4.1.1. Comparison to Prior Flux Estimates
 Most previous analyses of water budgets for oceanic crust do not quantify pore water; they focus instead on structural water. Estimates for the latter are higher than the 6.14 × 1014 g/yr of this study: 8.8 ± 2.9 × 1014 [Ito et al., 1983] and 8 × 1014 [Peacock, 1990]. Ito et al.  used assumed abundances and associated water contents for hydrous alteration minerals in an oceanic crust consisting of alteration-based layers: halmyrolysis, greenstone, amphibolite, and unaltered gabbro. Their structural-H2O values are similar to ours for the extrusives, higher for the dikes, and much lower for the gabbros. Peacock  assumed 2% structural H2O for basalts and 1% for gabbros, based on consideration of whole-rock analyses from three upper-crustal sites and the gabbros of Hole 735B and Oman. Peacock  noted that dredged rocks indicate much higher structural water: 1.7–4.9% (avg. 3.5%) for basalts and 1.6–5.8% (avg. 2.5%) for gabbros, based on the data of Anderson et al. . Bebout [1995, 1996] suggested that total water flux may be ∼18 × 1014 g/yr, about double the structural-H2O values of Ito et al.  and Peacock , to account for the high H2O+ contents of outcropping metamorphosed subduction-zone rocks such as the Catalina Schist. He noted, however, that this additional water may come from sedimentary or basaltic pore fluids. My structural-H2O flux for the extrusives, 1.58 × 1014 g/yr, is significantly higher than the 1.01 × 1014 of Staudigel et al. , due to a combination of higher structural H2O, thicker extrusive section (600 m versus 500 m), and lower porosity (7–13% versus 18.5%).
 Moore and Vrolijk  calculated global water fluxes for both subducting sediments and oceanic crust, using trench lengths and convergence rates from von Huene and Scholl , upper crustal porosity and hydrous minerals for Hole 504B [Becker et al., 1990], and hydrous minerals in sediments of Kastner et al. . They estimated fluxes (typo-corrected) of 1.8 × 1015 g/yr for sediment pores (versus my 7.7 × 1014 g/yr), 4.3 × 1014 g/yr for H2O+ in sediments (versus my 1.2 × 1014 g/yr), and 3.5 × 1014 g/yr for pore and structural water in the top 1 km of oceanic crust (versus my 3.7 × 1014 g/yr for the same portion).
 Nearly all of the water that enters subduction zones comes from seawater or – in the case of bound water in sediments – terrestrial hydrous minerals. Seven studies of fresh MORB glasses indicate about 0.15–0.25% H2O+ [Ito et al., 1983]. Sobolev and Chaussidon  argued that fluid inclusions in olivine phenocrysts provide a more reliable measure of initial MORB H2O+ (0.12 wt.%) than glass does. This value, in conjunction with a crustal consumption rate of 2.4 km2/yr, basaltic layer thickness of 2 km, and weighted average density of 2.76 g/cc, indicates that 1.6 × 1013 g/yr of subducted water, or 0.9% of the total subducted water, is primary water from crustal generation.
4.1.2. Slab Fluid Expulsion Pattern
 Subducting fluids control first-order structural and petrologic problems: structural style and evolution of accretionary prisms [Davis et al., 1983] and generation of arc magmas [Gill, 1981] and consequently long-term growth of continents [e.g., Reymer and Schubert, 1984]. Unfortunately, even rough fluid budgets demonstrate that our knowledge of slab fluid expulsion patterns still faces first-order uncertainties: present-day fluid expulsion rates from some subduction zones are an order of magnitude higher than fluid sources [Le Pichon et al., 1991; Kastner et al., 1991], and an order of magnitude more fluid enters subduction zones than is eventually released within arc magmas [Ito et al., 1983]. Yet, the 1.83 × 1014 g/yr loss of water to subduction zones must be nearly balanced by gains elsewhere, because even a sustained 20% imbalance implies a long-term sea level change of ∼1 m/Ma, which is unlikely on the time scale of >100 Ma.
 Figure 7 summarizes slab fluid expulsion patterns, based on the incoming global water budgets of Table 2 and dehydration evidence discussed in this section. A useful starting point is to stipulate that there are five types of subducting water, each of which might be released at a different point in the subduction process: sediment pore water (7.7 × 1014 g/yr), sediment structural water (1.2 × 1014 g/yr), igneous-crust pore water (3.2 × 1014 g/yr), igneous-crust loosely bound structural water (1.6 × 1014 g/yr), and igneous-crust firmly bound structural water (4.6 × 1014 g/yr). A likely sixth source, structural water in serpentinites, is ignored here because its magnitude is unknown. Structural water within the extrusive portion of oceanic crust is mostly in smectites, whereas that in the dikes and gabbros is in more temperature-resistant minerals such as actinolite and hornblende, so release of the former may be more closely related to smectite breakdown within sediments than it is to breakdown of lower crustal hydrous minerals. Structural water within the sediments is primarily in clays (particularly smectite) and secondarily in opal, and most smectite interlayer water is released at temperatures below 150°C.
Figure 7. Hypothesized water expulsion pattern in a generalized subduction zone. Bottom: cross-section of a subduction zone, with solid arrows showing escape paths of subducting water, and dashed lines showing mantle flow paths. Dehydration mechanisms are shown above the cross-section, with solid bars above the portion of slab undergoing each dehydration reaction. The plot of total global subducted water semiquantitatively illustrates expulsion magnitudes; starting values are from Table 2, but expulsion rates are only approximate, as discussed in text.
Download figure to PowerPoint
 Subduction at margins containing accretionary prisms expels water due to compaction within the accreting and underthrust sediments [Bray and Karig, 1985] and transformation of smectite to illite. Kastner et al.  and Le Pichon et al.  simultaneously demonstrated the discrepancy between present rates of water expulsion at Barbados, Nankai, and Peru prisms and the amount of water that can be accounted for by generation within the prism. Kastner et al.  calculated average water expulsion of ∼7 m3/yr per m of trench (global mass of ∼3 × 1014 g/yr) from internal fluid sources (compaction and dehydration), much less than the 100 m3/yr per m of trench presently being vented at the three margins. The difference was initially thought to be provided mainly by meteoric water [Kastner et al., 1991] and shallow seawater convection [Le Pichon et al., 1991].
 Isotopic studies and identification of low-chlorinity and high-methane anomalies demonstrated the significant contribution from deeper diagenetic processes, particularly smectite dehydration and hydrocarbon generation [Kastner et al., 1991, 1993]. In addition, the assumption of steady state compaction may be invalid [Moore and Vrolijk, 1992; Le Pichon et al., 1993]. Recent modeling of flow and solute transport at Nankai accretionary prism incorporated transient flow, caused by a hydrofracture-induced temporary increase in permeability along the décollement [Saffer and Bekins, 1998].
 Subducted sediments are initially overpressured and underconsolidated [Davis et al., 1983; Moore and Vrolijk, 1992]; rate of fluid expulsion depends on a combination of loading, convergence rate, and permeability [Saffer and Bekins, 2002]. Normal consolidation is achieved and most smectite dehydrates after perhaps 30–40 km of subduction (∼5 km burial) [Saffer and Bekins, 1998; Moore and Saffer, 2001]; near-complete loss of pore and structural waters is presumed to occur within a few additional kilometers of burial. It appears doubtful, however, that any significant net escape of fluids from oceanic crust occurs at these shallow depths, because the framework strength of basalts prevents compaction. This does not preclude fluid flow within the basalts, possibly including flushing of water whose chlorinity has been lowered by saponite breakdown, driven by the buoyancy force of warmer waters flowing updip from deeper sources. Vein and fabric studies of outcropping forearc sedimentary rocks indicate an evolution of fluid flow style during subduction: initial mud-filled veins, then cracks with calcite fill, then scaly fabric [Fisher, 1996]. Updip migration of fluids may occur mainly within the high-permeability scaly fabric of the upper portion of underthrust sediments, at least for burial depths of <15 km [Fisher, 1996].
 Studies of the Catalina Schist showed a slab-parallel melange fabric, suggesting that most fluid loss at depths of 15–45 km is similarly updip, toward the toe of the prism [Bebout, 1991]. At these depths, fluid loss from subducted sediments is nearly completed and slab devolatilization begins, driven by metamorphism rather than compaction. Bebout  documented this process for the metasedimentary and metamafic rocks of the Catalina Schist, including loss of both water and CO2 and associated homogenization of oxygen and hydrogen isotopic signatures. Bebout  extended these observations to other eclogite-facies subduction complexes. For both metasedimentary and metamafic rocks of the Catalina Schist, volatile loss progresses with increasing metamorphic grade, from H2O+ concentrations of 5–6% for lawsonite-albite to 1–2% for amphibolite-facies metamorphism [Bebout, 1995]. Bebout  noted that H2O+ contents of up to 10% for the low-grade metabasalts are higher and more homogeneous than those of DSDP/ODP basalts and ophiolites. He suggested that low-grade metamorphism transforms both original hydrous minerals and pore water into new hydrous minerals.
 Although the Catalina Schist is unlikely to be representative of all subducted sediments and basalts, its pattern of progressive water loss provides a first-order indication of devolatilization at moderate depths. By ∼15 km depth, a water content of 5% by weight for sediments, for a global sediment subduction of 1.7 × 1015 g/yr, implies 8.4 × 1013 g/yr of water subduction, only 9% of the pore and structural water within initially subducted sediments. By ∼40 km depth, reduction of this water content to ∼1.3% expels all but 2% of the initial water. Assuming that the Catalina Schist metabasalts are representative of the extrusive portion of oceanic crust, 5% water content at ∼15 km depth implies 2.1 × 1014 g/yr of water, or 72% of the originally subducted upper crustal waters, and 1.3% at ∼40 km depth implies a residual water content there of 5.6 × 1013 g/yr, or 19% of original subduction volumes (Figure 7). By ∼15 km depth, 50% of the water that entered the subduction zone has been expelled, mainly updip toward the prism toe. By ∼40 km, 60% has been expelled. Inclusion of water expulsion by metamorphic reactions within the dikes and gabbros presumably would raise this total substantially. For example, similar amphibolite-grade metamorphism of the dikes, reducing their water content from 2.8 to 1.3%, would expel 1.5 × 1014 g/yr, or an additional 8% of originally subducted water.
 Magnetotelluric surveys have detected high-conductivity zones attributed to free water in or near the top of the subducting slab at several subduction zones: at 25–50 km depth in Juan de Fuca [Wannamaker et al., 1989], at roughly 20–60 km beneath the Okinawa Trough forearc and somewhere within the interval 60–130 km depth beneath the arc [Shimakawa and Honkura, 1991], extending to >60 km depth in Izu-Bonin [Toh, 1993], and from 13 km to 30 km depth in Mexico [Arzate et al., 1995]. The excellent resolution of the Juan de Fuca resistivity model detected an order-of-magnitude conductivity decrease from the trench to 60 km inland, due to pore water loss, followed by a sudden increase in conductivity at 25 km depth, interpreted as dehydration of greenschist-facies minerals [Wannamaker et al., 1989].
 Updip reflux persists at least to 15 km and possibly to 45 km, based on the Catalina Schist melange fabric [Bebout, 1991, 1995]. The Juan de Fuca conductivity increase at 25 km [Wannamaker et al., 1989], however, does not suggest reflux. Whether subducting crust below 45 km depth retains sufficient fracture permeability for volatile escape is unknown. Dehydration may generate its own permeability, by inducing hydrofracture [Fyfe, 1997]. A magnetotelluric survey of the Vancouver Island portion of Cascadia subduction zone detected porosities of 0.5–4.0% (assuming seawater salinity) well above the slab, attributed to fluid rise from the slab to an impermeable barrier [Hyndman, 1988].
 Based on a variety of H2O+ studies of arc magmas, Ito et al.  calculated that the water flux at arcs is ∼1 × 1014 g/yr; major uncertainties in magma volumes result in a possible range of 0.95–2.0 × 1014 g/yr. Peacock  provided a similar estimate, 1.4 × 1014 g/yr, without mentioning data sources. A more appropriate measure of the volatile content of magmas may be fluid inclusions in olivine phenocrysts, which average 2.5% [Sobolev and Chaussidon, 1996]; in conjunction with an arc magmatism rate of 2.9–8.6 km3/yr [Crisp, 1984], resulting water flux is 2–6 times that of Ito et al.  or Peacock : 2–6 × 1014 g/yr. Of the ∼6 × 1014 g/yr of subducting water that appears to reach ∼45 km depth, at least a third and perhaps all is later released to the mantle wedge and reaches the arc. The remainder may be released into other parts of the mantle wedge, enter arc magmas but degas during volcanism, or be retained into the deep mantle. I emphasize, however, that the cumulative errors in this mass balance are huge and impossible to quantify. For example, Reymer and Schubert  calculated an arc magmatism rate of 1.1 km3/yr (1.3 km3/yr with a more accurate total trench length) from seismic profiles across arcs that is only 13–38% of Crisp's  rate from active arc volcanism; I use the latter. The unknown amount of water loss to the deeper mantle makes estimation of the mantle water budget problematic [Bell and Rossman, 1992; Thompson, 1992; Williams and Henley, 2001].
 Many authors have interpreted phase diagrams based on experimental petrology to infer the cause of the fluid release responsible for arc magmatism [e.g., Peacock, 1990, 1991, 1993, 1996; Pawley and Holloway, 1993; Pawley, 1994; Poli and Schmidt, 1995; Liu et al., 1996; Iwamori, 1998, 2001; Ono, 1998; Ernst, 1999]. A single volatile-expulsion event is not expected. A suite of hydrous minerals (lawsonite, chlorite, amphibole, epidote/zoisite, chloritoid, and others) is expected to break down in a series of overlapping depth zones [Schmidt and Poli, 1998; Kerrick and Connolly, 2001]. Systematic lateral variations in water-soluble elements in magmas from forearc to arc suggest a progressive decrease in water input from the slab [Ryan et al., 1996], possibly a change from amphibole breakdown near the trench to phlogopite decomposition in the backarc [Tatsumi and Kogiso, 1997].
 Even if a single volatile-release event were likely within each slab, it should not occur at similar depths in subduction zones with widely varying slab ages and subduction rates, based on pressure-temperature phase diagrams. Slab temperature at any depth is very sensitive to both subduction rate and slab age, as confirmed by thermal models. For example, for a constant subduction rate, the temperature of the top of upper crust at 100 km depth varies from 540°C for 145-Ma crust to 1180°C for 5-Ma crust [Peacock, 1990]. Accordingly, the same dehydration reaction that is expected to release water beneath the arc for intermediate geotherms releases that water beneath the forearc for high-temperature geotherms (slow subduction and/or very young crust) and retains that water to the deep mantle for low-temperature geotherms (rapid subduction and/or old crust) [Kerrick and Connolly, 2001]. Early release of volatiles from young, hot slabs may affect compositions of later, deeper arc magmas, based on a possible association between slab age and magma composition in only two subduction zones [Green and Harry, 1999; Harry and Green, 1999]. In contrast, ultrafast subduction, such as occurs today in Tonga, favors retention of some water to the deep mantle, possibly generating chemical anomalies in the mantle [Staudigel and King, 1992].
 Volatile release within the subducted slab may occur earlier within the extrusives than in the dikes and gabbros, not only because of more loosely held water in less stable minerals, but also because of larger-scale metamorphic phenomena. Hacker  concluded that conversion of the extrusives to eclogite is complete by ∼250°C, whereas eclogite conversion of the gabbros usually is complete by 550°C and may continue to >800°C.
 In contrast to the wide variability of depths for volatile release, both within and among slabs, volatile escape culminating in generation of arc magma is generally confined to ∼100 km depth. Sub-arc slab depths are readily computed from slab intermediate dip (0–100 km) and arc-trench gap, both tabulated by Jarrard  for 26 of the subduction zones of Table 2. All but two sub-arc depths are between 80 and 120 km. In modern subduction zones, subduction rates vary by a factor of 30, and average ages of the slab vary from 5 Ma to 145 Ma, yet no correlation is found between sub-arc depth and subduction rate, slab age, or their product, strongly implying that melting within the mantle wedge is insensitive to temperature of the adjacent oceanic crust. This observation is consistent with thermal models incorporating convection of the mantle wedge, whereas models lacking convection exhibit strong slab-induced cooling of the mantle wedge [Anderson et al., 1980]. Nor does fluid availability affect the depth of melting: sub-arc depth is not correlated with CO2 or any of the four types of water flux, nor with these fluxes per meter of trench length.
 Predictions of multiple volatile expulsions and thermally dependent expulsion imply large-scale transport prior to release beneath the arc. Even with available water to lower melting temperature, partial melting of the mantle wedge cannot occur until a temperature of at least 1000°C is reached. Water released between ∼50 km and 100 km may remain within the slab or migrate into the adjacent mantle that is dragged downward (Figure 7). In either case, the upper part of the slab and mantle wedge have insufficient permeability to permit large-scale escape of volatiles, until partial melting of the mantle wedge occurs, with associated volatile advection. Of the volatiles that are released after much deeper subduction, some may trigger slab or adjacent mantle melting and reflux along the slab top to the arc-magma generation zone, and some may be lost to the deep mantle.
 Because partial melting in the mantle wedge is triggered by volatile release, a correlation between rate of arc volcanism and amount of deeply subducted water may exist. Early explorations of this hypothesis, using convergence rate as a proxy for subducted water, found no correlation between convergence rate and either arc eruption rate [Gill, 1981] or seismic-based arc growth rate [Reymer and Schubert, 1984]. Subducted water per unit arc length (Table 2) is well correlated (R = 0.72) with arc growth rate if arc ages of Jarrard  are used to calculate the latter, but inversely correlated (R = −0.57) if arc ages from Reymer and Schubert  are used; in both cases, 1–2 extreme points dominate the correlation. A comprehensive analysis of these hypotheses, using water fluxes from Table 2, may be warranted but is beyond the scope of this paper.