February 2017 Issue Table of Contents
Magmas are erupted from a wide range of depths. Olivine compositions, for example, indicate magma storage in the lower crust and upper mantle, while clinopyroxene and amphibole record middle to upper crust storage. Pre-eruptive magmas also often cool by 100–300 °C, frequently at middle–upper crust depths, indicating clogged, ephemeral volcanic pathways. These coolings imply that mafic recharge is not a sufficient cause for eruption and that crystallization-induced vapor saturation is a more proximal eruption trigger. But an improved understanding of eruption mechanisms require precise identifications of what are herein termed “ultimate”, “proximal,” and “immediate” causes of eruption.
Keywords: thermometry, barometry, arc volcanoes, eruption triggers, magma storage, magma transport
Volcanism, and planetary evolution generally, involves the continual release of gravitational potential energy as buoyant materials rise and crystalline materials (or metallic liquids or subducted slabs) sink. Earth’s separation into a core, mantle, and crust, with oceans and an atmosphere, reflects this release, and volcanic eruptions are but a spectacular illustration of the process. However, the buoyant rise of magma (a silicate liquid that may or may not contain crystals or bubbles) does not always culminate in a volcanic eruption. As little as 3%–4% of arc magmas may reach Earth’s surface (Paterson and Ducea 2015), while lava flows comprise <10% of mid-ocean ridge crust. Most magmas apparently don’t erupt. Rather, they form dikes, plutons, and sill complexes (Fig. 1). It follows that to understand volcanoes we must understand magma storage, the pathways of magma movement, and the circumstances that countenance magma transport to a planetary surface.
The “mush column” of Marsh (1996), conceived to describe the Hawaiian plumbing system, provides an excellent scaffold to describe volcanic sources and pathways and has been adapted to large silicic systems (Bachmann and Bergantz 2004; Hildreth and Wilson 2007). “Mush” was defined by Marsh to describe magma with 25–55 volume % crystals: enough to increase viscosity but less than the 55% that causes a magma to become a rigid body. Later authors (e.g. Bachmann and Bergantz 2004) also used “mush” to describe more rigid magma systems. The mush column concept can, and should, replace earlier concepts of long-lived, liquid-dominated magma chambers. The more modern view is one of crust-wide dikes and sills, which may or may not connect scattered bodies of liquid (e.g. Annen et al. 2015), and mush zones (Bachmannn and Bergantz 2004) that segue into completely solid wall rock. It is from such systems that volcanic liquids erupt and plutons form. This mush model, inspired by field observations, is supported by calculated crystallization depths (Putirka 1997; Klugel and Klein 2006), a lack of seismic evidence for large batches of liquid sitting in the crust, and studies of plutons, which reveal that magma batches are assembled in episodic, small increments over millions of years (Paterson et al. 2011; Annen et al. 2015). It is also the model that most accurately connects plutons and volcanoes. Uneruptible mushes appear to be a typical igneous state, often destined to form plutons, while eruptible, liquid-dominated bodies exist ephemerally within such systems, within temporal windows lasting as little as 104 years (e.g. Barboni and Schoene 2014).
Our current view of magma storage builds upon one of the more important models developed in the last century: the MASH hypothesis of Hildreth and Moorbath (1988). The MASH acronym derives from “melting, assimilation, storage, and homogenization.” The “melting” process occurs in the mantle, while the “ASH” processes nominally occur together in a “MASH zone” in the lower crust. “Assimilation” refers to the reaction of basaltic magmas with adjacent wall rock; “storage” implies fractional crystallization; and “homogenization” refers to magma mixing, e.g. basaltic with rhyolitic, to make andesitic magmas. This model frames much volcanologic research, correctly emphasizing assimilation and mixing to form andesites and dacites.
But our understanding of magmatic systems has evolved. Wall-rock assimilation can be trivial (e.g. Putirka et al. 2014), and “ASH” processes can occur separately at quite different crustal levels. As will be shown, the lower crust can be mostly a site of mafic magma fractionation, whereas intermediate magmas are generated at a wide range of depths, often in the middle- and upper crust. We can still speak of MASH, but not necessarily of singular MASH zones. Additionally, other processes are no less crucial to understanding volcanic systems. Mafic “recharge”—the injection of basaltic materials into or beneath a felsic magma/mush—may be essential for revitalizing felsic systems (Wiebe 2016), for amalgamating plutons into batholiths (e.g. Coint et al. 2013), and, perhaps, for initiating eruptions. Additionally, crystallization-induced volatile saturation may trigger eruptions (Tait et al. 1989) or initiate deep-seated magma transport (e.g. Armienti et al. 2013). We might thus abbreviate our understanding of volcanic systems as “recharge, crystallization + mixing, and eruption (RCME)”, although it may be better to instead recognize that our understanding has evolved beyond acronyms.
The Deepest Roots Of Volcanic Plumbing Systems
The mantle—defined here as rocks with densities >3.0–3.25 g/cm3 or P-wave velocities >7.6 km/s—is often considered as no more than a source of heat and melt. But volcanic pathways can extend deep into the mantle. For example, in Hawaii (USA), mantle xenoliths derive from >50 km (Clague 1987), i.e. well below Moho depths of 15–20 km (Hill and Zucca 1987), and similar to maximum depths of magma-induced seismic tremor (Dvorak and Dzurisin 1993). Sub-Moho pressures have also been recorded in Hawaiian clinopyroxene phenocrysts (Putirka 1997) and fluid inclusions from the Madeira Archipelago (Klügel and Klein 2006). Geophysical studies further reveal that sill complexes—interpreted to be stalled mafic magmas with olivine precipitates—extend well below the Moho at Hawaii (ten Brink and Brocher 1987) and the Izu–Bonin arc (Sato et al. 2009). The crust/mantle transitions in these areas are not clearly defined, and sill complexes are the likely agents by which the Moho transitions are obscured (e.g. Sato et al. 2009). Together, these data provide compelling evidence for subcrustal magma storage.
Olivine compositions further show that volcanic pathways extend into the mantle. Olivine phenocrysts can contain high forsterite (Fo; Mg2SiO4) contents, irrespective of tectonic setting, reaching values of >91% Fo (noted as Fo = 91) both in ocean basin lavas (Putirka et al. 2007) and in continental arc lavas, such as the Cascades (Fig. 2A). Given that host lavas at arcs have up to 8%–9% FeO (Fig. 2B), these olivine compositions require parental liquids with 14%–18% MgO (Fig. 2B) (Putirka et al. 2007; Putirka 2016). These high-MgO magmas are called picrites, and can explain the subcrustal P-wave seismic velocities noted above. Moreover, if completely molten, picrites are as dense as, or even more dense than, the lower crust, and are denser still when carrying crystals (Fig. 3). This high density explains why picrites are common in the submarine portions of oceanic volcanoes (compared to subaerial eruptions) (Garcia 2002) and are rare in continental settings. Most volcanoes appear to be rooted in picritic sill systems in the lower crust and upper mantle.
Magma densities allow us to paint a broader picture of magma stagnation. As Lister and Kerr (1991) show, magmas can only overshoot their level of neutral buoyancy (LNB: where magmas and ambient wall rock have equal densities) by a few kilometers. This is not to say that tectonic forces are unimportant: ten Brink and Brocher (1987) show that regional stresses can profoundly influence volcanic activity. But such stress barriers only inhibit liquids from reaching their LNB—and so we can use calculated densities to estimate the depths to which tectonically uninhibited magmas (no crystals or bubbles) may rise. Silicate liquids form a density continuum, with no density minimum when P–T conditions are used as input (Fig. 3). However, there appears to be a distinct break in density at 10%–12% MgO, where buoyancy gains are more rapid with further fractionation (decreasing in MgO). Dry picrites (14%–18% MgO, no water) are neutrally buoyant in the uppermost mantle and lower crust, but can rise into the middle crust with an addition of 3% H2O (Fig. 3). More evolved dry basaltic magmas (5%–8% MgO) can stall within the middle crust, or, if hydrous, in the upper crust (Fig. 3). Of course, any of these melts can erupt—and do—if vapor saturated. But non-vapor-saturated basalts exhibit a wide range of densities, and so can pond and release heat throughout the crust. Volcanic pathways are “hot columns” with transient intervals of high melt fraction (“hot zones”; Annen et al. 2015), which can vary with depth and time, probably as a function of melt composition.
How fast might basaltic magmas transit the crust? Nickel-zonation in olivine yields mantle-to-surface transport speeds of 3.3 m/h (Ruprecht and Plank 2013)—within the 1–10 m/h range obtained by entirely different methods at Mt. Etna (Italy) (Armienti et al. 2013). The Etna study further implicates water saturation as an important accelerant in the middle to lower crust. These calculated magma transport rates probably reflect an average value that smooths over periods of rapid conveyance punctuated by interludes of stagnation. In a pulsed transport system, it may be no surprise that most magmas are unerupted (e.g. Paterson and Ducea 2015)—“interim” stagnation becomes permanent, possibly for lack of water, as magmas cool to the point of becoming uneruptible mush.
Magma Storage and Transport in the Middle and Upper Crust
Properly employed, igneous thermobarometers can detect both protracted and transient storage depths. Depth estimates are not necessarily depths of prolonged storage: individual P–T estimates may represent transient crystallization during transport (minor cooling or decompression-driven volatile release). And because silicate minerals have very slow internal diffusion rates, transient P–T conditions can easily be preserved. Yet another challenge for identifying prolonged storage sites is that a single chamber, within a dike-and-sill ladder, can collect crystals formed at many depths, only to be flushed out together in a later eruptive episode. However, prolonged storage depths might be obtained from P–T estimates in aggregate. For example, while individual clinopyroxene grains from Hawaii record crystal growth from the mantle to the near surface, depth estimates cluster at 11–16 km, and exhibit a low-T pool at 1,200–1,150 °C (Fig. 1B). This depth range matches an interval of high compressional wave velocities, which have been interpreted as gabbro sills (Hill and Zucca 1987). And gabbro xenoliths record crystal growth/reequilibration over this same depth range, but with an additional 200 °C of cooling. The 11–16 km interval thus appears to be a site of prolonged storage. It might also serve as a supply depot for sustained eruptions: a volcanic edifice inflates as magmas fill a shallow reservoir, and deflates during eruption. At Hawaii, eruptions continue well after deflation ends (Dvorak and Dzurisin 1987), draining in a top-to-bottom fashion progressively deeper parts of a hydraulically connected conduit that extends into the mantle (Putirka 1997). These postdeflation eruptions account for 80% of erupted volumes (Dvorak and Dzurisin 1987), although still unclear is the fraction of this 80% that is stored at 11–16 km, or below.
A loosely parallel story is told at volcanic arcs (Fig. 1D). For example, in the North American Cascades, clinopyroxene phenocrysts crystallize at depths ranging from near-surface down to 25 km. But most clinopyroxene crystallization in the Cascades occurs in the upper 5–11 km of the crust, suggesting that most basaltic magmas reach clinopyroxene saturation in the middle and upper crust. Amphiboles from Cascade magmas yield similar crystallization depths, recording up to 300 °C of additional cooling. These amphiboles hail from both mafic enclaves (the rock record of mafic recharge) and felsic host lavas, and support a model where clinopyroxene-saturated “recharge” magmas invade an already shallow, amphibole-saturated felsic magma/mush system, and then become amphibole saturated themselves. Of course, no single magma chamber extends from 5 km to 11 km; rather, a series of dikes, sills and mush bodies permeate this interval (Fig. 1C). The inferred dike–sill–mush system for the Cascades is clearly shallower than at Hawaii, probably reflecting the lower density of the andesitic Cascade magmas (60% SiO2, 2.8% MgO) which help them rise to higher levels in the crust (Fig. 3) compared to Hawaiian basaltic magmas. It is not yet known how the Cascades’ dike-and-sill ladder system is emptied: all in one eruptive episode, or mostly after magmas are collected in a shallow staging area?
How Volcanic Plumbing Systems Evolve
Plutonic rocks provide a glimpse into the evolution of volcanic pathways and hint at a key role for the middle crust. In one part of the Sierra Nevada (western USA), gabbros fractionate directly to form granite while successive recharge magmas are emplaced above their predecessors but below a growing, convective felsic magma body (Putirka et al. 2014). Coint et al. (2013) found similar patterns in the much larger Wooley Creek Batholith. Volcanic pathways are probably related to crustal evolution, where the middle crust acts as a transfer zone. Basaltic magmas repeatedly invade the middle crust and fractionate; their crystalline residues sink to form cumulates that add to the lower crust, and their felsic differentiates are extracted from a rigid mush to rise upwards (Bachmann and Bergantz 2004) and add to the upper crust.
Phenocrysts from different arc lavas also record crustal thickening. Beneath the Cascades (crust = 35 km), the median clinopyroxene crystallization depth is 5 km. In contrast, beneath the Central Andes (crust = 70 km), half of all clinopyroxene grains crystallize at >34 km (Fig. 1E). This contrast occurs despite Andean liquids being clinopyroxene-saturated at very similar SiO2 contents (59%), albeit higher MgO (4.5%) compared to Cascade lavas (60% SiO2; 1.8% MgO). In both cases, clinopyroxene rarely crystallizes beneath the middle crust (40 km at the Andes; 12 km at the Cascades) (Fig 1E). This seems to imply that clinopyroxene saturation is mostly a middle–upper crust phenomenon (Fig. 3): if arcs can erupt high-forsterite olivine grains, clearly they should also erupt high-P clinopyroxene grains, if such exist. Another possibility is that picrites only reach clinopyroxene saturation near their solidus, at which point they become uneruptible mushes. In either case, picrites, not pyroxenites, should dominate the lowermost mafic crust.
Further insights into magmatic plumbing systems derive from trace element ratios. Volcanic Sr/Y ratios, for example, are a rough measure of magma fractionation depths (Chapman et al. 2015). This is because at high pressures (>15–20 kbar), garnet precipitates from basaltic magma, or is a residual phase during partial melting of the crust, and garnet is a sink for Y but rejects Sr. Garnet fractionation, therefore, drives equilibrium liquids to high Sr/Y (40 to >200). In the middle and upper crust, however, plagioclase and amphibole replace garnet, and liquids equilibrated in these crustal regions have modest or very low Sr/Y (<40). Chapman et al. (2015) found that median Sr/Y ratios are low (<8) for arcs built on thin crust (e.g. the Izu–Bonin arc) and higher (>25) for arcs built on thicker crust (central Andes) (Fig. 4). A better proxy for crust thickness might be modal, rather than median, Sr/Y ratios (Fig. 4). In any case, Sr/Y ratios also record crustal growth: maximum Sr/Y values in the Sierra Nevada Batholith increase steadily with time (Fig. 4E). Importantly, however, low Sr/Y magmas are present at all time periods (Fig. 4E): within mature arcs, no single process (deep vs. shallow fractionation or melting) monopolizes felsic magma genesis (e.g. Coint et al. 2013). The middle crust also seems to behave as a leaky transfer zone, allowing some basaltic magmas to rise and differentiate at low P (Fig. 3) even when the upper crust is thick.
Why Magmas Erupt: a Muddle and a Suggestion For Progress
We have no shortage of working hypotheses regarding eruption triggering. Sparks et al. (1977) and Tait et al. (1989) posit that felsic magmas may be brought to an eruption-ready state by magma mixing, mafic recharge, or in situ partial crystallization. Any or all of these can drive a system to vapor saturation and increase magma buoyancy.
However, the above hypotheses are made more complex by recent advances (for details, see the other articles in this issue). Mineral-scale age dates show that felsic magmas are often 104–105 y older than their eruption ages. Thus, some felsic magmas/mushes lie in wait to be reactivated by mafic recharge magmas. Studies on plutons also reveal that large rigid magma bodies seem to have been formed from innumerable magma recharge events, which apparently rarely, or never, result in eruption (Coint et al. 2013; Putirka et al. 2014). Recharge, then, may be a necessary but still insufficient eruption condition. Perhaps, then magma mixing is an eruption trigger. Kahl et al. (2011) reported multiple magma mixing events that presaged the 1991–1993 eruptions at Mt. Etna. But which mixing event, if any, pushed the system to a point of eruptive instability? Recharge and mixing events are not singular: perhaps a particular event is crucial; perhaps none are. As an added complication, volcanoes, at least ephemerally, are hydraulically connected from an exhalative vent to a mantle source (e.g. Putirka 1997). We still need to know where along a volcanic pathway eruptible felsic systems lie in wait for a recharge/mixing event, and whether recharge/mixing depths are more important at certain depths compared to others, or whether eruption triggering is a conduit-wide process. A remaining challenge is to untie ourselves from acronyms and unfold these processes into causative sequences, which may vary between volcanoes, or between individual eruptions.
When thinking in terms or causative sequences, we are quite likely to find that eruptions are caused by multiple, cascading, triggers. For progress, we should begin to differentiate between causes that are “ultimate”, “proximal” or “immediate”, accepting that there may be a continuum. To illustrate these three causes, we’ll take the 18 May 1980 eruption of Mt. St. Helens (Washington, USA) as an example. The earthquake-triggered landslide that sparked the event may be its immediate cause; its proximal cause might be emplacement of low-density magma into the upper crust and/or the magma’s approach to vapor saturation; the ultimate cause—not at all clear—might involve partial melting in the mantle or a recharge event initiated by means yet uncertain from a depth unknown. None of these distinctions are easy. Was the landslide triggered by tectonic forces or by increased magma overpressure, driven by a crystallizing, vapor-saturated magma? In the latter case, proximal and immediate causes might be identical. Mantle partial melting might be an important trigger, or perhaps it is a trivial ultimate cause, if mantle melt supply rates are long compared to the timescales of eruptive cycles (e.g. Zellmer and Turner 2007). Perhaps mixing or recharge are more usefully viewed as “ultimate” causes. Can recharge also serve as an immediate cause? Perhaps not, when mafic enclaves (our record of recharge) record post recharge coolings of 100–300 °C (Fig. 1), but we have too few data to generalize. In any case, precisely articulated hypotheses on the causes and conditions of eruption are obligatory for progress.
Reasons for Optimism
The above-stated problems are tractable. For example, we can test whether crystallization-induced vapor saturation is a proximal or immediate cause of eruption by comparing cooling intervals for various systems, and determining whether magmas approach water saturation at pressure and temperature conditions that precede eruption. Or, if magma mixing is an eruption trigger, then mineral compositions should be strongly bimodal, reflecting the lack of time for newly created intermediate liquids to precipitate new minerals. We can also test hydraulic connectivity and storage. Clinopyroxene grains from the 2010 Merapi (Indonesia) eruptions record pressures ranging from 1 atm to 4.3 kbar (Putirka 2016b), values that are greater than the barometric 1σ error bounds (±2 kbar). If this P range reflects a magmatic system that is hydraulically connected at the time of eruption, then these crystals should yield sundry rim compositions that reflect a diversity of storage conditions. If, instead, a single chamber has collected crystals from various depths, and was evacuated from a once-but-no-longer connected pathway, then clinopyroxene grains should have convergent rim compositions, to the extent that such crystals share a late thermal history.
A further reason for optimism are the myriad opportunities to build on the excellent work of Kahl et al. (2011), to integrate petrological and geophysical data. Their study of a Mt. Etna eruptive sequence revealed magma-mixing one year before eruption, edifice inflation a month or two later, then additional mixing events 6 months prior to eruption. We can next examine whether, within such a sequence, any particular event pushes a system beyond a threshold, where an eruption becomes probable, perhaps even inevitable. Numerical models and mineralogical studies of active and ancient volcanic systems can be combined to establish threshold conditions and test our sundry working hypotheses on where magmas are stored and why volcanoes erupt.
I’m grateful to Kari Cooper for her thoughtful comments and thank both her and Gordon Brown for their editorial handling. I thank Katherine Cashman, Steve Sparks, and Colin Wilson for their very helpful, considerate and detailed reviews and criticisms. I highly appreciate the help of Jodi Rosso and Patrick Roycroft, whose edits greatly improved the clarity and accuracy of this article. Finally, I thank Calvin Barnes, Vali Memeti, Scott Paterson, Cecil Robinson, and Joshua Schwartz, for highly enlightening discussions on the workings of plutonic bodies. This work was supported by NSF grants EAR-1250322 and EAR-1250323.
Annen C, Blundy DJ, Leuthold J, Sparks RSJ (2015) Construction and evolution of igneous bodies: towards an integrated perspective of crustal magmatism. Lithos 230: 206-221
Armienti P, Perinelli C, Putirka KD (2013) A new model to estimate deep-level magma ascent rates, with applications to Mt. Etna (Sicily, Italy). Journal of Petrology 54: 795-813
Bachmann O, Bergantz GW (2004) On the origin of crystal-poor rhyolites: extracted from batholithic crystal mushes. Journal of Petrology 45: 1565-1582
Barboni, M, Schoene, B (2014) Short eruption window revealed by absolute crystal growth rates in a granitic magma. Nature Geoscience, doi: 10.1038/NGEO2185.
Chapman JB, Ducea MN, DeCelles PG, Profeta L (2015) Tracking changes in crustal thickness during orogenic evolution with Sr/Y: an example from the North American Cordillera. Geology 43: 919-922
Clague D (1987) Hawaiian xenolith populations, magma supply rates, and development of magma chambers. Bulletin of Volcanology 49: 577-587
Coint N, Barnes CG, Yoshinobu AS, Chamberlain KR, Barnes MA (2013) Batch-wise assembly and zoning of a tilted calc-alkaline batholith: field relations, timing, and compositional variation. Geosphere 9: 1729-1746
Dvorak JJ, Dzurisin D (1993) Variations in magma supply rate at Kilauea Volcano, Hawaii. Journal of Geophysical Research: Solid Earth 98: 22255-22268
Garcia MO (2002) Submarine picritic basalts from Ko`olau volcano, Hawai`i: implications for parental magma compositions and mantle source. In: Takahashi E, Lipman PW, Garcia MO, Naka J, Saramaki S (eds) Hawaiian Volcanoes: Deep Underwater Perspectives. Geophysical Monograph 128, American Geophysical Union, Washington, DC, pp 391-402
Hildreth W, Moorbath S (1988) Crustal contributions to arc magmatism in the Andes and central Chile. Contributions to Mineralogy and Petrology 98: 455-489
Hildreth W, Wilson CJN (2007) Compositional zoning of the Bishop Tuff. Journal of Petrology 48: 951-999
Hill DP, Zucca JJ (1987) Geophysical constraints on the structure of Kilauea and Mauna Loa volcanoes and some implications for seismomagmatic processes. U.S. Geological Survey Professional Paper 1350: 903-917
Kahl M, Chakraborty S, Costa F, Pompilio M (2011) Dynamic plumbing system beneath volcanoes revealed by kinetic modeling, and the connection to monitoring data: an example from Mt. Etna. Earth and Planetary Science Letters 308: 11-22
Klügel A, Klein F (2006) Complex magma storage and ascent at embryonic submarine volcanoes from Madeira Archipelago. Geology 34: 337-340
Lister JR, Kerr RC (1991) Fluid-mechanical models of crack propagation and their application to magma transport in dykes. Journal of Geophysical Research: Solid Earth 96: 10049-10077
Marsh BD (1996) Solidification fronts and magmatic evolution. Mineralogical Magazine 60: 5-40
Paterson SR, Ducea MN (2015) Arc magmatic tempos: gathering the evidence. Elements 11: 91-98
Paterson SR, Okaya D, Memeti V, Economos R, Miller RB (2011) Magma addition and flux calculations of incrementally constructed magma chambers in continental margin arcs: combined field, geochronologic, and thermal modeling studies. Geosphere 7: 1439-1468
Putirka K (1997) Magma transport at Hawaii: inferences based on igneous thermobarometry. Geology, 25: 69-72
Putirka K (2016a) Rates and styles of planetary cooling on Earth, Moon, Mars and Vesta, using new models for oxygen fugacity, ferric-ferrous ratios, olivine-liquid Fe-Mg exchange, and mantle potential temperature. American Mineralogist 101: 819-840
Putirka K (2016b) Amphibole thermometers and barometers for igneous systems and some implications for eruption mechanisms of felsic magmas at arc volcanoes. American Mineralogist 101: 841-858
Putirka KD, Perfit M, Ryerson FJ, Jackson MG (2007) Ambient and excess mantle temperatures, olivine thermometry and active vs. passive upwelling. Chemical Geology 241: 177-206
Putirka KD and 6 coauthors (2014) Pluton assembly and the genesis of granitic magmas: insights from the GIC pluton in cross section, Sierra Nevada Batholith, California. American Mineralogist 99: 1284-1303
Ruprecht P, Plank T (2013) Feeding andesitic eruptions with a high-speed connection from the mantle. Nature 500: 68-72
Sato T, Kodaira S, Takahashi N, Tatsumi Y, Kaneda Y (2009) Amplitude modeling of the seismic reflectors in the crust-mantle transition layer beneath the volcanic front along the northern Izu-Bonin island arc. Geochemistry, Geophysics, Geosystems 10, doi: 10.1029/2008GC001990
Sparks SRJ, Sigurdsson H, Wilson L (1977) Magma mixing: a mechanism for triggering acid explosive eruptions. Nature 267: 315-318
Tait S, Jaupert C, Vergniolle S (1989) Pressure, gas content and eruption periodicity of a shallow, crystallizing magma chamber. Earth and Planetary Science Letters 92: 107-123
ten Brink US, Brocher TM (1987) Multichannel seismic evidence for a subcrustal intrusive complex under Oahu and a model for Hawaiian volcanism. Journal of Geophysical Research: Solid Earth 92: 13687-13707
Wiebe RA (2016) Mafic replenishment into floored silicic magma chambers. American Mineralogist 101: 297-310
Zellmer GF, Turner SP (2007) Arc dacite genesis pathways: evidence from mafic enclaves and their hosts in Aegean lavas. Lithos 95: 346-362