Very few turbidites reaching the top of the swell during the Neogene are consistent with this proposal, implying that the initial uplift of the swell must have taken place earlier (Faugères et al. 1989). Volcanism subsequently became localized and further created individual volcanic islands on the swell . The oldest exposed volcanic rocks are in the eastern islands with thicker oceanic crust, whereas the recent volcanic eruption and seismic activity occurred in the western islands (e.g. Lodge and Helffrich 2006;Holm et al. 2008;Grevemeyer et al. 2010; Figure 1), which seems to show a vague westward age progression. However, many detailed 40 Ar-39 Ar analyses (Mitchell et al. 1983;Plesner et al. 2003;Duprat et al. 2007;Holm et al. 2008;Madeira et al. 2010) show that volcanic periods on the individual islands were sometimes overlapped with each other, suggesting that approximately synchronous eruptions took place on several volcanic islands of this hotspot.
All the above-mentioned features indicate that a hot mantle plume impacts the base of the oceanic lithosphere and supports the Cape Verde swell. The upwelling centre is located close to the Fogo active volcano, to the west of its initial position due to the very slow movement of the African plate during the past >20 million years (e.g. Morgan 1983;Pollitz 1991;Gripp and Gordon 2002). Partial melting probably occurred near the base of the lithosphere (Richards et al. 2013), and may further ascend through some channels in the lithosphere, possibly feeding individual volcanic islands synchronously. The recent volcanism is mainly restricted to Fogo, such as the known eruption in 1995, which coincides with the prominent low-V anomaly beneath it (Figures 5-7). Meanwhile, obvious seismic events in 1998 and 2004 were observed in a submarine volcanic edifice to the southwest of Fogo, which may reflect brittle rock failure associated with hotspot magmatism .
Here a question arises: does the plume feeding the Cape Verde hotspot originate in the lower mantle or just the upper mantle? Isotopic (Sr, Nd, and Pb) and elemental analyses indicate incorporation of ancient recycled crustal materials in Cape Verde magmas (Gerlach et al. 1988;Doucelance et al. 2003Doucelance et al. , 2010Barker et al. 2010;Mourão et al. 2012a), which may be related to a deep mantle plume, though it is hard to reach a conclusion with such isotope systems alone. Some authors (Bonadiman et al. 2005;Coltorti et al. 2010;Martins et al. 2010) have argued that part of the Cape Verde archipelago might be underlain by a fragment of ancient subcontinental lithospheric mantle left behind during the opening of the Atlantic Ocean, which may explain the unusual magma isotopic characteristics without recourse to recycled continental material in the sources of mantle plumes. However, many noble gas analyses Doucelance et al. 2003;Mata et al. 2010;Mourão et al. 2012b), which are probably helpful in the identification of lower mantle materials (e.g. Farley and Neroda 1998;Herzberg et al. 2013), revealed lower mantle materials (e.g. high 3 He/ 4 He ratio) contributing not only to the Cape Verde silicate rocks, but also to the oceanic carbonatites which are found only in the Cape Verde and Canary islands among oceanic islands. Global tomographic models (e.g. Zhao 2001Zhao , 2007Montelli et al. 2006; generally show low-V anomalies in the lower mantle under Cape Verde ( Figure S20), which may indicate a deep mantle plume. It is observed that many surface hotspots, including Cape Verde, are located above the large low shear-velocity provinces in the lowermost mantle (e.g. Thorne et al. 2004;Burke et al. 2008;Garnero and McNamara 2008;Lay and Garnero 2011;Steinberger and Torsvik 2012). However, Helffrich et al. (2010) obtained a normal thickness of the mantle transition zone and suggested that the Cape Verde hotspot does not require vertical thermal anomalies from the lower mantle. With more seismic stations deployed on Cape Verde, Vinnik et al. (2012) argued that the mantle transition zone is up to~30 km thinner than that of the ambient mantle due to a combined effect of a depression of the 410 km discontinuity and an uplift of the 660 km discontinuity, suggesting that a deep mantle plume exists under this hotspot.
Our detailed resolution tests (Figures 8, S4-S6, and S13-S19) show that our tomography model has a resolution of~100 km down to about 500 km depth. A continuous low-V column (diameter~100 km) can be well imaged down to that depth (500 km), but not deeper (Figures 8 and S16-S19) because of the limited aperture of the seismic network used in this study. Our results show that a hot plume exists at least in the upper mantle under the Cape Verde hotspot (Figures 5-7). To clarify the plume features in the deeper part of the transition zone and the lower mantle, we need a seismic network with a larger aperture to image the deeper structure under the Cape Verde hotspot, similar to that deployed in and around the Hawaiian hotspot (Wolfe et al. 2009(Wolfe et al. , 2011 and the USArray in and around the Yellowstone hotspot (e.g. Tian and Zhao 2012;Huang and Zhao 2013). Detection of a deep mantle plume is of fundamental importance for our understanding of not only the origin of hotspots and large igneous provinces, but also the nature of mantle convection and supercontinent assembly and break-up (e.g. Li and Zhong 2009;Tackley 2012;Ernst et al. 2013;).

Conclusions
We determined 3D P-and S-wave velocity images of the upper mantle beneath the Cape Verde hotspot using many arrival-time data collected from original seismograms of teleseismic events recorded by a portable seismic network and a GSN station. A columnar low-V anomaly was revealed from the uppermost mantle down to about 500 km depth under the Cape Verde hotspot, especially beneath the Fogo active volcano. Such a low-V feature in the upper mantle may reflect a hot upwelling plume that played a key role in the formation and continuing activity of the Cape Verde hotspot.