Tectonophysics 636 (2014) 325–337

Contents lists available at ScienceDirect

Tectonophysics

j ourna l homepage: www.e lsev ie r .com/ locate / tecto

Intraplate seismicity across the Cape Verde swell: A contribution from atemporary seismic network

Dina Vales a, Nuno A. Dias b,c,⁎, Inês Rio c, Luís Matias c,d, Graça Silveira b,c, José Madeira c,d, Michael Weber e,f,Fernando Carrilho a, Christian Haberland e

a Instituto Português do Mar e da Atmosfera — IPMA, Lisbon, Portugalb Instituto Superior de Engenharia de Lisboa — ISEL, Lisbon, Portugalc Instituto Dom Luiz— IDL, Lisbon, Portugald Faculdade de Ciências da Universidade de Lisboa, Lisbon, Portugale GeoForschungsZentrum — GFZ, Potsdam, Germanyf University of Potsdam, Postdam, Germany

⁎ Corresponding author at: Department of Physics, InstLisboa, Rua Conselheiro Emidio Navarro, 1, 1959-007218317135; fax: +351 218317138.

E-mail addresses: ndias@adf.isel.pt, ndias@dem.isel.ipl

http://dx.doi.org/10.1016/j.tecto.2014.09.0140040-1951/© 2014 Elsevier B.V. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Article history:Received 3 February 2014Received in revised form 9 September 2014Accepted 19 September 2014Available online 12 October 2014

Keywords:Intraplate seismicityClusteringLocal magnitude scaleActive volcanismCape VerdeAtlantic Ocean

We present an analysis and characterization of the regional seismicity recorded by a temporary broadbandseismic network deployed in the Cape Verde archipelago between November 2007 and September 2008.The detection of earthquakes was based on spectrograms, allowing the discrimination from low-frequencyvolcanic signals, resulting in 358 events of which 265 were located, the magnitudes usually being smaller than 3.For the location, a new 1-D P-velocitymodel was derived for the region showing a crust consistent with an oceaniccrustal structure. The seismicity is locatedmostly offshore the westernmost and geologically youngest areas of thearchipelago, near the islands of Santo Antão and São Vicente in theNWand Brava and Fogo in the SW. The SWclus-ter has a lower occurrence rate and corresponds to seismicity concentrated mainly along an alignment betweenBrava and the Cadamosto seamount presenting normal faulting mechanisms. The existence of the NW cluster, lo-cated offshore SW of Santo Antão, was so far unknown and concentrates around a recently recognized submarinecone field; this cluster presents focal depths extending from the crust to the upper mantle and suggests volcanicunrest. No evident temporal behaviour could be perceived, although the events tend to occur in bursts of activitylasting a few days. In this recording period, no significant activity was detected at Fogo volcano, the most activevolcanic edifice in Cape Verde.The seismicity characteristics point mainly to a volcanic origin. The correlation of the recorded seismicity withactive volcanic structures agreeswith the tendency for awestwardmigration of volcanic activity in the archipelagoas indicated by the geologic record.

© 2014 Elsevier B.V. All rights reserved.

1. Introduction

The Cape Verde archipelago comprises ten volcanic islands andseveral islets located 500 km off the coast of Senegal, West Africa. Thearchipelago stands on the southwest flank of the Cape Verde Rise, anelevated region of the ocean floor approximately 1200 km in diameter.The fact that the islands do not form a linear chain as observed in otherhotspots is interpreted as the effect of a long-livedmantle plumeunder-lying an almost stationary plate (Fig. 1).

The seismicity in Cape Verde is sparse and poorly known; most ofthe information is available only in old reports from 1947 to 1973.Previous data correspond to reports of felt earthquakes of low intensity

ituto Superior de Engenharia deLisbon, Portugal. Tel.: +351

.pt (N.A. Dias).

mainly in Fogo and Brava, in the southwest of the archipelago and typ-ically related to volcanic eruptions (Ferreira, 1956; Mendes, 1956;Neves, 1981), providing intensities and number of events but not loca-tions. Local temporary networks (Fonseca et al., 2003; Heleno, 2001;Heleno and Fonseca, 1999;Helffrich et al., 2006;Matias et al., 1997) pre-viously deployed in the archipelago had very focused objectives, such aslocal seismo-volcanic activity monitoring or teleseismic tomography,with limited results concerning the local seismicity in Cape Verde.Only very recently a permanent seismic network was installed in thearchipelago by the Cape Verdean authorities (National Institute forMeteorology and Geophysics — INMG; Faria et al., 2013; Fonseca et al.,2013; Faria and Fonseca, 2014), but their results are still limited to theassessment of volcanic hazard.

Within the framework of project CV-Plume, a cooperation betweenGeoForschungsZentrum — Potsdam and Instituto Dom Luiz — Lisbon,39 Broad-band (BB) stations were deployed on nine of the ten majorislands of the Cape Verde archipelago between 2007 and 2008

Fig. 1. The Cape Verde archipelago showing the location of the CV-Plume temporary broadband seismic network stations deployed in 2007–2008. Seismicity available in the ISC catalogue1938–2014 is resumed to 24 events (International Seismological Centre, 2011) and registered earthquakes felt in the islands during the 20th century. Top right: location of Cape Verde inthe Atlantic. Bottom right: presumed epicentral area of the 1980–81 seismic crisis and epicentres of the 1995 Fogo volcano eruption.

326 D. Vales et al. / Tectonophysics 636 (2014) 325–337

(cf. Fig. 1). While aimed mainly to image the seismic structure beneaththe archipelago, this dense seismic network covering it entirely, provideda rare opportunity to characterize the seismicity in the region.

Therefore, this work presents the analysis of the data collected bythe CV-Plume BB seismic network during the ten months of the project.From the numerous events detected during the processing stages de-scribed below several low-frequency signals, eventually associatedwith volcanic tremor or other similar sources, were detected but werediscarded for future analysis. We have focused on the signals thatcould be attributed to earthquakes in order to assess the nature of therecorded seismicity, ignoring other signals of possible volcanic origin.Thus, the analysis scheme followed is a typical one for local seismicity:detection stage, phase and amplitude picking, relocation tests with si-multaneous inversion of a new 1D model, waveform cross-correlationto assess the similarity between events, eventual location refinement,and calculation of focal mechanisms for source assessment and correla-tion with geological features.

2. Geological setting

The Cape Verde archipelago, located off the West African coastbetween 14.5° and 17.5° N, is a volcanic archipelago composed of tenislands (nine of which inhabited) and a few islets. The islands, present-ing a horseshoe shaped distribution opened to thewest, are built on topof the largest oceanic swell in the Atlantic Ocean – the Cape Verde Rise –and formed by persistent volcanic activity since the Oligocene (Torreset al., 2002) until present times. The swell and the long-lived volcanismare attributed to a persistent mantle plume underlying an almoststationary plate producing underplating and dynamic uplift (Pim et al.,2008; Ramalho, 2011; Ramalho et al., 2010c). The islands stand on150–122 Ma-old oceanic crust (M2–M16 anomalies; Heyes andRabinowitz, 1975; Stillman et al., 1982).

In terms of their state of evolution, the islands can be divided intotwo main groups: some islands still present youthful strong relief(Santo Antão, São Nicolau, Santiago and Fogo) while at others topogra-phy has been levelled by both subaerial erosion and marine erosion(São Vicente, Santa Luzia, Sal, Boavista, and Maio). Brava, although

presenting a strong relief, has been deeply eroded in a way that itssouthern half corresponds to exhumed magma chambers representedby a plutonic complex of pyroxenites, syenites and carbonatites thatcontacts to the north with pillow lavas and hyaloclastites formingan uplifted submarine sequence. Volcanic rejuvenation in Brava isrepresented by a Middle Pleistocene to Holocene post-erosive volcanicsequence (Madeira et al., 2010; Mourão et al., 2010).

Historical volcanism (postdating settlement in mid-15th century) isreported solely for the island of Fogo, forwhich coeval texts describe theoccurrence of 26 on-going eruptions (Ribeiro, 1960). Eleven of these de-scriptions present sufficient detail to allow mapping of the correlativecones and lava flows (including the most recent events in 1951 and1995; Torres et al., 1997, 1998), while the earliest references (since1500) are very laconic and in some casesmay refer to different observa-tions of the same eruptive episode by ships passing off the island, oreven to an almost continuous state of eruption of Pico do Fogo volcano(Torres et al., 1997; 1998). Geological evidence from the islands ofBrava and Santo Antão, where very recent (probably Holocene) volcanicevents occurred, also points to the presence of active volcanic systems(Holm et al., 2008; Madeira et al., 2010). The remaining islands, withthe exceptions of Boavista and Maio, present post-erosive volcanismof Pleistocene age (Duprat et al., 2007; Holm et al., 2008; Plesner et al.,2003; Torres et al., 2002).

Bathymetricmultibeam surveys of sea-floor areas around the islandsof Santo Antão, São Vicente, Fogo and Brava recognized the presence ofyouthful cone fields to the northwest and southwest of Santo Antão,south of São Nicolau, between the islands of Santiago and Fogo, andbetween Fogo and Brava (Masson et al., 2008), indicating apparentactive volcanic fields on the submarine flanks of these volcanic edifices.A growing seamount – Cadamosto – is located southwest of Brava(Grevemeyer et al., 2010).

Most islands present important but variable uplift resulting in thesubaerial exposure of sedimentary and volcanic submarine sequences(Madeira et al., 2010; Ramalho et al., 2010a, b; Ramalho, 2011). Thedifferent uplift rates of each island are explained by local processeswith a small contribution from a swell-related episodic component(Ramalho, 2011).

327D. Vales et al. / Tectonophysics 636 (2014) 325–337

Seismicity is scarce in the archipelago and has been reported mostlyin the islands of Fogo and Brava. According to published reports(Ferreira, 1956; Mendes, 1956) increased seismic activity occurred inthe 1941–1954 period, with 33 events with maximum macroseismicintensities of VI–VII (Modified Mercalli, MM) felt on the islands of Fogoand Brava in the SW and Santo Antão and São Vicente in the NW. TheNational Meteorological Service database (Serviço MeteorológicoNacional, 1947–1973) also refers to the occurrence of 88 earthquakesin the Fogo–Brava and São Vicente–Santo Antão areas, between 1955and 1971. Most of this seismicity corresponds to a seismic swarm in1963, during which 59 earthquakes were felt in Brava and seven inFogo. Instrumental data obtained during another seismic crisis fromDecember 1980 to May 1981 around the islands of Brava and Fogo,recorded 135 earthquakes presenting magnitudes up to 2.9 (Neves,1981). During the 1995 eruption, 58 events were recorded from April 9to 18, two of which were felt with intensities III–IV (MM) in Fogo(Matias et al., 1997). Other seismic swarms have been attributed tovolcanic activity in the Cadamosto seamount southwest of Brava(Grevemeyer et al., 2010). In summary, since 1941 there are reports offive earthquakes felt in São Vicente (Max. Int. IV–V), 14 in Santo Antão(Max. Int. V), 20 in Fogo (V–VI), and 83 in Brava (V). The data availablein the ISC database (International Seismological Centre, 2011), is veryscarce, since they rely on the global network due to the absence of a per-manent local seismic network, thus recording only some of the strongerevents (cf. Fig. 1).

Neotectonic deformation represented by fault scarps displacingHolocene and/or Late Pleistocene volcanic sequences in Brava, suggeststhat tectonic events cannot be precluded (Madeira et al., 2010). Struc-tural data from the southern group of islands indicate the prevalenceof tectonic or tectono-volcanic structures trending NW–SE, NE–SW,N–S and E–W (Madeira et al., 2010; Represas et al., 2012). Tectono-volcanic structures trending NW–SE are present in the islands of Santi-ago and Maio, including the alignment of volcanic edifices that shapesthe island of Santiago and the alignment of plutonic intrusions in theisland of Maio (Represas et al., 2012). The northwest flank of the islandof Fogo presents conspicuous NW–SE scarps interpreted as a grabenstructure (Brum da Silveira et al., 1997). The southernmost grabenfault is exposed in the caldera wall. NW–SE trending faults are alsopresent in Brava Island, namely the Cachaço and Vigia normal faultsdefining a graben that displaces the southern central area of the island(Madeira et al., 2010). Among other structures in the same island, amajor NE–SW fault zone separates the northwest half of Brava, wherethe submarine volcanic sequence crops out, from the southeast regioncharacterized by outcrops of exhumed magma chambers. Geologicalmapping suggests uplift of the southeast region relatively to thenorthwest, but fault outcrops present slickensides indicating dominantdextral strike slip. These structures displace Middle Pleistocene toHolocene volcanic deposits indicating active faulting. In Fogo Island, amajor N–S trending structure (the Sambango–Monte Vermelho fault;Brum da Silveira et al., 1997; Torres et al., 1998) controls the locationof several aligned Holocene to Upper Pleistocene volcanic centres,including the 2829 m-tall Pico do Fogo, and probably the head of theflank collapse that opened the caldera to the East. E–W faults wereobserved in the northwest coast of Brava (Sorno area) presentingdominant left-lateral displacement.

3. Broad-band seismic network and data recorded

Depending on the conditions at each site, stations of the CV-Plumenetwork were installed on open-field or in shelters (such as thoseprotecting communications antennas or electrical power distributionpoints). Some stations were installed near urban areas or near thecoast, whereas others were installed in the more central areas of theislands; thus, the recorded noise level in each station shows consider-able variability.

Each station was equipped with an Earth Data PR6-24 data loggerand a Guralp CMG-3ESP (60 s) seismometer that recorded data contin-uously with a sampling frequency of 100 Hz. This temporary networkoperated for ten months, from November 2007 to September 2008.During this period, instrumental problems such as timing or failingcomponents affected some stations, mainly in the southern islands,which may partially explain the smaller number of events recorded orlocatable in that area of the archipelago.

In order to detect the local small earthquakes, the continuous datastream was scrutinized using spectrograms: a spectrogram represents,for each seismic trace, an estimate of the average spectral contentaround each sample. Centred in each sample to be analysed, the applica-tion extracts 1024 points (10 s) and calculates the magnitude of theFourier transform of this series, resulting in 512 values between0.1 Hz and 50 Hz. The linear component of the series is removed beforethe spectral calculation. The output file is an image in bitmap format inwhich each strip of 50 pixels represents the desired frequency band. Inour case we analysed the frequencies between 1 and 25 Hz as being themost representative for the detection of local earthquakes, since itcontains their typical frequency content (Fig. 2, top).

For each station, the spectrograms were calculated over one daytime-windows; each daily spectrogramwas then assembled by the con-catenation of each station spectrum with the simultaneous plottingaiming to ease the search for coherent signals (Fig. 2, bottom). Althoughit is possible to simultaneously plot the entire network, to simplify theanalysis we divided the 39 stations in two overlapping groups of 24,one focused on the northern islands and the other on the southern,the overlapping occurring for the eastern stations on Sal and Boavistaislands.

The visual correlation of the records of several stations provides theenhancement of coherent signals; Fig. 2 (bottom) shows the typicalfeatures that can be found on a spectrogram. A local earthquake, or anactive source like an explosion, usually generates a signal with a largefrequency content and duration of a few to tens of seconds, which willbe recorded in several stations. On a spectrogram, this will result on avery marked vertical tight bar that is visible in several stations depend-ing only on their distance to the source. This narrow band in the spec-trum is usually more or less clear due to enhancement of its energy;the analysis of the raw seismograms usually proves to be much harderand quite often not allowing the identification of enough seismic phasesfor an event location. Thus, while one can detect several events on aspectrogram, the actual number of locatable earthquakes is usuallysmaller; the triangles in Fig. 2 (bottom) mark 3 detected events, butonly one effectively locatable (red triangle). Nevertheless, this provedto be a very robust technique that helped overcome the high short-period noise present at many stations, particularly during daytime.Other features also visible in the spectrogram correspond to localnoise produced near or at the station: several blurs, having limited fre-quency content and durations longer than seismic signals and usuallyconfined to a single station or nearby stations, may correspond toatmospheric phenomena or have an anthropic origin.

During the ten month period, 358 events were manually detectedfrom the visual inspection of the spectrograms calculated for the en-tire network, fromwhich it was possible to calculate 265 hypocentresolutions. The manual phase picking was performed by visual in-spection of a time window of about 10 s around the time determinedfrom the spectrograms, with weights attributed to each picking onan empirical basis according to the quality of the signals. Most sig-nals had a low signal-to-noise ratio, usually the S-wave packetbeing easier to identify than the P, resulting in several events beinglocated mainly by the S-arrivals. Therefore, the average accuracy ofthe P-arrivals picks was equivalent to the S-arrivals (∼0.1–0.2 s).First-motion polarities were determined only for signals with highsignal-to-noise ratios, but the problems with the P-arrivals resultedthat very few polarities were really available in the dataset. From therecorded events 19 were identified as corresponding to explosions,

Fig. 2. Top: Example of a spectrogram segment of 8th January 2008, for station CVSA1 located in Santo Antão, corresponding to a zoomingmarkedwith a green rectangle on the spectrogrambelow. The horizontal axis displays the time and the vertical displays the frequency in the range of 1–25 Hz. The colour scale is proportional to the amplitude of the signal, adjusted andnormalized according to thenoise level of each station, blue corresponding to lower and red to thehighest amplitude values. Bottom: Spectogram segment for a set of stations in thenortherngroup, where each station is represented by a horizontal coloured strip. A coherent common trace around 08:13, corresponding to the detection of a local earthquake, is marked by a redtriangle; other weaker coherent traces are marked by yellow triangles.

Fig. 3.Hypocentres of all locatable events registered between Nov. 2007–Sep. 2008 in theCape Verde archipelago (red); final area selected for derivation of the 1D model andevents complying with the selection criteria (black). To the bottom and left of the epicen-tral map are W–E and S–N vertical cross-sections.

328 D. Vales et al. / Tectonophysics 636 (2014) 325–337

218 earthquakes were located inside the study area whereas 28 werelocated outside.

Several records present signals of low-frequency content below5 Hz, typical of volcanic events andmay correspond to volcanic tremor.They were stored and left for a future analysis.

4. Seismicity analysis methods

Following the event detection and identification stage, several 1DP-velocity models were tested for hypocentre location, some ofwhich adapted from studies in the region: a model derived for theAzores crust (Hirn et al., 1980), the model derived for the 1995Fogo eruption (Matias et al., 1997); a model adapted from joint in-version of Ps and Sp receiver functions for the north-western islands(Vinnik et al., 2012); two models adapted from 2D wide-anglemodels (Pim et al., 2008; Wilson et al., 2010). The model that, in firstapproximation, had a lower root mean square (RMS) residualscorresponded to the 1D model for the Azores crust (Hirn et al., 1980),with 0.386 s while these values for the other models varied between0.514 s and 0.662 s. Therefore, thiswas themodel used to obtain the ini-tial hypocentral locations (cf. Fig. 3) using the hypocentre programme(Lienert and Havskov, 1995).

4.1. 1D inversion and event relocation

The image provided by the preliminary locations in Fig. 3 alreadygives an indication about themain seismogenic areas of the archipelago,despite having been obtained from a 1Dmodel that was not specificallyderived for this region. The fact that most of the seismicity is very deepand located already in the mantle raised the question on whether thismodel was the most adequate for the region.

Thus, in order to refine the hypocentral locations and increase theknowledge on the crustal structure in this region, we attempted toderive a new 1-D velocity model using the typical approach of VELEST(Kissling et al., 1994). This method performs a simultaneous inversionfor the 1D seismic velocity model with associated station correctionterms and hypocentral locations.

Since the 1D inversion method relies on stable hypocentre solutionsas input, the criteria adopted for data selection were the choice ofearthquakes recorded by at least five stations and having RMS lowerthan 0.6 (black events in Fig. 3). Considering that all events identified

as explosionswere located on São Vicente Island and that themain seis-micity was located offshore, they were excluded. Based on these condi-tions, two datasets emerged: one composed of 88 earthquakes locatedin the NW part of the archipelago, near Santo Antão, and 30 events inthe SW near Brava. The smaller size of the SW dataset coupled with in-stability of some of its hypocentres, which tended to further reduce thenumber of available events, led us to concentrate only in the group of 88events from the northwest area (Fig. 3).

329D. Vales et al. / Tectonophysics 636 (2014) 325–337

The optimal Vp/Vs ratio was derived from a Wadati diagram(Havskov and Ottemöller, 1999) applied to this set of 88 earthquakes.For a minimum correlation of 0.90, a Vp/Vs value of 1.75 was obtainedfor each event while for a minimum correlation of 0.80 a Vp/Vs ratioof 1.74 was obtained. We chose the last value, in spite the lowercorrelation, since a bigger number of events adjusted to it and also be-cause the numerical response of the several 1D models tested usuallyproduced smaller RMS.

For the 1D inversion, several initial velocity models were tested,based on the ones used for the preliminary location, and adaptedwhen needed (cf. Fig. 4a, left).

The strategy followed for the inversion was strongly conditioned bythe large azimuthal gaps resulting from the seismicity spatial distribu-tion and the seismic network layout, since themajority of the seismicity

Fig. 4. a) 1D inversionmodels. Left: initial inputmodels, with themodel used for the initial locatthe final adopted model (red); b) Spatial distributions of P-wave travel time delay computed fphase readings were not considered. CVSA6 was used as the reference station with a P-wave s

is outside the network (cf. Fig. 3). Moreover, since there are practicallyno events beneath the stations, and they are mainly concentrated SWof Santo Antão Island, the 1D model will correspond to an obliqueaverage of the crustal structure between the seismogenic areas andthe stations.

We started the inversion processwith a total of 1537 readings (1185P and 352 S) and a fixed Vp/Vs of 1.74. The conventional 1D inversionprocedure of VELEST resulted in unstable numerical solutions, beinghard to achieve a stable reduction of the RMS, a consequence of therelative low number of S-readings, essential to constrain focal depthsand therefore the vertical velocity distribution. Since the majority of theevents were deep (Fig. 3; 15 to 40 km), the inverted models were rela-tively insensitive to the mid-crustal input structure, frequently showingnumerical divergence of the RMS values on the 2nd or 3rd iterations.

ion (dashed red); Right: all outputmodels from the simultaneous inversion, together withrom the simultaneous inversion for the CV-Plume 1-D model. Stations with less than fourtation correction fixed to zero.

330 D. Vales et al. / Tectonophysics 636 (2014) 325–337

Therefore, we adopted a procedure that is slightly different from thetypical methodology of Kissling et al. (1994). Instead of testing eachinput model with several varying parameterizations and a high numberof iterations (eachmodel resulting from the inversion also serving as aninput for further iterations), we tested a limited number of variations, insome cases limiting the number of inversions for each initialmodel to beperformed. For the numerical stabilization of the outputmodels, besidesthe introduction of shallow layers to accommodate the island struc-tures, the damping factor of each layer increased with depth.

The numerous models resulting from the inversion are plotted inFig. 4a (right), with the final model adopted in red. The evaluation ofthe obtainedmodels was based on the lower RMS and visual inspectionof the event distribution, favouring the models that simultaneouslydecrease the RMS and that spatially concentrate the events. The initialRMS values for the input models (cf. Fig. 4a, left) varied between 0.5and 0.7 s, decreasing by about 20%–30% to values around 0.2–0.3 sfollowing the inversion; the model chosen as final (CV-Plume modelin Fig. 4a, right; RMS = 0.199 s) already includes the phase delay foreach station (Fig. 4b).

The comparison of the several models resulting from the inversion(plotted in Fig. 4a, right), shows larger variations at the upper andlower crustal levels and relatively smoother values in the mid-crust.The final model has a shallow 1-km thick sedimentary layer with aVp of 3.2 km/s, superimposed on a 1.5 km thick layer with 4.9 km/s;the mid-crust is modelled by a layer with thickness of 5.5 km and aVp of 5.9 km/s, while the lower crust corresponds to a 6 km thicklayer having a Vp of 7.1 km/s, overlying the top of the mantle whichhas a typical velocity of 8.0 km/s. These velocity values are consis-tent and within the range of values predicted for oceanic crusts(Karson, 1998), the model being composed respectively by the su-perposition of layers 2A, 2B, 2C and 3B, although the thickness valuesfor the mid-crust layers exceed the expected values. While thisinconsistency may be due to a real feature, more probably is a resultof the complexity of the 1D inversion and its difficulty in fixing a stablefinal model.

The new velocity model CV-Plume of Fig. 4a (right) was used torelocate the entire recorded seismicity. However, when applied to theevents recorded in the Cadamosto–Brava–Fogo sector it revealed to beinadequate: the RMS of the relocated events increased about 56% com-pared with the ones provided by the model used in the preliminary lo-cation. Also, the epicentral distribution tended to be much sparser andless consistent with the type of locations reported by Grevemeyeret al. (2010). As described below, the attempt to use a location basedon waveform clustering was not particularly successful, and as a conse-quence we opted to use two different velocity models for the finalrelocations: the seismicity recorded on the NW sector of Cape Verde,including the furthermost western events was relocated with the newCV-Plume model (see above), whereas for the events recorded on theSW and the few events dispersed in other areas of the archipelago wehave kept the initial locations provided by the model derived fromHirn et al. (1980).

4.2. Waveform cross-correlation and event clustering

It is widely recognized that earthquakes associated with the sametectonic structure usually present similar waveforms at all stationsrecording those events. To find “identical” events and to correlatethem with the known geological structures, a classification schemewas applied, quantifying the similarity and coupling the events into“clusters”. Such scheme was developed within our group and success-fully applied in the Azores (Dias et al., under review).

The basicmeasurement for the event resemblance is based onwave-form cross-correlation of the seismograms for pairs of events recordedat each station. While this could be performed over the completerecording, to better control the output we considered preferable toapply the cross-correlation to timewindows centred on specific seismic

phases. Considering the energy distribution between the three compo-nents of a seismogram, we used specific tapered time-window lengthsfor each seismic phase: 2.56 s for the P- and 5.12 s for the S-phases,starting respectively 0.5 and 0.8 s before the phase onset. The recordswere band-pass filtered between 2 and 10 Hz for the P- and 1–6 Hzfor the S-phases; in the case of the P-phase the time window wasapplied only to the vertical component, whereas for the S-wave it wasapplied to both horizontal components. Since the correlation may beaffected by several problems, such as data gaps, low signal-to-noiseratios, and spikes, the final similarity measure considered was basedon a weighted average of the cross-correlations computed for thewhole network, building a matrix such as the one shown in Fig. 5a.The partition of the events into clusters was based on that similaritycoefficient, with the exact value being selected upon a trial-and-errorprocess: several values were tested, ranging from 0.5 to 0.9 (only thecomparison of an event with itself reaching 1.0), with inspection ofthe resulting matrix aided also by dendrograms (see e.g. Dias et al.,under review, and references therein) and visual inspection of thewaveforms of the resulting clusters. In similar clustering studies con-ducted on tectonically generated events (e.g. Dias et al., under review,and references therein), this type of analysis is based on a sequentialclustering S–P procedure, i.e., a first partition based on the S-wave sim-ilarities followed by a refinement based on the P-wave. The basic reasonis that S-waves, being more complex and dependent on the fault struc-ture, are the main characterization property to be used to discriminatebetween events; the P-wave is then used merely for a refinement ofthe S-clusters, namely for the source type.

In our study of the Cape Verde seismic data, we found that the use ofthe S-wave as primary discriminator tended to split the events intoseveral small clusters, mainly doublets or single events. Therefore, thebest results were obtained by inverting the sequence analysis, i.e., byfirst computing the similarity of the P-phase, defining a first partitionof P-clusters, followed by a refinement in each P-cluster by computingthe S-phase similarity. All final clusters were thus obtained from a P–Ssequence analysis, ensuring that all the events in each cluster belongto the same family. In Fig. 5a to c we present the stepwise output ofthe P–S clustering procedure.

Fig. 5a shows the first step of the clustering process, with a graphicalrepresentation of the P-wave similarity analysis, with a minimumsimilarity cut-off value of 0.6 obtained by a trial-and-error process.The “hot-colour” squares, matching events with similar P-waves, corre-spond to preliminary clusters that are already close to the sequencesobtained in the final clustering process; however, since the P-wavehas amuch smaller complexity than the S-wave, there are still somedif-ficulties in discriminating events at this stage, resulting in somemixturebetween clusters.

Fig. 5b presents the final sequences after the S-phase similarityanalysis was applied to the preliminary P-clusters: the majority of theP–S sequences dispersed, mainly into doublets and single events,remaining only ten sequences with at least 3 events. Of those, onlyfour sequences remained intact through the P–S clustering process.The two sequences corresponding to explosions in São Vicente, S1 andS5, due to its size and identical P and S waveforms presented by allevents, were used as an output quality control and validation of thediscriminating capability of the method.

Sequence S2 gives an example of the changes occurring during thesequential P–S clustering process. Following the P-stage, this sequencecontained 19 events mainly concentrated near Brava Island, but pre-senting some scattered around the archipelago; following the sequen-tial S-stage, the number of events was significantly reduced from 19 to7 events with practically identical waveforms, eliminating the outliers,but at the cost of also eliminating a few close events (cf. Fig. 5b). Thisfinal sequence, S2, corresponds also to the more numerous final cluster,thus deserving special attention. Fig. 5c shows the P-waveforms for theevents of this sequence recorded by a station installed in Brava Island(CVBR1), all showing a compression first motion polarity. This

Fig. 5.Waveform similarity analysis and event clustering procedure: a) plot of the similaritymatrix for the P-wave of the 88 events analysed in Cape Verde. Thematrixwas built by cross-correlating P-pulses, over tapered timewindows on the verticalcomponent, between pairs of events and later averaging for the entire network. The nodes were later re-ordered according to the hierarchical classification obtained by the clustering procedure, with a minimum similarity value of 0.6 obtained bytrial-and-error; b) Epicentral map of the 10 clusters obtained from the P–S sequential clustering procedure, excluding all doublets and single events. The numbers between brackets correspond to the number of events contained in each sequence;c) P-waveforms for the best P–S final cluster located near Brava Island (S2), aligned on the P-onset.

331D.V

alesetal./Tectonophysics

636(2014)

325–337

332 D. Vales et al. / Tectonophysics 636 (2014) 325–337

set allowed the calculation of a composite focalmechanism. The generallow coherence of the waveforms even of neighbouring events (seeabove) did not allow using high-precision relative relocation based onwaveform cross-correlations.

4.3. Focal mechanisms

To compute focal mechanismswe used the FOCMEC programme, in-cluded in the SEISAN software package (Havskov andOttemöller, 1999),in which all the compatible solutions with a given P-arrival polaritydistribution are obtained. These best solutions were later used as initialsolutions in the MECSTA programme (Brillinger et al., 1980) to deter-mine themost viable solution for the focal mechanism (nodal plane ori-entations and also the orientation of the principal stress axes, P and T).

For the calculation of the focal mechanisms, the selection criterionfollowed was to use only events with a minimum of nine polaritiesreported, so a selection of just four events was obtained due to the dif-ficulty in marking the polarities. In order to increase the number ofsource mechanisms, we tried to calculate composite focal mechanismsfor the final clusters obtained fromwaveform cross-correlation process;however, these clusters suffered from the same problem of reducednumber of determined polarities, and only the biggest cluster S2 locatednear Brava had enough polarities to allow a composite focal mechanismto be calculated.

The four earthquake locations and their corresponding focal mecha-nism parameters are listed in Table 1, together with the compositefocal mechanism parameters. The nodal planes and the distribution ofP-wave polarities are represented in Fig. 6, where we present the calcu-lated single fault-plane solutions and a composite solution of the bestcluster obtained from the waveform similarity analysis. The compositefocal mechanism position corresponds to the cluster centroid of S2 ofFig. 5b.

4.4. New local magnitude scale

The local magnitude (ML) scale was calibrated using the approachdescribed in Hutton and Boore (1987) and Carrilho and Vales (2009).According to this method, anML= 2 earthquake is defined as a seismicevent that produces a maximum amplitude of 1 mm on a Wood–Anderson seismometer located 17 km from the epicentre. Therefore,to calculate the local event magnitudes for m stations, we used thefollowing formula (1):

ML ¼1m

Xmj¼1

log Ai j

� �þ n log

Δi j

17

� �þ k Δi j−17

� �þ 2:0þ Sj

� �ð1Þ

where A is the maximum amplitude in μm/s, Δ is the hypocentraldistance in km, n and k are respectively geometric and anelastic attenu-ation coefficients, S is a station correction, i is the earthquake numberand j the station number.

The attenuation coefficients are determined, using data recorded bythe CV-Plume network, by minimizing Eq. (2) with constraints on theparameter ranges:

σ2 ¼Xni¼1

Xmj¼1

log Ai j

� �þ n log

Δi j

17

� �þ k Δi j−17

� �þ Sj þ 2:0−ML

� �� �2: ð2Þ

Table 1Location and source parameters for four earthquakes and one composite focal mechanism.

Date Time Latitude (° N) Longitude (° E)

2008-04-15 00:33:54 14.201 −24.1762008-05-01 15:10:41 16.672 −25.4922008-05-14 06:51:06 16.702 −25.4102008-08-28 07:36:19 14.978 −24.993Composite

From this joint inversion, performed with σ = 0.266, the obtainedattenuation coefficients were n = 1.685 (geometric spreading) andk = 0.00 (anelastic attenuation). Thus, anelastic attenuation is notrelevant in the computation of local magnitude for this region. The coef-ficient of geometrical spreading is higher than the one inferred byHutton and Boore (1987) for California (n= 1.110) and anelastic atten-uation is lower (k = 0.00189). Fig. 7a shows a comparison of the MLattenuation correction, represented as −Log(A0), for the two regions.The local-magnitude station correction values are listed in Table 2.

Within the distance range of the data from Cape Verde, the majordifference in the attenuation correction occurs within the first fewkilometres, although not very well constrained due to few availableevent data for such distances. The impact of the new attenuationcurve on the MLmagnitudes can be observed in the histogram present-ed in Fig. 7b. Most events have higher ML values (average correctionML ~ 0.25). The few exceptions correspond to the few local eventslocated in Fogo Island. With the exception of Fig. 1, all other figureswith magnitude scale indication were performed already with thisnew magnitude scale.

5. Discussion

The obtained results show that the seismicity recorded during theten month period in 2007–2008 was mainly concentrated in two off-shore areas, roughly coincident with the western tips of the northernand southern groups of islands of the Cape Verde archipelago. No signif-icant seismicity was recorded in other areas of the archipelago, which isconsistentwith the activity previously reported for the region. Themostabundant seismicity occurred in the NW area around the islands ofSanto Antão and São Vicente, with a high number of events distributedover a wider area. In the south-western area, near the islands of Bravaand Fogo, the seismicity presents fewer events with a greater concen-tration in time and space (cf. Fig. 3 and Graphical abstract).

Due to the horseshoe shaped disposition of the islands, to the factthat only land stations were available, and because the majority of theevents occurred outside the archipelago, the azimuthal coverage ofthe recorded seismicity was not optimal and disturbed the relocationprocedure. Therefore, several methodologies were conducted in orderto characterize the recorded seismicity as best as possible.

The first step was to develop a new 1D velocity model suitable forthe region and simultaneously increase the quality of the hypocentralsolutions. Such a model could only be derived for the north-westernarea, due to the previously referred constraints and to the lowerseismicity rate recorded in the south-western area. Considering theuncertainties involved in the inversion process, the newly derived 1Dmodel – CV-Plume – should be considered a mathematical response tothe spatial disposition of the north-western seismicity, and its interpre-tation in terms of the crustal structure should be done with caution.Nevertheless, considering the obtained values and following Karson(1998) the values are consistent with the expected structure of an oce-anic crustal type: an upper crust composed of two layers of mainly ba-saltic composition, layers 2A and 2B, and a lower crust of gabbroiccomposition eventually interlayed with ultramafic rocks, layer 3B; themiddle-crust corresponds to a layer marking the transition betweenthese two domains, eventually more complex than the simple corre-spondence to layer 2C. While the CV-Plume model has the best

Depth (km) ML Strike (°) Dip (°) Rake (°)

48 3.9 163 18 6530.5 3.3 182 15 −2132.3 3.6 353 46 −1023 3.2 188 80 2

164 39 −83

Fig. 6. Top: P wave polarities and focal mechanism solutions obtained for the four single events as well as for the composite (in blue), which corresponds to a cluster near Brava. Bottom:location of the computed fault-plane solutions and the composite focal mechanism.

Fig. 7. a) Comparison of theML attenuation correction for Cape Verde and California; b) Distribution of the ML changes after applying the new set of parameters compared to Hutton andBoore (1987).

333D. Vales et al. / Tectonophysics 636 (2014) 325–337

Table 2Station corrections coefficients derived for the local magnitude scale of Cape Verde.

Santo Antão São Vicente São Nicolau Sal

Station code Mag correction Station code Mag correction Station code Mag correction Station code Mag correction

CVSA1 0.26 CVSV1 0.22 CVSN1 0.16 CVSL1 0.29CVSA2 0.14 CVSV2 0.17 CVSN2 −0.40 CVSL2 0.13CVSA3 0.05 CVSV3 0.00 CVSN3 −0.06 CVSL3 0.01CVSA4 0.28 CVSV4 0.17 CVSN4 0.14 CVSL4 0.40CVSA5 0.18CVSA6 −0.04

Brava Fogo Maio Santiago Boavista

Station code Mag correction Station code Mag correction Station code Mag correction Station code Mag correction Station code Mag correction

CVBR1 −0.18 CVFG1 −0.24 CVMA1 −0.17 CVST1 −0.18 CVBV1 0.29CVBR2 – CVFG2 −0.40 CVMA2 −0.04 CVST2 0.26 CVBV2 −0.28

CVFG3 −0.21 CVMA3 0.04 CVST3 0.11 CVBV4 0.22CVFG4 −0.40 CVMA4 0.01 CVST4 0.22 CVBV5 0.26CVFG5 −0.40 CVST5 0.20 CVBV6 –

334 D. Vales et al. / Tectonophysics 636 (2014) 325–337

numerical performance for the events located in the northern area, forthe southern seismicity the model derived by Hirn et al. (1980) fromthe Azores had the best numerical results in terms of RMS values andproduces a smaller spatial dispersion. Therefore, the final locations ofthe events were obtained using two different models, one for eacharea: CV-Plume in the north and the model from the Azores for thesouth.

Our attempt to discriminate similar events based on waveformcross-correlation and to try to associate the resulting groups or familieswith specific structures had ambiguous results. Unlike tectonic environ-ments, where each structure presents a constant behaviour, i.e., a givenfault tends to present the same type of slip producing waves with thesame basic characteristics (notwithstanding variations due to theevent magnitude), volcanic events may display considerable variation.For example, events associated with magma transfer within themagmatic plumbing system produce very different signals in the samestructure due to expansion and contraction. An example of differentwaveforms associated with two P–S sequences roughly at the same lo-cation, S8 and S10 in Fig 5b, can be observed in the supplementary ma-terial provided (cf. Supp. Fig. 1). The S-waveforms shownwere recordedat stationCVSA1 and band-passfiltered between 1 and 6Hzwhich is thesame used in the similarity analysis. The difficulty in aggregating eventsinto similar clusters, togetherwith the fact that different clusters locatedin a given position usually showdifferent waveforms, suggest a possiblevolcanic origin for some of the recorded seismicity.

The four individual focal mechanisms and the composite solutioncomputed for one of the sequences (cf. Fig. 6) show diversity in faultingstyles: normal, reverse and strike-slip. Given the large geographicaldispersion of the events, their low-magnitudes and the non-optimaldata coverage reliable conclusions on local and regional tectonics can-not be inferred from the results. The reverse solution is a particularcase, since it corresponds to an isolated event located SSE of Fogo Islandwhere no relevant information, such as high-resolution bathymetry, isavailable and therefore we discarded it.

Fig. 8 summarizes the results obtained in this work, showing boththe final locations of the seismicity in the NW (left) and the SW(right) areas, and corresponding focal mechanisms.

The north-western area (Fig. 8, left) is dominated by a major clusterlocated south-west of Santo Antão and São Vicente islands, with themajority of the focal depths between 10 and 30 km including P–S se-quences S8 and S10, as well as two focal mechanisms calculated forthe NW area. Both solutions present oblique normal and strike-slipmovement, but it is unclear which is the dominant component. Themain earthquake cluster coincides with a recently identified volcaniccone field in the multibeam bathymetry covering the slopes of thewestern Cape Verde Islands by Masson et al. (2008); this coincidence(stars in Fig. 8, left) may represent ongoing magmatic activity in this

volcanic system. This activity seems to be continuous, since it is alsoreported since 2008 (Faria et al., 2013), which may be an indicator ofvolcanic unrest at this submarine cone field. In addition, some of theevents with higher magnitude were also located in this cluster, the re-maining corresponding to the westernmost recorded events. Since weexcluded from the analysis all low-frequency signals, phenomena likevolcanic tremor are excluded and more probably these events are asso-ciated with fault slip affecting the volcanic system. The great depths ofthe hypocentres can, therefore, be due to effects induced by a magmachamber or by deep volcanic intrusions.

The alignments presented by some events inside this cluster areprobably due to the low azimuthal coverage and probably do not corre-spond to real features; alternatively, since the orientation of the nodalplanes of both focal mechanisms seems to be coherent with some ofthese alignments, theymay represent events at aligned volcanic edificesor planar feeding structures.

Additional smaller clusters can be perceived to the west, north-westandnorth of Santo Antão (Fig. 8, left and Supp. Fig. 2),whichmay also berelated to volcanic edifices, and coincident with P–S sequences S9, S6and S3 respectively (cf. Fig. 5b); however, since they correspond tosome of the cases of worst instrumental azimuth coverage, it is hardto be certain. Their focal depths of zero clearly show the uncertaintyof their location. The only certainty is that they do not seem to be asso-ciated with the Nola seamount nor to the submarine volcanic cones inits vicinity. The tight clusters located on São Vicente correspond to a se-ries of blasts associated with ongoing works for the enlargement of theairport. To the south of São Vicente there is a smaller group of events,corresponding to sequence S4, which is not related with either of theprevious clusters; they may also be associated with submarine volcaniccones, but they are already outside of the area surveyed byMasson et al.(2008).

In the south-western area (Fig. 8, right), besides the major NE–SWtrending cluster between Cadamosto Seamount and Brava, coincidentwith sequences S2 and S7, a conspicuous transverse NW–SE alignmentis evident immediately to the north of Cadamosto. The NE–SW align-ment may represent a volcano-tectonic structure, possibly a commonmagmatic plumbing system shared by three aligned volcanic edifices(Seco and Rombo Islets, Brava, and Cadamosto). The elongated shapeof Cadamosto seamount also trends NE–SW, corroborating a probablemagmatic system linking those volcanic edifices, which present agesprogressively younger towards the south-west. The transverse align-ment could correspond to a NW–SE trending fault, possibly belongingto the same fault system as Vigia Fault in Brava (see Section 2), a tectonicdirection common to Brava and Fogo Islands. Furthermore, severalepicentres just west of Brava seem to be aligned with the most promi-nent fault scarp (the Vigia Fault; Madeira et al., 2010; Fig. 8b). The com-posite focal mechanism suggests a pure normal movement, which is in

Fig. 8. Epicentral maps of the seismicity detected in Cape Verde during the CV-Plume experiment and associated focal mechanisms together with vertical W–E profiles with the depthdistribution. Left: NW areawith the events located with the new CV-Plumemodel derived in this study. The green stars correspond to the submarine volcanic cones identified byMassonet al. (2008). NS, Nola seamount; SA, Santo Antão; SV, São Vicente; SL, Santa Luzia; SN, São Nicolau; Right: SWareawith the events locatedwith the original 1Dmodel of Hirn et al. (1980),overlying the bathymetry of Grevemeyer et al. (2010). CS, Cadamosto seamount; B, Brava, and F, Fogo. Island topography from ASTER Global Digital Elevation Model v001 (NASA, 2009).

335D. Vales et al. / Tectonophysics 636 (2014) 325–337

agreementwith field evidence for normal faulting in the Vigia Fault. Thefocal depths also suggest two different sources for this seismicity: theNE–SW events associated with a shallower volcano-tectonic structurewhereas the deeper events around Cadamosto seamount may have apure volcanic origin. Surprisingly, no significant activity was recordedat Fogo volcano, the most active volcano in the archipelago. The singleevent source information available in this area points to a pure strike-slipmovement; anddue to its location outside thebathymetric surveyedarea by Grevemeyer et al. (2010) and to the fact that this focal mecha-nism solution corresponds to the onewith fewer polarities, no final con-clusion about its nature can be made.

An attempt was made to check whether this seismicity, and in par-ticularly the final sequences obtained, was clustered in time besidesbeing clustered in space. Based on the spatial distribution of therelocated seismicity (cf. Figs. 3 and 8) and on the position of the P–Ssequences (cf. Fig. 5b) several seismogenic sectors were empirically de-fined and the temporal evolution of the contained events analysed.While there are some bursts of activity, no obvious temporal evolutioncan be perceived. The results are summarized in the supplementarymaterial that accompanies this work (cf. Supp. Fig. 2).

6. Conclusions

In this work, we present the first evaluation of the seismicity acrossthe Cape Verde archipelago at a regional scale, using data recorded be-tween 2007 and 2008 by a temporary network of 39 BB stations. Forthe detection of seismic events, we applied a spectrogram technique,allowing a thorough screening of all seismicity detected in the regionduring that period. A standard approach was followed, based on

Geiger's method and inverting for a new 1D crustal model, calledCV-Plume, for the analysis of the detected events around the north-western islands. For a better characterization of the aggregation ofevents in time and space, a waveform similarity method was appliedto define clusters of similar events which, coupled with determinationof single and composite focal mechanisms, allowed a better constraintof the associated sources. In addition, attenuation coefficients requiredfor localmagnitude evaluationwere also derived, allowing the determi-nation of event magnitudes with a scale specific for this region.

During the survey period it is shown that themajority of the seismic-ity was concentrated in thewesternmost areas of the archipelago, at theterminations of the northern and southern groups of islands. No signif-icant activity was detected associated with the most active volcano ofFogo or the island of Santiago. As previous reports suggested, norelevant seismic activity is registered in the easternmost islands of Sal,Boavista andMaio. In terms of signal and occurrence, the recorded seis-micity presents characteristics that suggest a dominant volcanic origin.

In the south-western area of the archipelago, the seismicity presentsa lower occurrence rate and is concentrated mainly along an alignmentbetween the Cadamosto seamount and Brava. The distribution ofepicentres in this area suggests the presence of active linear tectono-volcanic structures trending NE–SW and NW–SE. These are interpretedas indicating the presence of an active volcanic system shared by Bravaand Cadamosto seamount and NW–SE trending active faults. Accordingto the composite focal mechanism, the movement associated with thisstructure points to a normal faulting type.

In the north-western area of Cape Verde, the recorded seismicitypresents a greater activity rate but is dispersed over a wider geograph-ical area. The most important clusters are concentrated south–

336 D. Vales et al. / Tectonophysics 636 (2014) 325–337

southwest of Santo Antão and São Vicente islands and the comparisonwith recent high resolution bathymetric surveys shows that they are as-sociated with a recently recognized submarine cone field, discoveredand named “Charles Darwin Volcanic Field” during a RRS CharlesDarwin 2005 cruise (Masson et al., 2008; Hansteen and Kwasnitschka,2013; Kwasnitschka et al., 2013). The clustering of seismicity in timeand space, suggests that the events in this cluster are the manifestationof unrest in that volcanic structure, with the focal depths indicating anorigin deep within the magmatic system.

Additional activity is occurring west of Santo Antão, probably amanifestation of similar volcanic signals, but no activity was observedrelated to the Nola seamount. This seismic activity in the NW hadbeen previously described, the few reports referring to rarely felt eventsin the volcanic edifices of Santo Antão and São Vicente islands.

The correlation with a volcanic origin suggests on-going magmaticactivity in the western tips of the two branches of archipelago, associat-ed with the westernmost islands and active seamounts or submarineconefields. This is in agreementwith the location of the youngest volca-nicmanifestations in thewesternmost islands, while the eastern islandsare dominantly older (Mitchell et al., 1983; Plesner et al., 2003; Dupratet al., 2007; Holm et al., 2006, 2008; Dyhr and Holm, 2010; Madeiraet al., 2010).

Supplementary data to this article can be found online at http://dx.doi.org/10.1016/j.tecto.2014.09.014.

Acknowledgements

This workwas supported by the Portuguese Science and TechnologyFoundation (FCT/Fundação para a Ciência e Tecnologia), under researchproject “CV-PLUME: An investigation on the geometry and deep signa-ture of the Cape Verde mantle plume” (PTDC/CTE-GIN/64330/2006).This is a contribution to PEST – OE/CTE/LA0019/2013–2014 – IDL. Theinstruments for the deployment were provided by the GeophysicalInstrument Pool Potsdam, Germany. The figures were produced usingthe software packages GMT (Wessel and Smith, 1998) and SEISAN(Havskov and Ottemöller, 1999).

We would like to acknowledge the effort of all the elements of theCV-Plume and COBO teams involved in the field operations of theseismic experiment that took place in the Cape Verde in 2007–2008.We are in debt with ELECTRA, the Cape Verde electrical company, bythe support provided and the invaluable aid of Professors AntónioLobo de Pina, Sónia Silva and Vera Alfama from the University of CapeVerde.

In addition, the authorswould also like to thankAgust Gudmundssonand an anonymous reviewer for their valuable revision and suggestionsto improve the quality of the initial manuscript.

References

Brillinger, D.R., Udias, A., Bolt, B.A., 1980. Probability model for regional focal mechanismsolutions. Bull. Seismol. Soc. Am. 70 (1), 149–170.

Brumda Silveira, A., Madeira, J., Serralheiro, A., 1997. A estrutura da ilha do Fogo, Cabo Verde.Proceedings of the 1st International Symposium “A erupção vulcânica de 1995 na ilha doFogo, Cabo Verde”, edited by Instituto de Investigação Científica Tropical and Ministérioda Ciência e Tecnologia 972-672-863-0,pp. 63–78, (in Portuguese).

Carrilho, F., Vales, D., 2009. Calibration of magnitude ML for Portugal mainland andadjacent areas. 6th Symposium of the Portuguese Association of Meteorology and,Geophysics. 48.

Dias, N.A., Matias, L., Téllez, J., 2014. Insight on the crustal strain in Faial and Pico Islands(Azores), from the 1998 aftershocks waveform similarities and shear-wave splitting.subm. Geophys. J. Int. (under review).

Duprat, H.I., Friis, J., Holm, P.M., Grandvuinet, T., Sørensen, R.V., 2007. The volcanic andgeochemical development of São Nicolau, Cape Verde Islands: constraints fromfield and 40Ar/39Ar evidence. J. Volcanol. Geotherm. Res. 162 (1–2), 1–19. http://dx.doi.org/10.1016/j.jvolgeores.2007.01.001.

Dyhr, C.T., Holm, P.M., 2010. A volcanological and geochemical investigation of Boa Vista,Cape Verde Islands; 40Ar/39Ar geochronology and field constraints. J. Volcanol.Geotherm. Res. 189, 19–32. http://dx.doi.org/10.1016/j.jvolgeores.2009.10.010.

Faria, B., Fonseca, J.F.B.D., 2014. Investigating volcanic hazard in Cape Verde Islandsthrough geophysical monitoring: network description and first results. Nat. HazardsEarth Syst. Sci. 14, 485–499. http://dx.doi.org/10.5194/nhess-14-485-2014.

Faria, B.V.E., Day, S., Fonseca, J.F.B.D., 2013. Insights from Geophysical Monitoring into theVolcano Structure and Magma Supply Systems at Three Very Different OceanicIslands in Cape Verde Archipelago. AGU2013 Fall Meeting, San Francisco, California,USA, pp. 9–13 (December, Abstract V41B-2786).

Ferreira, H.A., 1956. Macrosismos sentidos em Portugal no período de 1901–1945.Boletim da Ordem dos Engenheiros. 5(5), pp. 1–10, Lisboa.

Fonseca, J.F.B.D., Faria, B.V.E., Lima, N.P., Heleno, S.I.N., Lazaro, C., d'Oreye, N.F., Ferreira,A.M.G., Barros, I.J.M., Santos, P., Bandomo, Z., Day, S.J., Osorio, J.P., Baio, M., Matos,J.L.G., 2003. Multiparameter monitoring of Fogo Island, Cape Verde, for volcanic riskmitigation. J. Volcanol. Geotherm. Res. 125, 39–56.

Fonseca, J.F.B.D., Faria, B.V.E., Trindade, J., Cruz, G., Chambel, A., Silva, F.M., Pereira, R.L.,Vazão, T., 2013. “Last mile” challenges to in situ volcanic data transmission. Nat.Hazards Earth Syst. Sci. 13, 3419–3428.

Grevemeyer, I., Helffrich, G., Faria, B., Booth-Rea, G., Schnabel, M., Weinrebe, W., 2010.Seismic activity at Cadamosto seamount near Fogo Island, Cape Verdes-formationof a new ocean island? Geophys. J. Int. 180 (2), 552–558. http://dx.doi.org/10.1111/j.1365-246X.2009.04440.x.

Hansteen, T.H., Kwasnitschka, T., Klügel, A. (Eds.), 2013. Cape Verde Seamounts— R/V Meteor cruise No. 80 Leg 3, 29 December–1 February 2010, unpublishedreport.

Havskov, J., Ottemöller, L., 1999. Seisan earthquake analysis software. Seismol. Res. Lett.70, 532–534. http://dx.doi.org/10.1785/gssrl.70.5.532.

Heleno, S.I.N., 2001. Estudo Sismológico do Vulcão do Fogo(PhD Thesis) IST, Lisbon,(in Portuguese).

Heleno, S.I.N., Fonseca, J.F.B.D., 1999. A seismological investigation of the Fogo Volcano,Cape Verde Islands: preliminary results. Volcanol. Seismol. 20, 199–217.

Helffrich, G., Heleno, S.I.N., Fonseca, J.F.B.D., 2006. Hydroacoustic detection of volcanicocean-island earthquakes. Geophys. J. Int. 167 (3), 1529–1536. http://dx.doi.org/10.1111/j.1365-246X.2006.03228.x.

Heyes, D.E., Rabinowitz, P.D., 1975. Mesozoic magnetic lineations and the magnetic quietzone off northwest Africa. Earth Planet. Sci. Lett. 28, 105–115. http://dx.doi.org/10.1016/0012-821X(75)90217-4.

Hirn, A., Haessler, H., Hoangtrong, P., Wittlinger, G., Mendes-Victor, L.A., 1980. Aftershocksequence of the January 1st, 1980, earthquake and present-day tectonics inthe Azores. Geophys. Res. Lett. 7 (7), 501–504. http://dx.doi.org/10.1029/GL007i007p00501.

Holm, P.M., Wilson, J.R., Christensen, B.P., Hansen, L., Hansen, S.L., Hein, K.M., Mortensen,A.K., Pedersen, R., Plesner, S., Runge, M.K., 2006. Sampling the Cape Verde mantleplume: evolution of melt compositions on Santo Antão, Cape Verde Islands. J. Petrol.47, 145–189. http://dx.doi.org/10.1093/petrology/egi071.

Holm, P.M., Grandvuinet, T., Friis, J., Wilson, J.R., Barker, A.K., Plesner, S., 2008.An 40Ar–39Ar study of the Cape Verde hot spot: temporal evolution in asemistationary plate environment. J. Geophys. Res. 113, B08201. http://dx.doi.org/10.1029/2007JB005339.

Hutton, L.K., Boore, D.M., 1987. TheML scale in southern California. Bull. Seismol. Soc. Am.77 (6), 2074–2094.

International Seismological Centre, 2011. Internatl. Seis. Cent. On-line Bulletin. Thatcham,United Kingdom (http://www.isc.ac.uk).

Karson, J.A., 1998. Internal structure of oceanic lithosphere: a perspective from tectonicwindows. In: Buck, W.R., Delaney, P.T., Jeffery, J.A., Karson, A., Lagabrielle, Y. (Eds.),Faulting and Magmatism at Mid-Ocean Ridges. American Geophysical Union,Washington, pp. 177–218.

Kissling, E., Ellsworth, W.L., Eberhart-Phillips, D., Kradolfer, U., 1994. Initial referencemodels in local earthquake tomography. J. Geophys. Res. 99 (B10), 19635–19646.http://dx.doi.org/10.1029/93JB03138.

Kwasnitschka, T., Hansteen, T.H., Devey, C.W., Kutterolf, S., 2013. Doing fieldwork on theseafloor: photogrammetric techniques to yield 3D visual models from ROV video.Comput. Geosci. 52, 218–226. http://dx.doi.org/10.1016/j.cageo.2012.10.008.

Lienert, B.R.E., Havskov, J., 1995. A computer program for locating earthquakes bothlocally and globally. Seismol. Res. Lett. 66, 26–36.

Madeira, J., Mata, J., Mourão, C., Brumda Silveira, A., Martins, S., Ramalho, R., Hoffmann, D.,Masson, D.G., Le Bas, T.P., Grevemeyer, I., Weinrebe, W., 2010. Volcano-stratigraphicand structural evolution of Brava Island (Cape Verde) based on 40Ar/39Ar, U/Th andfield constraints. J. Volcanol. Geotherm. Res. 9, Q07015–Q07235. http://dx.doi.org/10.1029/2008GC001983.

Masson, D.G., Le Bas, T.P., Grevemeyer, I., Weinrebe, W., 2008. Flank collapse and large-scale landsliding in the Cape Verde Islands, off West Africa. Geochem. Geophys.Geosyst. 9, Q07015. http://dx.doi.org/10.1029/2008GC001983.

Matias, L., Cruz, J., Pena, J.A., Martins, I., Senos, M.L., 1997. A sismicidade registada na Ilhado Fogo durante os primeiros dias de actividade do vulcão na erupção de Abril de1995. In: Réffega, A., Figueiredo, M.O., Silva, L.C., Costa, F.L., Mendes, M.H., Torres,P.C., Silva, T.P., Correia, E. (Eds.), A erupção vulcânica de 1995 na ilha do Fogo, CaboVerde. Instituto de Investigação Científica Tropical and Ministério da Ciência eTecnologia, Lisboa. ISBN: 972-672-863-0, pp. 13–32 (in Portuguese).

Mendes, A.S., 1956. Contribuição para o conhecimento da sismicidade dos territóriosPortugueses. Boletim da Ordem dos Engenheiros. 5(3), pp. 1–4, (Lisboa.(in Portuguese)).

Mitchell, J.G., Le Bas, M.J., Zielonka, J., Furnes, H., 1983. On dating the magmatism of Maio,Cape Verde Islands, Earth Planet. Sci. Lett. 64, 61–76.

Mourão, C., Mata, J., Doucelance, R., Madeira, J., Brum da Silveira, A., Silva, L.C., Moreira, M.,2010. Quaternary extrusive calcio-carbonatite volcanism in Brava Island (CapeVerde): a nephelinite–carbonatite immiscibility product. J. Afr. Earth Sci. 56, 59–74.http://dx.doi.org/10.1016/j.afrearsci. 2009.06.003.

NASA, 2009. ASTER Global Digital Elevation Model v001, (at http://gcmd.nasa.gov/KeywordSearch/Metadata.do?Portal = GCMD&MetadataType = 0&MetadataView=Full&KeywordPath = &EntryId = ASTGTM1 (accessed in January 2014)).

337D. Vales et al. / Tectonophysics 636 (2014) 325–337

Neves, J.V., 1981. Relatório preliminar da missão geofísica a Cabo Verde no período de I0de Abril a I I deMaio de 1981. Unpublished report, Instituto Nacional deMeteorologiae Geofísica, Lisboa. (in Portuguese)

Pim, J., Peirce, C., Watts, A.B., Grevemeyer, I., Krabbenhoeft, A., 2008. Crustal structure andorigin of the Cape Verde Rise. Earth Planet. Sci. Lett. 272 (1–2), 422–428. http://dx.doi.org/10.1016/j.epsl.2008.05.012.

Plesner, S., Holm, P.M., Wilson, J.R., 2003. 40Ar/39Ar geochronology of Santo Antão, CapeVerde Islands. J. Volcanol. Geotherm. Res. 120, 103–121. http://dx.doi.org/10.1016/S0377-0273(02)00367-0.

Ramalho, R.A.S., 2011. Building the Cape Verde Islands(Springer Theses) 1st ed. Springer-Verlag, Heidelberg, Dordrecht 978-3-642-19102-2 http://dx.doi.org/10.1007/978-3-642-19103-9 (207 pp.).

Ramalho, R., Helffrich, G., Schmidt, D.N., Vance, D., 2010a. Tracers of uplift and subsidencein the Cape Verde Archipelago. J. Geol. Soc. Lond. 167, 519–538. http://dx.doi.org/10.1144/0016-76492009-056.

Ramalho, R.S., Helffrich, G., Cosca, M., Vance, D., Hoffmann, D., Schmidt, D.N., 2010b.Vertical movements of ocean island volcanoes: insights from a stationary plate envi-ronment. Mar. Geol. 275, 84–95. http://dx.doi.org/10.1016/j.margeo.2010.04.009.

Ramalho, R., Helffrich, G., Cosca, M., Vance, D., Hoffmann, D., Schmidt, D.N., 2010c.Episodic swell growth inferred from variable uplift of the Cape Verde hotspot islands.Nat. Geosci. 3 (11), 774–777. http://dx.doi.org/10.1038/NGEO982.

Represas, P., Catalão, J., Montesinos, F.G., Madeira, J., Mata, J., Antunes, C., Moreira, M.,2012. Constraints on the structure of Maio Island (Cape Verde) by a 3D gravitymodel: imaging partially exhumed magma chambers. Geophys. J. Int. 190 (2),931–940. http://dx.doi.org/10.1111/j.1365-246X.2012.05536.x.

Ribeiro, O., 1960. A Ilha do Fogo e as suas erupções, Memórias (série geográfica I), Junta deInvestigações do Ultramar2nd ed. (Lisboa, Portugal, 319 pp. (in Portuguese)).

Serviço Meteorológico Nacional, 1947–1973. Anuário Sismológico de Portugal, Year Bookfrom Serviço Meteorológico Nacional, No 1–N°27, (Lisboa. (in Portuguese)).

Stillman, C.J., Furnes, H., LeBas, M.J., Robertson, A.H.F., Zielonka, J., 1982. The geologicalhistory of Maio, Cape Verde Islands. J. Geol. Soc. Lond. 139, 347–361.

Torres, P.C., Madeira, J., Silva, L.C., Brum da Silveira, A., Serralheiro, A., Mota Gomes, A.,1997. Carta geológica das erupções históricas da ilha do Fogo: revisão e actualização.In: Réffega, A., Figueiredo, M.O., Silva, L.C., Costa, F.L., Mendes, M.H., Torres, P.C., Silva,T.P., Correia, E. (Eds.), A erupção vulcânica de 1995 na ilha do Fogo, Cabo Verde.Instituto de Investigação Científica Tropical and Ministério da Ciência e Tecnologia,Lisboa. ISBN: 972-672-863-0, pp. 119–132 (in Portuguese).

Torres, P.C., Madeira, J., Silva, L.C., Brum da Silveira, A., Serralheiro, A., Mota Gomes, A.,1998. Carta geológica da ilha do Fogo (República de Cabo Verde): erupções históricase formações enquadrantes. Edited by Laboratório de Tectonofísica e Tectónica Exper-imental (LATTEX), University of Lisbon, 1 sheet at the 1:25.000 scale. (in Portuguese)

Torres, P.C., Silva, L.C., Serralheiro, A., Tassinari, C., Munhá, J., 2002. Enquadramentogeocronológico pelo método K/Ar das principais sequências vulcano-estratigráficasda Ilha do Sal – Cabo Verde. Garcia de Orta. 18 (1–2), pp. 9–13, (in Portuguese).

Vinnik, L., Silveira, G., Kiselev, S., Farra, V., Weber, M., Stutzmann, E., 2012. Cape Verdehotspot from the upper crust to the top of the lower mantle. Earth Planet. Sci. Lett.319–320, 259–268. http://dx.doi.org/10.1016/j.epsl.2011.12.017.

Wessel, P., Smith, W.H.F., 1998. New, improved version of Generic Mapping Toolsreleased. Trans. Am. Geophys. Union 79, 579.

Wilson, D.J., Peirce, C., Watts, A.B., Grevemeyer, I., Krabbenhoeft, A., 2010. Uplift atlithospheric swells-I: seismic and gravity constraints on the crust and uppermostmantle structure of the Cape Verde mid-plate swell. Geophys. J. Int. 182 (2),531–550. http://dx.doi.org/10.1111/j.1365-246X.2010.04641.x.