Seismological evidence for subcrustal magmatic injection beneath Fogo volcano, Cape Verde hotspot

Fogo volcano belongs to the Cape Verde hotspot and its most recent eruption occurred from November 2014 to February 2015. From January to December 2016 we operated a temporary seismic network on Fogo and were able to locate 289 earthquakes in total. While most of the events occur at distances > 25 km near the neighboring island of Brava. However, on 15 August 2016 we recorded an isolated cluster of more than 20 earthquakes, 13 of which could be located beneath the southern part of Fogo. Differences between Sand P-wave arrival times at steep incidence clearly indicate focal depths between approximately 38 and 44 km, whereas the Moho discontinuity has been reported at about 14 km depth. This indicates that the earthquakes occur well within the upper mantle directly beneath Fogo. The observations suggest that the high strain rates required for the deep events result from fracturing induced by magma injection. The results further indicate that the magmatic source region of previous eruptions is indeed located at greater depth beneath Fogo and not Brava as previously hypothesized.


INTRODUCTION
The Cape Verde archipelago is located roughly 700 km west of the coast of Senegal in the Atlantic Ocean (see Fig. 1). The volcanism in this region originates from a mantle plume underlying an almost stationary tectonic plate (Courtney and White, 1986;Gripp and Gordon, 2002). Fogo, located in the SW of the archipelago, belongs to the younger islands (< 6 Ma, Holm et al., 2008) and is the only volcano of Cape Verde with reported historical eruptions (Faria and Fonseca, 2014). Its eruptions are dominated by effusive and strombolian activity.
The last one took place from November 2014 to February 2015 (González et al., 2015) and the time interval between eruptions is about 20 years (Faria and Fonseca, 2014).
Whereas Fogo is seismically rather quiet, high seismic activity has been reported around and beneath the neighboring island Brava (at a distance of about 20 km) and the Cadamosto seamount 20 km to the southwest of Brava (Grevemeyer et al., 2010;Vales et al., 2014). Little is known about the internal structure of Fogo volcano. In previous studies a possible link of the magma supply systems of Fogo and Brava has been discussed in relation to seismicity that occurs in between the islands (see e.g., Heleno and Fonseca (1999) and Hildner et al. (2011) for details). Specifically, it has been argued that the seismicity prior to the 1995 eruption of Fogo relates to magma transport from Brava towards Fogo. The orientation of eruptive fissures during this eruption supports this assumption. However, chemical analysis of phenocryst rims reveals a short term storage in shallow reservoirs of magma within the crust below Fogo a few hours before eruption (Hildner et al., 2011;Hildner et al., 2012). The results indicate that historic and prehistoric eruptions were fed by magma ascending from different small reservoirs within the mantle at depths ranging from 15 to 30 km (Hildner et al., 2012). This is supported by petrological and geochemical data of the most recent eruption 2014-2015, indicating magma storage within the mantle at a depth of about 25.6 ± 5.5 km (Mata et al., 2017). Furthermore, the island of Fogo has not experienced any significant deformation before and during this eruption, which also points to a deeper source, likely >15 km (González et al., 2015).
To better constrain the source region of the magma supply system beneath Fogo and its relation to seismic activity we deployed a seismic array at a distance of about 8 km from the southern rim of the Chã das Caldeiras from January to December 2016. To improve event detection and localization capabilities of the array, three additional broadband stations were deployed across the island (Fig. 1) in January 2016.  (Ryan et al., 2009).

DATA ANALYSIS AND METHODS
Our network consists of ten seismic stations, three of which are equipped with broadband Trillium Compact seismometers and seven with short-period (4.5 Hz) geophones.
The seven short-period instruments configure a seismic array with an aperture of 700 m, designed for the detection of local events with a mean frequency of about 7.5 Hz. The additional three broadband stations are distributed across the island, in the southern part of Chã das Caldeiras and in the northern and western parts of Fogo (e.g. Fig. 1). All stations were equipped with Omnirecs CUBE-dataloggers recording continuously with a sampling rate of 200 Hz. Data were collected from January to December 2016 with gaps of about 11 weeks in total due to storage limitations of the dataloggers. The longest gap of 4 weeks occurred in June and July.
Earthquakes that occurred outside of our network on Fogo are generally located using a time domain array analysis. By performing a time domain array analysis information from a wide frequency band, implicitly, is incorporated, as the complete waveform is used in forming the beam. Traces are shifted and stacked to determine the beam energy for horizontal slownesses (sx,sy) in the range of ±3 s/km. From the horizontal slowness components at maximum energy the backazimuth of the event is derived and the absolute slowness yields the apparent velocity of the wavefront. Uncertainties of backazimuth and slowness are estimated by considering resulting variations in slowness obtained from a 5% reduction of the maximum energy. Epicentral distances are calculated from arrival-time differences between S-and P-waves. For local events at greater horizontal distance from the array (close to Brava or the Cadamosto seamount, Fig. 1), we assume a simplified two-layer model for the crust and underlying mantle with appropriate P-wave velocities taken from Vales et al. (2014), particularly 6.1 km/s for the crust and 8.0 km/s for the mantle. A fixed vP/vS-ratio of 1.7X is used. The Moho depth is set to 14 km in agreement with previous studies (e.g. Vales et al., 2014), while the events are assumed to occur at a fixed depth of 5 km, which is the mean depth of hypocenters beneath Brava (e.g., Faria and Fonseca, 2014). Some authors (Vales et al., 2014) have reported deeper events of about 10 km depth. Although we cannot directly constrain the earthquake depth from the array analysis, we obtain information about the maximum depth of the ray path from the horizontal slowness (ray parameter) derived from the analysis. In a laterally homogeneous medium, the ray parameter is constant, and its inverse (as determined from the apparent velocity across the array) corresponds to the velocity at the turning point of the ray. Thus, for distant earthquakes, the apparent velocity allows to discriminate between ray turning points within the crust and the mantle. For earthquakes observed around Brava, the apparent velocity of about 6 km/s is consistent with a crustal turning point (depth < 14 km). The error that may result from the deviation between the true earthquake depth relative to the fixed depth of 5 km has been carefully investigated and is covered by the error that we assume in the analysis (see below). From the theoretical arrival times for P and S waves (in the model), we derive differential arrival times as a function of distance which we compare with the measured time difference. This procedure is applied to all stations of the array, such that a mean distance is obtained. We apply an error of 10% to this distance to account for the uncertainties related to the depth and velocities used in the model.
For locating the earthquakes directly beneath Fogo we apply a standard localization method, HYPOCENTER (Lienert et al., 1986) (without any a priori depth constraints). We use the velocity model of Vinnik et al. (2012) derived for Santiago and Fogo. This model has the advantage that it provides a S-wave velocity structure derived from receiver functions.
Details on the velocity model and its effects on the earthquake locations are given in the discussion below.

RESULTS
Most of the events detected by our network exhibit the characteristics of volcanotectonic (VT) earthquakes with clear P-and S-phases (see e.g. Wassermann (2012)) and occur beneath and around the island of Brava. There is also evidence for some long period volcanic events (naming in accordance to Fonseca et al., 2003) and more rarely harmonic "cigarshaped" tremors with a length of several minutes to hours, but these events have not yet been localized. Figure 1 shows the location of 289 earthquakes with magnitudes (ML) 0.3 to 3.3, recorded between January and December 2016 that were located using either classical network or seismic array analysis (depending on the epicentral distance). Assuming an error of 5% of the maximum stacked energy in the array analysis leads to average errors of about ±10° for the backazimuth and ±5-8 km for the distance estimations. Detectable earthquakes beneath Fogo were recorded on 15 August. More than 20 of these occurred within one hour, followed by some larger earthquakes during the next 10 hours. Around the time of this earthquake swarm we do not observe any long period seismicity or tremors. We were able to locate 13 earthquakes of this swarm for which the seismograms exhibited a sufficiently high signal to noise ratio. A table of the analyzed events is given in the supplements. As an example, we show 3-component recordings for one of the stronger earthquakes (ML 2.1) in Figure 2. The events are generally characterized by dominant frequencies between 10 and 20 Hz. As the earthquakes occurred at depth directly beneath the stations, their focal mechanism could not be determined. Figure 3 clearly indicates the clustering of the source locations, both, horizontally and vertically. The errors in latitude and longitude are of the order of ±4 to 5 km and ±5 to 6 km, respectively. The depth of the events ranges from 38.6 km to 43.4 km, with a mean depth of 40.8 km. This is in agreement with the observed time differences between P-and S-wave arrivals of about 4.3 s.
The error of the depth estimation is discussed below. There is no systematic change in depth and lateral position of the earthquakes over time. The earthquakes do not follow a typical mainshock-aftershock sequence. Magnitudes (ML) of the analyzed events range from 0.8 to 2.1. The events within the first hour reach a maximum magnitude of 1.2 whereas the events during the following 10 hours reach magnitudes of up to 2.1. Support for the location of the earthquakes directly beneath Fogo comes from seismograms recorded at the array stations. The traces show a negligible time shift and the maximum of the stacked energy corresponds to an apparent velocity that approaches infinity, consistent with a near-horizontal wavefront arriving at the array (Fig. 4).
In comparison, the typical apparent velocity for an event originating around Brava is around 6.4 km/s.

DISCUSSION
VT events beneath Fogo are relatively rare (Fonseca et al., 2003) and have been reported to occur at shallow depths <10 km (Heleno and Fonseca, 1999;Faria and Fonseca, 2014;Vales et al., 2014). The deep events beneath Fogo reported here are located south of Chã das Caldeiras and occur at a depth of about 40 km, whereas crustal thickness is only about 12 to 14 km (Pim et al., 2008;Vinnik et al., 2012). We therefore conclude that the earthquakes are located well within the uppermost mantle. The error in our depth estimate is in the range of ±1.2 to 2.7 km. However, systematic errors that may arise from the choice of the velocity model are not accounted for in this estimate. Therefore, to allow for a wider range of realistic velocities, we also consider velocity profiles that have been derived for some of the Canary Islands (Lodge et al., 2012). The velocity structure found there may generally be similar to that beneath Cape Verde. A standard deviation for the earthquake depth is then estimated from the different results obtained for all velocity models considered (for the Canary Islands and Cape Verde), including those used earlier (Vinnik et al. (2012) andVales et al. (2014)). The resulting uncertainties of the compilation of velocity models range from ±5.2 to 9.5 km in depth, depending on the event, and are given in Fig. 3. Additional application of a double-difference earthquake location method (hypoDD, Waldhauser (2001)) does not significantly change earthquake locations and confirms the previous depth estimates.
Due to its relatively high noise level, data of station CV11 had to be discarded from the analysis of some events. In Fig. 3 the color code indicates whether CV11 was included to determine a specific event location. To test the influence of station CV11, the localization procedure for all events was repeated by excluding data from this station. In this case, locations of the earthquakes are shifted systematically by about 1 km towards the North-East, while the effect on the depth estimates is negligible. However, the horizontal shift is well within the error range derived earlier.
To further confirm the subcrustal origin of the earthquakes we also performed a separate receiver function analysis for Fogo, for which we used the data collected at two of the broadband stations, CV10 and CV12, whereas CV11 was not used due to its higher noise level. In addition, we included the data of five temporary stations of the COBO network which was operating on Fogo during 2007 and 2008 (see Vinnik et al., 2012). We used the method of Zhu and Kanamori (2000) to constrain the crustal thickness H and the bulk vP/vS ratio of the crust beneath Fogo. For more information about data quality and preparation we refer to the supplementary material. The analysis leads to values of 11.1 ± 4.5 km for the crustal thickness and to 1.73 ± 0.12 for the velocity ratio (see Fig. S6). This is in agreement with previous results for Moho depths ranging between 12 and 14 km (Pim et al., 2008;Vinnik et al., 2012).
Previously, Vales et al. (2014) reported about different occurrences of deeper earthquakes in the region, some of them at depths close to 40 km. Several events occurred about 30 km offshore close to a field of volcanic cones southwest of Santo Antão, the northernmost island of Cape Verde (approximately 220 km north of Fogo), others in the vicinity of the Cadamosto seamount, southwest of Brava. Additionally, some isolated earthquakes at depths between 20 and 30 km occurred offshore, west of Brava. However, the corresponding depth estimates are not well constrained because of the offshore locations of these events resulting in a large azimuthal gap; a relation to the magmatic activity beneath Fogo cannot be established.
As a possible explanation for the subcrustal events beneath Fogo we suggest the rapid movement of magma into the shallower upper mantle, possibly into a reservoir where it accumulates before the next eruption. Similar processes have been observed for example in Hawaii (Wright and Klein, 2006) and at Piton de la Fournaise (Michon et al., 2015). At Bárðarbunga (Iceland) Hudson et al. (2017) report earthquakes in the ductile part of the crust, which they attributed to melt-induced brittle failure. Considering the location of the events directly beneath Fogo, a common magma supply system for Fogo and Brava, as discussed in previous studies (see e.g., Heleno andFonseca (1999), Fonseca et al. (2003), for details), seems rather unlikely. Observations of earthquakes in the region between Brava and Fogo, as well as the orientation of eruptive fissures led to this assumption. In contrast, two separated supply systems with an independent magma storage beneath Fogo, as also suggested by petrological studies (Hildner et al., 2011), better explain the isolated deep events. Based on thermobarometric analyses, Hildner et al. (2012) propose magma reservoirs related to historic and prehistoric eruptions at depths between 15 and 30 km.
In view of these estimates, the newly observed events may indicate magma transport from a deeper source into an accumulation zone at depths of about 40 km. Although these processes occur below the brittle crust, rapid movement of magma can lead to strain rates high enough to cause magmatic fracturing and seismicity in normally aseismic zones.
Possible processes can be the breaking up and pushing of plugs along a dyke or the occurrence of earthquakes at the tip of the dyke (White et al., 2011). Tarasewicz et al. (2012) suggest that overpressure in a sill causes magma to escape under high strain rates through a bottleneck, where seismicity originates. The deep events beneath Fogo show similar waveforms. Additionally, the observed distances between the individual deep earthquakes are likely overestimated due to uncertainties resulting from the localization procedure and hence the fracturing may be limited to a smaller domain. The scarcity of the deep events beneath Fogo can be explained if the molten material generally moves under aseismic conditions. Tarasewicz et al. (2012) argue that changes in flow conditions can result in clogged up pathways, leading to a transition from aseismic to seismic movement of magma.
A downward migration of relaxation through the plumbing system can also explain the occurrence of these events after the eruption (Michon et al., 2015). The absence of earthquakes beneath Fogo in the depth range between about 10 and 40 km may be related to aseismic movement of magma or an aseismic zone caused by melt accumulation. At Piton de la Fournaise, Michon et al. (2015) observe a seismic gap between 8 and 11 km and address this to a zone of magma storage. White et al. (2011) discuss a possible freezing of magma at depth and the aseismic storage in sills in the northern volcanic rift zone in Iceland. Tarasewicz