A multidisciplinary study of the final episode of the Manda Hararo dyke sequence, Ethiopia, and implications for trends in volcanism during the rifting cycle

Abstract The sequence of dyke intrusions between 2005 and 2010 in the Manda Hararo rift segment, Ethiopia, provided an opportunity to test conceptual models of continental rifting. Based on trends up to dyke 13 in the sequence, it was anticipated that, should magma supply continue, dykes would shorten in length and eruptions would increase in size and decrease in distance from the segment centre as extensional stress was progressively released. In this paper we revisit these predictions by presenting a comprehensive overview of the May 2010 dyke and fissure eruption, the 14th and last in the sequence, from InSAR, seismicity, satellite thermal data, ultraviolet SO2 retrievals and multiple LiDAR surveys. We find the dyke is longer than other eruptive dykes in the sequence, propagating in two directions from the segment centre, but otherwise fairly typical in terms of opening, propagation speed and geodetic and seismic moment. However, though the eruption is located closer to the segment centre, it is much smaller than previous events. We interpret this as indicating that either the Manda Hararo rifting event was magma limited, or that extensional stress varies north and south of the segment centre.

Dyke intrusion has been described as 'the quantum event of oceanic sea floor accretion' (Delaney et al. 1998). It is the process that initiates and maintains segmentation at rift systems once the crust is thin enough and the magma supply high enough that intrusion of dykes occurs preferentially to normal faulting for a given extensional stress (Hayward & Ebinger 1996;Buck 2004;Wright et al. 2006). The intrusion of dykes is also the principal mechanism by which magma is transported to and erupted at the surface in extensional settings (Head et al. 1996;Delaney et al. 1998). In addition, it can significantly modify the temperature and strength of the surrounding lithosphere through heating (Daniels et al. 2014). Dyke intrusion and related extrusion is thus responsible for constructing much of the Earth's oceanic crust; controlling the transition from continental rifting to seafloor spreading; governing late-stage rift morphology; and driving eruption dynamics. An understanding of dyke intrusion is essential for understanding these processes, as well as assessing seismic and volcanic hazard in extensional settings.
Dykes at rift segments frequently occur in swarms over a timescale of years separated by quiescent periods of hundreds to thousands of years, giving rise to the 'rifting cycle' (Ebinger et al. 2013). Globally, rifting events occur frequently, but typically at mid-ocean ridges making them difficult to observe. There have only been three wellobserved subaerial rifting events in the modern era, the Krafla Fires in Iceland (Bjornsson et al. 1977;Einarsson & Brandsdottir 1980), the Asal Ghoubett rifting event in Djibouti (Ruegg et al. 1979;Ruegg & Kasser 1987;Doubre et al. 2007a, b) and the Manda Hararo rifting event in northern Afar (Wright et al. 2006;Ayele et al. 2009;Grandin et al. 2009Grandin et al. , 2010aKeir et al. 2009;Ebinger et al. 2010). Recent events at the Manda Hararo rift provide a rare opportunity to study dyke intrusion and lava extrusion in a rift segment with the full suite of geophysical techniques available in a subaerial setting.
The Manda Hararo rift is one of the southernmost segments in the Red Sea rift system where it steps right-laterally across the Danakil Alps microplate into northern Afar, as shown in Figure 1a ). The segment is transitional between continental rifting and oceanic seafloor spreading Ferguson et al. 2013), and consists of a rifted central valley with a mid-segment central high containing a large collapse structure, and a significant stratovolcano, Dabbahu, at the northern end (Field et al. 2012a, b;Medynski et al. 2013). The segment can be considered transitional between the axial volcanic range morphology of the northern Afar rift segments (Erta 'Ale, Tat Ali, Alayta) and the rift valley graben morphology of southern Afar and the Main Ethiopian Rift (Bizouard et al. 1980;Hayward & Ebinger 1996;Field et al. 2012a, b). The current Manda Hararo rifting event began with the intrusion of an 80 km-long, 10 km-deep and 8 m-wide dyke running the full length of the segment between 20 September and 4 October 2005 (Wright et al. 2006;Ayele et al. 2007). This began with seismicity beneath Dabbahu at the segment tip, which then jumped to the segment centre and subsequently propagated north and south along the rift, indicating the involvement of both central and satellite magma chambers ). Thirteen more dykes were intruded including that of May 2010, with successive dykes typically propagating in one principal direction from the segment centre, in a complex pattern that has been interpreted as a function of (1) pre-existing heterogeneities in stress along the rift, and (2) 'unclamping' by previous dyke intrusions (Hamling et al. , 2010Grandin et al. 2010a). The earthquake swarms associated with the dykes originated at the segment centre in a region approximately 5 km in radius and .6 km depth (Belachew et al. 2011): satellite geodesy revealed little to no deformation during intrusion except for the November 2007 and October 2008 intrusions, where a Mogi source cannot account for the volume intruded ), while between intrusion deformation has been modelled as being consistent with re-inflation of a shallow composite sill -dyke magma body and a deeper body (Grandin et al. 2010b;Hamling et al. 2014).
The advance of the dykes, as recorded by propagating seismic swarms, decreased exponentially with time, which is consistent with decreasing pressure in the source magma chamber (Rivalta 2010;Belachew et al. 2011;Rivalta et al. 2015). The August 2007 and June 2009 dykes were both associated with basaltic fissure eruptions 10 km south of the segment centre in the Karbahi graben (Ferguson et al. 2010) and were unusually aseismic, shallower and smaller compared with the non-eruptive dykes (Belachew et al. 2011).
Based on the evolution of the Krafla dyke sequence, numerical models (Buck et al. 2006) and the Manda Hararo dyke sequence up to dyke 13, it was proposed that as extensional stress in the rift segment is released, dykes have less of a propensity to be drawn away from the segment centre to accommodate extension; instead becoming shorter and rising vertically, resulting in eruptions of increasing size and proximity to the segment centre, provided magma supply is maintained (e.g. Ferguson et al. 2010). The 2007 and 2009 eruptions established a trend of increasing erupted volume and increasing ratios of extruded to intruded magma, consistent with this model. In this paper we test this prediction by integrating interferometric synthetic aperture radar (InSAR), seismicity, light detection and ranging (LiDAR) and thermal and ultraviolet satellite observations of the May 2010 dyking event.

Methods
Here we outline the InSAR and seismicity methods used to recover the dyke dimensions and timing of emplacement, the thermal and LiDAR observations of the erupted lavas and the recovery of the total SO 2 emitted using ultraviolet satellite images.

InSAR
We used two interferograms derived from Envisat Advanced Synthetic Aperture Radar (ASAR) data covering the dyke intrusion on descending tracks 464 and 49. The radar acquisitions on track 464 were on 24 April and 29 May 2013 and on 30 April and 4 June 2013 for track 49. Both of the interferograms used in this study were processed using JPL/Caltech ROI PAC software (Rosen et al. 2004). The topographic phase was removed using a 3-arcsecond, 90 m resolution digital elevation model (DEM) generated by the NASA Shuttle Radar Topography Mission (SRTM) (Farr & Kobrick 2000) and a power spectrum filter was applied (Goldstein & Werner 1998). Interferograms were unwrapped using the branch-cut method of Goldstein et al. (1988), with errors fixed manually. To model the intrusion, we assumed the same geometry as Hamling et al. (2009Hamling et al. ( , 2010 and performed a joint inversion of both datasets to solve for the best-fitting distributed opening model.

Seismicity and moment tensor solutions
We used seismic data recorded during May 2010 in Afar by the 2009-11 Afar 0911 network of 12 Guralp ESPD seismometers (Fig. 1a). We used Hypo2000 (Klein 2002) to locate 308 earthquakes from arrival time measurements on clear P-and S-wave arrivals filtered using a bandpass filter of 1 to 15 Hz. We employed a crustal 1D seismic velocity model constrained by the results of wide-angle seismic experiments in Afar (Makris & Ginzburg 1987), but to avoid artificial clustering of earthquakes at sharp steps in the velocity model we included a linear increase in seismic velocity within each layer (Tarasewicz et al. 2012). The resulting locations have average horizontal errors of ,500 m and errors in depth of ,2 km. Local magnitudes (M L ) of the earthquakes were computed from the maximum zero-to-peak amplitude measured on simulated horizontal component Wood-Anderson displacement seismograms after removal of instrument response (Richter 1935). These measurements were used in conjunction with hypocentral distances to estimate M L using the distance correction applicable to the Main Ethiopian Rift (MER) , as employed in Keir et al. (2011). These analyses also showed that the magnitude of earthquakes ranges from M L 0.4 to 4. Seismic moment release (Mo) is determined using the empirical relationship derived between M L and Mo by Hanks & Kanamori (1979).
The Centroid Moment Tensor (CMT) algorithm based on that of Kao et al. (1998) is used to determine the source mechanisms for the largest earthquakes. The algorithm uses full waveforms from the vertical, radial and transverse components. Green's Functions for each station are calculated at an array of depths between 1 and 25 km in increments of 1 km beneath a fixed epicentre using a technique based on that of Zhu & Rivera (2002). An average 1D, layered velocity model is used. Moment-tensor inversion is performed for each trial focal depth in the depth array, and the misfit in waveform is calculated for each component of a station: where E i is the waveform misfit between the synthetic and the observation for each component of a given station, f i (t) and g i (t) are the amplitudes of the ith observed and synthetic waveforms, respectively, while f i (t) max and g i (t) max are the maximum amplitudes of these waveforms. E i ranges from zero (perfect match) to two (extreme mismatch). Misfit values for all the components from all the stations considered are combined to form an average waveform misfit (E ave ) at a given trial hypocentre: where N is the number of components used. From this a misfit-depth function is generated, and the moment tensor and hypocentre related to the minimum average misfit is considered to represent the best source mechanism estimate. We selected local earthquakes for CMT analysis that had a magnitude of more than three. In addition, a minimum of six stations was considered for the source inversion based on visual inspection of waveform quality and azimuth of the station with respect to the earthquake. A layered 1D P-wave velocity model of Jacques et al. (1999) derived from earlier seismic refraction studies in Afar (Makris & Ginzburg 1987) and V p /V s ratio of 1.8 determined from regression of travel times (Belachew et al. 2011) were used. To minimize the effect of the oversimplified velocity model and variation in the noise level at each station, the solutions were tested using three bandpass filters: 0.02 -0.06, 0.03-0.08 and 0.05 -0.1 Hz. The bandpass filter between 0.03 and 0.08 that resulted in a minimum average misfit among the three is chosen for all the events considered. The CMT solutions for ten events are given in Table 1.

Satellite measurements of SO 2 emissions
SO 2 emissions associated with the 21 May 2010 eruption in the Manda Hararo rift were detected from space by the Global Ozone Monitoring Experiment 2 (GOME-2) aboard the MetOp-A satellite and the Ozone Monitoring Instrument (OMI) aboard NASA's Aura satellite. GOME-2 and OMI are hyperspectral ultraviolet (UV)/Visible spectrometers with sensitivity to lower tropospheric SO 2 emissions. Infrared satellite instruments that have poor sensitivity to the lower troposphere, such as the Atmospheric Infrared Sounder (AIRS) on Aqua and the Infrared Atmospheric Sounding Interferometer (IASI) on MetOp-A, detected no SO 2 during the eruption, from which we infer a plume altitude of less than 5 km.
For our analysis we use the operational SO 2 products derived from each satellite instrument. GOME-2 SO 2 data were obtained from the Support to Aviation Control Service (SACS; Brenot et al. 2013), which archives daily GOME-2 SO 2 retrievals (e.g. Rix et al. 2009;SACS 2012). AIRS and IASI SO 2 data were also examined using the SACS data archive. The operational OMI SO 2 product is publicly available from NASA (2014) and provides vertical column amounts of SO 2 retrieved using the operational OMI SO 2 algorithm (Yang et al. 2007).

Airborne LiDAR
Two high-resolution airborne LiDAR datasets were acquired in October 2009 and November 2012 covering the central section of the Manda Hararo rift segment. From both datasets we generated DEMs of 0.5 by 0.5 m pixel resolution with a vertical accuracy better than 0.2 m. The change in height between the two surveys attributable to lava flow emplacement was calculated by subtracting the preeruption surface from the corrected post-eruption surface. We corrected the post-eruption surface by removing the dyke-induced uplift signal (0.2-0.3 m). This was done by fitting a linear regression plane to the fault block hosting the eruption, excluding the new lavas, and subtracting it from the surface.

Thermal remote sensing
We used thermal observations from two spaceborne sensors, the Advanced Land Imager (ALI) aboard NASA's EO-1 satellite (Ungar et al. 2003;Davies et al. 2006) and SEVIRI aboard EUMETSAT's Meteosat satellite (Aminou 2002). The ALI instrument collects images in nine bands in the visible to short-wave infrared sampled at 30 m intervals and one band in the panchromatic at 10 m, and can thus resolve geomorphological and high temperature features at sub-lava-flow scales. The satellite is in a low Earth orbit with a repeat period of 16 days. Here, we use an image acquired at 07:19 UTC on 22 May 2010, the first acquired after the eruption, to capture the final extent of the lava flow and show that the whole region of elevation change captured by the LiDAR surveys was anomalously hot and can be confirmed as emplaced lava. The SEVIRI instrument collects images in 11 bands across the visible and thermal spectrum sampled at 3 km intervals and one band at 1 km, with resolution decaying to the edge of the image of the Earth disc. The satellite is in a geostationary orbit and collects an image once every 15 min, capturing the variation in total radiant output of an eruption in great detail, and with a consistent view geometry, making successive measurements in a given pixel directly comparable.
The signal received at the SEVIRI sensor is a function of reflected and scattered sunlight from the ground, cloud and atmosphere, and thermal emission from the ground, cloud, hot lava and atmosphere, and we extract the volcanic thermal signal from this mixture using Independent Component Analysis (ICA) (Hyvarinen et al. 2001;Stone 2004), as shown in Barnie & Oppenheimer (2015). ICA is similar to Principal Component Analysis, except that instead of decomposing the radiance time series into linearly uncorrelated components we extract components that are statistically independent of each other by maximizing their non-Gaussianity. The assumptions of independence and non-Gaussian distribution prove sufficient to isolate the thermal signal from other processes contributing to the signal at the sensor. In this study we use ICA to extract the volcanic thermal signal from the short-wave infrared (1.6 mm) and mid-infrared (3.9 mm) channels. The short-wave infrared is pre-processed by finding linear relationships with other bands lacking thermal anomalies and subtracting the predicted radiance to suppress the background, making it easier for the ICA algorithm to isolate the volcanic signal. The mid-infrared is corrected for saturation effects by excluding saturated pixels from the ICA algorithm under the assumption that neighbouring pixels faithfully record the

InSAR and seismicity
The InSAR and seismicity data show the dyke was intruded along the same axis as the earlier dykes in the sequence (see Hamling et al. 2009;Belachew et al. 2011), with the seismicity lying mostly within the axial graben-bounding faults of the deformation model used by Hamling et al. (2009Hamling et al. ( , 2010 to model fault slip associated with earlier intrusions (Fig. 1). The distribution of opening inferred along the model by inverting the interferograms reveals the intrusion of two dyke segments north and south of the rift segment centre, each with a separate locus of maximum opening, as shown in Figure 2. The model predicts maximum opening of 1.6 m near the surface in the northern segment below the eruption site, with opening of up to 1 m occurring down to 4 km. An opening of c. 0.5 m is predicted in the southern segment at depths of between 5 and 9 km. The volume intruded by the dyke was 0.081 km 3 and produced a geodetic moment of 3.216 × 10 18 N m, over two orders of magnitude larger than the total seismic moment of 9.37 × 10 15 N m, indicating the majority of the dyke emplacement was accommodated aseismically, as occurred during previous events (Belachew et al. 2011). The seismicity is mostly contained within the region of significant opening, with the exception of a region at the segment centre that is largely aseismic. This aseismic region has been observed by previous authors (Belachew et al. 2011) and attributed to the presence of a shallow magma chamber or hot weak crust, consistent with it being the source region for dykes propagating north and south. Seismicity in the northern dyke segment extends from the surface down to roughly 6 km, while seismicity in the southern segment spans a depth range of 2-9 km, and in both cases is mostly confined to the region of significant opening revealed by InSAR observations. Earthquakes for which CMT solutions are available are all shallow (1-3 km depth), in the northern dyke segment, and consistent with extension slip on shallow normal faults (Fig. 2).
Earthquake depths constrained in the moment tensor inversion are consistent with the depths of the same earthquakes achieved during earthquake location.

Lava flow geomorphology
The eruption site has not been visited by any geoscientists, and the authors have been unable to find eyewitness accounts of the eruption itself, or of the lava flow post-emplacement, so the geomorphological interpretation is drawn from the difference between two LiDAR surveys and an ALI multispectral image, as shown in Figure 3. The difference between the before and after eruption DEMs reveals the thickness of the new material emplaced or removed. The positive topographic anomaly matches the thermal anomaly recorded in the short-wave infrared in the ALI image acquired 07:19 UTC 22 May 2010, showing that all of the new topography was emplaced as hot volcanic material. The eruption appears to consist of two lava flows, one situated on the footwall of a normal fault, which consists of a lava flow up to 2 m thick, sourced from a fissure about 600 m long, interspersed with low scoria cones and ramparts less than 2 m high. The southern end of the fissure terminates before the fault scarp, and below the fault scarp on the hanging wall is a second lava flow that appears disconnected from the first. This second flow abuts and follows the fault scarp; however, it is difficult to make out the source and direction of flow from the geomorphology as revealed by the LiDAR: there are no channelization features, and the flow surface is very smooth compared to the footwall flow. A plausible source is through the fault itself, perhaps through the collapse features revealed as a negative change in topography. The southernmost end of the fissure above the fault scarp and the entire lava flow below it exhibit higher radiance in the ALI image than the rest of the fissure and flow. This could be due to the material having been emplaced last and so still being hot. However, the image is acquired a full day after the eruption, which itself lasted only 6 h, and variation in surface temperature after a long interval is more a function of surface type (aa/pahoehoe) with smoother surfaces retaining heat, so this may indicate that the emplacement of the southern flow was a less intense eruption of pahoehoe lava. (Continued) distribution of openings on the face of the dyke, and the location of seismicity projected onto the dyke. All seismicity between 20 May 12:00 UTC and 23 May 00:00 UTC and within 3 km of the dyke is shown, with size proportional to magnitude and colour to date and time. The right and lower panels show the vertical and horizontal propagation of the seismicity during the first 6 and 12 hours, respectively. Isotachs of constant velocity are shown in grey, and the period of lava emission is shown as a red line. Note that seismicity originates at the centre of the segment, migrates upwards, then laterally and then down.

SEVIRI thermal observations and effusion rates
The onset of the eruption can be seen in both the 1.6 and 3.9 mm wavebands between 00:57 UTC and 01:12 UTC on 21 May 2010, while the signal drops to zero by 06:00 UTC at 1.6 mm and decays slowly over the next 24 h at 3.9 mm (Fig. 4). We interpret this difference in waning signal as a function of the non-linear relationship between temperature and radiance at different wavelengths such that shorter wavelengths are more sensitive to the hotter temperatures of freshly exposed material, while longer wavelengths are more sensitive to lower temperatures of older material, and thus are a measure of freshly exposed material and older cooling material (e.g. Davies et al. 2010). In other words, we can think of the radiance at short wavelengths as a function of the instantaneous rate of emplacement of hot material, and a convolution of that rate at longer wavelengths. We therefore take the short-wave radiance as some measure of the rate of exposure of fresh hot material within the field of view of the instrument. If we assume that there is a linear relationship between the effusion rate, rate of exposure of hot material, and radiance, we can normalize the short-wave radiance curve by the total volume of erupted material to give an estimate of the effusion rate. This gives a peak effusion rate of 24 m 3 s 21 .

Sequence of events
By integrating observations from seismicity and satellite thermal observations we can reconstruct the sequence of events during the dyke intrusion and eruption. In the following section the positions of events are given relative to the dyke as 'north' and 'south' referring to directions along the strike of the dyke on to which earthquake locations are projected. Times are in UTC.
Prior to the intrusion, between the 17 and 19 May, there was some minor seismicity (M L 1 -1.6) along the rift axis to the north of the May 2010 dyke intrusion. The initiation of the dyke injection is revealed by a cluster of M L 0.6-1.37 earthquakes on the 20 May between 18:09 and 18:47 located at the segment centre (Fig. 2b) that progressively shallow through time from 5.6 to 3.5 km. The low density of the earthquakes makes estimating propagation speeds difficult, but by plotting the seismicity over isotachs (lines of constant speed), the propagation speed can be estimated to the first order (Fig. 2b), giving a vertical motion of magma on the order of approximately 1 m s 21 . After a brief hiatus, earthquake activity recommences at 20:21 and is characterized by the simultaneous migration of seismicity north and south by approximately 8 and 10 km, respectively, reaching their maximum northerly and southerly extents at roughly 00:00 and 08:00, respectively, on 21 May, and giving northerly and southerly mean propagation speeds on the order of 0.35 and 0.25 m s 21 , respectively (Fig. 2b). During the propagation of the of the dykes north and south, the range of earthquake magnitudes rises to M L 2 to 4, with the majority of the seismic moment being released between 20 May 22:00 and 21 May 05:00 in the northern dyke segment, and 35% of the total moment released by a single earthquake of M L 4.28 at 3:13 on 21 May (Fig. 5).
The eruption began between 00:57 and 01:12 on 21 May, after the northern dyke appears to have reached most of its final extent, with effusion rates rising to a peak of around 24 m 3 s 21 at around 02:27 and dropping to zero by 06:27. The eruption coincides with a cluster of earthquakes that range from M L 2.7 to 4.28, occurring to the north of the eruption site. This cluster contains nine of the 10 events for which CMT solutions are available (see Fig. 1b, Table 1), with M L 3.3-3.7, depths between 1 and 3 km and solutions consistent with slip on shallow normal faults. The frequency of earthquakes drops rapidly at the end of the eruption. In contrast, in the southern dyke segment, seismicity persists until the morning of 22 May, with magnitudes dropping to approximately 1 (Fig. 5), and seismicity migrating to greater depths and back towards the centre during late 21 May and early 22 May (Fig. 2b).
The peak in seismic moment release during earthquake migration is consistent with the earthquakes being generated by localized extensional stresses near the tip of the propagating dyke (Roman & Cashman 2006). Earthquake magnitudes are greatest at the latter stages, or just after the end, of dyke growth, as noted for earlier dykes in the Manda Hararo sequence by Grandin et al. (2011), who explained this by a combination of growth of a seismogenic 'process zone' at the dyke tip and halting of dykes by inelastic deformation such as faulting. After the dyke stops propagating, there is no longer a mechanism for inducing large stress changes, but low magnitude earthquake activity continues around the stalled dyke tip perhaps due to small-scale faulting in response to processes such as dyke cooling, induced fault-related stress changes, or small volumes of continued intrusion  Barnie & Oppenheimer 2015) and the continuous line shows the 1.6 mm signal. Note that the 3.9 mm signal takes longer to decay to zero, which is a function of the signal being more sensitive to lower temperatures and thus picking up a cooling signal from the lava flow after emplacement. The 1.6 mm signal is only sensitive to the hottest temperatures, and so more faithfully tracks the rate of exposure of fresh hot material, which we take as a proxy for effusion rate.
into the already opened dyke. The seismic moment release associated with the dyke north of the segment centre is larger than that to the south: this may be due to the shallower northern dyke triggering faulting in cooler rock close to the surface that is more susceptible to brittle fracture, as well as due to the larger volume of the northern dyke.
In summary, the seismicity data suggest the dyke was intruded in two segments. The first propagated laterally 8 km north during 18:09 20 May to 00:00 21 May and was associated with an eruption in the period 01:00 to 06:27 21 May with peak effusion at 02:27 h, after which the frequency of earthquakes drops off rapidly. The second dyke propagated 10 km southward in the interval 18:09 20 May to 08:00 21 May, after which seismicity migrated downwards and back towards the segment centre while slowly decaying in frequency and magnitude.
Mass balance for sulphur emission GOME-2 detected SO 2 immediately SE of the eruption site at 6:54 on 21 May (approximately 6 h after onset of the eruption), which is apparent on SO 2 images produced by SACS (SACS 2010;Brenot et al. 2013). The SO 2 column amounts were too low and localized to trigger an SO 2 alert by the SACS system. However, daily GOME-2 SO 2 retrievals (e.g. Rix et al. 2009) available from SACS (2012) indicate two clearly anomalous (i.e. above background) pixels containing 9.2 and 7.4 Dobson Units (DU) of SO 2 (1 DU ¼ 2.69 × 10 16 molecules cm 22 ), assuming a lower tropospheric SO 2 plume altitude of approximately 2.5 km. Using the GOME-2 pixel dimensions of 80 km × 40 km, these column amounts yield a total SO 2 mass of approximately 1.5 Gg. Adjustment for a small background noise offset of 1 -2 DU would reduce this amount to approximately 1.15 -1.3 Gg. OMI SO 2 measurements at 11:21 UTC on 21 May confirm the presence of volcanic SO 2 . Column amounts retrieved using the operational OMI SO 2 algorithm (Yang et al. 2007), also assuming a c. 2.5 km plume altitude, yield a total SO 2 mass of c. 0.9 Gg. The difference between the GOME-2 and OMI SO 2 measurements may be explained by SO 2 removal (e.g. via oxidation or deposition) in the 4.5 h between satellite overpasses and also to the OMI row anomaly (OMI Row Anomaly Team, 2012), a data gap, which may have obscured some of the SO 2 cloud. Neither GOME-2 nor OMI measured significant SO 2 amounts above background on the following day, confirming the short timescale and low magnitude of the event.
The SO 2 emission can be used to estimate the volume of magma that degassed to produce it based on an assumed sulphur content of the melt before and after eruption. Following Ferguson et al. (2010), we take this value as 900 ppm by mass of S in the melt. The range of 1.15-1.3 Gg of SO 2 then suggests derivation from a magma volume of 0.26-0.29 × 10 6 m 3 (based on a density of 2500 kg m 23 and without accounting for crystal fraction). The differenced DEMs indicate a total emplaced lava volume of 0.23 × 10 6 m 3 . Given the uncertainties in the key parameters (notably sulphur content available for degassing and the density of the emplaced lava flow), the similarity between the volumes suggests that degassing of the erupted lava can account for the emitted SO 2 . Ferguson et al. (2010) reached the same conclusion for the earlier eruptions on the rift. The volume of the dyke modelled from the InSAR geodetic signal is 0.081 km 3 , nearly two orders of magnitude greater than the volume of erupted lava. The low ratio of erupted to unerupted magma breaks the trend of increasing ratios from the 2007 and 2009 Manda Hararo fissure eruptions expected if the rift cycle is following the same pattern as Krafla (Ferguson et al. 2010).

Discussion
Seismicity reveals the dyke initiated in the (relatively) aseismic centre of the segment at around 6 km depth on 18:09 20 May, propagating at speeds of 0.25-0.35 m s 21 laterally and 1 m s 21 vertically. The initial vertical propagation of the dyke involves some seismicity in an otherwise aseismic region (as noted for other intrusions, e.g. Belachew et al. 2011), and the seismicity in this region during the May 2010 dyke intrusion is (1) only during the initial injection, and (2) only tracking the vertical motion at the very centre, which is consistent with the seismicity tracking the propagation of the tip during the initial stages of dyke propagation Dahm et al. 2010;Belachew et al. 2011). Seismicity propagates north and south from the segment centre, emplacing two dyke segments of approximately 8 and 10 km length, which show differing final geometries and variations in seismicity.
The northern dyke is shallow, extending from the surface to 6 km depth with a maximum opening of 1.6 m near the surface. Seismicity suggests that it propagates faster than the southern dyke and produces an eruption after northward propagation has (mostly) finished. The majority of the seismic moment is released during the eruption, and seismicity almost ceases after the eruption stops.
In contrast, the southern dyke extends from 2 to 9 km. The migration of seismicity suggests that it propagates more slowly and there is no eruption associated with the intrusion. Seismicity continues for a day after seismicity at the northern segment has stopped and propagates downwards before migrating back towards the segment centre during emplacement of the southern dyke. It is unclear whether this back-propagation of seismicity represents dyke propagation or represents seismic activity in a previously aseismically intruded region. Comparing the distribution of seismicity with the distribution of opening, the latter scenario seems more likely than the dyke propagating by the tortuous route mapped out by the seismicity.
One explanation for the difference in behaviour between the two dykes is that the eruption in the northern segment 'depressurizes' the dyke, resulting in an end in seismicity after the eruption stops, whereas seismicity in the still-pressurized southern segment continues. However, the volume of the physically erupted lava is almost two orders of magnitude smaller than the volume of intruded material, and the mass of the SO 2 cloud can more or less be accounted for by degassing of the erupted lava, so it seems unlikely that the eruption of these comparatively low amounts of lava and gas had a significant impact on the dyking process. By the start of the eruption the dyke is nearly at its full northerly extent, and the magnitude of the earthquakes peak during the eruption and are mostly shallow (those with CMT solutions indicating extensional normal faults) so the earthquakes could be tracking the response of the shallow normal fault systems to stresses imparted by the final vertical intrusion and widening and narrowing of the vent over the course of the eruption. Alternatively, both the shallow earthquakes and evolution of the vent could be reflecting broader late-stage dyke intrusion at shallow levels.

Comparison with expectations based on previous Manda Hararo dykes, the Krafla rifting event and theoretical considerations
The May 2010 dyke is consistent with the sequence of dykes in the Manda Hararo rift so far, as shown in Figure 6. Like the other dykes associated with eruptions, it has a relatively low volume and low geodetic and seismic moments. The maximum dyke opening is larger than the previous two eruptive dykes and more similar to the average opening of non-eruptive dykes. The larger opening coupled with longer length due to the two dyke segments north and south of the rift centre results in a larger volume than the previous two eruptive dykes. Estimates of mean dyke propagation rates from seismicity for dykes with eruptions are not reported by Belachew et al. (2011)  The subaerial rifting event at Krafla during the period 1975-84 culminated in a significant increase in erupted volumes during dyke injection (e.g. Tryggvason 1984). This observation has been explained by a progressive decrease in extensional tectonic stress during the episode, with later dykes more likely to erupt, since insufficient extensional stress is available to drive complete subhorizontal propagation and intrusion (Buck et al. 2006). As a result, surplus magma reaches the surface closer to the magma chamber and in larger volumes. While an increase in eruptive behaviour of dykes has been observed during the Dabbahu rifting episode, the total eruptive volume remains far smaller than that observed during the Krafla rifting episode. This could indicate either that significant extensional tectonic stress remains within the Dabbahu segment or that magma supply has diminished. Alternatively, the small volume and duration of the May 2010 eruption compared to the August 2007 and June 2009 eruptions may be explained by greater extensional stress to the north of the segment centre, as proposed by Grandin et al. (2010a). The greater intrusive volume of northern dykes v. the greater extrusive volume of southern dykes may simply reflect a greater amount of  extension to be accommodated north of the segment centre, leaving little magma left over for extrusion. Additionally, the higher elevation of the May 2010 eruption site may result in magma being intruded in the north that could have reached the surface had the dyke propagated south.

Conclusions
In summary, the May 2010 Manda Hararo dyke is the 14th, and perhaps last, in the Manda Hararo rifting sequence that started in September 2005, giving a rift cycle length of 4 years and 9 months. The dyke mostly falls within the range of parameters (volumes, seismic and geodetic moment, maximum opening) of previous dykes in the series and was intruded along the same axis. The dyke takes the form of a northern and southern segment propagating away from a mostly (but not initially) aseismic central zone, as observed for previous dykes and, like earlier eruptive dykes, was relatively aseismic. The northern and southern dyke segments exhibit different patterns in seismicity and opening with time and space, which may be related to the eruption in the north, a function of local residual stresses from previous dyke intrusions in the sequence, or a function of differences in extensional stress north and south of the segment centre.
The May 2010 episode does not follow the pattern of increasing extrusive to intrusive ratios seen at Krafla (Tryggvason 1984) or modelled by Buck et al. (2006), or anticipated subject to sustained magma supply by Ferguson et al. (2010); however, the eruption did occur closer to the rift centre. This might be expected once extensional stress has been relieved and is no longer diverting magma laterally: however, the dyke was unusually long. In this case, the eruption occurring close to the centre may be more a function of the patterns of extensional stress just happening to provide a path of least resistance to the surface near the magma chamber. The low volume of erupted material may indicate that the rift cycle was ultimately magma limited, and extrusion stopped due to inadequate magma supply before the extensional stress was fully relieved, or it may reflect differing eruptive styles north and south of the segment centre due to differing crustal stresses.