Portal de Periódicos da CAPES

Subduction factory: 4. Depth-dependent flux of H 2 O from subducting slabs worldwide

2011; American Geophysical Union; Volume: 116; Issue: B1 Linguagem: Inglês

10.1029/2010jb007922

2156-2202

Peter E. van Keken, Bradley R. Hacker, E. M. Syracuse, Geoff Abers,

earthquake and tectonic studies

Journal of Geophysical Research: Solid EarthVolume 116, Issue B1 Geodesy and Gravity/TectonophysicsFree Access Subduction factory: 4. Depth-dependent flux of H2O from subducting slabs worldwide Peter E. van Keken, Peter E. van Keken [email protected] Department of Geological Sciences, University of Michigan, Ann Arbor, Michigan, USASearch for more papers by this authorBradley R. Hacker, Bradley R. Hacker Department of Earth Science, University of California, Santa Barbara, California, USASearch for more papers by this authorEllen M. Syracuse, Ellen M. Syracuse Department of Geoscience, University of Wisconsin–Madison, Madison, Wisconsin, USASearch for more papers by this authorGeoff A. Abers, Geoff A. Abers Lamont-Doherty Earth Observatory, Columbia University, Palisades, New York, USASearch for more papers by this author Peter E. van Keken, Peter E. van Keken [email protected] Department of Geological Sciences, University of Michigan, Ann Arbor, Michigan, USASearch for more papers by this authorBradley R. Hacker, Bradley R. Hacker Department of Earth Science, University of California, Santa Barbara, California, USASearch for more papers by this authorEllen M. Syracuse, Ellen M. Syracuse Department of Geoscience, University of Wisconsin–Madison, Madison, Wisconsin, USASearch for more papers by this authorGeoff A. Abers, Geoff A. Abers Lamont-Doherty Earth Observatory, Columbia University, Palisades, New York, USASearch for more papers by this author First published: 05 January 2011 https://doi.org/10.1029/2010JB007922Citations: 607AboutSectionsPDF ToolsRequest permissionExport citationAdd to favoritesTrack citation ShareShare Give accessShare full text accessShare full-text accessPlease review our Terms and Conditions of Use and check box below to share full-text version of article.I have read and accept the Wiley Online Library Terms and Conditions of UseShareable LinkUse the link below to share a full-text version of this article with your friends and colleagues. Learn more.Copy URL Abstract [1] A recent global compilation of the thermal structure of subduction zones is used to predict the metamorphic facies and H2O content of downgoing slabs. Our calculations indicate that mineralogically bound water can pass efficiently through old and fast subduction zones (e.g., in the western Pacific), whereas hot subduction zones such as Cascadia see nearly complete dehydration of the subducting slab. The top of the slab is sufficiently hot in all subduction zones that the upper crust, including sediments and volcanic rocks, is predicted to dehydrate significantly. The degree and depth of dehydration in the deeper crust and uppermost mantle are highly diverse and depend strongly on composition (gabbro versus peridotite) and local pressure and temperature conditions. The upper mantle dehydrates at intermediate depths in all but the coldest subduction zones. On average, about one third of the bound H2O subducted globally in slabs reaches 240 km depth, carried principally and roughly equally in the gabbro and peridotite sections. The predicted global flux of H2O to the deep mantle is smaller than previous estimates but still amounts to about one ocean mass over the age of the Earth. At this rate, the overall mantle H2O content increases by 0.037 wt % (370 ppm) over the age of the Earth. This is qualitatively consistent with inferred H2O concentrations in the Earth's mantle assuming that secular cooling of the Earth has increased the efficiency of volatile recycling over time. 1. Introduction [2] Large amounts of water enter the world's subduction zones at a trench. At shallow depths compaction causes fluids to be expelled, but a significant amount remains mineralogically bound in the sediments, crust, and potentially the uppermost mantle of the downgoing slab. When the slab comes in contact with the hot mantle wedge, a series of dehydration reactions (that depend on composition, pressure p, and temperature T) take place. This provides the main source for fluids that cause flux melting in the mantle wedge [e.g., Grove et al., 2006], which in turn causes the formation of volcanoes and the growth of arc crust. [3] Important questions remain regarding the mechanism and location of fluid release from subducting slabs and the amount of water that can be retained in the slab and subsequently can be recycled to the deep mantle. While improved analyses of fluid and gas inclusions have made inferences of volatile contents in the arc source more robust [e.g., Wallace, 2005; Shaw et al., 2008; Kelley et al., 2010], it remains difficult to provide precise global mass balance computations of volatile cycling in subduction zones due to incomplete sampling. The most recent global estimates generally provide extrapolations from the more robust observations of CO2 and SO2 outgassing [e.g., Hilton et al., 2002; Wallace, 2005]. Wallace [2005] provides an estimate of a global flux of 3 × 108 Tg/Myr outgassing of H2O and suggests that there is an approximate balance between the input of structurally bound H2O and that returned in arcs. This suggests that slabs are completely dehydrated and that minimal H2O is returned to the deeper mantle. It is important to constrain the flux of water to the deeper Earth, because even small amounts of water have significant effects on viscosity [e.g., Hirth and Kohlstedt, 1996; Jung and Karato, 2001] and melt productivity [Hirschmann, 2006], thereby directly influencing mantle convection, plate tectonics, and chemical differentiation, and ultimately the long-term chemical and thermal evolution of the Earth. [4] Seismic array studies have made it possible to illuminate the mantle wedge and subducting slab. The reduction in seismic velocities and quality factor Q make it possible to image regions of hydrated crust and mantle [Rondenay et al., 2008; Nakajima et al., 2009] and demonstrate the likely pathways of fluids and melt [Syracuse et al., 2008; Rychert et al., 2008], although effects of water on seismic observables remain incompletely calibrated [Aizawa et al., 2008]. Intermediate-depth seismicity in the crust and upper mantle has been linked to dehydration reactions [e.g., Kirby et al., 1996]. High-resolution images of the position of earthquakes in the slab provide intriguing support for this by demonstrating the similarity in position of intermediate-depth seismicity with the predicted location of the main dehydration reactions [Abers et al., 2006; Kita et al., 2006]. In order to quantitatively relate these observations to dehydration reactions, it is essential to develop petrological models that predict the location of the main dehydration reactions in specific subduction zones. [5] We present here a compilation of 56 subduction zone segments for which we predict temperature, mineral stabilities, and water content. The thermal models are based on those presented by Syracuse et al. [2010]. The petrological modeling is similar to that given by Hacker [2008] but incorporates significant improvements by using (1) an updated thermodynamic description (with the inclusion of melting) and (2) high-resolution finite element models for each subduction zone, which allows for a high-resolution determination of the metamorphic facies in the subduction slab. We use this global compilation to estimate the total flux of water into the mantle wedge and the deeper mantle. This paper is the fourth in a series discussing thermal, petrological, and seismological constraints on the “subduction factory” [Hacker et al., 2003a, 2003b; Hacker and Abers, 2004]. 2. Methods Finite Element Models for Subduction Zone Thermal Structure [6] We use the subduction zone models described by Syracuse et al. [2010]. The subduction zone models cover the world's major trenches in 56 segments and are based on 2-D cross sections normal to the trench (Figure 1). The locations of the segments are based on the global geometry compilation of Syracuse and Abers [2006] and have an average spacing of approximately 700 km. The geometry of each slab is based on an along-trench average for the segment. The temperature in the slab is predicted using a kinematic-dynamic approach similar to that of Peacock and Wang [1999] and van Keken et al. [2002] wherein the velocity in the slab is prescribed but the flow in the wedge is dynamically generated. The speed of the slab is given by the trench-normal component of the convergence velocity and is assumed to remain parallel to the slab surface. We base the wedge viscosity on the dry olivine flow laws of Karato and Wu [1993]. We ignore thermal buoyancy in the wedge. The principal driving force for wedge convection is the coupling to the downgoing slab. We use the “D80” models of Syracuse et al. [2010], in which the slab is fully decoupled from the overriding plate along the seismogenic zone (assumed to end at 40 km), partially coupled to a depth of 80 km, and fully coupled below this depth. The partial decoupling below the seismogenic zone is necessary to generate the “cold nose” of the wedge that is inferred for many subduction zones based on observations of seismic properties and heat flow [e.g., Furukawa and Uyeda, 1989; Hashida, 1989; Hyndman and Peacock, 2003; Stachnik et al., 2004; Yoshimoto et al., 2006; Rychert et al., 2008] and has been suggested to be a global phenomenon [Wada and Wang, 2009]. Figure 1Open in figure viewerPowerPoint Location map of the 2-D subduction zone models used in this study. The trench is colored with thermal parameter Φ defined as the product of convergence speed and age (km). Black lines indicate the location and azimuth of the 2-D cross sections. Numbers indicate the subduction zone (1, Alaska Peninsula; 2, Alaska; 3, British Columbia; 4, Cascadia; 5, Mexico; 6, Guatemala-El Salvador; 7, Nicaragua; 8, Costa Rica; 9, Columbia-Ecuador; 10, North Peru; 11, Central Peru; 12, Peru; 13, North Chile; 14, North Central Chile; 15, Central Chile A; 16, Central Chile B; 17, South Central Chile; 18, South Chile; 19, Northern Lesser Antilles; 20, Southern Lesser Antilles; 21, Scotia; 22, Aegean; 23, North Sumatra; 24, Central Sumatra; 25, South Sumatra; 26, Sunda Strait; 27, Java; 28, Bali-Lombok; 29, West Banda Sea; 30, East Banda Sea; 31, New Britain; 32, Solomon; 33, North Vanuatu; 34, South Vanuatu; 35, Tonga; 36, Kermadec; 37, New Zealand; 38, Southern Philippines; 39, Northern Philippines; 40, South Marianas; 41, North Marianas; 42, Bonin; 43, Izu; 44, Kyushu; 45, Ryukyu; 46, Nankai; 47, Central Honshu; 48, Tohoku; 49, Hokkaido; 50, Southern Kuriles; 51, Northern Kuriles; 52, Kamchatka; 53, Western Aleutians; 54, Central Aleutians; 55, Eastern Aleutians; 56, Calabria). [7] The governing equations are solved on a high-resolution finite element mesh (with 1 km resolution in the wedge tip and thermal boundary layers) using the finite element package Sepran [Cuvelier et al., 1986]. The subduction zone thermal models were benchmarked with several independent codes by van Keken et al. [2008]. With a few exceptions [see Syracuse et al., 2010] we model the evolution of each subduction zone for 20 Myr, which is sufficient to reach a quasi steady state thermal structure. For a full description of the model assumptions, equations, and solution techniques, see Syracuse et al. [2010]. [8] To expand the results for a single cross section to the corresponding subduction zone segment, we assign a length to each segment. This length is determined as follows. For trenches that are represented by a single subduction zone, we use the trench length measured from the maps by Syracuse and Abers [2006]. For segments with multiple subduction zones (e.g., Alaska-Aleutians, Tonga-Kermadec-New Zealand), we determine the length of the representative segment by the location of the cross section. For strongly arcuate trenches (Marianas, Scotia, Calabria) we estimate the effective trench length perpendicular to the convergence velocity. In Calabria and Scotia we exclude the segments that are subparallel to the convergence vector. We have excluded a few trenches for which we do not have representative models (e.g., Makran, Yap-Palau). The total effective length is approximately 38,500 km, which is more than 90% of the combined trench length reported by Clift and Vannucchi [2004] and Scholl and von Huene [2007]. The locations and assumed trench lengths are reported in Table 2. Petrological Modeling [9] According to Hacker [2008], the slab crust consists of a sediment layer of variable thickness and composition, a 300 m thick upper volcanic layer, a 300 m thick lower volcanic layer, 1.4 km of dikes, and 5 km of gabbro. We assume that at input the pore fluids have been expelled and that all water is chemically bound in hydrous phases. The input composition is based on the phases at 15 km depth and at the temperature predicted by the thermal models. The amount of sediment being subducted at each trench is based on the compilations of Clift and Vannucchi [2004] and Scholl and von Huene [2007] and includes the effects of compaction and the off scraping of sediments into the accretionary prism. The composition of the subducted sediment in each cross section was assigned one of six types (based on the methods of Plank and Langmuir [1998]), which are the four used by Hacker [2008] plus a diatomaceous ooze and a turbidite (Table 1). The bulk compositions for the upper volcanic layer, the lower volcanic layer, and dikes were chosen to be mid-ocean ridge basalt (MORB), with the slab-age-dependent K2O and H2O contents given by Jarrard [2003]. Table 1. Sediment Types Used in Modeling Rock Type ODP Site SiO2 TiO2 Al2O3 FeO MnO MgO CaO Na2O K2O H2O carbonate Guatemala site 495 13.3 0.0 0.7 3.7 1.3 44.5 0.4 0.3 0.5 chert Mariana site 800 88.7 0.1 2.3 1.3 0.3 0.7 0.4 0.5 0.6 0.8 terrigenous Antilles “pc” 55.2 0.9 20.8 6.0 0.1 2.2 0.5 0.6 3.0 5.2 diatomaceous ooze Kurile Site 579 70.8 0.5 12.2 4.9 2.2 0.7 3.5 2.3 2.3 pelagic Mariana site 800 49.8 0.6 14.7 7.3 2.1 3.1 3.5 3.1 3.6 6.5 turbidite Alaska Site 178 57.9 0.8 15.4 6.0 3.0 2.4 2.9 2.4 9.2 [10] The extent to which the upper mantle below the crust is hydrated is poorly known. Hacker [2008] summarized the uncertainties regarding hydration of the slab mantle, concluding that a limit of 2 wt % H2O in the upper 4 km of the mantle is a reasonable global average. Here we use three different assumptions for the hydration state of the upper 2 km: dry, 2 wt % H2O, and fully saturated. We will use the moderate hydration case as base model but will provide a discussion of the consequences of assumptions of zero or full hydration. As demonstrated earlier [e.g., Schmidt and Poli, 2003], the slab mantle may carry the bulk of the slab H2O to profound depths. Obtaining better constraints on slab-mantle hydration should be a priority. [11] In the modeling we ignore the contribution of continental crust subduction and erosion, because the overall contribution to the water budget is modest (∼16% [Hacker, 2008]) and we can add this as an a posteriori correction. Phase relations in the slab and overlying mantle wedge were calculated at subsolidus conditions following the methodology of Hacker [2008], using Perple_X version 7 [Connolly, 2009] and the 2004 version of the Holland and Powell [1998] thermodynamic database. The phase relations were calculated for H2O saturation. We then clipped the H2O contents using the limits specified in Table 1 for sediments, those given by Jarrard [2003] for the igneous crust, and the aforementioned three scenarios for the upper mantle section. Hypersolidus phase relations for crustal rocks were determined from the experimental literature (principally from Schmidt et al. [2004]). The addition of melting is a significant enhancement of our earlier models and results in more realistic dehydration of crustal lithologies at 700–900°C. 3. Results Thermal Structure of Subduction Zones [12] We summarize the thermal characteristics of the models by showing in Figure 2 the temperature at a depth of 125 km in three locations in the slab (the top of the sediments, the top of the volcanic rocks, and the top of the upper mantle) as a function of the slab thermal parameter Φ [Kirby et al., 1991]. This parameter is defined here as the product of slab age and convergence speed. It is small for slow subduction of young lithosphere (such as at Cascadia) and large for fast subduction of old lithosphere (such as at Tonga). The initial thermal structure has a strong influence on the Moho temperature, which ranges from 200°C to 900°C. At depths of >80 km, the top of the slab drags the mantle wedge down, and the return flow, which is strongly focused due to the temperature- and stress-dependent rheology, causes high slab surface temperatures, with only moderate sensitivity to the thermal parameter. Temperatures at the top of the volcanic section range from 700°C to 1000°C. While the thermal parameter clearly cannot describe the thermal structure of the slab to great precision, it is tempting to provide trend lines. A logarithmic fit for the temperature (in degrees Celsius) at 125 km depth yields for the top of sediments, for the top of the volcanic section, and for the Moho. The Moho fit is similar to that given by Hacker [2008], but the top of the slab is significantly warmer in the present models. Hacker [2008] used temperature data obtained from a model with the same geometry and parameters as that in the non-Newtonian benchmark (case 2b) of van Keken et al. [2008]. The benchmark model predicts lower temperatures due to the simple geometry (constant angle subduction), thick overriding plate (which forms due to the lack of crustal radiogenic heating and assumption of steady state), and lower potential temperature (1300°C instead of 1420°C). Figure 2Open in figure viewerPowerPoint Temperature at three positions in the slab at 125 km depth as a function of the thermal parameter. The temperature is shown at the top of the sediments (red circles), top of the oceanic crust (green triangles), and bottom of the oceanic crust (blue squares; the bottom of the oceanic crust is assumed to be at a constant depth of 7 km below the top of the oceanic crust). Best fits of a function T = a + b ln(Φ) are shown in corresponding colors. [13] The present thermal models provide a high-resolution temperature distribution, which allows us to more precisely determine the temperature in each compositional layer of the slab. For the thin sedimentary and upper crustal layers we assume that the temperature is that in the center of each layer. The thicker gabbro and uppermost mantle section are divided into 1 km thick layers. The use of more realistic and subduction zone specific models combined with the high-resolution temperature distribution in the slab contributes to significantly more robust predictions of temperature, and therefore of the mineralogy and H2O content, than those used by Hacker [2008]. Water Distribution in the Slab [14] From the petrological models we compute the global average H2O input, weighted by trench lengths. In the case of intermediate (2 wt % H2O) hydration of the uppermost mantle, we find that 7% of the H2O in the slab at input is in the sediment, 35% is in the upper crust (volcanic rocks and dikes), 28% is in the lower crust, and 31% is in the mantle. The flux into subduction zones from these models is 10.0 Tg/Myr, which is significantly lower than the value obtained by Hacker [2008] and at the lower end of the ranges assumed by previous global compilations (Table 3). [15] Inspection of the results for individual subduction zones shows a wide variety of patterns of water loss from slabs. Examples of the thermal structure for a hot (Cascadia), intermediate (Nicaragua), and cold subduction zone (Central Honshu) are shown in Figure 3a. The predicted metamorphic facies are shown in Figure 3b. In order to better display the fine details of the crustal and uppermost mantle structure, we transform the coordinates to a slab coordinate system where s is along the slab surface and t is perpendicular to the slab. This allows us to plot the facies with significant vertical exaggeration. The H2O concentration of the crust and uppermost mantle is shown in Figure 3c. Figure 3Open in figure viewerPowerPoint Cross sections of predictions for (a) temperature, (b) metamorphic facies, and (c) water content for Cascadia, Nicaragua, and Central Honshu as representatives for cold, intermediate, and hot subduction zones. Note the variable horizontal scale in Figure 3a. To display the facies and water content more clearly in the slab, we project the coordinates into a slab reference frame with s parallel to the slab and t perpendicular to the slab. The lower two rows show the facies and water content in this reference frame. The labels “80,” “100,” and “150” indicate the locations where these depths (in km) are reached. “hb sol,” hornblende solidus; “H2O-sat'd sol,” H2O-saturated solidus. The composition and temperature of the slab strongly influence the metamorphic facies and therefore H2O-carrying capability. Cold slabs such as Central Honshu can retain H2O to large depths, whereas warm slabs such as Cascadia dehydrate completely. Intermediate subduction zones such as Nicaragua show dehydration at various depth levels but have significant H2O-carrying capacity in the gabbros. [16] In Cascadia the full slab dehydrates at relatively shallow depths. The only water source beneath the volcanic arc is predicted to be the serpentinized uppermost mantle which dehydrates by 115 km depth. The entire crust at that depth is hotter than the water-saturated solidus. At greater depths the uppermost volcanic rocks and sediments can potentially carry H2O again because the solidus bends to a shallower (p, T) slope than the slab surface temperatures. Water fluxed from the serpentinite could potentially rehydrate the uppermost crustal sections. [17] In the intermediate subduction zone of Nicaragua, we predict that only a small section of the crust is above the water-saturated solidus. Whereas the uppermost crust dehydrates completely below the volcanic front, the deeper (and thus cooler) gabbro can retain H2O to >230 km depth. Shallower portions of the gabbro and the complete mantle section are predicted to dehydrate below ∼150 km depth, suggesting that H2O can be sourced from the slab at variable depths beyond the volcanic front. This prediction is consistent with Cl-isotope evidence for fluids being derived from serpentinite in the Nicaragua arc [Barnes et al., 2009] and the seismic evidence for significant serpentinization of the mantle due to normal faulting in the fore-arc high [Ranero et al., 2003; Ivandic et al., 2008]. Dehydration of much of the crust below the hot wedge here agrees with changes in observed boundaries in migrated scattered wave imaging [MacKenzie et al., 2010]. [18] For a cold subduction zone such as Central Honshu, we find that the uppermost crust dehydrates below the volcanic front, but that the entire lower crust and uppermost mantle have significant water-carrying capacity to profound depths. The location of the main dehydration reactions here is consistent with the location of in-slab seismicity [Kita et al., 2006] and the position of the low velocity layer seen in seismic tomography [Nakajima et al., 2009]. [19] We further note good agreement between our predicted slab surface temperature over a significant depth range for Kamchatka with the H2O/Ce derived temperatures of fluid release by Plank et al. [2009], who provide an explicit comparison between the geochemistry and the regional thermal model in their Figure 4. The predicted slab surface temperature for the Antilles is also in good agreement with the range inferred by comparing trace element geochemistry with experiments on clay melting [Skora and Blundy, 2010]. [20] The various layers in a subducting slab have different bulk compositions that lead to differences in mineralogy and thus in how much H2O is carried as a function of p and T. The layers also experience different (p, T) paths as a result of their different positions in the slab. The end result is that the magnitude and depths of H2O loss can vary dramatically from layer to layer. Figure 4 illustrates this using the (intermediate) southern Mariana subduction zone as example. The sediment at the top of the slab experiences high temperatures and reaches H2O-saturated melting conditions over a narrow depth range of ∼80–120 km, but the position and shape of the H2O-saturated solidus imply that ∼1 wt % H2O can still be retained to depths of >200 km. The volcanic hypabyssal section of the ocean crust is slightly cooler (Figure 4b), but the lower K2O content destabilizes hydrous phases, resulting in lower ( 200 km in gabbro and peridotite. Global Comparison of H2O Loss [22] Cold slabs show dehydration behavior that is markedly different from that of warm slabs. The reason for this is illustrated in Figure 5, which shows that the main dehydration reactions for mafic rocks have steep, positive Clapeyron slopes, whereas the main dehydration reactions for depleted mantle have steep negative slopes at 7 GPa (Figure 5a). In contrast, only in the coldest subduction zones can the uppermost mantle carry H2O to any significant depth (Figure 5b). This occurs in the central portion of the Indonesian arc, Tonga-Kermadec, south Philippines, Marianas, and the Tohoku-Kamchatka trench. Figure 5Open in figure viewerPowerPoint Moho (p, T) paths for global subduction zones on phase diagrams of (a) gabbro and (b) depleted mantle highlight the variation in subduction zone dehydration. In some cold subduction zones (circle) the slab mantle can undergo dehydration around 200 km depth due to the loss of chlorite, whereas H2O is retained in the gabbros to larger depths. In intermediate subduction zones (diamond) dehydration of the mantle is complete by ∼150 km whereas only minor dehydration occurs in the gabbros. In hot subduction zones (star) dehydration of the mantle is complete by ∼100 km depth and significant dehydration of the gabbros from the breakdown of K-white mica has occurred at depths as shallow as 80 km. [23] There is considerable variation in the magnitude and depth of H2O loss from slabs (Figure 6 and Table 2). A few warm slabs show considerable H2O loss at depths of 50–80 km. Most slabs, however, produce a large burst of fluid around 80 km depth, where the thermal models assume that the overlying mantle wedge couples to the slab. Beyond ∼100 km depth, some slabs show relatively continuous H2O loss with depth, whereas others show distinct pulses in H2O loss. Figure 6Open in figure viewerPowerPoint Diverse H2O loss as function of depth in each subduction zone. The hottest subduction zones lose significant H2O in the fore arc. All subducting slabs lose significant water as soon as the sla