Rhyolitic eruptions beneath Icelandic glaciers can be highly explosive, as demonstrated by Quaternary tephra layers dispersed throughout northern Europe. However, they can also be small and effusive. A subglacial rhyolitic eruption has never been observed, so behavioral controls remain poorly understood and the influence of pre-eruptive volatile contents is unknown. We have therefore used secondary ion mass spectrometry to characterize pre-eruptive volatile contents and degassing paths for five subglacial rhyolitic edifices within the Torfajökull central volcano, formed in contrasting styles of eruption under ice ∼400 m thick. This includes the products of the largest known eruption of Icelandic subglacial rhyolite of ∼16 km3 at ca. 70 ka. We find pre-eruptive water contents in melt inclusions (H2OMI) of up to 4.8 wt%, which indicates that Icelandic rhyolite can be significantly more volatile-rich than previously thought. Our results indicate that explosive subglacial rhyolite eruptions correspond with high H2OMI, closed-system degassing, and rapid magma ascent, whereas their effusive equivalents have lower H2OMI and show open-system degassing and more sluggish ascent rates. Volatile controls on eruption style thus appear similar to those for subaerial eruptions, suggesting that ice plays a subsidiary role in influencing the behavior of subglacial rhyolitic eruptions.


During subaerial eruptions, volatiles are considered a key factor in determining eruptive style, with a high pre-eruptive H2O and CO2 content and closed-system degassing leading to more-explosive volcanism (Eichelberger et al., 1986; Jaupart, 1998; Martel et al., 1998; Cashman, 2004). During subglacial eruptions, there are additional controlling factors that are poorly understood; e.g., a component of explosivity is thought to be influenced by the degree of magma-water interaction (Tuffen et al., 2001; Guðmundsson, 2005). However, experimental rhyolite-water interactions suggest that vesicles may actually hinder phreatomagmatic explosions (Austin-Erickson et al., 2008). Unlike subaerial eruptions, it is therefore unclear whether high volatile contents favor or inhibit explosive eruptions of rhyolite beneath ice.

Subglacial rhyolitic edifices have a wide spectrum of sizes, morphologies, and lithofacies, reflecting varying degrees of explosivity and the added complexities of a subglacial eruption setting (McGarvie, 2009). Eruptive products range from vesicle-poor quench hyaloclastites (Tuffen et al., 2001, 2008) to fine-grained, pumiceous pyroclastic deposits (Tuffen et al., 2002, 2008; Stevenson et al., 2011). Edifices range from small (<0.1 km3) mounds and ridges to large (∼1 km3) steep-sided, flat-topped tuyas, representing effusion-dominated and explosion-dominated activity respectively (Tuffen et al., 2007). However, eruption controls are poorly constrained, partly due to a lack of observed eruptions. Speculative models have suggested effusive activity is favored by either low initial volatile content (McGarvie et al., 2007; Stevenson et al., 2011), gas escape from magma (open-system degassing) (Furnes et al., 1980), the filling of subglacial cavities by erupted products (Tuffen et al., 2007, 2008), or thick overlying ice (Tuffen, 2010).

Iceland’s largest known subglacial rhyolitic eruption, the ∼16 km3 ring fracture eruption at Torfajökull (ca. 70 ka), mostly involved explosive tuya-forming activity (McGarvie et al., 2006). The eruption punctured an ∼400-m-thick ice sheet at a number of localities (McGarvie et al., 2006), generating widespread tephra layers, probably including the 6-cm-thick layer recently discovered in a marine core from the Norwegian Sea (Brendryen et al., 2010). It is presently unclear whether this was one continuous eruption (McGarvie, 1984) or several closely spaced events (Brendryen et al., 2010).

Better understanding of eruption controls is essential for hazard mitigation and reduction of socio-economic impact, especially given Iceland’s mid-Atlantic location. Explosive Icelandic eruptions can disrupt trans-Atlantic and/or European commercial flights, as demonstrated by the 2010 Eyjafjallajökull and 2011 Grímsvötn eruptions (Petersen et al., 2012).


We have determined the pre- and post-eruptive volatile contents, and reconstructed degassing paths, for five contrasting subglacial rhyolite edifices at Torfajökull (Fig. 1; Table 1) including four edifices from the ca. 70 ka ring fracture event: southeast Rauðfossafjöll and northwest Rauðfossafjöll, which were both explosive and burst through the ice to produce tuyas (Tuffen et al., 2002); Dalakvísl, which formed through mixed explosive-effusive activity (Tuffen et al., 2008); and Angel Peak, a small, effusively formed edifice. The fifth edifice is Bláhnúkur, a small effusively generated edifice (Tuffen et al., 2001), formed during a different eruption during the last glacial period (Owen et al., 2012). Grain-size distributions were acquired to confirm and quantify field observations relating to the degree of magma fragmentation and explosivity of these eruptions.

The volatile content of feldspar- and pyroxene-hosted melt inclusions (MI) and matrix glasses were analyzed using secondary ion mass spectrometry (SIMS). We used the “one-by-one approach” of Johnson et al. (1994) to identify MI that may have gained water. The matrix glass of every sample was analyzed using Fourier transform infrared spectroscopy (FTIR) to determine water speciation and therefore check for post-quenching hydration. Two samples identified as hydrated were discarded. Post-entrapment crystallization can cause volatile enrichment within MI, possibly leading to the formation of a vapor bubble (Steele-MacInnis et al., 2011). Therefore all bubble-bearing MI were discarded. Electron probe microanalysis (EPMA) data from MI and matrix glass suggest that post-entrapment crystallization played a minimal role in most of the remaining MI, as would be expected in rapidly quenched deposits (Lowenstern, 1995) emplaced beneath ice.

Our final data set consists of 62 analyses from 28 different MI within ten samples collected from five Torfajökull edifices. See the GSA Data Repository1 for additional geological background, sample descriptions, analytical and modeling methods, raw data, and data justification, including detail on identification of hydrated samples and post-entrapment modification processes.


Clear trends in volatile content are apparent (Fig. 2), with MI from effusive edifices (Angel Peak and Bláhnúkur) containing significantly less H2OMI (≤1.8 wt%) than those from explosively formed tuyas (southeast Rauðfossafjöll and northwest Rauðfossafjöll; ≤3.9 wt%). Dalakvísl (mixed effusive-explosive) spans the full range of water contents, including the highest measured value of 4.8 wt% H2OMI (Fig. 2). Low-H2OMI effusive samples are also Cl-rich, whereas H2O-rich, explosively-generated samples are Cl-poor (Fig. 2). The H2OMI contents of feldspar- and clinopyroxene-hosted MI are similar.

Matrix glasses contain 0.1–1.1 wt% H2OMI (Fig. 2), consistent with quenching at elevated pressures beneath ice hundreds of meters thick (see Tuffen et al. [2010] for detailed explanation of quenching pressure and ice-thickness reconstruction from H2O degassing). Inferred ice thicknesses (mostly ∼400 m) from lithofacies and degassing models (McGarvie et al., 2006; Owen et al., 2012) differ little between edifices and show no correlation with behavior (Table 1), so diverging eruption styles are not attributable to different ice thicknesses.

The major-element composition of melt inclusions and matrix glasses from Bláhnúkur and the ring fracture event are broadly similar (70–76 wt% SiO2), so compositional variation cannot explain the different eruptive styles. Some Bláhnúkur MI show SiO2 enrichment and alkali depletion, consistent with post-entrapment feldspar crystallization, and so Bláhnúkur H2OMI contents are maximum values. However, these are among the most H2O-poor MI (≤1.8 wt%), so any volatile enrichment due to crystallization does not mask differences between edifices. SiO2 enrichment is absent from Dalakvísl MI, suggesting that their high H2OMI contents are original.


Behavioral Control of Pre-Eruptive Water Content

The pre-eruptive water content (H2OMI) of the large-volume, predominantly explosive ca.70 ka event is considerably higher than that of Bláhnúkur (smaller, effusive). However, the ca.70 ka magma displays a range of H2OMI, being lower at the effusive edifice of Angel Peak (≤0.3 wt% H2OMI) than at the explosive edifices of Dalakvísl, southeast Rauðfossafjöll, or northwest Rauðfossafjöll (>2.9 wt% H2OMI). Perhaps surprisingly, our highest measured H2OMI came from the small-volume Dalakvísl edifice (<0.2 km3) rather than the larger-volume co-erupted (∼1 km3) tuyas. However, Dalakvísl has the finest-grained and most-vesicular ash of any of our sampling locations, suggestive of efficient magma fragmentation (Stevenson et al., 2011) in a powerful but perhaps brief explosive phase.

We have therefore found a strong positive correlation between H2OMI and the explosivity of eruptions. H2O is considered to be the most influential volatile species in terms of determining eruptive behavior during subaerial eruptions (Cashman, 2004); it may be equally important when eruptions occur beneath ice.

The differentiation between H2O-rich + Cl-poor, and H2O-poor + Cl-rich MI suggests that different edifices are recording different source magmas (whether separate magma bodies, a volatile-stratified chamber, or temporal gaps for melt evolution) rather than progressive degassing of a single homogenous supply.

Degassing Paths: Open- Versus Closed-System Degassing

Measured H2O-Cl trends (Fig. 2) have been modeled using H2O-Cl degassing systematics for rhyolitic melts (Villemant and Boudon, 1998; Villemant et al., 2008; Humphreys et al., 2009). Each edifice shows a single, distinct H2O-Cl trend, with the exception of Dalakvísl, which displays two different trends perhaps related to its bimodal eruptive behavior. Data scatter prevents discrimination between open- and closed-system degassing from degassing paths alone, but the chlorine distribution ratios (DCl) required to fit effusive sample data (≥50) greatly exceed those for explosive sample data (≤30). Microlite crystallization can drive an increase in DCl (Webster and De Vivo, 2002; Villemant et al., 2008), and effusive samples are significantly more microlite-rich than their explosive counterparts (Figs. 1E and 1F). Microlite crystallization occurs during magma ascent and degassing (Lipman et al., 1985; Sparks et al., 2000) and is favored by slow magma rise, which also favors open-system degassing (Jaupart, 1998). In contrast, less microlite crystallization typically occurs during closed-system degassing (Martel et al., 1998; Villemant et al., 2003, 2008). We therefore propose that our effusive samples experienced slow ascent rates and open-system degassing, whereas our explosive samples experienced fast ascent rates and closed-system degassing.

‘Wet’ Icelandic Rhyolite

Our data indicate far higher H2OMI than anticipated, as Icelandic rhyolite is often quoted as being “dry” in the absence of melt inclusion analysis (Sigurdsson, 1977; MacDonald et al., 1990; Jónasson, 2007). The only measurement to date of H2OMI in Icelandic rhyolite (Öraefajökull, 2 wt%; Sharma et al., 2008) is significantly lower than our measurements, as are matrix-glass H2O data from rhyolite tapped into by an exploratory well at ∼2 km depth (Krafla, 1.77 wt%; Elders et al., 2011). In addition, primitive Icelandic basalts are considered to be dry (<1 wt% H2O; Nichols et al., 2002). Although it is unclear whether the high H2OMI originates from partial melting of hydrated basalts at depth (Martin and Sigmarsson, 2007) or by another mechanism such as partial fusion of hydrothermally altered silicic material at shallower depths (Macdonald et al., 1987), our unexpectedly high H2OMI values clearly highlight how Icelandic central volcanoes, even those covered by thick ice, can generate highly explosive rhyolitic eruptions.


We have determined the pre-eruptive volatile content and degassing paths of subglacial rhyolitic magmas from five different edifices at Torfajökull, Iceland. Effusively erupted magmas have low H2OMI and show evidence for slow ascent rates and open-system degassing, whereas explosively erupted samples have high H2OMI associated with faster magma ascent and closed-system degassing. Volatile controls on eruption style during subglacial eruptions therefore appear similar to those for subaerial eruptions, suggesting that ice and meltwater played only a minor role in influencing the explosivity of these eruptions.

Our data show that Icelandic rhyolite can be significantly more water-rich (up to 4.8 wt%) than previously thought, highlighting the future potential for highly explosive rhyolitic eruptions from Icelandic central volcanoes, regardless of whether they are covered by ice.

Owen was funded by Natural Environment Research Council (NERC) studentship NE/G523439/1; Tuffen by NERC grants NE/G000654/1 and NE/E013740/1, and a Royal Society University Research Fellowship. McGarvie was supported by The Open University Staff Tutor Research and Scholarship Fund. We thank the Environment Agency of Iceland for allowing fieldwork and sampling from Fjallabak nature reserve and the Icelandic Museum of Natural History for sample export permission. Richard Hinton and Chiara Petrone assisted with SIMS and EPMA analyses, respectively. Reviews by two anonymous reviewers greatly improved the manuscript.

1GSA Data Repository item 2013060, additional geological background, sample descriptions, analytical and modeling methods, raw data, and data justification, including detail on identification of hydrated samples and post-entrapment modification processes, is available online at www.geosociety.org/pubs/ft2013.htm, or on request from editing@geosociety.org or Documents Secretary, GSA, P.O. Box 9140, Boulder, CO 80301, USA.