The Genesis of Intermediate and Silicic

Magmas in Deep Crustal Hot Zones

C.ANNEN 1*,J.

D.BLUNDY 2AND R.S.J.SPARKS 2

1SECTION DES SCIENCES DE LA TERRE,UNIVERSITE ′DE GENE `VE,13RUE DES MARAI ?CHERS,1205GENE `VE,SWITZERLAND

2DEPARTMENT OF EARTH SCIENCES,UNIVERSITY OF BRISTOL,WILLS MEMORIAL BUILDING,BRISTOL BS81RJ,UK

RECEIVED APRIL 14,2005;ACCEPTED OCTOBER 17,2005

ADVANCE ACCESS PUBLICATION DECEMBER 7,2005A model for the generation of intermediate and silicic igneous rocks is presented,based on experimental data and numerical modelling.The model is directed at subduction-related magmatism,but has general applicability to magmas generated in other plate tectonic

settings,including continental rift zones.In the model mantle-

derived hydrous basalts emplaced as a succession of sills into the lower crust generate a deep crustal hot zone.Numerical modelling of the hot zone shows that melts are generated from two distinct

sources;partial crystallization of basalt sills to produce residual H 2O-rich melts;and partial melting of pre-existing crustal rocks.

Incubation times between the injection of the first sill and generation of residual melts from basalt crystallization are controlled by the initial geotherm,the magma input rate and the emplacement depth.After this incubation period,the melt fraction and composition of residual melts are controlled by the temperature of the crust into which the basalt is intruded.Heat and H 2O transfer from the

crystallizing basalt promote partial melting of the surrounding crust,which can include meta-sedimentary and meta-igneous basement rocks and earlier basalt intrusions.Mixing of residual and crustal partial melts leads to persity in isotope and trace element chemistry.Hot zone melts are H 2O-rich.Consequently,they have

low viscosity and density,and can readily detach from their source and ascend rapidly.In the case of adiabatic ascent the magma attains a super-liquidus state,because of the relative slopes of the adiabat and the liquidus.This leads to resorption of any entrained crystals or country rock xenoliths.Crystallization begins only when the ascending magma intersects its H 2O-saturated liquidus at

shallow depths.Decompression and degassing are the driving forces behind crystallization,which takes place at shallow depth on timescales of decades or less.Degassing and crystallization at shallow depth lead to large increases in viscosity and stalling of the magma to form volcano-feeding magma chambers and shallow plutons.It is proposed that chemical persity in arc magmas is largely acquired in the lower crust,whereas textural persity is related to shallow-level crystallization.

KEY WORDS:magma genesis;deep hot zone;residual melt;partial melt;adiabatic ascent

INTRODUCTION

A key question in igneous petrology concerns the origin of intermediate to silicic magmatic rocks,such as volu-minous Cordilleran granite batholiths (diorites,tonalites,granodiorites and granites)and the evolved volcanic rocks (andesites,dacites and rhyolites)of destructive plate margins.The continental crust has an estimated

silicic andesite to dacite composition,with a vertical stratification from mafic lower crust to more evolved granite-dominated upper crust (Rudnick &Fountain,1995).The origin of intermediate to silicic igneous rocks is,therefore,central to understanding the evolution of the

continental crust.

In subduction settings melt is generated by partial melting in the mantle wedge where primary mafic mag-mas form by some combination of addition of H 2O-rich fluids or melts released from the subducted slab (e.g.Davies &Stevenson,1992;Tatsumi &Eggins,1995;

Schmidt &Poli,1998;Ulmer,2001;Grove et al .,2002;Forneris &Holloway,2003)and mantle decompression resulting from subduction-induced corner flow (e.g.Sisson &Bronto,1998;Elkins-Tanton et al .,2001;Hasegawa &Nakajima,2004).Experimental studies of mantle melting

(e.g.Ulmer,2001;Parman &Grove,2004;Wood,2004),

*Corresponding author.Telephone:tt41223796623.Fax:tt41223793210.E-mail:Catherine.Annen@terre.unige.ch óThe Author 2005.Published by Oxford University Press.All

rights reserved.For Permissions,please e-mail:journals.permissions@0025710d561252d381eb6e31 JOURNAL OF PETROLOGY VOLUME 47NUMBER 3PAGES 505–5392006doi:10.1093/petrology/egi084 at Institute of Geology and Geophysics, CAS on March 5, 20130025710d561252d381eb6e31/Downloaded from

and observations of the petrology and geochemistry of mafic arc magmas,indicate that primary,mantle-derived magmas range in composition from basalts to magnesian andesites (Tatsumi,1982;Tatsumi &Eggins,1995;Bacon et al .,1997;Conrey et al .,1997;Carmichael,2002,2004;Grove et al .,2002).In terms of liquidus temperatures and dissolved H 2O contents there is a range from dry and hot magmas to wet and cool varieties,even within a single volcanic arc (e.g.Sisson &Layne,1993;Baker et al .,1994;Elkins-Tanton et al .,2001;Pichavant et al .,2002a ).Volcanic rocks with MgO-rich compositions that could be in equilibrium with the man-tle wedge are rare in continental arcs and only a minor component of island arcs,an observation attributable to density filtering and intracrustal ‘processing’of ascending magmas.This processing accounts for the predominance of evolved (silica-rich)volcanic rocks and granitic plu-tonic rocks in continental and mature island arcs.The generation of intermediate and silicic arc magmas is widely attributed to two main processes:differentiation of primary magmas by crystallization within the crust or uppermost mantle (e.g.Gill,1981;Grove &Kinzler,1986;Musselwhite et al .,1989;Rogers &Hawkesworth,1989;Mu ¨ntener et al .,2001;Grove et al .,2002,2003)and partial melting of older crustal rocks (e.g.Smith &Leeman,1987;Atherton &Petford,1993;Tepper et al .,1993;Rapp &Watson,1995;Petford &Atherton,1996;Chappell &White,2001;Izebekov et al .,2004).These processes can occur simultaneously with the heat and volatiles (principally H 2O)liberated from the primary magmas triggering crustal melting (Petford &Gallagher,2001;Annen &Sparks,2002).Additionally crustal rocks can be assimilated into mantle-derived magmas (DePaolo,1981).The assimilated components may be much older than,and petrogenetically unrelated to,the arc magmas and possess distinctive trace element and isotope geochemistry.Partial melting can also occur in igneous rocks,including cumulates,that have formed from earlier arc magmas;in this case the assimilated components and arc magmas may have strong geo-chemical affinities (e.g.Heath et al .,1998;Dungan &Davidson,2004).Evidence for crustal assimilation and mixing of melts and crystals from different sources is common (Grove et al .,1988,1997;Musselwhite et al .,1989;De Paolo et al .,1992).These processes are central to models of assimilation and fractional crystalliza-tion (AFC;DePaolo,1981)and mixing,assimilation,storage and hybridization (MASH;Hildreth &Moorbath,1988).A key question is at what depth chemical differentiation occurs.Although the existence of shallow sub-volcanic magma chambers is indisputable,based on geophysical evidence as well as petrological and geological observa-tions,it is less clear that such chambers are the place where most chemical differentiation takes place.To produce igneous rocks that contain more than 60wt %SiO 2by fractional crystallization,60%or more crystal-lization of a typical primitive arc basalt is required (e.g.Foden &Green,1992;Mu ¨ntener et al .,2001).The vol-ume of parental mafic magma that crystallizes is,there-fore,typically twice as much as the evolved magma produced.As large granitoid batholiths and voluminous eruptions involve hundreds to thousands of km 3of silicic magma (e.g.Smith,1979;Crisp,1984;Bachmann et al .,2002),huge volumes of associated mafic cumulates are required.However,geological and geophysical evidence for the requisite large volumes of complementary dense mafic cumulates in the shallow crust is generally lacking.One resolution to this problem is density-driven sink-ing of mafic cumulate bodies into the lower crust (Glazner,1994).Alternatively,if differentiation of basalt occurs at deep levels in the crust then the complementary dense mafic cumulates will be located in the lower crust (e.g.Debari &Coleman,1989;Mu ¨ntener et al .,2001)where they may eventually delaminate into the mantle below (Kay &Kay,1993;Jull &Keleman,2001)thereby progressively driving the bulk crust towards andesite composition.The silica-rich residual melts generated by deep-seated basalt differentiation can be extracted and ascend,either to erupt immediately or to stall to form shallow magma chambers.If unerupted,such shallow chambers consolidate to form granite plutons,with mafic igneous rocks being a minor component or absent.Recent numerical simulations of heat transfer (Annen &Sparks,2002)and high-temperature experiments (Mu ¨ntener et al .,2001;Prouteau &Scaillet,2003)suggest a model whereby silica-rich magmas can be generated by incomplete crystallization of hydrous basalt at upper mantle and/or lower crustal depths.These observations motivate our development of a model in which basalt emplacement into the lower crust leads to generation of intermediate and silicic melts (Fig.1).Our model builds upon the concept of underplating (Raia &Spera,1997),expands on models of differentiation of basalt at high pressure (Gill,1981;Grove et al .,2002)and incorporates aspects of AFC (DePaolo,1981)and MASH (Hildreth &Moorbath,1988).We develop a quantitative model in which evolved melts are generated from H 2O-rich par-ental basalts both by partial crystallization of the basalts themselves and by partial melting of surrounding crustal rocks through heat and H 2O transfer from the cooling basalts.A key feature of our model is that melt composi-tions are determined by the depth of emplacement of inpidual basalt intrusions and thermal equilibration with the local geotherm.We refer to the site of basalt injection and melt generation in the lower crust as a deep crustal hot zone.Previous models of underplating (e.g.Huppert &Sparks,1988;Bergantz,1989;Raia &Spera,1997;Petford &Gallagher 2001;Jackson et al .,2003)have concentrated almost exclusively on melt generated

506JOURNAL OF PETROLOGY VOLUME 47NUMBER 3MARCH 2006

at Institute of Geology and Geophysics, CAS on March 5, 20130025710d561252d381eb6e31/Downloaded from

by heating of the crust,with less attention paid to the residual melt generated by partial crystallization of the underplated basalt intrusions.Here we develop the con-cepts proposed by Annen &Sparks (2002)and consider the full range of possible mechanisms of melt genera-tion in the hot zone,including residual melt from basalt crystallization and partial melting of surrounding crustal rocks (Fig.1).We then consider the evolution

of

Fig.1.Conceptual representation of a hot zone (not to scale).Sills of mantle-derived basaltic magma are injected at a variety of depths,including

(1)the Moho,(2)the lower crust and (3)the Conrad Discontinuity between lower and upper crust.Sills injected at the Moho displace older sills into the mantle,creating a contrast between the petrological Moho (base of sill complex)and seismological Moho (top of sill complex).Sills crystallize from their injection temperature to that of the geotherm,resulting in a wide variety of residual melt fractions at any given time,from near 100%(newly injected sill near Moho)to 0%(old sill injected into lower crust).The fraction of crustal melt varies throughout the hot zone according to the age and proximity of the basalt sills.Melts ascend from the hot zone to shallow storage reservoirs,leaving behind dense refractory cumulates or restites.Residual and crustal melts from different portions of the hot zone may be mixed together prior to ascent or within the shallow reservoir.

507ANNEN et al.DEEP CRUSTAL HOT ZONES

at Institute of Geology and Geophysics, CAS on March 5, 20130025710d561252d381eb6e31/Downloaded from

these melts as they are extracted from their source

rocks and ascend to shallow crustal levels,degassing

and crystallizing en route.The model is developed

primarily for application to the genesis of subduction

zone volcanic and plutonic rocks,and we will refer

collectively to this whole suite of intermediate and

silicic rock types as ‘andesite’,except where a composi-

tional or textural distinction is relevant.However,our

model has general applicability to other tectonic settings,

including continental rift zones where plume-related

basaltic magmas are intruded into the base of the

continental crust.

SOURCES AND MECHANISMS FOR

INTERMEDIATE AND SILICIC

MAGMA GENERATION

There are five currently popular models for the genera-

tion of andesites (sensu lato ),as follows.

Model I.Partial melting of harzburgite in the mantle wedge,fluxed by H 2O-rich fluids or melts liberated from the subducting slab (e.g.Tatsumi,1982;Hirose,1997;Blatter &Carmichael,2001;Carmichael,2002,2004;Parman &Grove,2004).Model II.Crystallization of mantle-derived basalt or basaltic andesite in shallow crustal magma chambers (e.g.Sisson &Grove,1993;Grove et al .,1997;Pichavant et al .,2002b ).Model III.Crystallization of mantle-derived basalt or basaltic andesite in the deep arc crust at or close to the Moho (e.g.Mu ¨ntener et al .,2001;Annen &Sparks,2002;Mortazavi &Sparks,2003;Prouteau &Scaillet,2003).

Model IV.Dehydration partial melting of meta-basalts

(amphibolites)in the lower or middle crust by intrusions of hot,mantle-derived magma (e.g.Smith &Leeman,1987;Petford &Atherton,1996;Jackson et al .,2003).Model V.Mixing between silicic magmas and mantle-derived mafic magmas (e.g.Heiken and Eichelberger,1980).In some cases the silicic component is generated by partial melting of crustal rocks (e.g.Druitt et al .,1999).In this paper we focus on Models III–V,which take place in the middle or lower crust.Models I and II are briefly considered first.Generation of andesite by mantle melting (Model I)has been demonstrated experimentally (Tatsumi,1982;Hirose,1997;Grove et al .,2002,2003;Parman &Grove,2004)and calculated thermodynamic-ally (Carmichael,2002,2004).The andesites produced in this way have elevated MgO contents and high mg-numbers,a requirement for equilibrium with the Mg-rich olivines of mantle harzburgite.Boninite series magmas are widely thought to originate by H 2O-fluxed melting of harzburgite (Falloon &Danyushevsky,2000;Parman &Grove,2004),whereas the generation of ‘high-Mg andesites’may involve reactions between ascending slab-derived silicic melts and mantle peridotite (Yogodzinsky &Kelemen,1998).However,high-Mg andesites and boninites are not the dominant rock types of volcanic arcs;typical arc andesites,with low mg-numbers,could not have been in direct equilibrium with mantle rocks.

Model II is widely favoured.Basalt and basaltic andesite lavas occur at many arc stratovolcanoes and occasionally contain xenoliths of cumulate origin (e.g.Arculus &Wills,1980).Several experimental studies demonstrate that andesite can be generated by fractional crystallization of H 2O-saturated basalts and basaltic andesites at p H 2O ?P tot of 200–400MPa and temperat-ures of 950–1050 C (Sisson &Grove,1993;Grove et al .,1999,2003;Pichavant et al .,2002b )by crystallizing an assemblage of plagioclase (An 60–90)tclinopyroxene tamphibole toxides ?orthopyroxene ?olivine.One constraint on the origin of andesites is that they typically contain <19%Al 2O 3(Fig.2),indicating that by the time residual melts have attained >57wt %SiO 2they have become saturated in an aluminous phase.In Model

II

0025710d561252d381eb6e31positions of experimentally produced residual melts from

crystallization of hydrous basalts in the lower crust.Squares denote

melt compositions from experiments on a primitive Mount Shasta basaltic andesite,sample 85-44(mg-number 0á71),from Mu ¨ntener et al .(2001)and Grove et al .(2003),at 0á8–1á2GPa,1045–1230 C

and with !2á5wt %added H 2O;filled circles denote experimental

melts from Kawamoto (1996)on a Higushi-Izu high alumina basalt,sample IZ27-2(mg-number 0á60),at 1á0GPa,1000–1150 C with 1wt %added H 2O.The compositions of the two different starting materials are indicated.Symbols that are filled or partially filled denote

glasses in equilibrium with an aluminous phase,as shown in the

legend.All of the IZ27-2glasses are saturated in plagioclase.For reference the compositional field defined by 387published analyses of Cascades andesites is shown.It should be noted that >96%of these andesites contain <19wt %Al 2O 3.

508JOURNAL OF PETROLOGY VOLUME 47NUMBER 3MARCH 2006

at Institute of Geology and Geophysics, CAS on March 5, 20130025710d561252d381eb6e31/Downloaded from

crystallization of plagioclase serves to limit Al 2O 3enrich-ment in residual melts.The lack of abundant dense com-plementary mafic to ultramafic cumulate rocks in the shallow crust is problematic for Model II unless the asso-ciated mafic cumulates are removed by sinking (Glazner,1994).Model III involves fractional crystallization of similar parental magmas to Model II,but at higher pressure,thereby obviating the problem of the missing mid-or upper-crustal mafic cumulates.Mantle-derived magmas intruded into the deep crust cool and crystallize produ-cing evolved residual melts.The principal difference between high-and low-pressure crystallization of hydrous basalt lies in the nature of the crystallizing assemblage.At higher p H 2O garnet (e.g.Wolf &Wyllie,1994;Rapp,1995)and aluminous amphibole (Grove et al .,2003)are stabilized and can contribute to minimizing Al 2O 3enrichment in residual melts.Conversely,plagioclase stability is reduced and liquidus plagioclase is anorthite-rich,a common finding in arc-related cumulate nodules (e.g.Arculus &Wills,1980).In terms of melt chemistry,it is very hard to distinguish between residual melts pro-duced by crystallization of An -rich plagioclase and pyrox-enes from H 2O-undersaturated basalt at $1á0GPa (Kawamoto,1996)and those produced from H 2O-saturated basalt at 0á2–0á4GPa (e.g.Sisson &Grove,1993;Pichavant et al .,2002b ).The appearance of garnet as the liquidus aluminous phase in andesite and dacite melts at pressures over $1á1GPa (Wolf &Wyllie,1994;Rapp,1995)imparts a distinctive trace element chemistry to residual melts (e.g.high Sr/Y),which provides a clear indication of high-pressure differentiation (e.g.Smith &Leeman,1987;Feeley &Davidson,1994;Feeley &Hacker,1995).In Models II and III,Al 2O 3enrichment in derivative melts is further minimized if the primitive basalt itself has relatively low Al 2O 3.Circumstances for generation of such magmas are inferred in many arcs with a relatively depleted mantle wedge (Grove et al .,2003;Parman &Grove,2004).For example,primitive arc basalts with only 14–15%Al 2O 3have been described for Klyuchevskoy volcano,Kamchatka (Ozerov,2000).When mafic magmas are intruded into the arc crust they transfer heat and volatiles (principally H 2O)into the surrounding crust,which can lead to partial melting of the wall-rocks.The deep crustal hot zone is,therefore,envisaged as a mixture of partially crystallized basalt,partially molten crustal rocks and H 2O liberated from the solidifying basalts (Fig.1).Geophysical evidence is consistent with these concepts.In the Cascades,for example,the release of significant volumes of H 2O from deeply intruded basalts may account for the presence of a highly electrically conductive layer at 10–30km depth (Stanley et al .,1990),and in the central Andes a broad conductive zone (Brasse et al .,2002)is associated with a low-velocity zone at depths of 20–40km (Yuan et al .,2000),interpreted as a laterally extensive region of partial melt,capped by a silicic magma body $1km thick (Chmielowski et al .,1999).Below volcanoes in the Japan arc broadband seismometers have recorded low-frequency tremors and micro-earthquakes at 30–50km depth (Obara,2002;Katsumata &Kamaya,2003).These can be explained by deformation associated with magma intrusions (S.Sachs,personal communication,2003)and their low frequency is consistent with the pres-ence of a fluid phase.Finally,beneath central North Island,New Zealand,a seismically highly reflective layer at 35km depth,interpreted as a body of partially molten rock (Stratford &Stern,2004),suggests that be-neath some arcs the hot zone may be located in the uppermost mantle,rather than within the crust,which is only 16km thick in this region.

The partially molten crust surrounding the basalt may be older intrusions of related mantle-derived hydrous basalt (or amphibolite)or unrelated metamorphic arc crust.This is Model IV.The volume and composition of the partial melt produced depends on the intrusion rate (heat flux)of the mantle-derived basalts,the prevailing geotherm and the extent to which the melting region is fluxed by H 2O liberated from the crystallizing basalt.Chemically hybrid melts can be formed if the residual melts from basalt crystallization are mixed with crustal partial melts during extraction,ascent and shallow intru-sion;this is Model V.

Models III and IV both involve partially molten hyd-rous basaltic rocks in the lower crust produced,res-pectively,by crystallization and melting.Deep-seated crystallization of hydrous basaltic magmas differs from dehydration melting of the lower crust,as modelled by Raia &Spera (1997),Petford &Gallagher (2001)and Jackson et al .(2003),in one fundamental regard,the availability of H 2O.In dehydration melting the H 2O content of the source rock is strictly limited by the amount of H 2O that can be structurally bound in hyd-rous minerals such as amphibole and mica.For a mafic amphibolite with 40%amphibole,this amounts to $0á8wt %H 2O.Greater quantities of H 2O can be involved only if the heat source efficiently fluxes the source region with H 2O.Although this is likely,no extant models of crustal melting consider this process,largely because it is uncertain whether H 2O passing through a low-porosity source rock triggers melting or is simply carried away along fractures.By contrast,deep-seated crystallization of hydrous arc basalt magmas has no such upper limit on H 2O content.Studies of melt inclusions in primitive arc magmas,together with high-pressure experiments,indicate dissolved H 2O con-tents from almost zero to 10wt %(e.g.Sisson &Layne,1993;Carmichael,2002;Pichavant et al .,2002a ;Grove

509ANNEN et al.DEEP CRUSTAL HOT ZONES

at Institute of Geology and Geophysics, CAS on March 5, 20130025710d561252d381eb6e31/Downloaded from

et al .,2003).The wide range of H 2O contents and bulk compositions of parental arc basalts ensures that crystal-lization of hydrous basalt can generate a wide persity of residual melt compositions,as demonstrated experiment-ally by Sisson et al .(2005).Dehydration melting (Model IV)requires a heat source.In arcs the widespread association of evolved igneous rocks with mantle-derived basalt strongly sug-gests that mafic magmas provide the heat source (Hildreth,1981).However,herein lies a problem:models of heat transfer show that arc basalts emplaced into the base of the crust at temperatures of 1100–1240 C (see Ulmer,2001;Pichavant et al .,2002a )cannot provide enough heat to melt amphibolite lower crust extensively (Petford &Gallagher 2001;Annen &Sparks,2002),because of the high dehydration melting temperature of amphiboles in mafic rocks ($950 C).More fertile upper crustal pelitic protoliths can be melted more effi-ciently,but large amounts of basalt are still needed as a heat source (Annen &Sparks,2002).In addition,silicic rocks in arcs are typically calc-alkaline and metaluminous,which places limits on the amount of pelite that can be melted.The isotopic and geochemical signatures of evolved plutonic and volcanic arc rocks clearly indicate contribution from pelitic crust in some cases (DePaolo et al .,1992),but significant amounts of basalt or meta-basalt (amphibolite)must be involved in their petrogenesis.The problem in arcs is how to generate large volumes of metaluminous,calc-alkaline evolved melts when the proposed amphibolite source is too refractory to undergo significant dehydration melting at plausible temperatures.This paradox can be solved if crystallization of H 2O-bearing mantle-derived basalt is the principal source of the evolved melts.CRYSTALLIZATION OF ANDESITE IN THE SHALLOW CRUST Once generated in the deep crust andesite and dacite residual melts can detach and ascend into the shallow crust.Subduction-related andesites and dacites are com-monly porphyritic,with phenocrysts of plagioclase plus various proportions of hornblende,clinopyroxene,orthopyroxene,biotite and oxides;the exact ferromagne-sian assemblage depends on magma composition,partial pressure of volatiles (especially p H 2O),oxygen fugacity (f O 2)and temperature (e.g.Rutherford et al .,1985;Rutherford &Devine,1988;Blatter &Carmichael,1998,2001;Moore &Carmichael,1998;Scaillet &Evans,1999;Pichavant et al .,2002b ;Izebekov et al .,2004).Invariably the groundmass or matrix glass in porphyritic andesites and dacites is rhyolitic in composition.The phenocryst assemblages commonly have complex textures and zoning patterns,which indicate that magmatic evolution can involve processes such as:repeated mixing of different batches of magma (e.g.Heiken &Eichelberger,1980;Clynne,1999);entrain-ment of old crystals from previously consolidated magma batches (Davidson et al .,1998,2001,2005;Heath et al .,1998;Cooper &Reid,2003;Reagan et al .,2003;Dungan &Davidson,2004)or from assimilation of crustal rocks (Ferrara et al .,1989);convective stirring (Couch et al .,2001);crystal growth induced by degassing (Blundy &Cashman,2001).Whereas some of these phenocrysts grew from the magma in which they are found,others are entrained xenocrysts from earlier magma pulses or from chemically unrelated wall-rocks (e.g.Izebekov et al .,2004;Davidson et al .,2005).Detailed studies of volcano evolution (e.g.Bacon,1983;Bacon &Druitt,1988;Druitt &Bacon,1989;Harford et al .,2002)and con-straints on timescales for crystallization (e.g.Zellmer et al .,1999,2003a ,2003b ;Harford &Sparks,2001)sug-gest that these various processes are the consequence of amalgamation of shallow magma bodies in the upper crust through many episodes of magma ascent from greater depths,sometimes accompanied by eruption.Field and geochronological evidence from calc-alkaline plutonic rocks (‘granites’,sensu lato )also supports their formation by amalgamation of many small intrusions,often of magmas with very similar bulk chemical com-position but subtle textural differences (e.g.John &Blundy,1993)or radiometric ages (e.g.Coleman et al .,2004;Glazner et al .,2004).

Our main concern here is to establish under what conditions the common phenocryst assemblages in andesites and granites are formed.Central to this issue are the H 2O contents and temperatures of andesite magmas.The importance of these two variables in inter-preting the phenocryst assemblages and compositions of

andesites has been investigated for over 30years in a

large number of experimental studies at p H 2O ( P tot )of 0á1to !400MPa (Eggler,1972;Green,1972;Eggler &

Burnham,1973;Maksimov et al .,1978;Sekine et al .,

1979;Rutherford et al .,1985;Rutherford &Devine,1988,2003;Luhr,1990;Foden &Green,1992;Sekine &Aramaki,1992;Sisson &Grove,1993;Kawamoto,1996;Grove et al .,1997,2003;Barclay et al .,1998;Blatter &Carmichael,1998,2001;Moore &Carmichael,1998;Cottrell et al .,1999;Martel et al .,1999;Sato et al .,1999;Scaillet &Evans,1999;Pichavant et al .,2002b ;Couch et al .,2003;Prouteau &Scaillet,2003;Barclay &Carmichael,2004;Costa et al .,2004;Izebekov et al .,2004).Although many of these studies are focused on rocks from a specific volcano,some general conclusions can be drawn regarding sub-duction-related andesites and dacites,as follows.(1)Eruption temperatures,as determined by geo-thermometry,are consistently less than low-pressure (<300MPa)andesite liquidus temperatures even under

510JOURNAL OF PETROLOGY VOLUME 47NUMBER 3MARCH 2006

at Institute of Geology and Geophysics, CAS on March 5, 20130025710d561252d381eb6e31/Downloaded from

H 2O-saturated conditions.In many cases the difference is several tens of degrees and can be as much as 200 C (e.g.Blatter &Carmichael,1998;Barclay &Carmichael,2004).(2)The liquidus phases at low p H 2O often include minerals (e.g.olivine,clinopyroxene)that are absent from the phenocryst assemblage in the natural rocks (e.g.Blatter &Carmichael,1998,2001;Scaillet &Evans,1999;Costa et al .,2004).(3)Although amphibole is a common phenocryst it rarely occurs on the andesite liquidus at p H 2O <400MPa even under oxidizing conditions;where it is stable,amphibole typically appears 100 C below the liquidus (e.g.Rutherford &Devine,1988,2003;Blatter &Carmichael,1998,2001;Moore &Carmichael,1998;Martel et al .,1999;Costa et al .,2004;Izebekov et al .,2004).(4)Plagioclase is stabilized only at low p H 2O and is rarely a true liquidus phase at p H 2O >100–200MPa,even though plagioclase is a ubiquitous phenocryst phase in most andesites (e.g.Eggler,1972;Maksimov et al .,1978;Sekine et al .,1979;Sekine &Aramaki,1992;Blatter &Carmichael,1998,2001;Moore &Carmichael,1998;Martel et al .,1999;Grove et al .,2003).(5)The anorthite (An )content of plagioclase increases with increasing p H 2O (at constant temperature)and increasing temperature (at constant p H 2O).For a given andesite,plagioclase phenocryst rims typically have considerably lower An contents (by up to 30mol %)than the experimentally determined liquidus or near-liquidus plagioclase (e.g.Rutherford et al .,1985;Scaillet &Evans,1999;Rutherford &Devine,2003;Costa et al .,2004).(6)The observed phenocryst assemblage,phase com-positions and crystallinity typically are consistent with H 2O-saturated conditions at pressures of 100–300MPa and at sub-liquidus temperatures consistent with those obtained from mineral thermometry on the natural rocks (e.g.Blatter &Carmichael,1998;Moore &Carmichael,1998;Martel et al .,1999;Costa et al .,2004).The similarity of phase proportions and compositions in both experiments and natural andesites (Fig.3)indic-ates that,to a first approximation,these magmas have undergone near-closed system crystallization from an initial fully molten state to a porphyritic magma under conditions of low-pressure H 2O-saturation.However,very few of the studied andesites contain their full com-plement of experimentally determined liquidus phases under these conditions,suggesting that either magma temperatures were never high enough to form a fully molten andesite liquid at low pressure or that the original liquidus phases were completely eliminated (or re-equilibrated)by reaction with the melt.The interpreta-tion we favour is that andesite liquids,once formed and extracted from the deep crust,typically crystallize under polybaric conditions,at temperatures that do not signi-ficantly exceed their eruption temperature.Thus the ini-tial fully molten state of an andesite is not a consequence of high temperature,but a consequence of high p H 2O.

All of the above observations [(1)–(6)]are consistent with this interpretation,as are the observed zoning patterns and rim compositions of plagioclase phenocrysts (Fig.3).For example,the phenocryst assemblage and proportions of the Colima andesite (Fig.3a)can be reproduced closely

at 950–960 C (consistent with mineral thermometry on the natural lava)and p H 2O from 70to 150MPa (Moore &Carmichael,1998).The very calcic cores of some plagioclase phenocrysts (An

85)were ascribed by Moore

&Carmichael (1998)to the onset of crystallization at

even higher p H 2O but at essentially the same 0025710d561252d381eb6e31ing analyses of phenocryst-hosted melt inclusions,Blundy &Cashman (2005)advanced a similar argument

for the silicic andesites of Mount St.Helens.They pro-posed that the observed phenocryst assemblage of the white pumice of 18May 1980crystallized in response to decompression from 233to 140MPa at a near-constant temperature of $900 C,whereas the sub-sequent microlite-bearing dome lavas continued to crystallize down to pressures as low as 9MPa with negli-

gible cooling.Another example is the Soufrie `re Hills andesite,Montserrat,where An 50–60plagioclase inclu-sions in the cores of amphibole phenocrysts (Higgins &Roberge,2003),combined with experimental data (Couch et al .,2003;Rutherford &Devine,2003),indicate protracted polybaric crystallization at temperatures suffi-ciently low to stabilize amphibole (840–880 C;Murphy et al .,2000;Devine et al .,2003;Rutherford &Devine,2003).Major element chemistry of whole-rocks,pheno-

crysts and groundmass glass (Murphy et al .,2000;Harford et al .,2002)is consistent with crystallization of predominantly amphibole and plagioclase from a liquid whose initial andesite composition evolved to rhyolite as crystallization proceeded.

All of the above examples suggest that decompression crystallization can play a major role in determining the crystallization sequence,assemblage and proportions.That is not to say that cooling is not important in some circumstances,nor that reheating caused by magma mix-ing does not occur:there is compelling evidence for both processes in many andesite magmas.

An attractive attribute of polybaric,decompression-driven crystallization is that it can be very rapid in com-parison with the slow rates of crystallization expected for cooling-driven crystallization caused by heat loss from shallow magma chambers.For example,consider the case of H 2O-saturated Colima andesite.To generate the observed phenocryst proportions by isobaric cooling alone would require a temperature drop of some 125 C at p H 2O ?70MPa (Moore &Carmichael,1998).To attain the same crystallinity by isothermal decompression

511ANNEN et al.DEEP CRUSTAL HOT ZONES

at Institute of Geology and Geophysics, CAS on March 5, 20130025710d561252d381eb6e31/Downloaded from

(at 960 C)would require a pressure drop of 60MPa,equivalent to an ascent of $2km.A pressure drop can be achieved much more rapidly than a temperature drop,as follows.Cooling of shallow magma chambers is controlled by conduction through the wall-rocks and convection within the magma body and the superjacent hydrothermal system (Carrigan,1988).The cooling timescale is con-trolled by the magma chamber size and the vigour of hydrothermal convection.The world’s most active geo-thermal systems associated with large silicic magma chambers have convective thermal fluxes of several W/m 2(Carrigan,1988).Assuming that the magma chamber convects internally,then the heat loss from the

chamber

0025710d561252d381eb6e31parison of experimental and natural whole-rock phase proportions (weight percent)for selected andesite compositions.(a)Volca ′n Colima,Mexico (Moore &Carmichael,1998);(b)Mont Pele ′e,Martinique (Martel et al .,1999);(c)Mount Pinatubo,Philippines (Scaillet &Evans,1999;B.Scaillet,personal communication,2004);(d)Valle de Bravo,Mexico (Blatter &Carmichael,2001).All experiments are H 2O-saturated at the pressure and temperature shown.Only experiments in which the temperature is close to that inferred from mineral thermometry of the whole-rock are shown.Also shown is the molar anorthite (An )content of plagioclase.gl,glass;plag,plagioclase;amph,amphibole;opx,orthopyroxene;cpx,clinopyroxene;ox,oxide.

512JOURNAL OF PETROLOGY VOLUME 47NUMBER 3MARCH 2006

at Institute of Geology and Geophysics, CAS on March 5, 20130025710d561252d381eb6e31/Downloaded from

can be converted into the time required to cool the chamber to a given temperature by a heat balance calculation.For example,for a cylindrical chamber with 1km radius and 1km depth (a volume of $3km 3),the time to cool the magma internally by $100 C is calculated at 8400years for a heat flow of 2á2W/m 2and a heat loss of 220kJ/kg assuming 20%crystalliza-tion,a latent heat of 419kJ/kg and heat capacity of 1361kJ/kg per 0025710d561252d381eb6e31rger chambers or lower hydro-

thermal heat fluxes would increase crystallization times

significantly.

In contrast,magma ascent into the shallow crust is envisaged to occur in dykes (Petford et al .,1993)at speeds of cm/s to dm/s.The time taken for an H 2O-saturated andesite melt to ascend 2km would be a matter of hours (Lister &Kerr,1991;Petford et al .,1993),thereby gener-ating a significant undercooling caused by gas exsolution,leading to rapid nucleation and growth of crystals.Rapid crystallization of phenocrysts in arc magmas is consistent with U-series data (Reagan et al .,2005)and diffusion dating studies of phenocrysts (Zellmer et al .,1999,2003b ;Costa et al .,2004),which suggest crystallization on time-scales that are far more rapid than would be expected for crystallization driven by cooling alone.Rapid crystalliza-tion also provides an effective means of generating the near-closed system crystallization inferred from experi-mental studies,because the timescales are too short to permit significant crystal–melt segregation,for example by crystal settling.The physical consequences of decom-pression crystallization are discussed further in a later section.Whatever the cause of crystallization,the experimental data present a compelling argument that the chemical composition of andesites is determined at depth,prior to magma emplacement in the shallow crust.Of course,this concept does not exclude subsequent processing of andesite magmas in shallow chambers,including magma mixing and more advanced fractional crystallization.For example,at Santorini,Greece,dacites and rhyolites can be demonstrably related to andesite by low-pressure frac-tional crystallization of orthopyroxene–clinopyroxene–plagioclase–oxide assemblages (Nicholls,1971;Druitt et al .,1999),whereas at Crater Lake,USA,rhyolite magma accumulated prior to the climactic eruption of Mount Mazama by repeated injection of andesite mag-mas into a shallow chamber,and extraction of residual rhyolitic melts by filter pressing (Sisson &Bacon,1999).Partially solidified andesitic bodies,or ‘proto-plutons’,with >50%crystals can also be remobilized by sub-sequent pulses of hot magma from below,as envisaged at Soufrie `re Hills (Couch et al .,2003)and Fish Canyon Tuff,USA (Bachmann &Dungan,2002).There are also examples of zoned plutons,such as Boggy Plain,Australia (Wyborn et al .,2001)where in situ fractionation from andesite to more evolved magmas has occurred.Additionally,such magma bodies are likely to develop incrementally over long periods of time so that mixing occurs between rising batches of andesite from depth (Fig.1).The key concept is that the starting point for shallow chamber processes (e.g.further fractionation,wall-rock assimilation,magma mixing,magma recharge,repeated remobilization,etc.)is andesite,itself generated at greater depths.

EVIDENCE FOR HIGH H 2O

CONTENTS IN ARC MAGMAS

Observations (Anderson,1979;Murphy et al .,2000;Cervantes &Wallace,2003)and experimental studies (e.g.Sisson &Grove,1993;Pichavant et al .,2002a ;Barclay &Carmichael,2004)indicate that many arc basalts have H 2O contents in the range 2–6wt %.Evolved residual melt obtained by crystallization of such basalts will be even more H 2O-rich provided that the pressure is high enough for H 2O to remain in solution.For example,60%crystallization of basalts with 2–6wt %H 2O can generate intermediate to silicic melts with H 2O contents of 5–15wt %.(The figure is only slightly less if amphibole or mica are crystallizing phases.)Estim-ates of H 2O contents in calc-alkaline intermediate and silicic magmas commonly yield values of 4–6wt %(Anderson,1979;Green,1982;Barclay et al .,1998;Devine et al .,1998;Carmichael,2002,2004;Blundy &Cashman,2005),although andesite melt inclusions with up to 10%H 2O have been reported (Anderson,1979;Grove et al .,2003).These estimates are principally based on comparison of natural phenocryst assemblages with experimental products and/or melt inclusion studies.Both approaches provide good estimates of pre-eruption H 2O contents during the later stages of magma crystal-lization,but do not necessarily constrain H 2O contents at earlier stages of magma genesis.For example,Carmichael (2002,2004)inferred from experimental phase equilibria and thermodynamic calculations that andesites erupted in west–central Mexico crystallized by decompression from a melt with an original H 2O content of at least 6wt %,and possibly as much as 16wt %,almost all of which was lost during magma ascent and eruption.

Additional experimental evidence for elevated H 2O contents in arc magmas comes from the presence of aluminous amphibole phenocrysts in andesites.At Mount Shasta,USA,Grove et al .(2003)showed that pargasitic amphibole (9–12wt %Al 2O 3)overgrowth rims on magnesian olivine and pyroxenes are con-sistent with amphiboles produced experimentally from H 2O-saturated magnesian basalt at 800MPa.At this pressure the dissolved H 2O content of the melts is estim-ated at $14wt %.At Mount Pinatubo,Philippines,

513ANNEN et al.DEEP CRUSTAL HOT ZONES

at Institute of Geology and Geophysics, CAS on March 5, 20130025710d561252d381eb6e31/Downloaded from

Prouteau &Scaillet (2003)observed aluminous cores (>11wt %Al 2O 3)to some amphibole phenocrysts in the 1991dacite.Amphiboles of similar composition were produced in H 2O-undersaturated experiments on the same dacite at pressures of 960MPa,under which conditions melt H 2O contents exceed 10wt %.Prouteau &Scaillet (2003)attributed the aluminous amphibole cores to generation of the 1991dacite by crystallization of a basaltic parent melt near the base of the arc crust.The lower Al 2O 3amphibole rims correspond to later crystallization at $200MPa in the sub-volcanic magma chamber.

In summary,the available petrological and experi-mental data are consistent with the derivation

of

Fig.4.Schematic representation of the evolution of the modelled hot zone.Basalt sills are emplaced at a fixed depth (a,b)or at random throughout the lower crust (c,d).The system is shown at the onset of intrusion (a,c)and after the emplacement of a series of sills (b,d).Each new sill volume is accommodated by downward displacement of the crust,previous sills and mantle below the injection level.Temperature is represented by the dashed line.The initial temperature is determined by a geothermal gradient of 20 C/km.The temperature of inpidual sills evolves with time and with their evolving position along the geotherm.The temperatures at the Earth’s surface and at 60km depth are fixed.

514JOURNAL OF PETROLOGY VOLUME 47NUMBER 3MARCH 2006

at Institute of Geology and Geophysics, CAS on March 5, 20130025710d561252d381eb6e31/Downloaded from

H 2O-rich andesites

by crystallization of hydrous,mantle wedge-derived basalt in a lower crustal hot zone.MODELLING DEEP CRUSTAL HOT ZONES I—METHODS We address the thermal development of a crustal hot zone with specific attention paid to all potential sites of melt generation,including partially crystallized basalt and partially melted crust.In our model a hot zone develops by injection of numerous discrete basaltic sills at the Moho or within the crust (Fig.4).The thermal evolution of the hot zone,as a result of heat transfer between successive basaltic intrusions and country rock,is computed using the heat balance equation r C p @T @t t@X @t r L ?k @2T

@x 2e1Twhere r is density,C p is specific heat capacity,T is temperature,t is time,X is melt fraction,L is latent heat of fusion,k is thermal conductivity and x is (vertical)distance.The system is discretized into a one-dimensional array of cells,and equation (1)is solved by forward finite difference and iterative methods.The code was written with Delphi 4óin Object Pascal language.The resolution of the finite difference cells is 25m.Between the liquidus and solidus the finite differ-ence equivalent to equation (1)is solved by iterative approximation:

r c T p t1i àT P

i D t tL X p t1i àX p

i

D t

?k T p i à1àT p i D x 2tk T p i t1àT P

i D x 2

e2T

where D t is the time step between the time p and the time

p t1.Cells i –1and i t1are below and above cell i ,

respectively.D t is limited by the cell dimension,D x ,and

rock diffusivity to <8á5years.Above the liquidus or

below the solidus,the latent heat is zero and the

temperature of cell i at time p t1is

T p t1i ?

D t r c k T p i à1àT p i D x tk T p i t

1àT P

i D x !

tT P i e3T

The values of the parameters used in the model are given in Table 1.For sills with a horizontal dimension of 20km or more,the neglected lateral heat loss at the boundary

of the system does not significantly affect the outcome for

timescales of <4Myr (Annen &Sparks,2002).The

basalt emplacement rate,the fertility of the crust,the

temperature of the injected basalt,and the sill injection

level all control the thermal evolution of the system,and the amount and composition of the melt generated

(Annen &Sparks,2002).The model is entirely conduct-

ive and static.The issue of melt segregation process is discussed below,but it is instructive to consider the static

case first.

Temperature and melt fraction

Application of equation (2)requires knowledge of the variation of melt fraction X with temperature T .The liquidus temperature (T L ;X ?1)of a basalt varies with dissolved H 2O and MgO contents (Ulmer,2001;Wood,2004).The relationship between X and T was paramet-erized using experimental data.There is no single set of experiments on a primitive basalt with fixed H 2O content against which to calibrate X –T relationships from solidus

to liquidus.Instead we have spliced together two datasets,

one at low temperature and one at high temperature,for two different basalt compositions obtained at slightly dif-

ferent pressures (Fig.5).For high temperatures (X >0á5)

we have used the 1á2GPa experimental data of Mu ¨ntener et al .(2001)for a Cascades basaltic andesite (sample 85-44;10á8wt %MgO,mg-number 0á71)with initial H 2O contents of 5,3á8and 2á5wt %(Fig.5a),run at f O 2close to QFM (the quartz–fayalite–magnetite buffer).For lower temperatures (X <0á4)we used the 0á7GPa experi-ments of Sisson et al .(2005)on Cascades basalt (87S35A;6á5wt %MgO,mg-number 0á54)with 2á3wt %H 2O (Fig.5b).To match the f O 2of the two datasets we have only used those experiments of Sisson et al .(2005)that are within 0á5log units of QFM.

Table 1:Parameters used in the model r Density (kg/m 3)Injected basalt 2830Lower crust 3050Upper crust 2650C p Specific heat capacity (J/kg)Injected basalt 1480Lower crust 1390

Upper crust 1370L Specific latent heat (J/kg per K)Injected basalt 4.0·105Lower crust 3.5·105Upper crust 2.7·105k 0Thermal conductivity at surface Injected basalt 2.6temperature and pressure Lower crust 2.6(J/s per m per K)Upper crust 3.0

k Thermal conductivity (J/s per m per K)k ?k 0(1t1.5·10à3z )/(1t1.0·10à4T )Sources:r ,Holbrook et al .(1992)and Kay et al .(1992);C p

and L ,Bohrson &Spera (2001);k 0

and k ,

Chapman &

Furlong (1992).In the expression for thermal conductivity,z is the depth in kilometres,and T is the temperature in Kelvin.515ANNEN et al.DEEP CRUSTAL HOT ZONES

at Institute of Geology and Geophysics, CAS on March 5, 20130025710d561252d381eb6e31/Downloaded from

For the modelled basalt we assumed linear X –T rela-

tionships between T L and the onset of amphibole crystal-lization,T a ,and between T a and the solidus,T s :

X ?1T >T L e4a T

X ?3á25·10à3eT àT L Tt1T L T T a e4b T

X ?X a

T a àT s eT àT s TT a T T s

e4c TX ?0T

T L was extrapolated from the 3á8wt %H 2O experiment of Mu ¨ntener et al .(2001)to 1261 C (Fig.5a).Following Wood (2004)T L varies with H 2O according to the relationship

D T L ?80ewt %H 2O T0á4e5T

where D T L is the liquidus depression relative to an anhydrous basalt of the same composition.T L was calculated with equation (5)to be 1225,1285and 1302 C for total H 2O contents of 5,2á5and 1á5wt %,respectively (Fig.5a)in good agreement with the variation in T L of the experiments of Mu ¨ntener et al .(2001).The solidus temperature (T s ;X ?0)of basalt also depends on H 2O content.However,for any basalt that contains more H 2O than can be accommodated in sub-solidus hydrous phases,principally amphibole for arc basalts,the appropriate solidus is that for H 2O saturation.Amphibole contains $2wt %H 2O and 40wt %amphibole is an approximate upper limit for a basaltic composition amphibolite.Thus,for basalts with !0á8wt %we have adopted the experimentally determ-ined H 2O-saturated solidus of 720 C (Liu et al .,1996).The change of slope in the linear function at temper-atures below the appearance of amphibole T a [Fig.5b;equation (4c)]is consistent with experimental data (e.g.

Foden &Green,1992;Kawamoto,1999;Sisson et al .,2005).In equation (4c),X a is the melt fraction at T a ,calculated using equation (4b).We chose T a ?1075 C based on the data of Mu ¨ntener et al .(2001)data—the 1GPa data of Foden &Green (1992)on a slightly more evolved arc basalt indicate a similar T a of 1040 C.The values of T L and T a are for a pressure of 1á2GPa

and are modified by 90and 120 C/GPa,respectively (Foden &Green,1992).We have not included here the effect of f O 2on melt fractions and compositions,al-

though we recognize that f O 2can play a key role in controlling the stability of an oxide phase and the con-sequent SiO 2enrichment at a given X (Osborn,1957;Sisson et al .,2005).Of course,at other subduction

zones,

(b)

Fig.5.Modelled melt fraction (X )vs temperature (T )curves for basalts.(a)At 1á2GPa with,from left to right,initial H 2O contents of 5,3á8,2á5and 1á5%.Symbols show experimental data from Mu ¨ntener et al .(2001).The liquidus temperature (T L )of basalt with 3á8wt %initial H 2O is estimated by extrapolation.T

L for basalts with 1á5,2á5and

5wt %initial H 2O is calculated with equation (5).The curves have a slope of 0á325 C à1between T L and T a .They kink at T a to fall linearly to T s ,the H 2O-saturated basalt solidus.(b)At 1GPa,the modelled basalt curve with 2á5wt %initial H 2O shows a good fit to the experi-

mental data points of Sisson et al .(2005).The dashed curve is the X –T variation for amphibolite lower crust (after Petford &Gallagher,2001).At low temperature the basalt produces more melt than the amphibol-ite because H 2O concentrates in the residual melt.At higher temper-atures the higher fertility of the amphibolite is attributed to its slightly more differentiated composition (see Petford &Gallagher,2001).Dashed lines in (a)pide fields in which the residual melt composition is,broadly speaking,basaltic andesite,andesite and dacite.516JOURNAL OF PETROLOGY VOLUME 47NUMBER 3MARCH 2006

at Institute of Geology and Geophysics, CAS on March 5, 20130025710d561252d381eb6e31/Downloaded from

with different primary magma compositions,a different X –T parameterization may be more useful.However,at present there are insufficient experimental data available to make this worth while.As X decreases so the SiO 2content of the residual melts increases and MgO content decreases.An accurate para-meterization of SiO 2content would require additional experimental data as well as consideration of f O 2and is beyond the scope of this paper.None the less we are able to sketch indicative contours of melt composition onto Fig.5a as an approximate guide to how melt composition varies during basalt crystallization.Andesitic residual melts are generated at 0á35

X ?1T >T L e6a T

X ?6á36·10à3T à5á17T L T 900 C e6b TX ?1á75·10à3T à1á017900 C T T s e6c T

X ?0T

The solidus temperature,T s ,is taken to be 812 C.The model lower crust is an amphibolite for which we have adopted the experimentally constrained T –X rela-tionship of Petford &Gallagher (2001).It should be noted that their model amphibolite has a more fertile bulk composition than the amphibolite modelled by Annen &Sparks (2002).Consequently,at a given temperature the amphibolite generates more melt than the hydrous man-tle basalt,except close to the solidus where basalt melt

fractions are higher (Fig.5b).Melt generated by dehyd-ration melting of pelite or greywacke is peraluminous in composition (Montel &Vielzeuf,1997)and differs signi-ficantly from that generated by basalt crystallization or by

amphibolite dehydration melting.

Basalt emplacement rate We use an emplacement rate of 5mm/year,correspond-ing to an addition rate of one 50m basalt sill every 104years.This value is representative of typical estimates of magma productivity in arcs (Crisp,1984).We also tested intrusion rates of one 50m basalt sill every 5·103and every 25·103years,corresponding to average intrusion rate of 10and 2mm/year,respectively.High intrusion rates result in fast rates of melt accumulation and high melt fractions,whereas low intrusion rates reduce or inhibit melt production.For a given emplace-ment rate,if the sill thickness is small compared with the total thickness of intruded basalt,the exact dimensions of sills only modify the details of the temperature and melt fraction on short timescales.As long as the repose period between sill intrusions (104years)is much shorter than the total duration of basalt emplacement (106years),the long-term evolution of the system is controlled by the average emplacement rate and is not affected by the details of sill thickness and injection frequency (Annen &Sparks,2002).An intrusion rate of 5mm/year is equival-ent to 5km of new crust per million years.Over several million years crustal thicknesses of tens of kilometres could be generated.However,our model does not con-sider the counteracting thinning processes such as spread-ing of hot thickened crust and delamination of dense lower

crust.

Fig.6.Modelled melt fraction (X)–T curves for greywacke and pelite under upper crustal conditions.The greywacke curve is based on experimental data of Patin ?o-Douce &Beard (1996)and Montel &Vielzeuf (1997).The pelite curve is based on the model of Clemens &Vielzeuf (1987).517ANNEN et al.DEEP CRUSTAL HOT ZONES

at Institute of Geology and Geophysics, CAS on March 5, 20130025710d561252d381eb6e31/Downloaded from

Boundary and initial conditions The boundary conditions are constant temperatures at the surface (T ?0 C)and at 60km depth (T ?1200 C).The initial temperature variation with depth is determ-ined by the geothermal gradient.Here we chose an ini-tially linear geothermal gradient of 20 C/km.The upper crust,which is greywacke or pelite,is taken to be that part of the crust initially situated above 20km depth;the amphibolite lower crust is initially situated between the Conrad Discontinuity at 20km and the Moho at 30km (Fig.4).We have investigated intruding basalt magmas with 1á5and 2á5wt %H 2O.Emplacement depth Because the temperature through the hot zone varies with depth according to the geotherm,the basalt emplacement depth controls the temperature of equilib-ration of the intrusions and the consequent melt genera-tion.We present results for two fixed emplacement levels,30km and 20km (Fig.4a and b),which model the Moho and Conrad discontinuities,respectively.We also tested a model where basalt sills are randomly emplaced in the lower crust (Fig.4c and d).In all models,the thickness of each intruded sill is accommodated by downward dis-placement of the sequence below the sill resulting in a thickening of the crust and a downward displacement of the Moho.This type of accommodation is an approxima-tion of isostatic equilibration.MODELLING DEEP CRUSTAL HOT ZONES II—RESULTS Sills are emplaced at the basalt liquidus temperature (T L )appropriate for the chosen H 2O content.Each sill trans-fers its heat to the country rocks and equilibrates with the surrounding temperature,which depends on em-placement depth and the geotherm.With successive in-trusions the temperature of the system progressively increases,the hot zone develops and the geotherm slowly evolves.Eventually new sills equilibrate above their sol-idus temperature and start to retain residual melt.Figures 7and 8show the evolution of temperature and melt fraction over time for a selection of 50m sills em-placed every 10kyr (Fig.7)and every 25kyr (Fig.8)at 30km depth (Figs 7a and 8a)and at 20km depth (Figs 7b and 8b).Figures 7and 8depict the first sill (emplaced at time 0),the 50th sill (emplaced at time 500kyr),the 100th sill (emplaced at time 1Myr),and so on.The temperature and melt fraction of each sill evolve with its position on the geotherm,which changes shape with time as heat is supplied by successive sills (Fig.4b and d).When the solidus of the surrounding crust is reached it begins to undergo partial melting.Thus,melt in the hot zone ultimately comes from two distinct sources:from crystallization of basalt,which we refer to as residual melt ;and from partial melting of crust,which we refer to as crustal melt .The crustal melt may derive from upper crust (greywacke or pelite)or lower crust (amphibolite).The relative contribution from each source depends on the thermal profile of the hot zone and the depth of basalt sill injection.In some cases the earliest intruded basalt sills will cool below their solidus,only to be heated above their solidus by subsequent basalt sills (Figs 7and 8).In this case previously solidified basalt sills can be remelted.In detail these remelts may differ in composition from residual melts because of loss of volatiles on complete

solidification.We have not,however,taken this subtle

effect into account because a further parameterization is required (dehydration melting of hydrous mantle basalt),which is not well constrained by experiments.In practice,relatively little melt in the hot zone derives from remelt-ing,and in the interests of simplicity we group all remelts with residual melts.Incubation times The incubation time is defined as the time between the emplacement of the first sill,which typically cools below its solidus,and the generation of the first silicic melt,whether as residual melt or crustal melt.The incubation time for residual and crustal melt strongly depends on basalt emplacement rate and emplacement depth

(Fig.9a).For a fixed emplacement depth at 30km,the incubation time for residual melt generation varies from

260kyr for an emplacement rate of 2mm/year to 20kyr for an emplacement rate of 10mm/year (Fig.9a).At

shallow level the ambient temperature is lower and the

incubation time for a given emplacement rate is signific-antly longer (Fig.9a).In the case of sills randomly emplaced in the lower crust between the Conrad and Moho discontinuities the heat advected by the basalts is distributed over the entire thickness of the lower crust,rather than being focused at one depth.Consequently,the incubation time is longer than in the case of fixed emplacement at 30km or 20km,except for a low emplacement rate of 2mm/year (Fig.9a).The incubation time for crustal melts depends on the depth of sill injection (Fig.9b),but is typically greater than the corresponding incubation time for residual melt generation.One exception is for basalt emplaced in dir-ect contact with fertile pelitic upper crust.Hot zone efficiency and melt productivity Figures 10and 11illustrate the efficiency of the system in generating melt for different emplacement levels and emplacement rates.Figure 10shows the productivity of accumulated melt,i.e.the total thickness of residual or crustal melt generated for each 1m of basalt injected in

518JOURNAL OF PETROLOGY VOLUME 47NUMBER 3MARCH 2006

at Institute of Geology and Geophysics, CAS on March 5, 20130025710d561252d381eb6e31/Downloaded from

the system since the beginning of intrusions,whereas Fig.11shows the production rate,i.e.the amount of melt generated per unit of time.Figure 10a confirms that it is a lot more efficient to generate residual melt from the basalt at deep rather than at shallow level.For example,with an emplacement rate of 5mm/year,after 3á2Myr of basalt injection each 1m of intruded basalt has generated 0á35m of residual melt if emplaced at 30km depth,0á22m if emplaced at 20km depth,and only 0á14m if emplaced at 10km depth.If the sills are ran-domly emplaced in the lower crust,after 3á2Myr,each 1m of intruded basalt has,on average,generated 0á23m of residual melt.Thus,to generate a given quantity of melt less basalt needs to be intruded at deep level than at shallow level.The residual and crustal melt production rates strongly depend on the emplacement rate (Fig.11).The production of residual melt continuously increases with time after the initial incubation period because more residual melt is generated with each new crystallizing basalt injection (Figs 10a and 11a).In contrast,the pro-ductivity of crustal melt is limited by the thickness of crust that can be partially melted.For a fixed intrusion depth,crustal melt productivity and production rates reach a maximum and then decrease (Figs 10b–d and 12b).This is because within the crust heat from the underlying basalt is transferred by conduction whereas the crust is cooled from above (fixed temperature at the Earth’s surface).Thus the thickness of the partially melted crust is limited by heat diffusivity in the crust and cannot grow indefinitely.The situation is somewhat dif-ferent for randomly emplaced sills because screens of crust are sandwiched between hot sills,thereby providing heating from below and above.Consequently,although it requires the longest incubation time,random sill injection is the most efficient way to partially melt the lower crust on a long timescale (Fig.10b).Another effici-ent way to produce crustal melt is to inject basalt at 20km in contact with a fertile upper crust (Fig.10d).At still shallower depths the efficiency of this process

is

Fig.7.Hot zone temperature (left)and melt fraction (right)evolution with time for a selection of basaltic sills injected into the lower crust.Sills 50m thick are injected every 10kyr,i.e.at an average emplacement rate of 5mm/yr.The initial H 2O content of the basalt is 2á5wt %and its injection temperature is 1285 C.The basalt is injected at a fixed depth of:(a)30km (Moho discontinuity);(b)20km (Conrad Discontinuity).The upper crust (above 20km depth)is a pelite;the lower crust is amphibolite.The basalt cools very rapidly after injection to equilibrate thermally with the surrounding crust.Successive intrusions elevate the hot zone temperature.In the first tens of thousand years after injection,sill temperatures oscillate in response to the intrusion of subsequent sills.Eventually their temperature stabilizes when they are displaced sufficiently far below the injection level.

519ANNEN et al.DEEP CRUSTAL HOT ZONES

at Institute of Geology and Geophysics, CAS on March 5, 20130025710d561252d381eb6e31/Downloaded from

diminished because of lower ambient temperatures (Fig.10c and d).Temporal and spatial controls on melt compositions In the classical concept of evolved melt generation by basalt differentiation,a body of basalt slowly cools and crystallizes.The residual melt generated in this way becomes more evolved with time.In contrast,in our hot zone model the reverse is true:the system is heated with time as long as basalt continues to be injected.The formation of a residual melt from each rapidly cooling sill is instantaneous,but as the whole hot zone is heated,the generated melt becomes,on average,less evolved with time (Figs 7and 8).The first residual melts to be produced in the hot zone correspond to the most silicic compositions (rhyolite)and lowest temperatures,evolving with time through dacite to andesite;a sequence opposite to that predicted by models of fractional crystallization in magma chambers.Only when basalts cease to be injected into the hot zone does the system cool and the sequence of melt evolution reverse.

Because sill temperature and melt fraction depend on

their position on the geotherm,the compositional vari-

ation is also a function of depth.At any given time following the incubation period,a variety of residual

and crustal melt compositions coexist across a range of

depths in the hot zone (Figs 12–15).The shape of the geotherm and the consequent melt compositional pers-ity depend on the basalt emplacement rate.For example,in Figs 7and 8basalt is intruded with emplacement rates of 5and 2mm/year,respectively.After a total emplace-ment of 16km of basalt at 30km depth and an intrusion duration of 8Myr,the melt fraction for an emplacement rate of 2mm/year is 0á25–0á26,corresponding to dacitic melt compositions (Fig.8).With an emplacement rate of 5mm/year and after 3á2Myr,which also corresponds to 16km of intruded basalt,the melt fraction varies with depth from 0á25to 0á54,corresponding to melt com-position from dacite to basaltic andesite (Fig.7).In the case of randomly emplaced intrusions,screens of amphibolite are sandwiched between basalt sills (Fig.

15).

Fig.8.Hot zone temperature (left)and melt fraction (right)evolution with time for a selection of basaltic sills injected into the lower crust.Basalt sills 50m thick are injected every 25kyr (average emplacement rate of 2mm/yr)at (a)30km and (b)20km.The initial H 2O content,injection temperature and crustal compositions are as in Fig.7.In this case the temperature in the hot zone grows more slowly than in Fig.7because of the lower emplacement rate;melt fraction (and composition)are more homogeneous.

520JOURNAL OF PETROLOGY VOLUME 47NUMBER 3MARCH 2006

at Institute of Geology and Geophysics, CAS on March 5, 20130025710d561252d381eb6e31/Downloaded from

At high temperature the amphibolite has a higher melt fraction than the basalt,whereas close to the solidus the melt fraction is higher in the basalt.Similarly,the ages of inpidual melts are spread out across the hot zone.In Figs12–15,young melt that has differentiated from the last injected sill coexists with melts that were generated more than3Myr earlier.This has important implications for the apparent timescales of magmatic differentiation.In the likely case that melts from different depths within the hot zone are mixed together during ascent(e.g.Fig.1),unravelling the timescales of differ-entiation from a single rock sample may be especially complicated.

Melt H2O content

Basalt crystallization at high pressure concentrates H2O in residual melts(Figs12–14)even in the case where hydrous minerals such as amphibole are crystallizing. For relatively low melt fractions and high initial H2O contents of the intruding basalt,the residual melt can be extremely H2O-rich.For example,65%crystallization of a basalt with an initial H2O content of2á5wt%will lead to an andesitic residual melt with$7wt%H2O.Elevated H2O contents result in low melt viscosity and density (Figs12–14).Thus,although the residual melt from basalt is rich in silica it is both buoyant and mobile.In contrast, the H2O concentration of the crustal melt is limited by the H2O content of the hydrated minerals in the proto-lith.As a consequence,crustal partial melts are much more viscous than the residual melts in the basalt (Figs13d and14d).This has implications for the relative extractability of crustal vs residual melts.

For low residual melt fractions in the basalt,the H2O concentration can reach sufficiently high levels to become saturated,resulting in exsolution.This situation occurs if the parent basalt has more than$1wt%H2O.Exsolved volatiles can flux the overlying crust and induce further melting.The transfer of H2O from the basalt into the crust is difficult to constrain in low-porosity crust and we have,therefore,not explicitly modelled flux melting of the crust.For this reason,the amounts of crustal melt generated in our models should be considered as minima. H2O exsolution may also enhance the extraction of melt by gas-driven filter pressing(Sisson&Bacon, 1999).Seismic observations are consistent with fluid ex-solution and movement in the deep arc crust(Obara, 2002;Katsumata&Kamaya,

2003).

Fig.9.Incubation time between the first intruded basalt and the beginning of melt accumulation for different average basalt emplacement rates

and emplacement depths.(a)Residual melt;(b)crustal melt.The basalt initial H2O content is2á5wt%,inpidual sill thickness is50m and the

different emplacement rates correspond to different time intervals between intrusions.

521

ANNEN et al.DEEP CRUSTAL HOT ZONES

at Institute of Geology and Geophysics, CAS on March 5, 2013

0025710d561252d381eb6e31/

Downloaded from

CONTRASTS BETWEEN DIFFERENTIATION IN DEEP HOT ZONES AND SHALLOW MAGMA CHAMBERS Generation of evolved andesite and dacite melts in deep crustal hot zones can be contrasted with melt differenti-ation in shallow magma chambers.In the former case sills consolidate quickly and the evolved residual melt is gen-erated on (geologically)short timescales.The residual melt can then be stored for long periods without further differentiation,because on a large scale the thermal pro-file of the hot zone evolves very slowly,on timescales governed by thermal conduction.In contrast,a shallow magma chamber requires high rates of magma input to be maintained in a molten state,as the surroundings are cold.Unless the heat input from the new magma input balances the heat loss from the chamber walls,then temperature and melt composition will evolve continu-ously.U-series data for evolved arc rocks commonly suggest that,following U–Th fractionation,magmas

have residence times of the order of 104–105years

(Reagan et al .,2003;Zellmer et al .,2003b ).We suggest that such long residence times are hard to reconcile with a shallow magma storage system because of the require-

ment of stable thermal conditions.In a deep hot zone the

residual melt is stored at temperatures governed by the geotherm,allowing melt to remain compositionally stable for long periods.

The high H 2O content of residual melts also has important implications for their physical properties,spe-cifically reduced density and viscosity,which mean that melts can be readily concentrated by compaction,facilit-ating extraction,and rapid ascent toward the surface.Carmichael (2002)arrived at a similar conclusion regard-ing the H 2O-rich andesites of the Mexican volcanic arc.He argued that their low viscosity and density allowed magma ascent to be near-adiabatic.The adiabat for hydrous andesite melts is of the order of 25–50 C/GPa (Mastin &Ghiorso,2001;Carmichael,2002).Consequently,there may be only 50–100 C

difference

Fig.10.Productivity for (a)residual melt from basalt crystallization,and crustal melts from:(b)amphibolitic lower crust;(c)greywacke upper crust;(d)pelite upper crust.Productivity is defined as the thickness of the accumulated melt pided by the total thickness of the intruded basalt.Basalt initial H 2O content is 2á5wt %.One 50m thick sill is injected every 10kyr.Results are shown for fixed injection depths of 10,20and 30km and for random intrusion between 20and 30km.

522JOURNAL OF PETROLOGY VOLUME 47NUMBER 3MARCH 2006

at Institute of Geology and Geophysics, CAS on March 5, 20130025710d561252d381eb6e31/Downloaded from

between andesite generation temperatures in the lower crust and low-pressure equilibration temperatures inferred from mineral thermometry or phase equilibria.A further implication of our model is the significant thickening of the crust that results from basalt emplace-ment.If melt is efficiently extracted then the residual basalts will have a cumulate character,with associated

high densities and seismic velocities,as observed in

exposed deep crustal sections through arcs (Debari &Coleman,1989).In some cases,the processes of delamination and recycling of dense mafic cumulates in the less dense underlying mantle can be invoked to limit the extent of crustal thickening (Kay &Kay,1993;Jull &Kelemen,2001).In the Sierra Nevada,USA,for example,seismic refraction and gravity data (Fliedner et al .,1996,2000)show that the granitic rocks of this young mountain belt are not supported by an isostatic root,which Fliedner et al .(1996,2000)attributed to delamination of the mafic counterpart to the granites.Such an interpretation is consistent with rapid Pliocene uplift in the region and the change in the petrology of xenoliths in basalt lavas from predominantly lower crus-tal granulites to predominantly mantle peridotites between 10and 3Ma (Ducea &Saleeby,1996,1998a ,1998b ;Ducea,2002;Farmer et al .,2002).A more recent seismic experiment across the southern Sierra Nevada,by Zandt et al .(2004),using receiver functions,has identified a welt of thickened crust and a ‘hole’in the Moho,which those workers ascribed to asymmetric flow of dense lower crust into a delaminating mantle drip beneath the Great Valley.In other arcs,for which there is less compelling evidence for delamination,the apparent lack of a deep cumulate root may simply be a consequence of the difficulty of seismically distinguishing pyroxenites and/or garnet-rich mafic rocks from mantle peridotite.For example,Fliedner &Klemperer (2000)proposed that beneath the Aleutians volcanic arc some 10km of ultramafic cumulates lie below the geophysical Moho.MELT SEGREGATION

The storage of evolved residual melts in the deep crustal hot zone depends critically on the dynamics of cooling and crystallization of each sill.Two end-member scen-arios can be envisaged,and mirror observations of differ-entiation in high-level intrusions.In one end-member the magma body retains its suspended crystals and consolid-ates as a physically undifferentiated layer of partially molten rock.Crystallization is near equilibrium and porous media processes are required to segregate the evolved residual melts.In the other end-member,crystals and melt are efficiently separated during cooling;for example,by crystal settling and floor crystallization with compositional convection (Tait et al .,1984).The sill thus becomes strongly physically differentiated and evolution may be closer to fractional crystallization.In either case the sill may,in principle,develop into a lower layer of cumulates and an upper layer of buoyant crystal-free evolved residual melt,which may detach immediately or shortly after sill consolidation.Studies of shallow sills suggest that both these end-members,and intermediate situations,can occur depending on many parameters,including sill thickness,density and viscosity of melt,and whether convection

develops.

Fig.11.Melt production rates (mm/yr)for different basalt intrusion rates (2,5,10mm/yr),for (a)residual melt,and (b)crustal melt as a function of the thickness of the intruded basalt.The production rate is defined as the thickness of melt generated per year.The basalt emplacement depth is 30km.The curves were smoothed to eliminate peaks caused by model discretization.

523ANNEN et al.DEEP CRUSTAL HOT ZONES

at Institute of Geology and Geophysics, CAS on March 5, 20130025710d561252d381eb6e31/Downloaded from

Several mechanisms have been proposed for the segregation of buoyant partial melts;namely,compaction of partially molten rock,large-scale Rayleigh–Taylor instabilities of the entire hot zone,and tectonic deforma-tion.The rate of segregation by compaction is very sensitive to melt viscosity (McKenzie,1985).The high H 2O content of hot zone residual melts will result in low viscosity (Figs 12–14)and low density (<2300kg/m 3).For a typical viscosity of 103Pa s compaction-driven segrega-tion times are estimated to be in the range 104–106years for a porosity of the order of 10%(McKenzie,1985).Long residence times of evolved melts in the deep crust are consistent with data from studies of 230Th/226Ra disequilibrium in many intermediate and silicic arc vol-canic rocks (Reagan et al .,2003;Zellmer et al .,2003b ).Jackson et al .(2003)coupled heat transfer from basaltic sills,partial melting of crustal rocks and melt segregation by compaction processes.Their model focuses on the segregation of partial melts from heated crustal rocks,but the same principles can be applied to residual melts in basaltic sills.Melt segregates to produce porosity waves,which move upwards because of buoyancy and start to accumulate at depths just above the solidus.Unmelted rocks above the depth at which the solidus is reached are considered impermeable,so melt cannot ascend by compactional mechanisms.Thus the depth in the crust at which the geotherm reaches the solidus tem-perature is highly significant,because residual melts from basalt and partial crustal melts can only exist below this depth.Jackson et al .(2003)also showed that segregated melts in high-porosity zones can have the geochemical attributes of highly evolved melts.

Melts can segregate more rapidly when the partially molten rock is deformed (Petford,2003).Tectonic processes and large-scale Rayleigh–Taylor buoyancy instabilities in the entire hot zone can cause deformation and melt segregation (De Bremond d’Ars et al .,1995).Interaction between extracted melts from one layer

and

Fig.12.Variation of (a)temperature,(b)melt fraction,(c)melt H 2O content,and (d)melt viscosity as a function of depth,taken as a snapshot 3á2Myr after initiation of the hot zone shown in Fig.7a.The total added thickness of intruded basalt at this stage is 16km;the original crustal thickness was 30km.Melt viscosity is calculated using the equation of Baker (1998).Continuous line shows basalt with initial H

2O content of 2á5wt %and an injection temperature of 1285 C;dashed line shows basalt with initial H 2O content of 1á5wt %and injection temperature of 1302 C.The amphibolite lower crust has become partially melted.Its melt fraction is high but its H 2O content is low because of low initial H 2O content.Thus,the viscosity of the crustal melt is high compared with the viscosity of the H 2O-rich residual melt.

524JOURNAL OF PETROLOGY VOLUME 47NUMBER 3MARCH 2006

at Institute of Geology and Geophysics, CAS on March 5, 20130025710d561252d381eb6e31/Downloaded from

melts from other layers,with lower or higher melt frac-tions,will result in chemical mixing between more and less evolved melts,leading to a wide spectrum of melt compositions extracted from the hot zone (e.g.Fig.1).These hybrid melts should define linear chemical trends in contrast to the curved trends diagnostic of fractional crystallization alone.PETROLOGICAL CONSTRAINTS ON THE ASCENT OF ANDESITE MELT We now consider the ascent of evolved residual melts segregated from a deep hot zone.This exercise requires information on the liquidus temperature (T L )of a particu-lar andesite composition as a function of H 2O content from pressures between those of the lower crust and the depth of H 2O saturation.Unfortunately,to date,no such study has been carried out experimentally,and the desired information can only be compiled for a range of experi-mental studies,and must be considered semi-quantitative.Figure 16shows the liquidus surface,contoured for H 2O content,of a typical silicic andesite.This was con-structed from available phase equilibria experiments in which the residual melt composition was within 2SD of the average 1980–1986Mount St.Helens silicic andesite

for the components SiO 2,Al 2O 3,MgO,FeO,CaO and

Na 2O tK 2O (see Fig.16caption for average values).

We assume that the minor components TiO 2,MnO and P 2O 5have negligible effect on phase relations.We have

also used some experimental data from older near-

liquidus experimental studies (e.g.Eggler,1972;Green,1972)in which the residual melt composition was not analysed,but where the starting composition is within 2SD of the Mount St.Helens average.Our approach involves a number of approximations.We assume that f O 2has only a small effect on phase relations,which is not strictly true for amphibole (Allen &Boettcher,

1983;

Fig.13.Variation of (a)temperature,(b)melt fraction,(c)melt H 2O content,and (d)melt viscosity as a function of depth,3á2Myr after initiation of the hot zone shown in Fig.7b.The total added thickness of basalt,original crustal thickness and melt viscosity are as in Fig.12.Continuous line shows basalt with initial H 2O content of 2á5wt %and an injection temperature of 1285 C;dashed line shows basalt with initial H 2O content of 1á5wt %and injection temperature of 1302 C.The pelitic upper crust has become partially melted.The low H 2O initial content of the pelitic melt accounts for the low viscosity of the crustal melt relative to the H 2O-rich residual melt.Saturation of the melt in H 2O is reached at the base of basalt column.Saturation values are taken from Zhang (1999).

525ANNEN et al.DEEP CRUSTAL HOT ZONES

at Institute of Geology and Geophysics, CAS on March 5, 20130025710d561252d381eb6e31/Downloaded from