Introduction

The Afar depression lies at the triple junction between the Red Sea, Gulf of Aden and East African rifts, and has been produced by rifting between Africa and Arabia over the past ∼30 Myr (Wolfenden et al. 2005). The southern Red Sea rift has become punctuated over the past ∼3 million years with ∼60-km long magmatic segments as faulting and volcanic activity have become more focused into localised regions (Barberi and Varet 1977). Erta Ale is a basaltic shield volcano (13°36′11.41′′N, 40°39′50.08′′E) situated within the Erta Ale magmatic segment (Fig. 1) which comprises six main volcanic centres (Barberi and Varet 1970). The Erta Ale segment is the northernmost in Afar and forms the sub-aerial southwards continuation of the Red Sea rift (Prodehl et al. 1997). The last eruption in this segment occurred in November 2008 when basaltic lava covered 15 km2, originating from a fissure between Dalaffilla and Alu volcanoes (Pagli et al. 2012).

Fig. 1
figure 1

Sketch map indicating the location of Erta Ale and satellite image of the Erta Ale crater, with the extent of the November–December 2010 eruption lavas [∼0.4 km2: GPS track courtesy of James Hammond (February 2011)]. Sample locations from 22 November are shown, which were subsequently covered by further overflows. Image taken by the PRISM instrument on ALOS satellite, with an image resolution of 2.5 m. Image acquired on 19 November 2008. Data acquired from ESA

Full size image

Erta Ale is known for its active lava lakes situated in the northern and southern pits within a main 0.7 × 1.6-km elliptical summit crater (Fig. 1; a summary of recorded activity is provided in Supplementary Table 5). The Erta Ale lava lake is one of just four long-lived lava lakes existing across the world; others are Erebus (Antarctica), Nyiragongo (DR Congo) and Ambrym (Vanuatu). Shorter-lived lava lakes also appear elsewhere periodically, e.g. Kilauea, Hawaii and Villarrica, Chile.

The first systematic study of the Erta Ale lava lake began in 1967 (Barberi and Varet 1970), but the lake is likely to have existed from much earlier than this, as reports of a ‘smoking mountain’ and characteristic red summit glow were recorded by early visitors to the area (e.g. Nesbitt 1935; Dainelli and Marinelli 1907; Barberi and Varet 1970). Erta Ale volcano is remote and difficult to access, and as a result, observations have been sporadic: it was not until the CNR-CNRS campaigns of the late 1960s that it began to be studied in detail (e.g. Barberi and Varet 1970; Tazieff 1973). At this time, a lava lake existed in both the northern and southern pits, but the northern lake solidified between 1988 and 1992 following emplacement of two lava flows during this period (Vetsch et al. 1992). The southern lake's surface solidified briefly between late 2004 and April 2005 (Bardintzeff and Gaudru 2004; Yirgu and Philpotts 2005). It was reactivated in September 2005, with lava visibly overturning at the edges of the lava lake, within the pit, possibly in response to the volcano-seismic crisis in Afar (Ayele et al. 2007). The level of the southern lake has fluctuated (Le Guern et al. 1979), but the majority of recent observations indicate the lake level was generally tens of metres below the floor of the main crater, nestled at the base of the southern pit (e.g. Oppenheimer and Francis 1997). Eruptions occur from these lakes from time to time, causing lava to overflow from the pits onto the main crater floor, but they are rarely observed. Varet (1972a), reported the first recorded eruptive event from the southern pit in March 1972, following the emplacement of two new lava flows into the main crater which emanated from the southern side of the south pit; one sometime prior to 21 and one on 23 of March. This activity continued with three additional new lava flows, visible when the volcano was visited in May 1972 (Varet 1972b). Tazieff (1973) reported overflows which overran the main crater rim in March 1973. From this occurrence until November 2010, no further recorded observations were made of the southern pit filling and overflowing into the main crater in an eruptive event, or of the appearance of any fresh lava flows.

Ground observations

Ground observations began on 21 November 2010 at ∼1600 hours UTC (Co-ordinated Universal Time). The team were unaware on approaching the volcano of any eruptive activity, and no visible signs, e.g. plume, could be seen on the 8-km hike up to the crater; however, on arrival, the eruption was underway (Table 1). Observations from the previous few years located the lava lake at the base of the pit, ∼20 m depth. However, the pit had filled with lava, level with the main crater floor on the western side (Figs. 1 and 2). The first overspill of lava onto the main crater floor occurred at ∼1650 hours UTC (Fig. 3a). An embryonic scoria ring cone had built up around the lake, similar to that observed in 1973 (Tazieff 1973). Overspills out of the scoria cone into the pit, and occasionally on to the main crater floor, continued at approximately 2-h intervals until ∼1430 hours UTC on 22 November. Video footage indicates the lava was of low viscosity (Supplemental Video 1). Lava flows were inflated by volatiles and thus had hollow interiors. For the remainder of the ground observation, the lake was contained within the scoria ring cone, which by 0300 hours UTC on 23 November had grown to ∼4 m high on the southern side. The lava lake level rose and fell by ∼4 m on a ∼20-min cycle, with agitated Strombolian activity at the peak of each cycle. Between the peaks, activity at the lake surface was more subdued and constrained within the scoria ring with sporadic bubble-bursting, rather than larger-scale fire-fountaining. On the morning of 23 November, the northern side of the scoria ring cone suffered partial collapse. Blue-green gas jets could be seen periodically at various points across the lava lake, some remaining stable for several hours. New lava flows were visible within the northern pit, together with some incandescence during the night of 21 November. After departing from the summit on 23 November, a small but distinct plume was visible above the volcano from the surrounding plain.

Table 1 Summary timetable of ground observations
Full size table
Fig. 2
figure 2

Comparison of the lava levels in the south sink between (a) February 2010 and (b) 22 November 2010. The lava has filled a depth of ∼20 m. The February 2010 level is similar to that observed on the 11 and 14 November (J. Wilkinson, personal communication). c–f Cartoon cross-sections of the Erta Ale lava lake from observations by members of the field team: (c) October 2009, (d) February 2010 and (e, f) stages in the eruption of November 2010 indicating the formation of the scoria ramparts and infilling of the southern pit and sink

Full size image
Fig. 3
figure 3

a Emplacement of first observed overspill (sample EA001) ∼1650 hours UTC on 21 November 2010. b Close-up of a section through the fresh lava showing the glassy surface, plagioclase-rich layer and vesicular core. c Close-up of the surface of the new lava. d BSE image of oscillatory-zoned plagioclase rim from sample EA001 with quenched glass matrix. e Euhedral olivine and melt inclusion from sample EA003. f Sample EA002, the underlying lava flow, which has a fine-grained groundmass. In g and h Kα X-ray intensity maps in an olivine from sample EA001 are shown. g Combined phosphorous image from four spectrometers shows faint zoning throughout the crystal. h Example major element (Fe) map shows no visible zoning

Full size image

Satellite imaging

Method

The SEVIRI instrument is a multispectral imager aboard the second generation of Meteosat geostationary weather satellites. The Meteosat 9 satellite is positioned over 0°E, 0°N, and the on-board SEVIRI instrument images the full Earth disc visible from this position every 15 min in 12 bands. Four of these bands are useful for monitoring thermal radiation from volcanic products, at wavelengths of 1.6, 3.9, 10.8 and 12.0 μm (e.g. Hirn et al. 2009). The resolution of the image at the sub-satellite point is 3 km, but this increases towards the edge of the image due to the constant angular sampling of the instrument and the curvature of the Earth, so that at Erta Ale, the resolution is approximately 4 km in the E–W direction and 5 km in the N–S direction (Müller 2010). Due to the stable viewing geometry of the instrument, each pixel images the same area of the Earth's surface at the same view angle in successive images, so a sequence of radiance measurements from a given pixel is directly comparable. This is in contrast with instruments on low Earth orbit satellites such as AVHRR and MODIS, where for a given volcanic feature, the observing pixel location, size and view geometry vary substantially with each acquisition, making inter-image comparisons difficult.

We acquired 9,802 Meteosat 9 SEVIRI images of the Erta Ale region between 1 October 2010 and 10 January 2011. The images were converted to radiance using the gain and offset values stored in the image headers, and the time series of radiance values for the pixel covering the Erta Ale lava lake was extracted, giving a 15-min time resolution record of the radiant output of the eruption. Thermal anomalies were only observed in the 3.9-μm band, consistent with the presence of a very small surface area at volcanic temperatures. Larger areas would register at the 10.8- and 12.0-μm bands, saturate the 3.9-μm band and, if there was enough incandescent material (e.g. cracks, fire fountains and shear zones), also register at the 1.6-μm band (Davies et al. 2010). We subtracted the radiance from a neighbouring cold pixel to obtain the excess radiance emitted by the hot volcanic material, over and above that radiated and reflected by a region with similar reflective and emissive properties at background temperatures. This gives us a time series of the thermal anomaly associated with the eruption, which we interpret qualitatively. The lava lake and background pixel do have subtle differences in radiance patterns in the absence of hot material in the lava lake pixel, and as a result, there is a regular small dip in the thermal anomaly at dawn; however, this effect is small and does not invalidate our observations, which are first order and qualitative in nature.

Interpretation

The Erta Ale lava lake does not typically register in any SEVIRI band, despite being frequently observed by MODIS (e.g. Ayele et al. 2007; Wright and Pilger 2008). This is a function of look angle; the MODIS instruments view the lava lake from a range of angles, while SEVIRI views from a constant angle of ∼42°E, at which the lake is completely occluded from the sensor field of view by the crater rim when the lava lake is at a low stand (typically ∼20 m below the surface). As the lava lake level rises, or overflows onto the pit floor, more of the surface area of the lava comes into view of the sensor. This increases the measured radiance, resulting in a thermal anomaly when compared with neighbouring pixels containing no hot lava. The thermal anomaly varies over two time periods, one on a timescale of days, the other on a timescale of hours (Fig. 4).

Fig. 4
figure 4

SEVERI thermal signal of Erta Ale compared to contemporaneous ground observations. a Detail from when field team were present at Erta Ale showing the radiance from lava and background, and the difference, i.e. the excess thermal radiation. Time in UTC. b Overall image of the eruption showing the excess thermal radiation graph only

Full size image

The variation on a timescale of hours is characterised by sharp increases in radiance over a period of <15 min, followed by an exponential decay over a period of around an hour. These spikes in radiance have been correlated with lava overflows observed in the field, so we interpret the spikes as lava overflows: either initially onto the ‘terrace’ abutting the lake within the pit as the lake level rises or over the scoria ramparts surrounding the high stand lake, onto the new infilled surface of the pit and eventually onto the main crater floor. The rapid rise in radiance over 15 to 30 min associated with each spike is interpreted as the advance of the overflow; the subsequent exponential decay in radiance over a period of a few hours is the cooling of the surface. Each overflow can be considered a temporary increase in the area of the ‘lava lake’. Between spikes, the background radiance from the lava lake varies little as the lava lake level remains relatively constant, which we interpret to mean the surface temperature distribution varies little.

The second thermal anomaly variance is on a timescale of days. A thermal anomaly is present during two periods (Fig. 4b). The first starts with an isolated spike in radiance at 0212 hours UTC on 11 November, which decays to the background level after about an hour. The anomaly reappears slightly later at 1557 hours UTC, peaks around midnight on 11 November and then decays to 0657 hours UTC on 13 November. This overall trend is punctuated by numerous spikes in radiance. The second period begins on 1442 hours UTC on 19 November and is marked by numerous spikes in radiance. It continues until ∼1457 hours UTC on 22 November. The baseline anomaly then slowly increased with markedly fewer radiance spikes to approximately midnight on 30 November. After this time, the background radiance drops to a lower level and remains at approximately this level until early morning on 7 December, when the lava lake became obscured by cloud. The thermal anomaly decays to background between 8 December and 15 December.

This pattern in the long timescale variation in the thermal anomaly could be due to changes in lava lake height (and therefore occlusion by the rim), changes in lava lake area, surface temperature distribution, Strombolian activity, overflows or renewal of activity in the solidified northern lake. Comparison with ground observations allows us to discriminate between these possibilities. From field observations, we know the period over which the lake level rose from the former position of ∼20 m below the rim to the surface, and this corresponds with a period of steadily increasing background radiance. The final area of the lava lake at high stand is also similar to the initial area (∼500 m2). We therefore infer that the long timescale variation in radiance is primarily a function of varying degrees of occlusion of a lava lake of a roughly constant area as a result of changes in lake height, rather than a variation in total lake area as a result of increasing width with height within the pit, or freezing of the margins of the lake.

Radiance spikes during the period of ground observations correlate with overflows and collapses, while a steady low-level radiance anomaly between spikes correlates with background lake activity. The sampling interval of 15 min is insufficient to resolve the observed 20-min variation in lake height. We therefore interpret the radiance spikes as overflows, as the lava lake progressively builds itself up by first episodically flooding the pit, and then constructing the scoria ramparts by successive overflows and spatter (see Fig. 2e, f).

Petrology

Method

Three samples were collected for analysis: EA001 collected from the margin of the first overflow onto the main crater floor (Figs. 1 and 3a) and sampled approximately 12 h after emplacement; EA002 collected from the main crater floor (most likely erupted in 1973), i.e. the flow underlying the newly erupted flows; and EA003 from the last overflow observed by the field team, collected <1 h after emplacement. Whole rock major and trace elements were analysed by X-ray fluorescence (XRF) using fused glass discs and pressed powder pellets, on a PANalytical Axios-Advanced XRF spectrometer at Leicester University.

Polished thin sections were carbon-coated and imaged on a Hitachi S-3500N scanning electron microscope (SEM) prior to major element analysis using a CAMECA SX-100 electron microprobe (EMP) instrument at the University of Bristol. A 15-kV accelerating voltage, 4-nA beam current and defocused 10-μm beam were used for matrix glasses and feldspars, and 20-kV accelerating voltage, 10-nA beam current and a focused beam (1 μm) for olivines and cpx (clinopyroxene). A minimum of six analyses per phase were measured in each thin section.

Olivine crystals were handpicked from a crushed sample of EA003. The samples lose their glassy appearance towards the centre of the flow, and the matrix becomes dull more than ∼2-cm depth from the surface. Phenocrysts picked for analysis were taken from the rapidly cooled, glassy upper 2 cm of the samples, mounted in resin and cold-cured for 2 weeks prior to polishing. Petrographic observation ensured all mounted grains contained melt inclusions (Fig. 3e). These were examined and mapped on the SEM uncoated, under variable vacuum, to minimise carbon contamination. Any melt inclusions that showed obvious signs of post-entrapment crystallisation, fractures or re-entrants were rejected. All melt inclusions were bubble-free. Eleven spot analyses from five separate melt inclusions were made, analysing for CO2, H2O and a number of trace elements by secondary ion mass spectrometry (SIMS) using a Cameca ims4f instrument at the University of Edinburgh Ion Microprobe Facility. Two analyses of matrix glass were made for CO2 and H2O. We used a 10-kV (nominal) O− primary beam and 4,500 eV positive secondary ions with offset applied to reduce transmission of molecular ions. To enhance relief, grain mounts were polished with 0.3-μm alumina for 30 s prior to gold coating. CO2 (12C+) was determined first at sufficient mass resolution (∼1,600) to separate 24Mg2+ from 12C+ because both ions have the same mass-to-charge ratio. A 5-nA primary beam, with a 15-μm focused spot was used, with 50 eV offset. Twenty-five- and two hundred fifty-micrometre lenses were activated to give a 60-μm image field with field aperture No. 2 providing a ∼15–20-μm restricted central area for analysis. Entrance slits were set at ∼110 (entrance) and 112 (exit). A 2-min pre-sputter with a 15-μm raster removed surface contamination prior to using a static spot for analysis. For subsequent H2O and trace element determinations, a 5-nA primary beam with 75 eV offset and 30 s pre-sputter was focused in the same location as the CO2 analysis. Secondary ions of 1H+, 9Be+, 11B+, 12C2+, 24Mg2+, 26Mg+, 30Si+, 42Ca+, 44Ca+, 45Sc+, 47Ti+, 88Sr+, 89Y+, 90Zr+, 93Nb+, 138Ba+, 139La+, 140Ce+, 141Pr+, 143Nd+, 149Sm+, 159Tb+, 161Dy+, 165Ho+, 166Er+, 171Yb+, 178Hf+, 181Ta+, 232Th+ and 238U+ were collected at low mass resolution, with a 25-μm image field and field aperture No. 2 providing ∼10 μm area of analysis. Calibration and secondary standards included a range of glasses of known trace element, H2O and CO2 values (basaltic glasses of Shishkina et al. (2010), BCR2g, TPA, TPF, SRM610 and CFD), together with an olivine crystal from Erta Ale as a monitor of background consistency.

Mid-IR spectra for H2O and CO2 were obtained at Bristol using a Nicolet Nexus 670 FTIR spectrometer with an IR PLAN microscope attachment, Globar source, KBr beamsplitter and MCT detector (512 scans at 8-cm−1 resolution). Double-polished plates of around 50-μm thickness were prepared and measured with Mitutoyo Digital Micrometer to ±1 μm. This thickness allowed an 80-μm beam to pass through the melt inclusions (or the matrix glass) uninterrupted. Spectra were compared to a volatile-free basalt to estimate peak heights, and these were converted to concentrations using extinction coefficients of 63 L mol−1 cm−1 for water at 3,550 cm−1 and 375 L mol−1 cm−1 for one of the carbonate doublet peaks at 1,375 cm−1 (from Dixon et al. (1988) and Fine and Stolper (1986), respectively). The errors in the measured IR peak intensities (due to background subtraction), the extinction coefficient, the thickness measurement and the density estimate lead to an uncertainty of about ±20 % for water and ±35 % for CO2 in the inclusions. For the matrix glass, the CO2 peak is not visible, suggesting it is below the detection limit of ∼50 ppm for this sample thickness.

Major elements were determined quantitatively by EMPA on the same melt inclusions previously analysed by SIMS and on phenocryst minerals. High precision S and Cl concentration were also determined for matrix glass, one melt inclusion (J), and two small, previously unanalysed melt inclusions using two spectrometers for each element to increase total counts. The peak count was 120 s on each spectrometer. The spectrometers were calibrated using BCR2 [Basaltic glass, Columbia River (Wilson 1997)]. Oxygen was calculated by stoichiometry, and weight percent oxide totals for SO2 are reported. Kα X-ray intensity maps for phosphorous, Ca and Fe in olivine phenocrysts in EA001 were obtained in two stages. The first mapped P and Fe with a 256 × 256 pixel image and dwell time of 100 ms per pixel. A second run at higher resolution used dwell times of 200 ms per pixel and a 512 × 512 pixel image, imaging P and Ca. For both runs, a beam current of 400 nA, an accelerating voltage of 20 kV and beam diameter of 1 μm was used. Phosphorous was mapped on four of the five spectrometers in both cases, and the final images were summed.

Results

The lavas are porphyritic (∼20 % by volume) and vesicle-rich (up to 60 % volume in EA003, calculated by point counting; see Fig. 3b), and can be classified chemically as transitional basalts (Irvine and Baragar 1971). There are surprisingly few published chemical analyses of basalts from Erta Ale volcano. To our knowledge, the only published values come from Barberi et al. (1974) and Bizouard et al. (1980) who present only major elements on two samples from Erta Ale volcano (one from the summit crater wall, and one from a western flank fissure) and Barrat et al. (1998) who present major and trace elements of four samples from the volcano. The 2010 lavas (EA001 and EA003) are olivine-normative, whereas the older underlying lava (EA002: most likely from the 1973 eruption) is quartz-normative (Table 2). The sample taken from the underlying lava has a higher silica content than the 2010 samples, comparable to the Erta Ale sample (F12) analysed by Barberi et al. (1974) from the main crater wall. This suggests the recent input to the system of less evolved but chemically similar magma. EA002 also shows significant differences in texture (Fig. 3f); a crystalline groundmass with large glomerocrysts of plagioclase, cpx and olivine, compared to the quenched glass matrix and discrete euhedral crystals found in EA001 and 3 (Fig. 3d, e). The silica contents of the 2010 samples are comparable (<50 wt.%) to those of samples analysed by Barrat et al. (1998), although the exact locality of those samples has not been ascertained. The mineral assemblage in the 2010 samples is plagioclase + clinopyroxene + olivine. No oxides or other accessory phases are present in any sample. Euhedral plagioclase phenocrysts, up to 4 mm, show normal zoning from core to rim, but within some rims, oscillatory zoning is present (Fig. 3d), and there is evidence of resorption between some of the zones. Plagioclase is the dominant phenocryst comprising ∼90 % of all phenocrysts. It is also found as small inclusions in both clinopyroxene and olivine, with anorthite (An) content in the range An83−74. Anorthite content in plagioclase cores from the 2010 lavas (Table 3) is An85−69; the range within an individual core is more restricted, e.g. An85−82, An71−69. Within the zoned rims, An values show a wide range from An80−66, with the higher An values found in those phenocrysts with higher An cores. A very similar range of An contents is found in EA002 (An82−75 in cores, An76−66 in rims). Plagioclase groundmass microlites are only present in EA002 and range from An71−63. Clinopyroxene crystals are subhedral, small (<1 mm) augites, often forming glomerocrysts with plagioclase (Table 3). Mg# (100 × molar Mg/[Mg + Fetot]) values range from 84 to 72. One tiny (∼50 μm) euhedral clinopyroxene inclusion (Mg# 82) was found in an olivine. Olivines are euhedral with Fo84−78 content in phenocrysts and Fo80−73 in rims and EA002 groundmass (Table 3).

Table 2 Whole rock XRF analyses
Full size table
Table 3 Representative phenocryst analyses
Full size table

The raw analyses of the melt inclusions indicate compositions that are close to those of the matrix glass, suggesting limited post-entrapment modification (Table 4). Using an olivine-melt KdFe-Mg of 0.28, we estimate that post-entrapment crystallisation is less than 6 %, which would have negligible effect on the measured volatile or trace element contents. It is much more difficult to make a correction for open-system processes such as Fe diffusion through the host olivine. Consequently, the MI data have not been corrected for post-entrapment processes, and all data are reported as analysed. The H2O content of the five inclusions ranges from 867 to 1,329 ppm; CO2 contents range from 33 to 194 ppm (Table 4). Multiple analyses of single melt inclusions show them to be homogeneous for H2O and CO2. Matrix glass contains ∼560 ppm H2O and <50 ppm CO2. FTIR values for CO2 agree with those determined by SIMS (<50 ppm in matrix glass and 94–136 ppm in melt inclusions). FTIR H2O values are systematically higher than SIMS, with melt inclusion values of 1,700–2,000 and 1,700 ppm in matrix glass). The source of this systematic offset is unknown, but may originate in uncertainty in the FTIR extinction coefficients or ion-yield difference for 1H between Erta Ale glasses and the standards used. SO2 contents of melt inclusions, measured by EMP (Table 4), are 247–419 ppm; chlorine contents are 158–368 ppm. Average SO2 and Cl values from 25 analyses in matrix glass were determined as 91 and 195 ppm, respectively.

Table 4 Averaged melt inclusion major, trace and volatile analyses with host olivine analyses, and matrix glass majors and volatiles
Full size table

Temperatures were calculated using various mineral/liquid thermometers with an average of ∼1,150 °C from all methods (Table 5). Pre-eruptive temperatures derived from crystal rim/matrix combinations in plagioclase/liquid (Eq. 24a, Putirka 2008) are in the range 1,171–1,180 °C, clinopyroxene/liquid (Putirka et al. 1996) temperatures are 1,148–1,175 °C, and olivine/liquid temperatures are 1,151–1,158 °C (Putirka 2008) and 1,157 °C (Beattie 1993). Temperatures derived from melt inclusion/host crystal combinations using the method of Putirka (2008) are in the range 1,094–1,148 °C, with similar temperatures (1,104–1,148 °C) obtained from the method of Beattie (1993).

Table 5 Calculated thermometry
Full size table

Volatile saturation pressures were calculated using the method of VolatileCalc (Newman and Lowenstern 2002) and range from 7 to 40 MPa. To convert these pressures to depth, we have assumed a crustal density of ∼2,850 kg m−3 based on gravity studies in the Main Ethiopian Rift (e.g. Mahatsente et al. 1999). The calculated saturation depths are shallow: 0.2 to 1.4 km.

Trace element contents of melt inclusions (Table 4) are broadly similar to those of the host lavas with slightly elevated incompatible elements and lower compatible elements, consistent with ∼20 % crystallisation, i.e. the phenocryst content. Incompatible elements, e.g. Ba, are positively correlated with H2O and negatively correlated with CO2, despite a relatively small variation in all parameters (Fig. 6). The trace element content of the underlying lava (EA002) differs from the 2010 lavas, most noticeable in the elevated levels of incompatible Ba which may reflect the greater crystallinity of this sample.

Kα X-ray mapping of an olivine phenocryst from EA001 indicates homogeneity throughout the crystal for major elements such as Fe (Fig. 3h). However, feint concentric, oscillatory zoning is visible in phosphorous (Fig. 3g).

Integration of ground observations, satellite imagery and petrology

Ground observations and satellite imagery

In a location as remote as Erta Ale, eruptions can occur unnoticed and unrecorded. The November 2010 eruption was short lived, which is typical of recent eruptions in Afar (e.g. Ferguson et al. 2010; Wright et al. 2006), and ordinarily, the brevity would not allow scientists sufficient time to travel to the eruption site to make observations. However, from the interpretation of the SEVERI features of the radiance time series, together with serendipitous ground observations, we can make inferences about the eruption chronology.

The SEVIRI data suggest the first thermal anomaly period occurred between 11 November and 14 November (Fig. 4), and records overflows onto the southern pit floor only. By the end of the first period, the pit floor had been covered with fresh lava flows, but the lake level was still low enough that the rim of the pit obscured it from view by SEVIRI. Ground observations from 11 November confirm that the southern pit was still >20 m deep, comparable to its state 6 months earlier in February 2010 (Fig. 2a, b). However, by 14 November, the base of the pit was covered in fresh lava (J. Wilkinson, personal communication, December 2010). There then followed a hiatus of overflows, up until the start of the second thermal anomaly period, as seen by the lack of excess radiation (Fig. 4).

It was during this period of quiescence that seismic activity occurred in the Gulf of Aden [14–17 November—European-Mediterranean Seismological Centre (EMSC 2010)] and around the town of Chiro (formerly Asebe Terferi) at the northern end of the Main Ethiopian Rift. Disturbance was also reported in the Danakil region with the opening of large sinkholes and unusually high groundwater levels.

The field team arrived at ∼1600 hours UTC on 21 November when the second phase of the eruption was well underway, and the southern pit had filled. SEVERI imaging suggests this second phase of activity began during 19 November, which indicates a rapid infilling of the southern pit between the 19th and 21st (Figs. 2 and 4). Assuming little or no activity between 15 and 19 November, a flow rate of approximately 2 m3 s−1 would be required to fill the area of the pit between 19 and 21 November (53 h). Oppenheimer and Francis (1997) estimated an average magma discharge rate (MDR) of ∼400 kg s−1 for February 1972–March 1973. Assuming the same density for comparison (2,700 kg m−3), the calculated MDR for 19–21 November is ∼5,600 kg s−1. The first overspill outside of the southern pit onto the main crater floor occurred at ∼1750 hours UTC (Fig. 3a). Throughout the eruption, the lake was surrounded by a scoria ring cone which continued to build, similar to that observed in 1973 (Tazieff. 1973). Overspills continued at approximately 2-h intervals until ∼1430 hours UTC on 22 November (Supplementary Video 1).

For the remainder of the ground observations, the lake was contained within the scoria ring, which had grown to ∼4 m high by 0300 hours UTC on 23 November. A 20-min cycle followed, with lava lake levels rising and falling by >4 m, during which, peak levels were marked by Strombolian activity. The northern part of the ring suffered a partial collapse on the morning of 23 November, which corresponds to a sharp peak at 0400 hours UTC in the SEVERI imaging (Fig. 4). Periodically, incandescent blue-green gas jets were observed similar in description to gas flames reported in January 2003, assumed to be SO2 (Vetsch et al. 2003). The northern pit also showed signs of activity with fresh lava flows erupted during the night of 21 November.

After the field team departed on 23 November, the eruption continued, with new overflows covering over the sampled flows, and nearly reaching the western rim of the caldera (Fig. 1; J. Hammond and J. Wilkinson, personal communication). The spacing between successive overflows became long enough that periods of sustained radiance can be identified, indicating that the lava lake had gained sufficient height to come into view of the sensor. SEVERI indicates lake activity persisted with sparse superimposed overflows for 7 days, followed by a period of sustained lake activity with no overflows for 6 days, and within a further 4 days, after 30 November, the lake signal disappeared, indicating a drop in lake level and occultation of the lake by the rim once again. Cloud cover prevents us determining the exact end of the eruption, but activity was confirmed to have ceased prior to 15 December (J. Wilkinson, personal communication, December 2010). In summary, the pattern appears to be (1) first period of lake level rise, (2) hiatus/drop in lake level, (3) lake level rise including lava overspills into the southern pit, (4) lake level stabilisation at high stand including lava overspills into the main crater and (5) lake level drop. The estimated total volume of both the erupted surface lava flows and the volume of lava required to fill the pit from the previous level is ∼0.006 km3, based on an estimated pit depth of 20 m.

Thermometry

There is good agreement between temperatures determined in February 2001 by Burgi et al. (2002) using thermocouples and pyrometers (1,145–1,187 °C), and the petrological estimates reported here (1,094–1,180 °C). The temperatures are within error indicating a temperature for the system of ∼1,150 °C, although there are likely to be some fluctuations within the conduit caused by magma circulation between magma storage, the conduit and the lake itself (Harris et al. 2005). Temperatures determined by Le Guern et al. (1979) in December 1971, January 1973 and 1974 provided a range of 1,100 °C through 1,210 °C. The Le Guern field team directly inserted thermocouples into active hornitos, which may account for the differences between their study and the study of Burgi et al. (2002).

Volatiles

The petrology indicates that crystallisation is taking place in the upper ∼1.5 km of the conduit. The highest H2O content measured in the melt inclusions is low, >0.13 wt.%, lower than the water content proposed for a typical alkali basalt (∼0.5 wt.%; e.g. Lowenstern and Mahood 1991). The low water values in the melt inclusions are very similar to basaltic melt inclusions from Hawaii (Fig.5a, Hauri 2002), but significantly lower than MORB (e.g. Pineau and Javoy 1994), Piton de la Fournaise, Réunion (Bureau et al. 1998) and Iceland (Nichols et al. 2002; Nichols and Wysoczanski 2007), which yield H2O contents up to 0.9 wt.%. Low H2O and CO2 content can result from leakage, e.g. through cracks, and diffusive loss can occur through the host olivine. However, no cracks or bubbles were present in any of the inclusions studied. The positive correlations between trace elements and volatiles (Fig. 6) could not be preserved if significant H2O had diffused out of the inclusions as H2O, CO2 and trace elements have very different diffusivities through olivine. The sole exception is melt inclusion L which has a slightly higher Ba than other melt inclusions (Fig. 6) and may represent entrapment of a slightly different melt composition. The overall trajectory of melt inclusion volatiles, namely, the large range in CO2 with little range in H2O (Fig. 5a), suggests that, although the magma became extensively degassed during ascent, the original magmatic H2O content was not significantly higher than 0.15 wt.%.

Fig. 5
figure 5

a Plot of CO2 vs. H2O in melt inclusion and matrix glasses compared to other basaltic systems. Note broad agreement between Erta Ale (green triangles and yellow circles) and Hawaii (Hauri 2002). b Close-up of the data from Erta Ale (same symbols as a) indicating possible volatile evolution paths in melt inclusions (solid lines), based on Blundy and Cashman (2008). Dotted lines indicate schematic isobars for pressure P1 > P2 (their exact positions are poorly constrained at these low pressures). Path a indicates vapour-saturated decompression without crystallisation, path b indicates isobaric vapour-saturated crystallisation and path c indicates vapour-saturated decompression crystallisation. Percentages in c indicate crystallisation. Path c best describes the Erta Ale data. Values indicate approximate percentage of crystallisation calculated from the Ba values of averaged MI, normalised to whole-rock Ba

Full size image
Fig. 6
figure 6

Melt inclusion analyses of H2O (a, c) and CO2 (b, d) plotted against concentrations of the incompatible elements Ba and Zr. These plots suggest that the November 2010 lavas experienced vapour-saturated crystallisation, as both Ba and Zr are positively correlated with H2O and negatively correlated with CO2. Green triangles show individual SIMS analyses; yellow circles indicate the averaged analysis for each melt inclusion studied. Melt inclusion (MI) L lies distinct from the main melt inclusion trend and may represent the presence of a slightly different magma composition within the system

Full size image

There are several lines of evidence that the Erta Ale magma was volatile-saturated prior to eruption. Our CO2–H2O data define an array which is oblique to the isobars, consistent with decompression-driven degassing (Fig. 5a, b). However, the negative slope of the array cannot be ascribed to degassing alone, which would produce a steep positive (sub-vertical) array, due to the much lower solubility of CO2 compared to H2O. The positive correlation of an incompatible element, such as Ba or Zr, with H2O (Fig. 6), and negative correlation with CO2 are consistent with vapour-saturated crystallisation (Blundy and Cashman 2008). A possible explanation is that crystallisation occurred concomitantly with decompression in Erta Ale magmas. Simple mass balance calculations for EA003 indicate ∼19 vol% crystallisation is required to obtain the melt inclusion composition from the bulk rock, which is comparable to the modal volume for this sample of ∼18 vol%, obtained by point counting. Note that if the inclusions had lost water by diffusion, the H2O (which diffuses more rapidly than CO2) would be displaced to low values at a given CO2 content, which is not observed (Fig. 5a).

The conclusion that Erta Ale magmas were volatile-saturated at shallow levels allows us to estimate, using VolatileCalc (Newman and Lowenstern 2002), the composition of the coexisting exsolved gas for comparison to previous measurements from Erta Ale. The estimate for exsolved gas in November 2010 is 0.2–2.9 mol % H2O and 97.1–99.8 mol % CO2, i.e. molar H2O/CO2 ratios of 0.002 to 0.03. There is appreciable uncertainty in this estimate due to the paucity of experimental data on basalts at low pressures. Nonetheless, these estimated volatile compositions are significantly more H2O poor than those determined by Gerlach (1980) and Sawyer et al. (2008) during passive degassing of the Erta Ale lava lake, who give molar H2O/CO2 ratios of 4–7 and 25, respectively (Table 6). As CO2 is considerably less soluble in basalt than H2O, the elevated H2O/CO2 suggests that passive degassing involves volatiles released at shallower depths than those at which melt inclusions were trapped. One way to evaluate this hypothesis is to calculate the composition of the total gas released by the ascending magma from the difference between melt inclusion and matrix glass volatile contents. This is, in effect, an integration of the gas released from the magma as it ascends to the surface. As the matrix glass has CO2 contents below detection, we assume all CO2 has been lost by degassing making our H2O/CO2 estimates maxima. Our calculated values for molar H2O/CO2 from individual melt inclusions lie in the range 4–57 with an average for the entire population of 11 ± 7. These H2O/CO2 ratios fall within the range determined by Sawyer et al. (2008) and Gerlach (1980). This suggests that the composition of the gas released from the magma is consistent, quantitatively, with exsolution of pre-eruptive volatiles contained within the magma; if a significant population of bubbles of deeper, magmatic gas with elevated CO2/H2O was present in the magma prior to ascent, this would reduce the H2O/CO2 ratio of the gas below the values observed by Sawyer et al. (2008) and Gerlach (1980).

Table 6 Calculated molar gas ratios
Full size table

The calculated molar SO2/Cl ratio of the released gas is in the range 1.0–2.0 (Table 6) compared to 12–19 measured in field studies (Gerlach 1980; Sawyer et al. 2008). As chlorine is the last volatile species to be exsolved from magma, the SO2/Cl ratio is a sensitive indicator of degassing depth (Lesne et al. 2011): higher ratios denote shallower degassing. Our low SO2/Cl ratios are consistent with complete degassing upon eruption in November 2010. Conversely, the lava lake gas measurement of Gerlach (1980) and Sawyer et al. (2008), which were made at a time when the lava was contained within the pit and not overflowing into the main crater, likely records a slightly deeper level of degassing, probably due to circulation within the conduit, giving rise to higher SO2/Cl ratios. These small changes in degassing depth will not have significant effect on the H2O/CO2 ratios, as almost all CO2 is lost prior to the onset of Cl degassing (Lesne et al. 2011). A schematic illustration of the Erta Ale shallow plumbing system that is consistent with the observed differences in gas composition between eruptive and non-eruptive (passive) degassing episodes is shown in Fig. 7.

Fig. 7
figure 7

Schematic of the Erta Ale shallow plumbing system in November 2010 based on integrated ground and satellite observations and petrology (not to scale). The arrows on the right hand side illustrate how gas extraction from different depths influences gas composition, with elevated Cl characteristic of very shallow degassing. Passive degassing does not appear to require an additional source of gas to explain the measured gas chemistry; it is sufficient that magma ascends from the storage region to the lake itself releasing its volatiles en-route. However, it is possible that vigorous Strombolian activity is triggered by the ascent of large bubbles from depth as shown schematically

Full size image

The difference in SO2 contents of melt inclusions and matrix glass is in the range 160 to 330 ppm (Table 4). Assuming that the November 2010 magma had a similar composition to that in the lava lake in 2005 when studied by Sawyer et al. (2008), we calculate the total magma flux through the conduit required to sustain their observed SO2 flux (0.69 ± 0.17 kg s−1) to be in the range 2,090 ± 520 to 4,310 ± 1,060 kg s−1. A broadly similar magma flux, calculated in the same way, can sustain the range of CO2 fluxes measured by Sawyer et al. (2008). However, a higher flux, by a factor of two, is required to sustain the observed H2O flux. This mismatch may be rooted in the combined uncertainties on the analytical and gas measurement data, or reflect, as suggested above, small differences in the degassing depths between and during eruptions. Sawyer et al. (2008) noted that the H2O in the volcanic plume had significantly increased in the 30 years since Gerlach's (1980) work, and suggested this may be related to volatile depletion of the deep source magma and/or to fractional magma degassing. Alternatively, variation in the H2O content of Erta Ale gases may reflect systematic changes in the depth within the conduit at which gases are extracted. In any case, our calculations lend support to the idea that the observed gas flux can be provided by near-complete degassing of magma passing through the shallow conduit system at rates of ∼3,000–7,000 kg s−1. It does not appear that any substantial additional source of gas, i.e. gas-fluxing, is required to account for the field measurements of gas flux at Erta Ale.

The Erta Ale lavas show a high vesicularity (up to ∼60 %), which may initially seem surprising given the low volatile contents indicated by the melt inclusions. Using the ideal gas law, the average value for lost H2O of 450 ppm (i.e. melt inclusion minus groundmass), a temperature of 1,150 °C, standard atmospheric pressure and an assumed density for basalt of 2,700 kg m−3 (e.g. Oppenheimer and Francis 1997), a gas volume of 2.92 × 10−3 m3 is obtained, ∼8 times more gas than melt. This indicates that if 10 % degassing were to take place, ∼40 % vesicularity would result, and 20 % degassing would result in ∼60 % vesicularity. Thus, releasing <448 ppm gas at, or close to, the surface at the eruption temperature is sufficient to generate the observed high vesicularities. Melt inclusion volatile data are also not suggestive of the presence of significant pre-eruptive bubbles in the magma during passive degassing episodes (see above). However, we cannot rule out the possibility that an influx of additional gas drove the Strombolian activity observed in 2010. We have no field measurements of this degassing episode with which to test our hypothesis, but it does seem likely that an additional gas source was required to sustain the observed eruptive vigour.

During eruption, the viscosity of the lava appeared to be very low (Supplementary Video 1). The calculated viscosity of the melt at eruption temperature using the model of Giordano et al. (2008) is 70 Pa s. The small fraction of crystals (∼18 vol% in EA003) is unlikely to have had any major impact on the viscosity (Caricchi et al. 2007). However, the vesicularity would have had a greater effect. The vesicles observed in the Erta Ale samples are highly elongate, which would have resulted in a lowering of viscosity, whilst spherical vesicles would instead have acted like phenocrysts to raise the viscosity (Llewellin and Manga 2005). Consequently, the calculated viscosity of the vesicular magma is reduced to 5 Pa s (Pal 2003) or 15 Pa s (Llewellin and Manga 2005).

Geochemistry

A key question is whether the geochemistry of the 2010 lavas is sufficiently different from those of earlier Erta Ale lavas to indicate that recharge of the magma system occurred subsequent to 1973. We have already shown that the 1973 lavas appear to be texturally and chemically distinct from those of 2010. However, these differences can be readily reconciled with a slightly increased degree of fractionation of EA002, as evinced by higher contents of incompatible elements, and lower temperatures as suggested by their more crystalline, microlite-rich texture. Overall, however, the 1973 and 2010 lavas are sufficiently similar to each other and within the range of Erta Ale Range lavas that it seems unlikely that any recharge of the system since 1973 involved a truly distinct magma type. Nonetheless, the renewal of activity, the reduction in differentiation and crystallinity strongly suggest that the sub-volcanic reservoir was refilled by hotter, compositionally similar basaltic magma.

Trace element systematics

Trace elements concentrations normalised to primitive mantle (Sun and McDonough 1989) are plotted in Fig. 8. Whole rock samples from Barrat et al. (1998) are shown for comparison. The melt inclusions have a similar overall pattern to the whole rocks, but are displaced to the high concentration end of the whole-rock range, consistent with a modest amount of crystallisation (∼20 %) of trace element poor phenocrysts. The small negative anomaly for Sr, in both melt inclusions and whole rocks, is consistent with crystallisation of plagioclase. The presence of plagioclase inclusions in both cpx and olivine crystals indicates plagioclase was already crystallising when melt inclusions were entrapped in olivine. Europium, which also partitions strongly into plagioclase, was not measured in the melt inclusions due to the large isobaric interference of BaO on Eu at the mass resolution used for trace element analyses.

Fig. 8
figure 8

Trace element compositions normalised to primitive mantle (Sun and McDonough 1989) of melt inclusions, with whole rock data from an Erta Ale basalt of Barrat et al. (1998) for comparison. The broadly similar patterns are consistent with crystallisation of an olivine-dominated phenocryst assemblage. The small negative Sr anomaly is consistent with plagioclase crystallisation

Full size image

The relationship between the whole rocks and the melt inclusions is further illustrated using log–log plots in Fig. 9. The melt inclusions define linear trends that extend away from the 2010 whole rocks along vectors consistent with ∼20 % crystallisation of an olivine-dominated assemblage with plagioclase and cpx. The range in trace element contents of the suite of melt inclusions records continued crystallisation of the same assemblage. The correlations between incompatible elements and volatile species (Fig. 6) suggest that this crystallisation occurred concomitantly with degassing. The subtle geochemical differences between the 2010 lavas and those of the lava floor are suggestive of magma refill of slightly different composition, although there is insufficient data to fully evaluate this possibility.

Fig. 9
figure 9

Log–log trace element plots for individual melt inclusion analyses and whole rock. Vectors show the effect of fractionation of plagioclase, clinopyroxene and olivine using partition coefficients from Rollinson (1993). The filled circles denote 20 wt.% fractionation. Data from Barrat et al. (1998) from both the Erta Ale volcano and the Erta Ale range has been included for comparison

Full size image

Magmatic timescales

Kα X-ray mapping revealed phosphorous zoning in the olivines, which was not apparent for other elements, e.g. Fe (Fig. 3h). Phosphorous diffuses extremely slowly compared to Fe–Mg and Ca in olivine (Milman-Barris et al. 2008; Spandler and O'Neill 2010). Hence, the P zoning is preserved, whilst Fe–Mg and Ca are homogenised during residence in hot magma. An estimate of the minimum time it would take to homogenise the crystal with respect to Fe–Mg and Ca at magmatic temperatures can be obtained from the characteristic diffusion length (x = (Dt)1/2; where x = distance, D = diffusivity, t = time) of each element of interest. Diffusion coefficients for Ca and Fe–Mg, parallel to the c-axis, were calculated using the models of Coogan et al. (2005), Dohmen et al. (2007) and Dohmen and Chakraborty (2007) at a magmatic temperature of 1,170 °C and an oxygen fugacity equivalent to the QFM buffer. For a typical 200-μm olivine crystal, Fe–Mg would have homogenised in approximately 0.2 years, and Ca in 10 years, providing a minimum estimate for the residence time of a crystal within this magma. Phosphorous diffusion rates have not been constrained, but are at least two orders of magnitude slower than other major cations (Spandler and O'Neill 2010). Studies of experimental and natural olivines have shown P and Cr to be correlated (Milman-Barris et al. 2008). Cr diffusion coefficients (Ito and Ganguly 2006) were therefore used to give an indication of the minimum timescales required to eradicate the P zoning. This gives a minimum timescale of ∼94 years for homogenisation of P within a typical 200-μm olivine crystal, as modelled here. As P is not homogenous in the studied olivines, this timescale provides an upper limit for residence of this crystal. Although crystals can be retained in the sub-volcanic system following an eruptive episode, it is considered that this is unlikely as the texture and appearance of crystals in the 2010 samples (EA001 and 3) are very different from the underlying lava (EA002) from the previous overspill (compare Fig. 3e, f). The last observed eruptive episode at Erta Ale was in 1973. The fact that the estimated timescales are of the order of decades, a period comparable to that elapsed since Erta Ale last erupted, is consistent with our earlier inferences that the magma system was recharged by a new pulse of hotter magma subsequent to 1973 (refer to Supplementary Table 5), we think these eruptive episodes are sufficiently infrequent that the period between the eruptions acts as an upper bound and therefore suggest that the crystal residence time was between ∼10 and ∼37 years.

Conclusions

We present an integrated field, satellite and petrological study of the November 2010 eruption of the Erta Ale volcano. The timeline of this small, but rare, eruption has been reconstructed from a combination of ground observations and SEVERI imaging. Two phases occurred, beginning on 11 and 19 November 2010. Pristine melt inclusions have shown the Erta Ale magma to be relatively dry (<0.15 wt.% H2O), although several petrological lines of evidence suggest the magma was volatile-saturated prior to eruption. The low volatile contents within the melt inclusions are sufficient at eruption temperature and low pressure, to account for the vesicularity of the erupted lava. The eruption and overflow of magma from the lava lake in November 2010 facilitated almost complete magma degassing. The calculated composition of the released gas is broadly consistent with measured passive emissions of volcanic gas from the Erta Ale lava lake, although the higher measured SO2/Cl ratios suggests that degassing during non-eruptive episodes did not proceed to the same extent as it did during the 2010 eruption. Coupled trace and volatile element systematics indicate that crystallisation took place at shallow levels, in the top ∼1.5 km of the conduit, largely in response to degassing. We tentatively conclude that slightly hotter, but broadly similar, magma was injected into the magmatic system after 1973, and that typical phenocryst residence times indicated by phenocrysts in the 2010 samples were between ∼10 and 37 years. The 2010 samples are shown to be quite distinct from the underlying lava from the previous overspills in 1973.

The November 2010 eruption was coincident with a seismic swarm in the Gulf of Aden; increased activity in 2005 was also noted to be coincident with the 2005 volcano-seismic crisis in the Manda-Hararo magmatic rift segment. It is possible that Erta Ale is particularly sensitive to tectonic activity in the region, but future monitoring will be required to confirm this.