Received 4 June 2014Accepted 15 September 2014Available online 6 October 2014

Keywords:Explosive eruptionsCalderasMagma storageSyn-eruptive melt extraction

. . . . . . . . . . 32

. . . . . . . . . . 35

. . . . . . . . . . 35

Journal of Volcanology and Geothermal Research 288 (2014) 2845

Contents lists available at ScienceDirect

Journal of Volcanology and Geothermal Research4.2. Time scales of magma accumulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 354.3. Eruption triggers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

4.4. Eruption dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373.2. Caldera formation from complex mac magma reservoirs an example from Colli Albani . . . . . . . . . . . . . .4. Storage and eruption from large silicic systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

4.1. Pre-eruptive magma storage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1. Introduction why review calderas? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292. Caldera-forming eruptions: an overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303. Mac magmatic systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

3.1. Stacked sill models of mac magmatic systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314.5. Caldera collapse . . . . . . . .5. Implications for recognizing eruption poten

Corresponding author at: University of Bristol, SchoolE-mail address: glkvc@bristol.ac.uk (K.V. Cashman).

http://dx.doi.org/10.1016/j.jvolgeores.2014.09.0070377-0273/ 2014 Elsevier B.V. All rights reserved.Contentsgists have used the eruptive products to probe conditions of magma storage (and thus processes that drivemagma evolution), while volcanologists have used them to study the conditions under which large volumes ofmagma are transported to, and emplaced on, the Earth's surface. Traditionally, both groups have worked onthe assumption that eruptible magma is stored within a single long-livedmelt body. Over the past decade, how-ever, advances in analytical techniques have provided new views of magma storage regions, many of which pro-vide evidence of multiple melt lenses feeding a single eruption, and/or rapid pre-eruptive assembly of largevolumes of melt. These new petrological views of magmatic systems have not yet been fully integrated into vol-canological perspectives of caldera-forming eruptions. Here we explore the implications of complex magma res-ervoir congurations for eruption dynamics and caldera formation. We rst examine mac systems, wherestacked-sill models have long been invoked but which rarely produce explosive eruptions. An exception is the2010 eruption of Eyjafjallajkull volcano, Iceland, where seismic and petrologic data show that multiple sills atdifferent depths fed a multi-phase (explosive and effusive) eruption. Extension of this concept to larger maccaldera-forming systems suggests a mechanism to explain many of their unusual features, including theirprotracted explosivity, spatially variable compositions and pronounced intra-eruptive pauses. We then reviewstudies of more common intermediate and silicic caldera-forming systems to examine inferred conditions ofmagma storage, time scales of melt accumulation, eruption triggers, eruption dynamics and caldera collapse.By compiling data from large and small, and crystal-rich and crystal-poor, events, we compare eruptions thatare well explained by simple evacuation of a zoned magma chamber (termed the Standard Model by Gualdaand Ghiorso, 2013) to eruptions that are better explained by tapping multiple, rather than single, melt lensesstoredwithin a largely crystallinemush (whichwe term complexmagma reservoirs).We then discuss the impli-cations of magma storage within complex, rather than simple, reservoirs for identifying magmatic systems withthe potential to produce large eruptions, and for monitoring eruption progress under conditions where succes-sive melt lenses may be tapped. We conclude that emerging views of complex magma reservoir congurationsprovide exciting opportunities for re-examining volcanological concepts of caldera-forming systems.

2014 Elsevier B.V. All rights reserved.Article history: Large caldera-forming eruptions have long been a focus of both petrological and volcanological studies; petrolo-a b s t r a c ta r t i c l e i n f oReview

Calderas and magma reservoirs

Katharine V. Cashman a,, Guido Giordano b

a University of Bristol, UKb Universit Roma Tre, Italy

j ourna l homepage: www.e lsev ie r .com/ locate / jvo lgeores. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38tial of large magmatic systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

of Earth Sciences, Bristol BS8 1RJ, UK.

Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41Appendix A. Supplementary data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

However, there is a pleasure in recognizing old things from a newpointof view. Richard Feynman, 1948

1. Introduction why review calderas?

The importance of studying caldera-forming eruptions cannot beunder-estimated. Caldera-forming eruptions include someof the largestvolcanic events ever to affect the Earth. Many of these have producedlarge volumes (N100 km3) of highly evolved crystal-poor melt. As a re-sult, an enduring paradigm of both igneous petrology and volcanologyhas been oneofmelt accumulation, evolution and eruption froma singlemelt-dominated magma chamber. This conceptual metaphor has pro-vided a framework for petrological models of magma evolution and dif-ferentiation, and for volcanological models of eruption initiation,magma withdrawal and caldera collapse (Fig. 1). Key features include:(1) development of stably zoned magma chambers by crystal fraction-ation, where crystal-liquid separation is driven by settling of individual

crystals or crystal plumes within a large batch of liquid that cools fromthe margins inward; (2) eruption initiation by injection of a verticaland pressurized dyke, located either in an axisymmetric position or atthe chamber margin; (3) magma withdrawal starting from the top ofthe melt lens and propagating downward, as evidenced by depositsthat are reversely zoned in composition and/or pre-eruptive tempera-ture and pressure; and (4) caldera formation by collapse of an under-pressured magma chamber after some fraction of magma has beenwithdrawn, with the timing of collapse determined by the strengthand thickness of the overlying country rock relative to width of themagma chamber.

Over the past few decades, however, detailed volcanological, petro-logical and geophysical studies of individual (intermediatesilicic)mag-matic systems have shown that (1) magma storage regions arecomposed primarily of crystallinemush (crystals plus interstitial liquid;Fig. 2; e.g., Hildreth, 2004; Hildreth and Wilson, 2007; Lipman, 2007;Bachmann and Bergantz, 2008; Reid, 2008; Bachmann, 2010; Deeringet al., 2011; Walker et al., 2013; Simon et al., 2014), (2) large melt vol-umes may be assembled rapidly (Charlier et al., 2005; Wilson andCharlier, 2009; Druitt et al., 2012; Allan et al., 2013; Gebauer et al.,2014; Simon et al., 2014; Wotzlaw et al., 2014), (3) caldera-forming

42

29K.V. Cashman, G. Giordano / Journal of Volcanology and Geothermal Research 288 (2014) 2845Fig. 1. The Standard Model of caldera formation. (A) Stably stratied magma chamberforms over thousands of years by crystal settling and upward migration of volatiles.(B) Eruption starts with Plinian activity through a single vent, driven primarily by volatileexsolution. (C) Evacuation of magma causes under-pressurization and destabilization of

Fig. 2. End member models of caldera-producing magmatic systems. (A) Crystal-poor(CR), and often zoned, eruptions are fed from a single melt body contained within amuch larger and more crystalline system comprising a crystal mush (~50% crystals) andsurrounding rigid sponge (N65% crystals); modied from Hildreth (2004). (B) Crystal-rich (CR) eruptions occur by rejuvenation of a near-rigid crystal mush (by input of meltthe reservoir. (D) Caldera forms by roof collapse. and/or gas); modied from Bachmann and Bergantz (2006); Huber et al. (2011).

eruptions may be triggered either internally (chamber triggered) orexternally (roof triggered) depending on the tectonic setting, accumu-latedmagma volume and roof aspect ratio (Lindsay et al., 2001; Jellinekand DePaolo, 2003; de Silva et al., 2006; Gudmundsson, 2008, 2012;Gottsmann et al., 2009; Gregg et al., 2012), and (4) large eruptionsmay tap multiple and distinct melt sources and crystal populations (in-cluding phenocrysts, antecrysts and xenocrysts, e.g., Maughan et al.,2002; Ellis et al., 2010; Wright et al., 2011; Allan et al., 2012; Cooperet al., 2012; Ellis and Wolff, 2012; Vinkler et al., 2012; Gualda andGhiorso, 2013; Fig. 3). At the same time, geophysical studies of activevolcanic systems have failed to locate large volumes of crystal-poormelt (e.g. Iyer et al., 1990; Masturyono et al., 2001; Schilling andPartzsch, 2001; Sherburn et al., 2003; Zandt et al., 2003; Lees, 2007;Chu et al., 2010). These observations are difcult to reconcile with aclassical magma chamber model sensu stricto, that is, a single, verylarge, long-lived melt-dominated magma chamber (termed the Stan-dard Model by Gualda and Ghiorso, 2013).

Herewe review explosive caldera-forming eruptions and their prod-ucts, paying particular attention to new conceptual models of complexmagma storage regions. By complex we refer to magma reservoirscomprisingmultiplemelt lenseswithin a partially to completely crystal-line framework; such reservoir congurations have been advanced by

Excellent reviews of magmatic systems that produce large silicic

are most commonly associated with stratovolcanoes, where storage re-gions tend to be vertically elongated.

The very largest eruptions almost always involve MI magma that is(relatively) homogeneous in both composition and phenocryst content(Fig. 4A). Resulting ignimbrites are typically composed of ash and bro-ken crystals, with only scarce pumice clasts (b10%; e.g., Carter et al.,1986; Bachmann et al., 2002; Gottsmann et al., 2009; Wright et al.,2011). CPR eruptions, in contrast, initiate with pumice-bearing depositsof crystal-poor (and often high-SiO2) rhyolite magma that may vary inboth crystallinity and bulk composition throughout the course of aneruption (e.g., Hildreth, 1981; Deering et al., 2011; Pamukcu et al.,2013; Fig. 4B). The high-SiO2 rhyolite melt often lacks a counterpart incorrelative plutonic sequences (although there are exceptions,e.g., Walker et al., 2007), but overlaps the composition of matrix meltsin MI magmas (e.g., Lindsay et al., 2001; Lipman, 2007; Bachmannet al., 2007; Fig. 4). For this reason, both magma types are inferred tohave a similar origin, with the difference being that MI eruptions evac-uate the entire (rejuvenated) magma storage region, while CPR erup-

30 K.V. Cashman, G. Giordano / Journal of Volcanology and Geothermal Research 288 (2014) 2845caldera-forming eruptions are provided by Hildreth (1981, 2004) andLipman (2007); here we summarize some pertinent points from theseand related studies. Large (100 km3 DRE) silicic eruptions can be clas-sied by dominant magma type as either crystal-rich (35%) dacite(often termed monotonous intermediates, MI) or crystal-poor (15%)rhyolite (CPR). Magma within these systems is typically stored in sill-like bodies, that is, they have horizontal dimensions that greatly exceedthe vertical dimension. Smaller (b100 km3) caldera-forming eruptions

Fig. 3. Schematic view of the magmatic system that fed the very large (~1200 km3) rhyo-litic Kidnappers eruption, Mangakino volcano, Taupo Volcanic Zone, New Zealand. Theeruption tapped three separate melt bodies distributed laterally along the rift.recent petrological studies of mac systems, but are consistent withemerging views of some silicic systems (e.g., Fig. 3). We take this viewone step further by exploring the implications of complexmagma reser-voir geometries for syn-eruptive melt extraction, eruption dynamicsand caldera collapse. We end by examining the implications of differentmagma storage models for hazard assessment, including geophysicalprospecting for magma reservoirs capable of producing very largeeruptions.

2. Caldera-forming eruptions: an overviewModied from Cooper et al. (2012).tions are dominated by the (segregated) melt phase (e.g., Bachmannet al., 2007). Accumulation of rhyolitic residual melt prior to CPR erup-tions is inferred to occur by crystal settling, compaction and/or lterpressing of the crystal mush (e.g., Sisson and Bacon, 1999; Bachmannand Bergantz, 2004; Bea, 2010; Dufek and Bachmann, 2010).

MI and CPR eruptions differ in eruption style and timing of calderaformation. MI eruptions commonly lack an early Plinian (single vent,high plume) phase and initiate, instead, with eruption of pyroclasticdensity currents from faults along the calderamargin.Where the timingof caldera collapse can be determined, it is coincident with the start ofthe eruption (Sparks et al., 1985; Lindsay et al., 2001; de Silva et al.,2006; Gregg et al., 2012; Willcock et al., 2013). As a result, associateddistal ash deposits derive mostly (or exclusively) from the co-ignimbrite plume (e.g., Chesner et al., 1991). CPR eruptions, in contrast,typically start with a Plinian (high plume) phase, as recorded in wide-spread and voluminous fall deposits. With time, the vent widens(often by propagating ring faults) and pyroclastic ows comprise an in-creasing proportion of the erupted volume. Caldera collapse occurs onlyafterwithdrawal of a critical volume ofmagma that can be related to thedepth and geometry of the reservoir (e.g., Roche andDruitt, 2001; Geyeret al., 2006; Geshi et al., 2014).

Smaller caldera-forming eruptions (~100 km3) encompass a widerange of magma compositions and crystallinities (e.g., Hildreth, 1981)and are typically associated with stratovolcanoes. When classied bymatrix glass (rather than bulk) compositions, these eruptions can beassigned to one of three groups: rhyolite (SiO2 N 70%), intermediate(phonolite/trachyte; 55% b SiO2 b 70%), or mac (ultrapotassic;SiO2 b 55%; Fig. 4B). These melt compositions are often buffered atpseudo-invariant points (Fowler et al., 2007; Boari et al., 2009; Gualda

Fig. 4. Bulk rock and matrix glass compositions of CP and CR ignimbrites as a function ofDRE erupted volume. (A) CR data; yellow squares show matrix glass, orange bars showbulk compositional range. (B) CP data; blue circles showmatrix glass composition of ear-liest erupted magma, blue bars show bulk compositional range. Data sources are listed in

Table S1.

et al., 2012). By analogy with the larger systems, it is commonly as-sumed that evolved crystal-poor (CP)magma is segregated into a singlelarge body prior to eruption, and that late-erupted crystal-rich (CR)magma is mobilized by recharge melts (e.g., Bacon and Druitt, 1988;Pallister et al., 1992; Allen, 2001; Bachmann, 2010). When crystal-richmagmas are erupted early (for example, Pinatubo 1991, Philippines;Quilotoa 800 ybp, Ecuador), erupted magmas appear similar to largerMI eruptions in their (general) homogeneity, bulk composition andhigh crystallinity (e.g., Polacci et al., 2001; Rosi et al., 2004). Caldera col-lapse in these systems is attributed to magma withdrawal and under-pressurization, and may happen at some point during the eruptive se-quence (e.g., Druitt and Sparks, 1984).

Of the caldera-forming ignimbrite family, the smallest, and in manyways the oddest, group is that of ultrapotassic (SiO2 b 55%) ignimbritesfound primarily in the Quaternary Roman Magmatic Province (QRMP;Italy). The QRMP comprises four major caldera complexes that haveproduced recurrent eruptions of tephritic to tephri-phonolitic ignim-brites with DRE (dense rock equivalent) volumes of 150 km3 (Fig. 5;Giordano et al., 2006; Boari et al., 2009; Masotta et al., 2010; Fredaet al., 2011; Vinkler et al., 2012; Acocella et al., 2012). The recurrencetimes for caldera-forming eruptions at each caldera complex are 40 to50 ka, and caldera areas range from 30 to 300 km2. These eruptionsare curious because ultrapotassic magmas have low viscosities(104.5 Pa s, even accounting for up to 30% syn-eruptive crystallization;Vinkler et al., 2012; Campagnola, 2014) and are therefore susceptibleto gas escape during magma ascent. The eruption sequence ofultrapotassic ignimbrites is, however, comparable to that of their silicic

counterparts, with an early fallout phase from a central single ventfollowed by climactic, syn-collapse highlymobile ignimbriteswith asso-ciated proximal collapse breccias (e.g. Giordano and Dobran, 1994;Watkins et al., 2002). As is typical for other stratovolcano eruptions,none of these caldera complexes has undergone post-collapse resur-gence; instead, post-collapse subsidence, intracaldera volcanism andsedimentation are common (Giordano et al., 2006; Acocella et al., 2012).

3. Mac magmatic systems

To explore relations between conditions of magma eruption andstorage, we rst briey reviewmac and mac ultrapotassic magmaticsystems, where both recent observations of eruptions (Tarasewicz et al.,2012; Carbotte et al., 2013) and detailed petrological (e.g., Marsh, 2013;Neave et al., 2013) and volcanological (e.g., Brown et al., 2014; Vinkleret al., 2012) studies all point to syn-eruptive, and sometimes explosive,tapping of multiple melt lenses stored within complex magma plumb-ing systems. Geochemical data suggest, moreover, that these meltlenses may be vertically distributed throughout the crust. In fact, isoto-pic evidence for crustal assimilation of ood basalt magmas (e.g., Wolffet al., 2008; Vye-Brown et al., 2013) requires that very large volumes ofmac melt accumulate within the crust prior to eruption.

3.1. Stacked sill models of mac magmatic systems

Mac magmatic systems are commonly envisioned as sequences ofmelt lenses or stacked sills within largely to completely crystalline

TheLion=

31K.V. Cashman, G. Giordano / Journal of Volcanology and Geothermal Research 288 (2014) 2845Fig. 5. The Quaternary RomanMagmatic Province. (A)Main caldera-producing centers. (B)tion unit: black lines = isopachs of the basal scoria fall (shown in yellow); orange= Tufoposition of VSN units: orange eld = Tufo Lionato (from Freda et al., 1997); pink squares

collapse breccia (from Conticelli et al., 2010).Colli Albani volcano with the extent of the 355 ka Villa Senni (VSN) caldera-forming erup-ato pre-collapse ignimbrite; pink= Pozzolanelle climactic ignimbrite. (C) Chemical com-Pozzolanelle ignimbrite (from Conticelli et al., 2010); purple circles = spatter in caldera

zones (e.g., Marsh, 1996; Annen et al., 2006; Gudmundsson, 2012). Sillsformwhen repeatedmagma injections are sufciently spaced in time toallow complete cooling between injection events (e.g., Annen et al.,2006; Burchardt, 2008); melt lenses dominate when new inputs areadded to systems that are still partially molten, or within sills that aresufciently thick to allow internal redistribution of melt (e.g., Marsh,2002, 2013). Variants of stacked sill models have been invoked to ex-plain the petrologic diversity ofmacmagmaerupted during individual,often long-lived, eruptive episodes (e.g., Aarnes et al., 2008; Kelley andBarton, 2008; Erlund et al., 2010; Dahren et al., 2012; Passmore et al.,

32 K.V. Cashman, G. Giordano / Journal of Volcanology and Geothermal Research 288 (2014) 2845Fig. 6. Time-progressive unloading of a stacked sill sequence beneath Eyjafjallajkull vol-cano, Iceland, 2010; illustrates response to a downward propagating decompression frontof a partially molten crustal magmatic system.

Redrafted from Tarasewicz et al. (2012).2012; Neave et al., 2013; van der Zwan et al., 2013). In these conceptualmodels, each thermally zoned mac melt lens evolves by the progres-sion of a solidication front that is multiply saturated, such that meltcomposition is buffered along a eutectic/cotectic. At the same time, crys-tallization generates highly differentiated melt compositions that may(1) segregate by compaction at intermediate crystallinities, (2) becometrapped within small pores in regions of high crystallinity (N70%;e.g., Dufek and Bachmann, 2010), or (3) segregate into lenses, pockets,and bulbous masses (Marsh, 2002; Masotta et al., 2012).

The volcanological consequences of stacked sill models have notbeen thoroughly explored. Direct evidence for syn-eruptive tapping ofpreviously intruded sills is provided by precursory and syn-eruptivegeophysical data for the remarkable pattern of downward-propagating seismicity that accompanied the 2010 Fimmvrduhls-Eyjafjallajkull eruption in Iceland (Tarasewicz et al., 2012; Fig. 6). Evi-dence for at least two sill intrusions in the 1990s (Sigmundsson et al.,2010) provides support for this interpretation. An important conse-quence was an eruption that involved several distinct explosive epi-sodes separated by times of lava effusion; another consequence wasa wide range of erupted compositions and groundmass textures(e.g., Cioni et al., 2014). This well documented example shows thatmelt does not have to be assembled pre-eruptively into a single largebody to contribute to a single eruptive episode, and supports previousinterpretations of individual macintermediate eruptions fed fromcomplex reservoirs (e.g., Yoshimoto et al., 2004; Roman et al., 2006;Erlund et al., 2010; de Angelis et al., 2013; Neave et al., 2013). Thesedata also show that changes in eruptive activity (such as pauses andtransitions between explosions and lava effusion) may reect not onlyconditions of magma transport to the surface (e.g., Melnik et al., 2005)but also conditions of magma storage, including the over-pressuremaintained within the magma reservoir. For example, pressure withinindividual melt lenses may be modulated dynamically by the interplaybetween gravitational instability of the solidication front (Marsh,2002; Humphreys and Holness, 2010), local gas build-up caused byvolatile-saturated crystallization (Tait et al., 1989), changes in melt vol-ume caused by the balance between crystallization and interstitial meltsegregation (e.g. Sisson and Bacon, 1999; Bachmann and Bergantz,2004), compressibility of magma with an exsolved volatile phase(Johnson, 1987; Voight et al., 2010), and/or intrusion of rechargemelt and/or gas from deeper melt lenses.

3.2. Caldera formation from complex mac magma reservoirs anexample from Colli Albani

The Eyjafjallajkull eruption produced a small volume of magma,was only moderately explosive, and did not produce a caldera. Could amagma system comprising multiple melt lenses produce a large,caldera-forming eruption? To address this question we examine themac ultrapotassic volcano Colli Albani (QRMP; Fig. 5), which has pro-duced several large (3050 km3) caldera-forming eruptions (Giordanoand the CARG team, 2010). Best characterized is the 355 ka Villa Sennieruption (VSN; Vinkler et al., 2012; Figs. 5, 7). The architecture of theVSN deposit is similar to those of many silicic caldera-forming erup-tions, with a basal fall deposit of 0.3 km3 (all volumes in dense rockequivalent, or DRE) followed by the N10 km3 tephri-phonolitic,crystal-poor Tufo Lionato ignimbrite (VSN1). VSN1 is overlain by theN20 km3 Pozzolanelle ignimbrite (VSN2), which is tephri-phonolitic totephritic in composition and (mostly) crystal-rich (Watkins et al.,2002). Evidence of caldera collapse can be found in the sudden appear-ance between VSN1 and VSN2 of an intercalated co-ignimbrite brecciawith up to 30% deep-seated lithics (from ~1 to 6 kmdepth, including in-trusives and cumulates). The breccia also contains an isotopically dis-tinct K-foidite juvenile magma that does not appear elsewhere in theVSN stratigraphy (Conticelli et al., 2010). The deposit characteristics,aswell as the ubiquitous vapor-phase lithication of the lower VSN1 ig-

nimbrite, suggest that a (short?) hiatus preceded caldera collapse.

33K.V. Cashman, G. Giordano / Journal of Volcanology and Geothermal Research 288 (2014) 2845One unusual feature of the eruptive deposit relates to the pyroclastvesicularities, which are generally very low (b~40% bulk, and b ~60%,melt-referenced; Fig. 7), and contrast with typical crystal-poor silicicpumice vesicularities of N70% (e.g., Klug et al., 2002; Gurioli et al.,2005; Adams et al., 2006; Houghton et al., 2010). At the same time, ves-icle number densities of 108 cm3 reach those of other mac plinianeruptions (Vinkler et al., 2012). These vesicle characteristics suggest ex-tensive syn-eruptive outgassing, although concomitant with high ratesof magma decompression. And yet tens of cubic kilometers of low vis-cosity magma were erupted explosively to form two separate, large-volume ignimbrites. This apparent conundrum strongly suggests thatvolatile exsolution is not necessarily the primary driving force for erup-tive activity.

At the same time, geochemical examination of the eruptive prod-ucts shows that extensive (60%) crystallization of clinopyroxeneand leucite are required to produce Colli Albani magma (Fredaet al., 1997, 2011; Peccerillo, 2005; Boari et al., 2009). This impliesthat the bulk of the magma storage system was highly crystalline,even though both early-erupted (pre-collapse) and very evolved(syn-collapse foidite) magma are crystal-poor (Fig. 8a, b). Furthersupport for a largely crystalline magma reservoir can be found in

Fig. 7. Stratigraphic and textural characteristics of VSN (compiled from Vinkler et al., 2012); cololines for vesicularity and vesicle number density (VND) are from Rust and Cashman (2011). Arcollapse deposits.the ejection of both cognate cumulates and loose cumulate crystals(up to 35% of mm- to cm-sized leucite, clinopyroxene and biotite;Fig. 8c, d) during and after caldera collapse. Incorporation of cumu-late crystals into the erupted melt explains the observed change inbulk composition from tephri-phonolite to phono-tephrite, andsuggests that progressive disruption of the crystal frameworkaccompanied caldera collapse. Additionally, eruption of K-foidite inassociation with syn-caldera breccias shows that at least one isolat-ed melt lens was tapped at the time of caldera formation.

The data reviewed above low pyroclast vesicularity, time gaps,multiple explosive episodes, variable melt and bulk compositions are difcult to reconcile with magma extraction from a single, pressur-ized, well mixed, melt-dominated magma chamber. Instead, theeruptive sequence has many elements that suggest involvement of acomplexmagma reservoir composed of both laterally and vertically dis-tributedmelt lenseswithin a variably crystallinemush (Fig. 9).Most im-portantly, both the explosive nature of the eruptive activity and thepoorly vesicular scoria appear to require sustained overpressures todrive the eruption.We suggest that a complexmagma reservoir can ex-plain these observations if overpressure is accommodated within indi-vidual melt lenses (e.g., Bagdassarov and Dorfman, 1998). In this

r scheme used is the same as in Fig. 5. Juvenile types (by number) refer to Fig. 8. Referencerow on right hand side labeled f represents the volumetric ratio of pre-collapse and post-

cocroc

34 K.V. Cashman, G. Giordano / Journal of Volcanology and Geothermal Research 288 (2014) 2845Fig. 8. Photomicrographs of component types in VSN deposit (labeled by number to(2) Microphenocryst-rich scoria with irregular vesicles. (3) Phenocryst-rich scoria with mileucite, clinopyroxene and biotite. All images are 3.3 mm across.scenario, elastic relaxation of the crystalline framework could maintaina sufciently high driving pressure to sustain fast withdrawal of evenvolatile-poor magma. As a consequence, sustained melt extractionwould not require volatiles to provide the only driving force for erup-tion. Instead, as for oil or water extraction from an over-pressured res-ervoir, melt out-ow could be driven initially by release of storeddeformational energy within the (bubble-bearing?)magma, the (visco-elastic) crystalline parts of the reservoir, and the country rock. Vesicula-tion triggered by magma ascent and decompression would thenenhance magma ascent, particularly at low pressures. In this complex

Fig. 9. Conceptual model for the VSNmagma reservoir, which is composed of CPmelt lenses wieruption, isolatedmelt pockets may be tapped either as intervening septa are ruptured or wheninitially fromwithin CP lenses, where viscosity is lower, and progressively involves larger portioleft depicts the possible geometry of an upper and a lower solidication front with embeddedshow the inferred original positions ofmagma parcels erupted sequentially in the Villa Senni ignrrespond with labels in Figs. 6 and 8). (1) Crystal-poor scoria with round vesicles.rystalline matrix and highly irregular vesicles. (4) Cognate xenolith containing crystals ofmagma reservoir model, time gaps reect modulation of magma with-drawal by the strength of the crystal framework (Fig. 10), and thetime required for connection of melt lenses to the main conduit(e.g., Fig. 6).

To summarize, magma storage within a complex melt/mush reser-voir helps to explain many unusual characteristics of the Villa Sennieruption, including the stability afforded by distributing, rather thanconcentrating melt, the potential for isolated lenses to develop bothhigh overpressures and highly evolved compositions, and the opportu-nity for sequential tapping of melt lenses to sustain explosive activity

thin a CRmagmamatrix. Melt lenses are variably interconnected prior to eruption; duringintercepted by propagation of the caldera-bounding fault. During eruption, magma ows

ns of the reservoir eventually including the crystal rich framework. The enlarged box to themelt accumulation. Numbers 1 to 4 are the same as the zoned juvenile types in Fig. 7, andimbrite succession, from crystal poor type (1) to progressively crystal richer types (2, 3, 4).

35K.V. Cashman, G. Giordano / Journal of Volcanology and Geothermal Research 288 (2014) 2845throughout an eruptive sequence. Deposits from the VSN eruption fur-ther suggest that the timing and style of caldera collapse may be con-trolled by processes within the reservoir (specically, failure of partsof the crystal network), in addition to processes external to the reservoir(such as the geometry, thickness and mechanical properties of the roofrock and caldera faults). More broadly, we highlight emerging views ofmac magma reservoirs as vertically extensive and comprising bothmelt-rich and melt-poor (or melt-absent) regions. This view derivesnot only from the geophysical, petrologic and volcanologic studiesreviewed above, but also fromnew thermalmodels that examine condi-tions required to develop complex storage regions (e.g., Annen et al.,2006; Annen, 2011; Solano et al., 2012). More important from avolcanological perspective, however, are the implications for conditionsleading to, and evolving during, volcanic eruptions from magmareservoirs that contain multiple and variably connected melt lenses(e.g., Gudmundsson, 2012).

4. Storage and eruption from large silicic systems

Wenow address the question of the extent towhich complex (melt-lens-dominated), as compared to simple (single melt body), magmareservoirs can provide insight into processes that contribute to themuch more common eruptions of intermediate to silicic magmas. Ourgoal is not to dismiss the StandardModel, but instead to evaluate the ex-tent to which emerging, and sometimes conicting, observations aboutvery large explosive eruptions can be reconciled by broadening ourviews of magmatic systems.

Fig. 10. Changes inmagma strength as a function ofmelt volume fraction. Dashed lines aret to experimental data on the Westerly granite (upper) and Delegate aplite (lower).Adapted from Rosenberg et al. (2007).4.1. Pre-eruptive magma storage

Several recent studies of crystal-poor rhyolitic ignimbrites suggestthat multiple, rather than single, melt batches were tapped during indi-vidual caldera-forming eruptions. Evidence for multiple melt batches isparticularly common in extensional environments such as the SnakeRiver Plain (US; Ellis et al., 2010; Ellis and Wolff, 2012) and the TaupoVolcanic Zone (TVZ, New Zealand; Brown et al., 1998; Charlier et al.,2003; Gravley et al., 2007; Wilson and Charlier, 2009; Bgu et al.,2014), and demonstrates the importance of crustal forcing on bothmagma storage and eruption (e.g., Lindsay et al., 2001; Gottsmannet al., 2009; Cooper et al., 2012; Allan et al., 2013). Two laterallydisplaced (and non-communicating) melt lenses may also have beentapped during the 600 km3 Bishop Tuff eruption from the Long Valleycaldera (Gualda and Ghiorso, 2013), which lies within a trans-tensional setting at the eastern edge of the Sierra Nevada and has longbeen considered the iconic example of the Standard Model of a singlezoned magma chamber. In all cases, melt lenses are similar in bulk,but distinct in trace element and isotopic, composition and were storedin laterally extensive (rather than vertically elongated) reservoirs(e.g., Fig. 3).

Monotonous (crystal-rich) ignimbrites (MI) may also preserve evi-dence of melt segregation. In fact, although the very name monoto-nous denotes homogeneity, careful examination of some large MIdeposits has shown that these systems may also be spatially heteroge-neous. The best-documented example is the Lund Tuff (USA), wherestudies of individual pumice clasts show that the erupted magma wasinhomogeneous in temperature, phenocryst proportions, and mineralcompositions (Maughan et al., 2002). Compositional heterogeneity hasalso been documented in the Cerro Galan ignimbrite, NW Argentina,where different sectors of the ignimbrite outow sheet have distinctcompositional characteristics (Francis et al., 1989; Kay et al., 2011),and compositionally distinct white and gray pumice clasts provide evi-dence for at least two magmas involved in the eruption (Folkes et al.,2011; Wright et al., 2011). In fact, discrete evolved melt pockets are in-ferred even for systems that lack evidence for diversemelt compositions(e.g., Huber et al., 2012; Willcock et al., 2013).

Compositionally zoned ignimbrites are perhapsmost representativeof the Standard Model, in that they are interpreted to record top-downevacuation of individual magma chambers. They also demonstrate thefundamental role of mac magma in providing either heat (to partiallymelt roof rocks) or evolved melt (from cooling and crystallization). Ex-amples of the former include Gran Canaria (Freundt and Schmincke,1995) and Iceland (Askja, Sigurdsson and Sparks, 1981), where macinputs are hot and water-poor. Examples of the latter are common inwater-rich environments, where (often cooler) crystal-poor silicicmagma overlies hotter (often more crystalline) mac magma(e.g., Hildreth, 1981; Bacon and Druitt, 1988; Druitt and Bacon, 1989;Deering et al., 2011; Pamukcu et al., 2013). As both lower temperaturesand lower PH2O promote crystallization (Blundy and Cashman, 2008;Cashman and Blundy, 2013), the lower temperature of the dominantcrystal-poor rhyolite requires commensurate zoning in water unlesscompositional differences are large. Zoned magmas are particularlycharacteristic of (although not unique to) vertically elongatedmagmat-ic systems that feed arc stratovolcanoes, perhaps because verticallyelongated (andwater-rich) reservoirs are less susceptible to convectivemixing than sill-likemelt bodies (e.g., Maughan et al., 2002; Blundy andCashman, 2008).

Amore unusual example of a zonedmagma reservoir is provided bythe NovaruptaKatmai eruption of 1912, which involved 7.5 km3 ofcrystal-poor high-silica rhyolite and 5.5 km3 of crystal-rich continuouslyzoned dacite to andesite magma. The unusual aspect of the eruption isnot the compositional zonation but the observed caldera collapse atMt. Katmai, which lies 10 km from the eruptive vent of Novarupta. De-tailed studies by Hildreth and Fierstein (2012) demonstrate that therhyolite was extracted from the intermediate composition magma andthat extraction and lateral transport of crystal-poor rhyolite from thestorage region beneath Mt. Katmai was necessarily rapid, perhaps oc-curring during the 5 days of recorded precursory activity. This raisessome interesting questions. First, to what extent does zoning observedin ignimbrites provide direct evidence of zoning within magma reser-voirs? Second, how can large amounts of rhyolite melt be extractedfrom crystal mush zones both efciently (without accompanying crys-tals) and rapidly (in days)? An alternative suggestion is that the rhyolitemagma intruded as an ascending rhyolite dike that intersected the res-ident intermediate composition magma reservoir under Mount Katmai(Eichelberger and Izbekov, 2000). Neithermodel provides a good expla-nation for the lateral displacement of the eruptive vent from themagmastorage region.

4.2. Time scales of magma accumulation

The time scale of magma accumulation prior to large eruptions hasbeen the subject of numerous recent studies that apply new diffusion

chronometers to observed crystal zoning patterns. A surprising result

of these studies is the suggestion that large volumes of silicic melt mayaccumulate over short (centuries or less) time scales (e.g., Charlieret al., 2007; Gualda et al., 2012; Allan et al., 2013; Fig. 11). Other studiesdescribe very short time scales (decades to years or even months) oflate-stage crystallization (e.g., Wark et al., 2007; Saunders et al., 2010;Druitt et al., 2012; Gualda et al., 2012; Matthews et al., 2012) and/or in-corporation of xenocrysts (e.g., Gardner et al., 2002) prior to large erup-tions. Short timescales of magma accumulation are difcult to reconcilewith long times (104105 years) required for compaction-driven meltextraction (e.g., McKenzie, 1984; Bachmann and Bergantz, 2004; Dufekand Bachmann, 2010). An alternative mechanism for segregating silicicmagma invokes formation ofmelt channels or dikeswithinmore crystal-line parts of the magma reservoir; melt extraction in this scenario isdriven by either pore pressure response to an anisotropic stress eld(e.g., Eichelberger et al., 2006; Allan et al., 2013) or rapid inter-connection of isolated melt lenses (e.g., Eichelberger and Izbekov,2000; Fig. 6). Rapid extraction and shallow accumulation of melt mayalso be driven by perturbations of local stress elds surrounding crystalmush zones. Perturbations could be caused by the arrival of newmagma inputs, gas exsolution (Sisson and Bacon, 1999), or tectonicstresses (particularly extension related to rifting; Allan et al., 2012).

4.3. Eruption triggers

36 K.V. Cashman, G. Giordano / Journal of Volcanology and Geothermal Research 288 (2014) 2845Fig. 11. Time constraints on melt accumulation prior to the Oruanui eruption, TaupoVolcanic Zone, New Zealand. Orange curve represents the time required to construct theprimarymagma reservoir (high-Si rhyolite), using FeMg interdiffusion in orthopyroxene(opx). Purple curve represents late-stage re-equilibration of opx incorporated into themain magma body. Green inset curve represents the distribution of opx diffusion ageswithin low-Si rhyolite magma, which is interpreted as growth in isolated melt pocketsthat were intersected during the eruption.Critical to understanding caldera-forming eruptions is the consider-ation of processes that cause a stable melt/mush system to destabilize.By denition, an eruption starts when the magmatic system becomesconnected to the surface. This connection can be established byupward-propagating dikes driven by magmatic overpressure (an inter-nal trigger), or downward-propagating faults generated by thermo-mechanical instabilities in the roof rocks (an external trigger;e.g., Gudmundsson, 2008; Gregg et al., 2012). Upward-propagatingdikes are commonly invoked when there is evidence for intrusion ofhotter rechargemelt (and/or volatile phase). Evidence ofmagma inuxmay be preserved in the formofmagmatic inclusions, banded pumice ormultiple pumice populations (e.g., Hildreth, 1981; Pallister et al., 1992;Polacci et al., 2001; Rosi et al., 2004). Although most easily recognizedwhen mac magma is intruded into a silicic system, the intrudingmagma may be silicic, in which case it is typically hotter and lessComplied from Allan et al. (2013)).evolved than the resident magma in the uppermost part of the system(Eichelberger et al., 2006; Hildreth and Wilson, 2007; Wright et al.,2011). The geometry of the magma reservoir may also control boththe extent of interaction of new hotter melt inputs with cooler mush(e.g., Humphreys et al., 2009) and the interaction of volatiles with themush (Wright et al., 2012).

Late-stage disturbances to magmatic systems may also be recordedas selective crystal dissolution (e.g., feldspar and/or quartz; Bachmannet al., 2002), phenocryst rim growth (Wark et al., 2007; Saunderset al., 2010; Druitt et al., 2012; Matthews et al., 2012; Allan et al.,2013) or microlite formation (Pamukcu et al., 2012). Both dissolutionand phenocryst rim growth are commonly interpreted to reect intru-sion ofmacmagma into the system; in the latter crystal growth resultsfrom cooling of themac input. In the absence of evidence for mac in-puts, however, selective dissolution and new crystal growth can also beexplained by changes in PH2O in response to decompression or additionof volatiles (Bachmann et al., 2002; Wark et al., 2007; Blundy andCashman, 2008;Matthews et al., 2012; Cashman and Blundy, 2013). Ev-idence for volatile transfer underlies the concept of gas sparging,where-by sufcient heat to unlock crystal networks is transferred by an inuxof volatiles to the system (Bachmann and Bergantz, 2006). The timescales required for unlocking by heat transfer alone are long andare similar to those required for melt extraction by compaction(e.g., Gottsmann et al., 2009). Fluxingwith H2O-rich uids could unlockcrystal networks more rapidly, however, by resorbing anhydrousphases. Introduction of CO2-richuids, in contrast, would promote crys-tal growth, particularly of feldspar (e.g., Cashman and Blundy, 2013),thereby strengthening crystal networks.

Early (precursory) phases of eruptive activity provide insight intoconditions required to initiate and sustain an eruption. Interestingly,many eruptions are preceded by leaks from the magma reservoir. Ex-amples include the eruption of the 200 km3, largely degassed, PagosaPeak dacite just before the very Fish Canyon Tuff eruption (FCT;e.g., Bachmann et al., 2000), the explosive-to-effusive Cleetwood erup-tion that preceded the c. 50 km3 caldera-forming (zoned) eruption ofCrater Lake, USA by weeks to months (Bacon, 1983; Kamata et al.,1993), and the small (~0.3 km3) explosive eruption that preceded,probably by months, the 530 km3 (crystal-poor) Oruanui eruption inNew Zealand (Allan et al., 2012). In all three cases, precursory eruptionsclearly tapped the primary magmatic system, and yet did not immedi-ately trigger the climactic eruption. In both the FCT and Oruanui exam-ples, precursory activity has been linked to tectonism in the form ofblock faulting (FCT) or rifting (Oruanui), with the latter inducing lateralmelt migration from an isolated melt lens. The dynamics of theCleetwood eruption have not been explained, although Crater Lakealso lies within an extensional region (Bacon et al., 1999) and mayhave been subject to tectonic stressing.

What, then, causes transitions from precursory activity to climacticeruptions? Interestingly, the crystal-rich FCT preserves evidence ofpre-eruptive crystal breakage interpreted to record rapid decompressionof the magma storage region (and explosive expansion of phenocryst-hosted melt inclusions) during either the Pagosa Peak eruption orearly ignimbrite eruptions from the southern part of the FCT caldera(Lipman et al., 1997). Pre-eruptive crystal breakagemay have been nec-essary to fully mobilize magma from the crystal-rich reservoir. Pre-eruptive crystal breakagemay also occur in response to migration of re-charge melt through overlying crystal mush (e.g., Pallister et al., 1992),as illustrated by the association of broken crystals with a geochemicallydistinct and partially degassedmagma in the crystal-rich Cerro Galan ig-nimbrite (Wright et al., 2011). Heating accompanying melt migrationcan also cause crystal rupture by volatile expansion within melt inclu-sions (Gualda et al., 2004; Bindeman, 2005). Finally, it has been sug-gested that crystal breakage could be a response to seismic shaking(Gottsmann et al., 2009). In all cases, physical breakage of the crystalframework would help to mobilize magma preparatory to the climactic

event.

4.4. Eruption dynamics

The triggering event can also determine the nature of initial eruptiveactivity. For example, very largeMI eruptions inferred to be triggered byroof collapse (e.g., Jellinek and DePaolo, 2003; Gregg et al., 2012; deSilva and Gregg, 2014) lack an early single vent (Plinian) phase(e.g., Druitt and Sparks, 1984; Sparks et al., 1985; de Silva et al., 2006;Cas et al., 2011; Chesner, 2012). They initiate instead with eruption ofpoorly expandedpyroclastic density currents along bounding ring faults(Willcock et al., 2013) that are sustained by very high mass uxes (Caset al., 2011; Lesti et al., 2011). In contrast, (often smaller) eruptions ofcrystal-poor magma typically have protracted single vent phases priorto caldera collapse (e.g., Crater Lake, USA (Bacon, 1983); AD 161Taupo, New Zealand (Wilson and Walker, 1985); Minoan Santorini,Greece (Druitt and Bacon, 1989; Sparks and Wilson, 1990); 39 kaCampanian Ignimbrite, Italy (Rosi et al., 1999)). Initial vents may belocated either on marginal ring faults when eruptions tap large sills(e.g., Hildreth and Mahood, 1986), or the summits of stratovolcanoes.In these eruptions, the mass eruption rate probably increases withtime, particularly after caldera collapse, and pressure changes withinthe magma reservoir may be preserved in pyroclast textures(e.g., Bacon, 1983; Klug et al., 2002; Gurioli et al., 2005).

Themode of eruptionwill also affect the nature of the eruptive prod-ucts. Ignimbrite deposits from large MI eruptions contain mostly ashand pervasively shattered individual crystals, with limited abundanceof (often low vesicularity) pumice (e.g., Carter et al., 1986; Bachmannet al., 2002; Gottsmann et al., 2009; Wright et al., 2011). Shattering ofcrystals provides evidence of extensive disruption of magma during ex-traction from the reservoir (e.g., Maughan et al., 2002).When combinedwith the low vesicularity of rare pumice clasts, these textural character-

horizontally propagating yield surfaces (Karlstrom et al., 2012) andaround subsiding caldera blocks (Kennedy et al., 2008). These processeswould promote extensive syn-eruptivemixing, which could explain theapparent dichotomy between the broad homogeneity of MI pyroclasthand samples and the extreme complexity commonly recorded withinphenocryst populations. Alternatively, mobilization of crystal-richmagma is common in late stage (syn- or post-collapse) eruptive activityfrom (inferred) vertically zoned magma reservoirs (e.g., Bacon andDruitt, 1988; Druitt and Bacon, 1989; Deering et al., 2011; Hildrethand Fierstein, 2012; Pamukcu et al., 2013), where eruptions arewell ex-plained by the Standard Model of top-down magma withdrawal(e.g., Bacon and Druitt, 1988; Allen, 2001; Mandeville et al., 2009). Inthese examples, however, crystals are not pervasively broken.

The scarcity of pumice clasts in MI ignimbrites makes it difcult toestablish details of magma extraction. Deposits from crystal-poor rhyo-lite eruptions, in contrast, often produce abundant pumice that allowsindividual parcels of magma to be related directly to their pre-eruptive storage conditions. Most informative are detailed studies ofphenocryst-hosted melt inclusions, which preserve dissolved volatilesthat can be used to estimate entrapment pressures, and major andtrace element compositions that can be used to trackmagma evolution.From a volcanological perspective, an interesting observation is thatmelt inclusion studies often indicate magma extraction from a largepressure range, even very early in the eruptive sequence (e.g., Wallaceet al., 1999; Liu et al., 2006; Mangiacapra et al., 2008; Roberge et al.,2013). One explanation for this could be pre-eruptive mixing of crystalsfrom throughout the magma storage region. Alternatively, magmacould be extracted from a large depth (pressure) range syn-eruptivelyby lateral melt migration to a vertically extensive feeder dike.

Themost thoroughly documented example is that of the Bishop Tuff,

gmand lferree m

37K.V. Cashman, G. Giordano / Journal of Volcanology and Geothermal Research 288 (2014) 2845istics suggest that vesiculation played a limited role in the eruption pro-cess, which was probably dominated instead by catastrophicdecompression (Gottsmann et al., 2009). This scenario is similar tothat outlined above for Colli Albani. At the same time, large shear strainsmay be imposed by magma mobilization along vertically extensive and

Fig. 12.Melt inclusion constraints andmagma storage andwithdrawal of the Bishop Tuffmavent, dark red = ignimbrites erupted from the southeastern margin of the caldera, dark arespectively (modied fromWilson and Hildreth, 1997). (B) Magma storage pressures inelement ratio U/Ce in melt inclusions as a function of eruption time. Colors are coded to th

Melt inclusion data fromWallace et al. (1999); Roberge et al. (2013); time constraints fromWwhere detailed stratigraphic and volcanological studies of the eruptivedeposits (Wilson and Hildreth, 1997) provide exceptional constraintson both the timing and location of magma extraction from the underly-ing reservoir (Fig. 12A).Here quartz-hostedmelt inclusions record pres-sures of b100 to N200MPa throughout the eruption, although a slightly

, LongValley. (A)Map of the aerial distribution of Bishop Tuff deposits; right red= Plinianight blue represent ignimbrites erupted from the northwest and north rim of the caldera,d from melt inclusion data as a function of eruption time. (C) Variations in incompatibleap in (A).

ilson and Hildreth (1997).

deeper magma level may have been tapped late in the eruption(Fig. 12B). Phase equilibrium constraints, in contrast, allow both earlyand late Bishop magma to span the entire (100250 MPa) pressurerange (Gualda and Ghiorso, 2013). Here continuous magma extractionfrom a large pressure range can be explained by lateral magma supplyto vertically extensive feeder dikes located on the caldera margin(e.g., Gardner et al., 1991). Horizontally directed melt ow is also sug-gested by lateral propagation of ring faults during caldera collapse(Wilson and Hildreth, 1997). Corresponding trace element analysesshow that late (post-collapse) eruptive activity tapped magma thatwas both more and less evolved than prior to collapse (Fig. 12C),which suggests late stage involvement of both less evolved melt lenses(as also suggested from zircon data; Chamberlain et al., 2014) andmoreevolved matrix melt, the latter perhaps extracted during calderacollapse.

4.5. Caldera collapse

Comprehensive reviews of caldera collapse are provided in Lipman

38 K.V. Cashman, G. Giordano / Journal of Volcanology and Geothermal Research 288 (2014) 2845(1997); Cole et al. (2005); Acocella (2007);Marti et al. (2008). These re-views focus largely on structural controls on caldera collapse styles, atopic that is beyond the scope of this review. Insteadwe explore the re-lation between melt storage and caldera formation, as indicated by var-iations in the timing of caldera formation within an eruptive sequence.Caldera formation after evacuation of substantial magma volumes hasbeen interpreted as a consequence of under-pressurization of themagma storage region (Druitt and Sparks, 1984; Scandone, 1990;Mart, 1991; Branney, 1995; Lipman, 1997; Cole et al., 2005). Calderacollapse coincident with the onset of eruptive activity, in contrast,suggests that pressurization and pre-eruptive doming caused by shal-low magma accumulation may trigger collapse of large calderas(e.g., Gudmundsson, 2008; Gregg et al., 2012; de Silva and Gregg,2014). Also important is the tectonic stress eld, which can controlthe location of caldera-bounding faults (e.g., Holohan et al., 2008a).

Models of caldera collapse (e.g., Roche and Druitt, 2001; Geyer et al.,2006; Stix and Kobayashi, 2007; Geshi et al., 2014) typically measurethe timing of collapse by f, the fraction of the total (DRE)magma volumethat is erupted prior to the onset of collapse. The value of f can be relatedto the roof rock strength and magma chamber aspect ratio (or roof as-pect ratio R = thickness/width; Roche and Druitt, 2001; Fig. 13). Col-lapse is assumed to occur when the roof can no longer support the(under-pressured) magma chamber. Theoretical analysis suggests thatcollapse begins earlier (smaller f) for magma chambers that are shallowand wide compared to those that are deep and narrow; these modelsalso predict that collapse will be piston-like when R b ~1.4, and

Fig. 13. Volume ratio f of pre-collapse to total eruption volume as a function of R, the ratioof roof thickness to roof area. High fmeans that caldera collapse was late in the eruptivesequence, under these conditions it is likely that only part, rather than all, of the magmareservoir was evacuated. Specic eruptions are labeled.

Modied from Roche and Druitt (2001).incoherent when R N ~1.4. Conditions of f = 0, that is, where collapseis synchronous with the start of an eruption, require coupling betweenshallow, laterally extensive magma chambers and the overlying roofrocks (Gregg et al., 2012). Model predictions have been tested experi-mentally with analogue magmas that are withdrawn steadily fromchambers. Fluids used in analogue experiments include air ( =105 Pa s), water ( = 103 Pa s), and silicone ( = 104 Pa s), andthus span a range of viscosities. All experiments, however, investigatesteady withdrawal of uid from a single (simple) reservoir; caldera for-mation by uid extraction from complex reservoirs has not yet beenexplored.

Variations in f can be evaluated as a function of both eruption mag-nitude (DRE volume) and eruption type (crystal-rich [CR] or crystal-poor [CP]; Fig. 14A; Table S1 in Supplementary material). These datashow that f is small (or 0) in large CR eruptions (e.g., Cerro Galan [CG],La Pacana [LP] and Fish Canyon Tuff [FCT]), and variable (but non-zero) in moderate to large CP eruptions (e.g., Long Valley [LV], Taupo[TP] andVilla Senni [VSN]). Collapsemay occur very late (large f) in stra-tovolcano eruptive sequences (e.g., Vesuvius [VS], Tambora [TMB] andAso [AS]), even for CR magma (e.g., Pinatubo [PN]) or substantialerupted volumes (e.g., AS). This compilation shows that collapsecan occur over a wide range of f for similar total erupted volumes(i.e., after very different volumes of magma extraction from the reser-voir), and at similar f for erupted volumes that vary over two orders ofmagnitude. Some of this variation can be explained by differences inmagma chamber geometry (R), but the extreme variability must reectother controlling factors, including the distribution of melt within acrystalline reservoir.

To relate f directly to conditions of magma withdrawal requires thatwe know the thickness (pressure) of themelt lens evacuatedduring col-lapse. If magma is assumed to be withdrawn from a single, sill-like res-ervoir, and if the surface expression of the caldera can be taken as thefootprint of that reservoir, then the total thickness of magma extract-ed during an eruption can be estimatedusing the erupted (DRE) volumeand caldera area, a value commonly referred to as the collapse height.Early studies identied a linear relation between DRE volume and col-lapse area (Smith, 1979; Spera and Crisp, 1981) that suggested a con-stant thickness of magma evacuation characterized many collapseevents. This thickness could be related to the tensile strength of thecrust (Walker, 1984; Scandone, 1990). Our re-examination of calculatedcollapse heights for all sufciently well characterized eruptions (120 intotal) in the Collapse Caldera Data Base (Geyer and Marti, 2008;Table S1; Fig. 14B) shows that many eruptions do have collapse heightsthat cluster at a single value ~ 1000 m, although the data are highlyvariable. In general, collapse heights 2000 m are relatively rare andoccur primarily in large CR eruptions, where total evacuation of themagma reservoir is expected. Calculated collapse heights for manylarge (N100 km3) CP eruptions, in contrast, are surprisingly small(b1000 m), possibly because the measured caldera areas exceed themagma reservoir footprint. Where f (and thus the volume of magmawithdrawn prior to caldera collapse) is known (Fig. 14A), the thicknessofmagmawithdrawn from the reservoir prior to collapse can also be in-ferred. Eruptions with large f have correspondingly large pre-eruptivemagma extraction depths, with a maximum value of ~1000 m (TMBand Campanian [CMP]). Not surprisingly, eruptions with small f havesmall pre-eruptive magma extraction depths (b0.2 km for Aira [AR],KOS, VSN and Santorini Minoan [SAN]).

The compilation shown in Fig. 14B also provides insight into erup-tion triggers. The dashed line on the diagram denes a caldera areaA=100 km2, which has been identied as a thermomechanical bound-ary that separates eruptions triggered by (external) roof collapse or (in-ternal) chamber collapse (Gregg et al., 2012; de Silva and Gregg, 2014).CR eruptions tend to lie above this boundary (in the roof-collapse re-gion), a placement that is consistentwith the observed synchroneity be-tween eruption initiation and caldera collapse. Data from many CP

eruptions also lie above this line, however, even for eruptions with

39K.V. Cashman, G. Giordano / Journal of Volcanology and Geothermal Research 288 (2014) 2845very large f values (e.g., Campanian [CMP] andAso [AS]) that denote latestage collapse (which is not consistent with a roof trigger). These dis-crepancies indicate that caldera area is not the sole control on eruptiontriggering, which will also be affected by magma storage depth, reser-voir conguration, magma input and tectonic triggering (e.g. Lindsayet al., 2001; Allan et al., 2012). Support for the latter includes the relativeplacement of small to moderate (b100 km3) stratovolcano eruptions(e.g., Tambora [TM], Krakatau [KR] Crater Lake [CL] and Santorini [SN];A b 100 km2) compared with eruptions of similar sizes in extensionalsettings (e.g., Taupo [TP]; A N 100 km2).

Ideally, the entire vertical extent of magma extraction (the drainageheight) should be recorded by the volatile contents of phenocryst-hostedmelt inclusions (Wallace et al., 1999) and/or the stability of phe-nocryst assemblages (e.g., Hammer et al., 2002; Gualda and Ghiorso,2013). There are very few caldera-forming eruptions, however, forwhich the drainage height is well constrained. An exception is the Bish-op Tuff, where both melt inclusion and phase equilibria data suggestmagma withdrawal over 130 MPa (~5000 m; Fig. 12B). This value

Fig. 14. Relations between erupted volume (DRE), fraction of magma erupted prior to col-lapse (f) and inferred collapse height (calculated as erupted volume/caldera area). (A) f vs.volume; (B) volume vs. inferred collapse height. Symbols are the same in both, with bluecircles CPmagma and yellow squares CRmagma. Lines in (A) show pre-eruptive DRE vol-umes (labeled, in km3). Dashed line in (B) shows the contour for a caldera area of 100 km2,which Gregg et al. (2012) suggest as the bounding limit between chamber-trigged androof-triggered (yellow shading) eruptions. Labeled eruptions are as follows: AR = Aira,AS = Aso, CEB = Ceboruco, CG = Cerro Galan, CL = Crater Lake, CMP = Campanian(Campi Flegrei), KT= Katmai, KOS= Kos, KRA= Krakatau, LG= La Garita (Fish CanyonTuff), LV = Long Valley (Bishop Tuff), NYT = Neapolitan Yellow Tuff (Campi Flegrei),PN = Pinatubo, SAN = Santorini (Minoan), TMB = Tambora, TP = Taupo (181 AD),VICO = Vico, VS = Vesuvius (79 AD), VSN = Villa Senni (Colli Albani). All data fromTable S1.exceeds by a factor of 3 the collapse height of 1500 m calculated forthe entire erupted volume (Table S1), and by a factor of ~2 the collapseheight of 23 km indicated by analysis of drill core samples taken fromwithin the caldera (Hildreth andMahood, 1986). At face value, this dis-crepancy implies that on average, melt was extracted from less than 1/21/3 of the magma reservoir. Possible explanations for this mismatchinclude under-estimation of collapse heights, prior incorporation ofdeep crystals (and their melt inclusions) into the eruptible melt lens,or withdrawal of magma from stacked lenses that comprise only partof the total magma reservoir (Fig. 15). We prefer the latter explanation,as data presented in Hildreth (1979) argues against mixing by large-scale convective overturn of the magmatic system prior to eruption.

If collapse height records the combined thickness of individual meltlenses and compaction, but not evacuation, of the crystal framework,and if the structural support provided by the reservoir plays a role incontrolling collapse height, thenwewould expect the largest discrepan-cies between caldera collapse heights and magma drainage heights inCP eruptions that tap complex reservoir geometries, where the strengthof the crystal framework would allow decoupling between melt drain-age and collapse. By contrast, collapse height and drainage heightshould coincide in eruptions that involve complete evacuation of thereservoir (f = 0). This prediction is supported by data that indicatethat subsidence associated with large volume CR eruptions is typically34 km, consistent with complete evacuation of these crystal-richmagma reservoirs (Lipman, 1997). In CP eruptions, in contrast, partialinvolvement of the crystal framework is suggested by a transitionduring and after caldera collapse from initially crystal-poor magmato crystal-rich magma. The latter often contains both cognateglomerocrysts and antecrysts with melt inclusions that are moreevolved in composition than the matrix glass (that is, from a cooler, ormore evolved, part of the magma reservoir; Beddoe-Stephens et al.,1983; Wolff et al., 1999; Charlier et al, 2007; Saunders et al., 2010;Roberge et al., 2013).

In summary, there is growing evidence that caldera-forming erup-tions are not all fed by single magma bodies, and that accumulationsof eruptiblemelt do not necessarily require assembly over long time pe-riods. Instead, some systemsmay store melt withinmultiple sills (with-in a rigid framework) or lenses (within a crystal mush) that may eitheramalgamate into a single melt body shortly before eruption, or may betapped syn-eruptively, particularly in extensional environments. Inthese complex magma reservoirs, pulsed interconnection of isolatedmelt lenses promoted by syn-eruptive depressurization could both pro-long explosive activity and explain commonly observed hiatuses ineruptive sequences (e.g., Aramaki, 1984; Allen, 2004; Palladino andSimei, 2005; Bear et al., 2009; Vinkler et al., 2012), particularly if adjust-ments within the reservoir are required to mobilize more crystalline(and viscous) magma to newly formed caldera-bounding fractures.The internal geometry and properties of a complex magma reservoirmay also bear an unexplored inuence on the development of ringfaults (e.g., Kennedy et al., 2004; Holohan et al., 2008b; Burchardt andWalter, 2010) and extent of collapse in different sectors of a caldera.In fact, it seems likely that the spatial distribution of melt within a com-plex reservoir will contribute to the identied spectrum of piston,downsag, trapdoor and piecemeal collapse styles (e.g., Cole et al.,2005). Taken together, we suggest that the full range of eruptive condi-tions (precursors, triggers and eruption dynamics) created by tappingcomplex storage regions has yet to be explored, and represents excitingopportunities for future research.

5. Implications for recognizing eruption potential of largemagmatic systems

Effective volcano monitoring requires identication of systemscapable of producing very large eruptions. For this reason, severalpotentially active volcanic regions have been the recent targets of

geophysical surveys to search for large melt bodies. With only a

40 K.V. Cashman, G. Giordano / Journal of Volcanology and Geothermal Research 288 (2014) 2845few exceptions (e.g., Zollo et al., 2008), interpretations of resultinggeophysical images assess melt contents at ~30% (e.g., Schillingand Partzsch, 2001; Zandt et al., 2003; Chu et al., 2010; Luttrellet al., 2013), a number that appears safely at odds with the 50%melt considered necessary for melt to be eruptible (e.g., Bachmannand Bergantz, 2008). New views of magmatic systems, however,show that large volumes of melt may accumulate rapidly, and thatmultiple magma lenses may be tapped during a single eruptive epi-sode. Syn-eruption melt extraction from a largely crystalline reser-voir seems likely for eruptions such as Aso, where a large volume

Fig. 15. Schematic diagram showingmagma evacuation from a complex reservoir. (A) Prior to c(the drainage depth), and magma migration is lateral, as well as vertical. (B) Collapse initiatereduced and the framework itself partially disrupted. Here the collapse height is less than theantecrysts from the framework; transition from a single vent to a ring vent phase is often accomalong with any constituent melt lenses, the collapse height equals the drainage height (the cas(~200 km3) of magma was erupted under conditions of f N 0.9 butR b 1 (estimated from reported caldera area of 200 km3 and H2O con-tents ~57 wt.% (Kaneko et al., 2007), which places the magma atdepths no greater than 810 km, R ~ 0.50.67). Also suggestive is ev-idence (from both melt inclusion and phase equilibria) for magmaextraction over much larger pressure (depth) ranges than the col-lapse height calculated from erupted volume and caldera area. With-in this framework, we suggest that magma reservoirs such as thoseimaged beneath Yellowstone may actually be capable of producinga large eruption (e.g., Wotzlaw et al., 2014).

aldera collapse, magma is extracted frommelt-dominated lenses throughout the reservoirs when sufcient melt has been withdrawn that the strength of the crystal framework isdrainage depth (as seen in many CP eruptions), and the erupted magma often containspanied by a hiatus in eruptive activity. (C)When the crystalmush is completely evacuatede for many MI eruptions).

sion). These observations require newmodels to explain and anticipatetriggering and eruption of magma from complex storage reservoirs.

41K.V. Cashman, G. Giordano / Journal of Volcanology and Geothermal Research 288 (2014) 2845Another challenge relates to monitoring complex magmatic sys-tems. An increasingly important and effective volcanomonitoring tech-nique is measurement of surface deformation, particularly usingsatellite-based Interferometric Synthetic Aperture Radar (InSAR;e.g., Sparks et al., 2012; Pyle et al., 2013). Surface deformation overlarge, shallow and sill-like magma bodies (that is, those susceptible toroof triggers) should be substantial; this makes them particularly goodtargets for monitoring by InSAR. One complication, however, is thatlarge magmatic reservoirs often have well-developed active hydrother-mal systems that may show extensive deformation related to shallowchanges in pore-pressure and water levels (e.g., Chiodini et al., 2003;Husen et al., 2004; Chang et al., 2007). Pore-pressure-generated defor-mation signalsmay eithermask ormimicmagmatic activity. Surface de-formation can also provide evidence of deep intrusive activity, such asthat currently ongoing at Uturuncu volcano, Bolivia (e.g., Sparks et al.,2008), and thus provides a potential tool for tracking long-term migra-tion of magma inputs at different crustal levels.

Magmatic activity that precedes internally triggered eruptions (thatis, triggers involving intrusion of gas or hot melt from below, or over-pressurization caused by crystallization and associated gas exsolution)may be more difcult to recognize. One potential precursor is earlymagma leakage from large reservoirs (e.g., Bacon, 1983; Dufeld,1990; Dufeld et al., 1990; Hildreth, 2004; Fabbro et al., 2013). An inter-esting observation is that precursory leaks from largemagmatic systemsare often sourced from shallow levels and may produce either unusuallow energy fountains (e.g., Dufeld, 1990; Bachmann et al., 2000) orlava ows (e.g., Bacon, 1983), despite tapping volatile-rich componentsof themagmatic system (e.g., Dufeld andDalrymple, 1990; Bacon et al.,1992; Mandeville et al., 2009). A good illustration of this phenomenon,and a cautionary tale for event-tree-based hazard analysis, is providedby the eruptive sequence at Crater Lake, OR (Bacon, 1983). Here a com-posite eruption (Llao Rock; 1.7 km3 DRE pumice fall and 0.5 km3 DRElava ow) tapped the main magma reservoir about 200 years beforecaldera formation. Another composite eruption (Cleetwood; 1.5 km3

DRE pumice fall and 0.6 km3 DRE lava ow) preceded the main (caldera-forming) phase of the eruption by only weeks to months (Kamataet al., 1993). In both cases, the magma apparently came from theclimactic reservoir (Bacon et al., 1992; Mandeville et al., 2009) and yeteach eruption transitioned from explosive to effusive. Why, then, wasthe Cleetwood eruption followed so promptly by a very large(~50 km3 DRE) explosive eruption from the same magmatic system?One possible explanation is that the precursory eruptions tapped rela-tively shallow and partially to fully isolated melt lenses within a largerreservoir. Magma withdrawal from the Cleetwood melt lens couldthen have triggered the climactic event by either downward or lateralpropagation of a decompression wave capable of connecting the isolat-ed lens to the larger reservoir. This scenario illustrates the importance ofdevelopingmethods tomonitor processes internal tomagma reservoirs(e.g., induced seismicity; Catalli et al., 2013) during, as well as prior toeruptions, and to distinguish between precursors and the main event(e.g., Allan et al., 2012).

Summary

It has long been known that caldera-forming eruptions may evacu-ate large volumes of either crystal-rich or crystal-poor magma(e.g., Hildreth, 1981). The past ten years have seen a growing numberof studies that relate the chemical and physical conditions in magmastorage regions to the conditions underwhich different parts of the sys-tem may be erupted (e.g., Jellinek and dePaolo, 2003; Bachmann andBergantz, 2004; Gottsmann et al., 2009; Bachmann, 2010; Allan et al.,2012; Cooper et al., 2012; Druitt et al., 2012; Ellis and Wolff, 2012;Gregg et al., 2012; Hildreth and Fierstein, 2012; Huber et al., 2012;Karlstrom et al., 2012; Gualda and Ghiorso, 2013). Key observationsarising from these studies include: (1) many large eruptions tap multi-

ple melt sources, (2) large melt bodies are probably transient features,We address this question by rst considering eruptions from macsystems, which are commonlymodeled as stacked sills. Here both directobservations and petrologic studies of recently active systems show thatsingle eruptions may tap multiple melt lenses. One consequence ispulsatory (and often protracted) eruptive activity and alternation be-tween explosion and lava effusion; another is eruption of a range ofmagma compositions. We extend these modern examples to larger,caldera-forming eruptions using an example from the ultrapotassicQuaternary Roman Magmatic Province (QRMP), where we suggestthat apparently contradictory observations of protracted explosivityand low pyroclast vesicularity can be reconciled if the eruption tappeda complex reservoir containing multiple melt lenses.

We then turn to the more common, and larger, eruptions of siliciccrystal-poor (CP) and crystal-rich (CR) magma, and review conditionsof magma storage, time scales of melt accumulation, eruption triggers,eruption dynamics and conditions of caldera collapse. We show wherethe Standard Model (single magma chamber) appears consistent withthe nature and stratigraphy of the eruptive products, as well as exam-ples where a complex (melt-lens-dominated) magma reservoir maybetter explain both petrological and volcanological observations. Mostimportant is the growing evidence that inmany systems that erupt (ini-tially) crystal-poor silicicmagma, particularly those in extensional envi-ronments, melt may be stored within multiple isolated and/or partiallyconnected lenses. This type of complex storage geometry could providea mechanism for rapid assembly of large melt bodies, as well as a localsource of triggering (recharge) magma. In fact, Eichelberger et al.(2006) extend this concept to suggest that a combination of porousmedia and triggered dike ow could allow sufciently rapid syn-eruptive melt extraction to feed a silicic Plinian eruption, as we have in-ferred for the QRMP example. From a hazard perspective, eruptions fedfrom complex magma reservoirs may show abrupt changes in eruptiveprocesses (such as pauses and transitions between explosive and effu-sive activity) that pose challenges for volcano monitoring and forecast-ing efforts.

From aheuristic perspective,magma storage in complex, rather thansimple, magma reservoirs resolves several existing paradoxes aboutconditions of both pre-eruptive magma storage and syn-eruptivemagma withdrawal and caldera collapse. This perspective also repre-sents a logical extension of recent attempts to reconcile petrologicalviews of incremental assembly of plutons with volcanological require-ments of instantaneous availability of very large melt volumes(Hildreth, 2004; Bachmann et al., 2007; Lipman, 2007; Walker et al.,2007). Melt accumulation within, and syn-eruptive extraction from, in-terconnected melt lenses also alleviates problems related to maintain-ing large and stable volumes of crystal-poor melt in the upper crust(Freda et al., 2011; Gualda et al., 2012; Vinkler et al., 2012) and placeswithin a single coherent framework apparently disparate observationsrelated to magma extraction and caldera collapse. Finally, such amodel is consistent with geophysical observations that even activemagma storage reservoirs often contain 30% melt (Zandt et al.,2003; Chu et al., 2010); the latter observation has importantimplications for recognizing the eruption potential of large magmareservoirs.

Acknowledgments

This work was supported by the AXA Research Fund through aResearch Professorship to KC and by Regione Lazio (818000-2009-R-(3) crystals carried by the transporting melt have been stored at arange of pressures and temperatures, and (4) eruptions of crystal-richmagma are probably driven by roof collapse and fragmented by suddendecompression (with a limited role for volatile exsolution and expan-M-R.N.C.T_001) for GG. We are grateful for the very thoughtful reviews

42 K.V. Cashman, G. Giordano / Journal of Volcanology and Geothermal Research 288 (2014) 2845by G. Gualda and S. Burchardt and for the encouragement of the editor(L. Wilson) to write this review.

Appendix A. Supplementary data

Supplementary data to this article can be found online at http://dx.doi.org/10.1016/j.jvolgeores.2014.09.007.

References

Aarnes, I., Podladchikov, Y.Y., Neumann, E.-R., 2008. Post-emplacement melt ow inducedby thermal stresses: implications for differentiation in sills. Earth Planet. Sci. Lett. 276(12), 152166.

Acocella, V., 2007. Understanding caldera structure and development: an overview of an-alogue models compared to natural calderas. Earth Sci. Rev. 85 (34), 125160.

Acocella, V., Palladino, D.M., Cioni, R., Russo, P., Simei, S., 2012. Caldera structure, amountof collapse, and erupted volumes: the case of Bolsena caldera, Italy. Geol. Soc. Am.Bull. 124 (910), 15621576.

Adams, N.K., Houghton, B.F., Hildreth, W., 2006. Abrupt transitions during sustained ex-plosive eruptions: examples from the 1912 eruption of Novarupta: Alaska. Bull.Volcanol. 69 (2), 189206.

Allan, A.S.R., Wilson, C.J.N., Millet, M.-A., Wysoczanski, R.J., 2012. The invisible hand: tec-tonic triggering and modulation of a rhyolitic supereruption. Geology 40 (6),563566.

Allan, A.R., Morgan, D., Wilson, C.N., Millet, M.-A., 2013. From mush to eruption in centu-ries: assembly of the super-sized: Oruanui magma body. Contrib. Mineral. Petrol. 166(1), 143164.

Allen, S.R., 2001. Reconstruction of a major caldera-forming eruption from pyroclastic de-posit characteristics: Kos Plateau Tuff, eastern Aegean Sea. J. Volcanol. Geotherm. Res.105 (12), 141162.

Allen, S.R., 2004. Complex spatter- and pumice-rich pyroclastic deposits from an andesiticcaldera-forming eruption: the Siwi pyroclastic sequence, Tanna, Vanuatu. Bull.Volcanol. 67, 2741.

Annen, C., 2011. Implications of incremental emplacement of magma bodies for magmadifferentiation, thermal aureole dimensions and plutonismvolcanism relationships.Tectonophysics 500, 310.

Annen, C., Blundy, J.D., Sparks, R.S.J., 2006. The genesis of intermediate and silicic magmasin deep crustal hot zones. J. Petrol. 47, 505539.

Aramaki, S., 1984. Formation of the Aira Caldera, Southern Kyushu, 22,000 years ago. J.Geophys. Res. 89, 84858501.

Bachmann, O., 2010. The petrologic evolution and pre-eruptive conditions of the rhyoliticKos Plateau Tuff (Aegean arc). Cent. Eur. J. Geosci. 2 (3), 270305.

Bachmann, O., Bergantz, G.W., 2004. On the origin of crystal-poor rhyolites: extractedfrom batholithic crystal mushes. J. Petrol. 45 (8), 15651582.

Bachmann, O., Bergantz, G.W., 2006. Gas percolation in upper-crustal magma bodies as amechanism for upward heat advection and rejuvenation of silicic crystal mushes. J.Volcanol. Geotherm. Res. 149, 85102.

Bachmann, O., Bergantz, G.W., 2008. Rhyolites and their source mushes across tectonicsettings. J. Petrol. 49, 22772285.

Bachmann, O., Dungan, M.A., Lipman, P.W., 2000. Voluminous lava-like precursor to amajor ash-ow tuff: low-column pyroclastic eruption of the Pagosa Peak Dacite,San Juan volcanic eld, Colorado. J. Volcanol. Geotherm. Res. 98 (14), 153171.

Bachmann, O., Dungan, M.A., Lipman, P.W., 2002. The Fish canyon magma body, San JuanVolcanic Field, Colorado: rejuvenation and eruption of an upper-crustal batholith. J.Petrol. 43 (8), 14691503.

Bachmann, O., Miller, C.F., de Silva, S.L., 2007. The volcanicplutonic connection as a stagefor understanding crustal magmatism. J. Volcanol. Geotherm. Res. 167 (14), 123.

Bacon, C.R., 1983. Eruptive history of Mount Mazama and Crater Lake caldera, CascadeRange, U.S.A. J. Volcanol. Geotherm. Res. 18, 57115.

Bacon, C.R., Druitt, T.H., 1988. Compositional evolution of the zoned calcalkaline magmachamber of Mount Mazama, Crater Lake, Oregon. Contrib. Mineral. Petrol. 98 (2),224256.

Bacon, C.R., Newman, S., Stolper, E., 1992. Water, CO2, Cl, and F in melt inclusions in phe-nocrysts from three Holocene explosive eruptions, Crater Lake, Oregon. Am. Mineral.77, 10211030.

Bacon, C.R., Lanphere, M.A., Champion, D.E., 1999. Late Quaternary slip rate and seismichazards of the West Klamath Lake fault zone near Crater Lake, Oregon Cascades. Ge-ology 27, 4346.

Bagdassarov, N., Dorfman, A.M., 1998. Viscoelastic behaviour of partially molten granites.Tectonophysics 290, 2745.

Bea, F., 2010. Crystallization dynamics of granite magma chambers in the absence of re-gional stress: multiphysics modeling with natural examples. J. Petrol. 51 (7),15411569.

Bear, A., Cas, R., Giordano, G., 2009. Variations in eruptive style and depositional processesassociated with explosive, phonolitic composition, caldera-forming eruptions: the151 ka Sutri eruption, Vico Caldera, central Italy. J. Volcanol. Geotherm. Res. 184(3), 225255.

Beddoe-Stephens, B., Aspden, J.A., Shepherd, T.J., 1983. Glass inclusions and melt compo-sitions of the, Toba Tuffs, northern Sumatra. Contrib. Mineral. Petrol. 83 (34),278287.

Bgu, F., Deering, C.D., Gravley, D.M., Kennedy, B.M., Chambefort, I., et al., 2014. Extrac-tion, storage and eruption of multiple isolated magma batches in the paired Mamakuand Ohakuri eruption, Taupo Volcanic Zone, New Zealand. J. Petrol. 55, 16531684.Bindeman, I.N., 2005. Fragmentation phenomena in populations of magmatic crystals.Am. Mineral. 90, 18011815.

Blundy, J., Cashman, K., 2008. Petrologic reconstruction of magmatic system variables andprocesses. Rev. Mineral. Geochem. 69, 179239.

Boari, E., Avanzinelli, R., Melluso, L., Giordano, G., Mattei, M., De Benedetti, A.A., Morra, V.,Conticelli, S., 2009. Isotope geochemistry (SrNdPb) and petrogenesis of leucite-bearing rocks from Colli Albani volcano, Roman Magmatic Province, Central Italy:inferences on volcanic evolution. Bull. Volcanol. 71 (9), 9771005.

Branney, M.J., 1995. Downsag and extension at calderas: new perspectives on collapse geom-etries from ice-melt, mining, and volcanic subsidence. Bull. Volcanol. 57 (5), 303318.

Brown, S.J.A., Burt, R.M., Cole, J.W., Krippner, S.J.P., Price, R.C., Cartwright, I., 1998. Plutoniclithics in ignimbrites of Taupo Volcanic Zone, New Zealand; sources and conditions ofcrystallisation. Chem. Geol. 148 (12), 2141.

Brown, R.J., Blake, S., Thordarson, T., Self, S., 2014. Pyroclastic edices record vigorous lavafountains during the emplacement of a ood basalt ow eld, Roza Member, Colum-bia River Basalt Province, USA. Geological Society of America Bulletin 126, 875891.

Burchardt, S., 2008. New insights into the mechanics of sill emplacement provided byeld observations of the Njardvik Sill, Northeast Iceland. J. Volcanol. Geotherm. Res.173, 280288.

Burchardt, S., Walter, T.R., 2010. Propagation, linkage, and interaction of calderaringfaults: comparison between analogue experiments and caldera collapse atMiyakejima, Japan, in 2000. Bull. Volcanol. 72 (3), 297308.

Campagnola, S., 2014. Large scale Plinian eruptions of the Colli Albani and the CampiFlegrei volcanoes: insights from textural and rheological studies. Universit RomaTre, (Unpublished PhD thesis).

Carbotte, S.M., Marjanovi, M., Carton, H., Mutter, J.C., Canales, J.P., Nedimovi, M.R., et al.,2013. Fine-scale segmentation of the crustal magma reservoir beneath the East Pacif-ic Rise. Nat. Geosci. 6, 866870.

Carter, N.L., Ofcer, C.B., Chesner, C.A., Rose, W.I., 1986. Dynamic deformation of volcanicejecta from the Toba caldera: possible relevance to Cretaceous/Tertiary boundaryphenomena. Geology 14 (5), 380383.

Cas, R.A., Wright, H.M., Folkes, C.B., Lesti, C., Porreca, M., Giordano, G., Viramonte, J.G.,2011. The ow dynamics of an extremely large volume pyroclastic ow, the 2.08-Ma Cerro Galn Ignimbrite, NW Argentina, and comparison with other ow types.Bull. Volcanol. 73 (10), 15831609.

Cashman, K., Blundy, J., 2013. Petrological cannibalism: the chemical and textural conse-quences of incremental magma body growth. Contrib. Mineral. Petrol. 166 (3),703729.

Catalli, F., Meier, M., Wiemer, S., 2013. The role of Coulomb stress changes for injection-induced seismicity: the Basel enhanced geothermal system. Geophys. Res. Lett. 40, 7277.

Chamberlain, K.J., Wilson, C.J.N., Wooden, J.L., Charlier, B.L.A., Ireland, T.R., 2014. New per-spectives on the Bishop Tuff from zircon textures, ages and trace elements. J. Petrol.55 (2), 395426.

Chang, W.L., Smith, R.B., Wicks, C., Farrell, J., Puskas, C.M., 2007. Accelerated upliftand magmatic intrusion of the Yellowstone caldera, 2004 to 2006. Science 318,952956.

Charlier, B.L.A., Peate, D.W., Wilson, C.J.N., Lowenstern, J.B., Storey, M., Brown, S.J.A., 2003.Crystallisation ages in coeval silicic magma bodies: 238U230Th disequilibrium evi-dence from the Rotoiti and Earthquake Flat eruption deposits, Taupo Volcanic Zone,New Zealand. Earth Planet. Sci. Lett. 206 (34), 441457.

Charlier, B.L.A.,Wilson, C.J.N., Lowenstern, J.B., Blake, S., Van Calsteren, P.W., Davidson, J.P.,2005. Magma generation at a large, hyperactive silicic volcano (Taupo, New Zealand)revealed by UTh and UPb systematics in zircons. J. Petrol. 46 (1), 332.

Charlier, B.L.A., Bachmann, O., Davidson, J.P., Dungan, M.A., Morgan, D.J., 2007. The uppercrustal evolution of a large silicicmagma body: evidence from crystal-scale RbSr iso-topic heterogeneities in the Fish Canyon magmatic system, Colorado. J. Petrol. 48(10), 18751894.

Chesner, C.A., 2012. The Toba caldera complex. Quat. Int. 258, 518.Chesner, C.A., Rose, W., Deino, A., Drake, R., Westgate, J., 1991. Eruptive history of Earth's

largest Quaternary caldera (Toba, Indonesia) claried. Geology 19 (3), 200203.Chiodini, G., Todesco, M., Caliro, S., Del Gaudio, C., Macedonio, G., Russo, M., 2003. Magma

degassing as a trigger of bradyseismic events: the case of Phlegrean Fields (Italy).Geophys. Res. Lett. 30 (8).

Chu, R., Helmberger, D.V., Sun, D., Jackson, J.M., Zhu, L., 2010. Mushy magma beneathYellowstone. Geophys. Res. Lett. 37, L01306.

Cioni, R., Pistolesi, M., Bertagnini, A., Bonadonna, C., Hoskuldsson, A., Scateni, B., 2014. In-sights into the dynamics and evolution of the 2010 Eyjafjallajkull summit eruption(Iceland) provided by volcanic ash textures. Earth Planet. Sci. Lett. 394, 111123.

Cole, J.W., Milner, D.M., Spinks, K.D., 2005. Calderas and caldera structures: a review.Earth Sci. Rev. 69 (12), 126.

Conticelli, S., Boari, E., Avanzinelli, R., De Benedetti, A.A., Giordano, G., Mattei, M., et al.,2010. Geochemistry, isotopes and mineral chemistry of the Colli Albani volcanicrocks: constraints on magma genesis and evolution. The Colli Albani Volcano. Spec.Publ. IAVCE I 3, 107139.

Cooper, G.F., Wilson, C.J.N., Millet, M.-A., Baker, J.A., Smith, E.G.C., 2012. Systematic tap-ping of independent magma chambers during the 1 Ma Kidnappers supereruption.Earth Planet. Sci. Lett. 313314, 2333.

Dahren, B., Troll, V., Andersson, U., Chadwick, J., Gardner, M., Jaxybulatov, K., Koulakov, I.,2012. Magma plumbing beneath Anak Krakatau volcano, Indonesia: evidence formultiple magma storage regions. Contrib. Mineral. Petrol. 163 (4), 631651.

de Angelis, S.H., Larsen, J., Coombs, M., 2013. Pre-eruptivemagmatic conditions at AugustineVolcano, Alaska, 2006: evidence from amphibole geochemistry and textures. J. Petrol.54, 19391961.

de Silva, S.L., Gregg, P.M., 2014. Thermomechanical feedbacks in magmatic systems: Im-plications for growth, longevity, and evolution of large caldera-formingmagma reser-voirs and their supereruptions. J. Volcanol. and Geotherm. Res. 282, 7791.

43K.V. Cashman, G. Giordano / Journal of Volcanology and Geothermal Research 288 (2014) 2845de Silva, S., Zandt, G., Trumbull, R., Viramonte, J.G., Salas, G., Jimenez, N., 2006. In: Troise, C.,De Natale, G., Kilburn, C.R.J. (Eds.), Large ignimbrite eruptions and volcano-tectonicdepressions in the Central Andes: a thermomechanical perspectiveMechanisms of Ac-tivity and Unrest at Large Calderas vol. 269. Geological Society of London, London, pp.4763.

Deering, C.D., Bachmann, O., Vogel, T.A., 2011. The Ammonia Tanks Tuff: erupting a melt-rich rhyolite cap and its remobilized crystal cumulate. Earth Planet. Sci. Lett. 310(2011), 518525.

Druitt, T., Bacon, C., 1989. Petrology of the zoned calcalkaline magma chamber of MountMazama, Crater Lake, Oregon. Contrib. Mineral. Petrol. 101 (2), 245259.

Druitt, T.H., Sparks, R.S.J., 1984. On the formation of calderas during ignimbrite eruptions.Nature 310 (5979), 679681.

Druitt, T.H., Costa, F., Deloule, E., Dungan, M., Scaillet, B., 2012. Decadal to monthly time-scales of magma transfer and reservoir growth at a caldera volcano. Nature 482(7383), 7780.

Dufek, J., Bachmann, O., 2010. Quantummagmatismmagmatic compositional gaps gener-ated by melt-crystal dynamics. Geology 38 (8), 687690.

Dufeld, W.A., 1990. Eru