Form and growth of an embryonic continental rift: InSAR observations and modelling of the 2009 western Arabia rifting episode

Summary

A magma-driven rifting episode occurred at the Harrat Lunayyir (Harrat Al-shaqa) volcanic field, western Arabia, between 2009 April and July. It was accompanied by a swarm of more than 4000 M > 2 earthquakes, the largest ever documented in that region, with a peak M_w 5.7 shock on May 19. We combine Interferometric Synthetic Aperture Radar (InSAR) measurements and elastic modelling with seismic moment calculations to resolve the evolution of surface deformation associated with this event. Phase discontinuities and low-coherence lineaments are used to infer the location of the main active structures during the various deformation stages and descending-track interferograms that span the entire period are inverted to resolve the slip and opening distributions along two graben-bounding normal faults and a dyke, respectively. Assuming negligible rift-parallel displacements, we combine ascending- and descending-track interferograms to derive the vertical and rift-perpendicular deformation, which add up to a maximum surface extension of 1.5 m across the rift and subsidence of 0.8 m. The far-field deformation is dominated by the dyke opening, whereas the near-field displacements are mostly associated with movements along the faults. The cumulative seismic moment released during the entire swarm period accounts for about 14 per cent of the total geodetic moment, compared to about 55 per cent at the 2007 Gelai (Tanzania) and about 8 per cent at the 2005 Manda Hararo–Dabbahu (Afar) events. We propose that the differences in moment partitioning ratios are due to the different crustal and seismogenic layer thicknesses in the three regions and represent different stages in the evolution of a volcanic rift. The Gelai event represents the most juvenile stage of rifting, the Dabbahu event represents the most evolved and the Harrat Lunayyir event represents a rift that is intermediate between the two in its degree of maturity.

Radar interferometry, Volcano seismology, Intraplate processes, Continental tectonics: extensional, Asia

1 Introduction

Volcanic rift zones are characterized by arrays of normal faults, extensional fractures, eruptive fissures and vents. They are common along divergent plate boundaries (Mid-oceanic ridges and their continental extensions), island arcs and discrete volcanic edifices such as Hawaii, Mt. Etna and others. Volcanic rift zones form through a combination of dyke intrusion at depth, slip along normal faults that form graben structures, surface subsidence and volcanic eruptions (e.g. Abdallah et al. 1979; Pollard et al. 1983; Rubin & Pollard 1988). Rifting episodes are often associated with increased seismicity in the form of swarms that may last from several hours to several years (Benoit & McNutt 1996; Vidale et al. 2006). Despite the large number of volcanic rift zones worldwide, the processes that govern rifting in the different tectonic settings and that discriminate mature from immature rifts are still poorly understood. A significant step towards a better understanding of these underlying processes has been made with the introduction of space geodetic techniques: the Global Positioning System (GPS), and particularly, Interferometric Synthetic Aperture Radar (InSAR). Two recent dyke-induced swarms were unravelled by InSAR in east Africa. The first occurred in 2005 September along the Manda Hararo–Dabbahu rift segment in Afar. It was associated with the intrusion of an ∼8-m thick dyke and was followed by a minor explosive eruption at the nearby Da'Ure vent (e.g. Wright et al. 2006; Ayele et al. 2009). The second occurred in Oldoinyo Gelai, northern Tanzania between 2007 July and September, and showed a similar sequence of events, starting with the intrusion of an ∼1- to 2-m thick dyke and followed by ash eruptions from the nearby carbonatitic Oldoinyo Lengai volcano (Baer et al. 2008; Calais et al. 2008; Biggs et al. 2009). The two episodes differ in their magnitude, duration and degree of partitioning between the seismic and aseismic processes (e.g. Biggs et al. 2009).

In 2009 mid-April, an earthquake swarm struck the Harrat Lunayyir (HL) volcanic field, western Arabia (Fig. 1). The geographic association of the swarm with the volcano (Fig. 1b), which probably erupted in historic times (Van Padang 1963; Simkin et al. 1981), the formation of open fissures in several locations in the volcanic field [Saudi Geological Survey (SGS) website reports] and the absence of a major shock at the beginning of the sequence, suggest that the swarm was also of magmatic origin. We present here first InSAR measurements and elastic modelling of the surface deformation, which strongly support its magma-driven origin, yet so far there has been no sign of eruptive activity in HL. Due to the remoteness of the area and the scarcity of published ground measurements during the swarm period, we use InSAR measurements to locate the surface ruptures and determine the spatial and temporal evolution of surface deformation. We then evaluate the independent contribution of a dyke and two graben-bounding faults to the overall vertical and extensional deformation. Finally, we discuss the geometric and evolutional characteristics of volcanic rifts by comparing the HL event with the two previous dyke intrusion episodes in Afar and Tanzania.

Figure 1

(a) Location map of Harrat Lunayyir (HL) (black rectangle) on ASTER GDEM map. White rectangles mark the ENVISAT ASAR and ALOS PALSAR scenes used for interferometry in this study. (b) Landsat image of HL volcano. Dark areas are lava flows and red circular bodies are volcanic cones. Epicentre locations and times for the period 2009 May 25–2009 August 5 (coloured circles) are taken from the SGS website reports. Prior to May 25 daily reports are incomplete or entirely missing. The reported error in location is 0.5–1 km. A-A′ marks the projection line for the hypocentral depths profile (Fig. 2). White lines mark the surface traces of the InSAR-determined faults (see below). (c) Map showing the volcanic fields in Arabia (grey patches) and historical eruptions (black circles; after Van Padang 1963; Simkin et al. 1981). The major volcanic fields lie along the Mekkah-Madinah-Nufud (MMN) line (see text).

2 The Western Arabia Volcanic Province

Two separate phases of continental volcanism are distinguished in western Arabia. The first, primarily tholeiitic in composition, formed between 30 and 20 Ma along NW-trending structures, and was attributed to passive mantle upwelling associated with the opening of the Red Sea (e.g. Almond 1986; Coleman & McGuire 1988). The second, of transitional to strongly alkaline composition, has been active over the past 12–14 Myr and is attributed to the 1200-km wide ‘West Arabian Swell’ (WAS, Camp & Roobol 1992). This broad continental uplift has an overall N–S trending structural axis that deviates from the Red Sea trend by about 25°, intersects it between 15°N and 19°N (Fig. 1c) and continues southwards into the Danakil Depression in Afar (Camp & Roobol 1992). Most of the post-12 Ma linear vent systems of the western Arabia lava fields (‘Harrats’) are subparallel to this trend, yet locally, northwesterly orientations are also common. The major Harrats in this province lie along the 600-km long Mekkah-Madinah-Nufud (MMN) volcanic line (Fig. 1c). Recent eruptions along this line occurred in 641, 1256 (the ‘Madinah eruption’) and possibly also in 1800 AD; the ‘Madinah eruption’ was extruded from a 2.25-km long fissure oriented 343° (Camp et al. 1987). The most recent eruption in the Arabian Peninsula was in 1937 near the town of Dhamar in North Yemen and the latest eruption along the Red Sea axis occurred in 2007 in Jabal al-Tair Island off the coast of Yemen. Due to the lack of reliable documentation, not all the historical volcanic and seismic events were recognized. According to the available records, however, there have been at least 21 eruptions on the Arabian Peninsula during the past 1500 yr (Van Padang 1963; Simkin et al. 1981; Camp et al. 1987).

3 Harrat Lunayyir and the 2009 Earthquake Swarm

HL (also named Harrat Al-shaqa) is one of the smallest of the Holocene lava fields of Arabia, located about 60 km east of the Red Sea coastline, 150 km east of the Red Sea spreading centre and about 300 km west of the MMN line (Fig. 1). It contains more than 50 volcanic cones that formed over Precambrian crystalline rocks along a generally NNW-trending axis (Fig. 1b). One of the cones may have erupted around 1000 AD (Van Padang 1963; Simkin et al. 1981). The crustal structure is transitional between a highly attenuated, 5–8.5-km thick crust at the Red Sea spreading centre (Cochran 2005), and an ∼40-km thick continental crust beneath the Saudi shield (Mechi et al. 1986). The crustal thickness gradient between the two provinces is exceptionally high across Red Sea coastal plain and the HL lava field. The region is dominated by NW-striking dykes and faults, and by NS- and EW-striking fault systems (Brown 1972).

An earthquake swarm, consisting of tens of M < 4 earthquakes occurred at HL volcano on 2007 October. Seismic activity resumed on 2009 April 18 and culminated on May 17–19 with a sequence of seven M > 4 events including an M_w 5.7 earthquake. Open fissures, reaching places of widths of a few metres and a total length of about 8 km formed during this period (SGS website reports). In total, over 4000 earthquakes of M > 2, about 200 M > 3 earthquakes, and 11 M > 4 earthquakes were recorded so far (United States Geological Survey, European–Mediterranean Seismological Centre and SGS website reports). The reliability, details and completeness of the reported seismicity improved during the swarm period. Starting on May 25, daily press releases, accompanied occasionally by maps, were issued by the SGS and were posted on the SGS official website. From June 26, these reports also included daily tables of earthquake coordinates, magnitudes and depths, with a reported uncertainty of 0.5–1 km in location and 0.8–1.5 km in depth. The epicentres are aligned NNW (Fig. 1b) and according to the published data (SGS website), hypocentres seem to be confined to 5–10 and 15–20 km depth intervals (Fig. 2), with deeper earthquakes in the southern part of the region and shallower earthquakes in the north. The shift between the two depth provinces coincides with the southern termination of a graben that was formed during the event (Fig. 1b).

Figure 2

Hypocentral depths projected along line A-A′ (Fig. 1b). Data from SGS website updates. The reported error in depth is 0.8–1.5 km.

4 InSAR Observations and Modelling

4.1 Methods and results

Over the past two decades, InSAR has become a widespread tool to measure subtle displacements at the ground surface (e.g. Massonnet & Feigl 1998). We use SAR images from the European Space Agency (ESA) ENVISAT satellite (C-band, 5.6 cm wavelength) that were acquired in beam mode IS2 (mean incidence angles of 23°), and ALOS PALSAR images of the Japan Aerospace Exploration Agency (JAXA) (L-band, 23.6 cm wavelength and mean incidence angle of 34°). Interferograms were made using the JPL/Caltech ROI-PAC software (Rosen et al. 2004). We remove the topographic phase from the phase changes due to ground displacements and geocode the interferograms using NASA's 3 arc second Shuttle Radar Topographic Mission (SRTM) digital elevation model (DEM) (Farr & Kobrick 2000; Jarvis et al. 2006) and the 1 arc second ASTER Global Digital Elevation Model (GDEM), developed by NASA and the Ministry of Economy, Trade and Industry of Japan (METI).

Altogether we generated 10 interferograms spanning the periods before, during and after the main May 17–19 events (Fig. 3; Table 1). The deformation fields are shown in the form of unwrapped interferograms that display the absolute movement along the satellite line-of-sight (LOS) direction. For the main deformation phases, we superimpose the unwrapped interferograms on their respective wrapped phase fringes (Figs 3a, b, d and f), where each fringe cycle corresponds to 28 mm of LOS movement. Due to the extremely arid conditions in this region, the interferograms provide good coherence, despite the relatively long spatial and temporal baselines of some pairs (Table 1). Using the 1 arc second ASTER GDEM we generated full-resolution interferograms for most of the periods. This increased the fraction of near-field measurements that were unwrapped unambiguously, despite the occasionally high deformation gradients (Fig. 3). Pre-2009 interferograms show no deformation, indicating that the 2007 swarm was too small to produce detectable movements by InSAR. Three interferograms that capture the main swarm period (Figs 3a, b and d) show significant deformation, and two interferograms that span the late stages of the swarm (Figs 3c and e) show minor deformation. A post-July 17 interferogram (not shown in Fig. 3) shows negligible deformation indicating that the deformation terminated gradually towards the beginning of August.

Figure 3

(a–e) ENVISAT ASAR interferograms and (f) ALOS PALSAR interferogram for the 2009 HL swarm period. For the major deformation periods (a, b, d and f) the unwrapped phase is draped upon the corresponding wrapped fringes. Each fringe cycle corresponds to 28 mm displacement in the satellite to ground direction. For the late stages (c and e) we show only the unwrapped phase. Positive values (red) mark movement towards the satellite. White lines in (a) mark the faults. Arrows show satellite view directions and incidence angles. Note the different displacement scales for each period.

Table 1

Interferometric pairs used in this study.

The surface traces of the major fractures were detected by phase discontinuities. The general structure formed during the swarm period is a wedge-shaped graben bounded by N–S and NW–SE striking faults (Figs 1b and 3a). To identify fractures that did not produce differential movements we used coherence maps that measure the stability of phase contribution from the individual scattering cells within each pixel between the two satellite passes. Low coherence is thus expected in pixels that were internally changed by fracturing or other surface processes. Coherence loss is observed in all coherence maps, thus, to isolate the swarm-related low-coherence faults and surface fractures from other low-coherence areas, and enhance their signature we subtract coherence maps of the swarm period from those of the pre- or post-swarm period. By this procedure we also distinguish between fractures that formed during the period May 27 – July 1 (Fig. 4a) and those that formed during the entire period (Fig. 4b). In addition to the major boundary faults, we find NW-striking lineaments at the central and eastern parts of the graben, forming an ∼2-km wide fracture zone. Our modelling (see the next section) suggests that the central lineament roughly coincides with the location of a subsurface dyke. Earlier geological maps of this region (Brown 1972; Jokisch et al. 1985) do not show any faults that may correspond to these InSAR-identified fractures.

Figure 4

‘Differential’ coherence maps generated by (a) subtraction of an ENVISAT coherence map of the period 090527–090701 from a pre-swarm ENVISAT coherence map and (b) subtraction of an ALOS coherence map of the entire swarm period from a post-swarm ALOS coherence map; the maps show incoherence along the major faults and fissures. (c) Incoherent lineaments for the entire period (Fig. 4b, in black), and for the period 090527–090701 (Fig. 4a, in white), on a shaded Aster GDEM background.

Some of the ruptures persist through the sequential interferograms (Figs 3 and 4), indicating ongoing deformation, rather than short-term events. The gradual subsidence of the graben is clearly seen in both the ascending and descending sequences (Fig. 3). About 80 per cent of the subsidence occurred during the first week following the peak event (before May 27), and about 95 per cent of the subsidence occurred during the first month, before June 17 (Fig. 3).

4.2 Inversion for fault-slip and dyke-opening models

InSAR observations are typically investigated using simple models based on solutions for dislocations in an elastic half-space (e.g. Okada 1985). We inverted the InSAR measurements to assess the opening and slip distribution on the dyke and faults. The inversion scheme is based on a least-squares minimization of the displacement misfit with iterations for the faults and dyke geometry (Fialko 2004; Hamiel & Fialko 2007; Baer et al. 2008). Throughout the interferogram data points are subsampled across a uniform grid with spacing of ∼3 km. At distances less than ∼20 km from the main structures, the sampling algorithm selects additional points that are essential for a high resolution description of the original data sets using the quadtree method (e.g. Jónsson et al. 2002). In that way we account for the observed deformation pattern and avoid most of the noise. The initial fault geometry in our models was chosen using discontinuities in the InSAR phase and low coherence lineaments (Figs 3–5). Dyke location and strike was initially determined by the coherence map and refined by iterative search during the inversion process (Fig. 5). To account for the non-planar geometries of the faults and dyke, we divide their traces into several segments, allowing for pure opening along the dyke and normal slip along the faults. Dyke segments were chosen to be vertical rectangles having a downdip dimension of 15 km. Fault segments were chosen to extend from their location at the surface down to their intersection with the dyke. Subsequently, we refined the fault plane geometries by allowing changes in their dip and length, to minimize the data misfit in a non-linear least-squares inversion. Fault dips were found to be 60°, with variations in downdip widths from about 2 km at the southern part of the graben to about 5 km at the north, intersecting the dyke plane at depths between 1 and 3.5 km, respectively (Fig. 6).

Figure 5

Left-hand panels: Original unwrapped interferograms used for the inversion (local UTM coordinates, origin at N25 E37). Fault and dyke traces, marked by black and red lines, respectively, follow phase discontinuities and low coherence lineaments (Figs 3 and 4). Centre panels: Model (synthetic) interferograms obtained by inversion of the corresponding data interferograms. Right-hand panels: Residuals between interferograms and models. Positive values (red) mark movement towards the satellite.

Figure 6

Best-fit fault slip and dyke-opening models obtained by inversion of the descending track InSAR measurements. The dyke (red trace) is modelled as a pure opening dislocation; faults are allowed only normal slip.

To evaluate the slip and opening distributions, each fault and dyke segment was further subdivided into patches with constant slip or opening. Patches were chosen to have a size of 1 × 1 km² at the surface, increasing with depth. As the seismic velocity structure of the crust in this region is poorly constrained, we calculated the slip distribution by assuming a homogeneous half-space model with Poisson ratio of 0.25. The models were smoothed to moderate the opening and slip gradients and positivity was imposed. For each stage the preferred slip model is selected based on minimal rms using the highest possible smoothness.

To resolve the deformation history of the swarm, we inverted two sequential interferograms (2007 August 1 to 2009 May 27 and 2009 May 27 to 2009 July 1), which capture the main deformation of the event (Figs 5a and b). As the entire event includes some minor deformation after July 1, we invert an additional descending track interferogram that spans the entire period from 2007 August to 2009 August 5 (Fig. 5c). Our best-fit models account for 89–95 per cent of the observed deformation with respective rms values between 1.3 and 0.7 cm (Fig. 5). The best-fit inverted model for the first stage, until 2009 May 27 (Figs 5a and 6a), consists of a wedge-shaped graben bounded by two shallow normal faults diverging towards the NNW and a dyke bisecting the angle between the faults. The dyke shows up to a 2 m opening at depth of about 5 km below the narrow part of the graben. During the following period (May 27 to July 1) the dyke shows additional opening of about 50 cm and the graben faults slipped at almost the same locations as during the previous period.

4.3 Vertical and rift-perpendicular displacements

The InSAR measurements along the descending and ascending tracks (Fig. 3) show ground displacements in two different LOSs. To derive the complete 3-D deformation field, an additional, independent component is needed. Possible data sets that may fulfill this requirement are the subpixel azimuthal offsets, along the satellite track that are calculated by cross correlating pre- and post-event amplitude images (e.g. Michel et al. 1999; Fialko et al. 2001; Wright et al. 2006; Grandin et al. 2009). Alternatively, the subpixel offsets in optical satellite images taken before and after the event also provide an independent third component (Van Puymbroeck et al. 2000; Michel & Avouac 2002; Barisin et al. 2009; Grandin et al. 2009). A third alternative is the Multi-Aperture SAR Interferometry (MAI) (Bechor & Zebker 2006). The measurement accuracies in these methods are much smaller than the LOS accuracy. The estimated along track accuracy in the MAI is about 5 cm, and the azimuthal offset accuracy is typically 10–30 cm (Fialko et al. 2001; Grandin et al. 2009). They are thus applicable only to relatively large displacements in the track-parallel (∼north) direction, which are not expected in this case (see the following sections). The reported optical correlation accuracies are 30–40 cm. As the maximum peak-to-peak cumulative LOS displacement during the HL swarm period is about 1 m (Fig. 3), all radar amplitude and optical cross-correlation methods are close to the accuracy limit and are thus not explored in this study.

We derive the horizontal and vertical deformation fields of the HL event using LOS measurements alone by assuming negligible rift-parallel displacements. Since the boundary faults are not parallel to each other (Figs 3–7), we choose the earthquake epicentre cloud (Fig. 1b) as a proxy for the rift direction. Our model results show dyke opening and normal faulting deformation perpendicular to the rift (Figs 5 and 6), with very minor rift-parallel deformation (as expected in typical rifts; see also figure 5b of Grandin et al. 2009). Furthermore, as the rift is subparallel to the ascending track and at about 25° to the descending track, any rift-parallel displacement would be unresolved on the ascending LOS and only about 15 per cent of its value would be resolved on the descending LOS. These two arguments justify the assumption of plane-strain deformation in the vertical and rift-perpendicular directions. The descending and ascending track interferograms of the entire period (#5 and #8 in Table 1) are thus combined to derive these two displacement components (Fig. 7). A maximum surface extension of 1.5 m is calculated across the central part of rift and a maximum subsidence of 0.8 m is calculated at the deepest part of the graben.

Figure 7

(a) Horizontal displacements perpendicular to the rift and (b) vertical displacements, calculated from the ascending and descending interferograms of the entire period (#5 and #8, Table 1). Black lines mark the faults; white lines mark the location of rift-perpendicular profiles shown in Fig. 8.

To evaluate the independent contribution of the dyke and the graben faults to the overall deformation we compare our vertical and extensional calculated displacements with dyke-alone, faults-alone and total (dyke and faults) modelled displacements in 2 rift-perpendicular profiles across the graben (Figs 7 and 8). The model inputs are the slip and opening distributions obtained by our inversion (Fig. 6). The fault-alone models seem to explain most of the vertical displacements within the graben (Figs 8a and b). On the other hand, the dyke-alone models explain fairly well the deformation at the margins of the graben (Figs 8c and d). Combined, the two models show a good agreement with both vertical and horizontal displacements in the two profiles across the graben (Figs 8e–h). Thus, it is apparent that the dyke dominates the far field deformation, whereas the faults dominate the near-field deformation (as previously noted by Rubin (1992) and Barisin et al. (2009)).

Figure 8

Vertical (left-hand panels) and rift-perpendicular (right-hand panels) displacements along profiles A and B across the graben (see Fig. 7 for location). Blue lines are observed values, orange lines are modelled values. (a and b) Comparison between observations and faults alone models for profile A–A′. (c and d) Comparison between observations and dyke alone models for profile A–A′. (e–h) Comparison between observations and total displacement models (dyke and faults) for profiles A–A′ and B–B′. Note the location of the graben at the crest of the volcanic ridge (shaded profiles).

We note, however, some asymmetry in the displacement profiles (and to a less extent also in the model profiles). These different forms of asymmetry lead to misfits between the models and the data in some regions, mainly along the western margin of the graben (Fig. 8). The misfit may be a result of some incorrectly determined phase ambiguities in the interferograms or minor rift-parallel deformation.

5 Discussion

5.1 Seismic versus aseismic moment release during the HL event

Seismic moment release during the entire HL swarm, calculated from all published earthquake catalogues, is about 5.3×10¹⁷ Nm (Fig. 9a; Table 2). Due to the lack of systematic reports until May 25, some 2 < M < 4 events were not included in our moment calculation. Thus, taking into account the reported approximate numbers of 4000 M > 2 and 200 M > 3 earthquakes during the entire swarm period (SGS website), our calculated seismic moment may be lower than the actual moment by about 10 per cent. During the first InSAR interval (until May 27) about 20 per cent of the moment was released seismically, and as the event continued a lower contribution of seismic moment is evident (Fig. 9a; Table 2). During the first InSAR interval, about 78 per cent of the aseismic deformation was released by dyke opening and 22 per cent by aseismic slip along the faults, evolving during the second interval to 67 per cent by dyke opening and 33 per cent by aseismic slip (Table 2). This large component of aseismic slip can be attributed either to fault creep or to a mixed opening and slip modes along the graben faults. Both mechanisms have been proposed for other volcanic rifts. Creep along the faults was observed during the Gelai event (Baer et al. 2008); wide open cracks accompany many of the faults in volcanic rift zones worldwide (e.g. Tentler 2005; Rowland et al. 2007; Baer et al. 2008; Grandin et al. 2009).

Figure 9

(a) Cumulative seismic moment, fault-slip geodetic moment and total (fault-slip + opening) geodetic moment released during the 2009 HL episode. (b) Cumulative seismic moment (solid lines) and total geodetic moments (dashed lines) released during the 2009 HL (red), 2007 Gelai (green) and 2005 Dabbahu (black) swarms.

Table 2

Seismic and geodetic moments released during the HL event (Fig. 9a).

5.2 Comparison with other volcanic rifts

The 2009 HL swarm shows a similar temporal pattern of moment release as those observed at the 2007 Gelai (Tanzania) and the 2005 Dabbahu (Afar) rifting episodes (Baer et al. 2008; Grandin et al. 2009; Fig. 9b). Each swarm can be divided into two major stages: an initial, mostly seismic stage, that lasted 2 d at HL, 4 d at Gelai and about 11 d in Dabbahu, followed by months of stable, relatively low, seismicity, interrupted by short (up to a few days) seismic bursts. The total seismic moment released accounts for about 14 per cent of the total geodetic moment at HL, 55 per cent at Gelai and 8–10 per cent at the 2005 Dabbahu rift (the first of a series of dyke intrusion events that is still ongoing, Hamling et al. 2009; Fig. 9, Table 3). To understand this different partitioning between seismic and aseismic moment components we compare the major geometric and tectonic parameters characteristic to each of the three rifting events (Table 3).

Table 3

Geometric, seismic and tectonic parameters of the HL, Gelai and Dabbahu events.

The three rifts are quite similar in some of the parameters whereas in others they differ significantly. For example, the cumulative seismic moment released in the Gelai and Dabbahu events are comparable in size, whereas the HL event is an order of magnitude lower. On the other hand, the total geodetic moment is comparable in size in the HL and Gelai events however higher by an order of magnitude in the 2005 Dabbahu event. As both seismic and geodetic moments in Dabbahu are an order of magnitude higher than in HL, the seismic/geodetic moment ratios are approximately equal in these two rifts. The Gelai event shows a significantly higher seismic/geodetic ratio (Table 3). The faults in Gelai intersect the dyke at 12 km depth (Baer et al. 2008), whereas the faults intersect the dyke of HL at 1–3.5 km depth (Fig. 6) and at 1–2.5 km in Dabbahu (e.g. Wright et al. 2006). The graben at Gelai is only slightly wider than at Dabbahu, however, the boundary faults are steeper than those of Dabbahu and HL. The Gelai event is thus characterized by deeper faults (including a blind normal fault at its initial stage), which are capable of releasing a larger seismic moment per unit length (Table 3). Magma transport below Gelai originated at depths probably greater than 10 km (Baer et al. 2008), and while the dyke ascended, it moved past the bottom level of the graben faults until reaching its final shallow crustal level.

The depth of the seismogenic layer (determined here by the depth above which 99 per cent of the earthquakes occur) may be used as a proxy for the thickness of the brittle crust (e.g. Lamontagne & Ranalli 1996; Albaric et al. 2009) in each of the three rifts. A thicker seismogenic layer will result in a higher ratio of brittle to ductile deformation. The crustal and seismogenic layer thicknesses are highest at Gelai (37 and 26 km respectively), intermediate at HL (30–35 and 20 km) and lowest at Dabbahu (13–16 and 8 km) (Table 3). These different values are in good agreement with the different partitioning ratios between seismic and total geodetic moments in the three rifting episodes. We thus propose that the higher seismic to aseismic moment ratio of the Gelai rifting event compared to the HL and Dabbahu events is a result of its thicker crust and deeper levels of magma chamber and dyke.

Biggs et al. (2009) showed empirical relationship between dyke length and spreading rates in four volcanic rifts. Following the model of Segall et al. (2001), they suggested that the order of magnitude difference in length between the 2005 Dabbahu and the 2007 Gelai dykes can be attributed to shallower and larger magma chambers in Afar, compared to the smaller and deeper chamber in Gelai. Solomon et al. (1988) showed an inverse correlation between spreading rates and the thickness of the seismogenic crust, inferred by the centroid depth of mid-ocean ridge earthquakes. Combined, these two studies suggest empirically that slower spreading ridges, such as the Gelai rift, are associated with thicker seismogenic crusts, deeper faults and deeper magma chambers. Our results lend additional observational support to these propositions.

The observations from the three events represent different stages in the evolution of volcanic rifts. At the embryonic stage of a rift, magma is transported upwards from deep-seated chambers. Faulting is triggered at relatively large depths and the seismic/aseismic moment ratio is high. The rift does not show any topographic depression at that stage. As the rift evolves, the crust thins, heat flow increases and magma is transported to shallower chambers. Laterally propagating dykes trigger shallow earthquakes along their paths, and the seismicity accompanying these dykes is significantly lower than that associated with the deeper dykes of the embryonic stage. At late stages a topographic depression is developed. The Gelai event thus represents the more nascent of the three rifts, the Dabbahu represents the most evolved, and the HL event represents an intermediate maturation stage, in which the rift is limited to a single volcanic edifice, it does not form any surface depression (Fig. 8) yet the faults are relatively shallow, and closer in their form to the more evolved Dabbahu faults.

6 Summary and Conclusions

The 2009 HL rifting event is the third of its kind along the Red-Sea–East African Rift plate boundary in the past 4 yr that was captured by InSAR. As with the previous 2005 Dabbahu (Afar) and 2007 Gelai (Tanzania) events, a temporally continuous, almost complete InSAR data set enabled analysis of the entire event. The ∼NS orientation of the rift and of the satellite tracks allow us to assume negligible rift-parallel components of deformation in the satellite LOS and resolve the vertical and rift-perpendicular deformation using InSAR data alone. We find that the deformation during the rifting event was controlled by an ∼12-km long dyke, with maximum thickness of 2.5 m, and two normal faults that bound a wedge-shaped graben above the dyke. The dyke dominated the far field deformation whereas the faults dominated the near-field deformation. We compared the geometric, structural and seismic characteristics of the three rifting events. The HL and Gelai rifts are almost similar in size (dyke and fault lengths, graben width) and in their geodetic moment, however differ in their seismic/geodetic moment partitioning. The 2005 Dabbahu event is different in size (thicker dyke and longer and narrower graben) however shows an almost similar seismic/geodetic moment partitioning as the HL event. We suggest that the differences in moment partitioning are related to the different crustal thicknesses. The Gelai faults are deeper and steeper than faults in the other two rifts, and were thus capable of releasing a relatively larger seismic moment. These and other observations suggest that the Gelai event represents a more juvenile stage of rifting, in which the crust is the thickest, magma chambers are the deepest, extension rate is low and rifting is restricted to a single volcanic edifice. The Dabbahu event represents the most evolved rifting stage of the three: the crust is attenuated, extension rate is the highest, magma chambers become shallower, rifting extends along the plate boundary, and dykes propagate laterally along the rift. The HL event is juvenile in the sense that rifting is still limited to a single volcanic edifice, extension rate is low and the crust is only slightly attenuated, yet the shallow faults and the low seismic/geodetic moment partitioning indicate that some progress towards maturity has already been made.

Acknowledgments

We are grateful to Meir Abelson, Vladimir Lyakhovsky and Gadi Shamir for their comments and discussions on an earlier version of this paper. GJI Editor Michel Diament, and the two reviewers Juliet Biggs and Cécile Doubre are thanked for their thorough reviews, which improved this paper significantly. ENVISAT ASAR and ALOS PALSAR images were provided by the European Space Agency (ESA) under project C1P.5544. YH acknowledges the support of US–Israel Binational Science Foundation (BSF) grant number 2008248.