Geochemistry, Geophysics, Geosystems

Global analysis of the effect of fluid flow on subduction zone temperatures



[1] Knowledge of the controls on temperature distributions at subduction zones is critical for understanding a wide range of seismic, metamorphic, and magmatic processes. Here, we present the results of ∼220 thermal model simulations covering the majority of known subduction zone convergence rates, incoming plate ages, and slab dips. We quantify the thermal effects of fluid circulation in the subducting crust by comparing results with and without advective heat transfer in the oceanic crustal aquifer. We find that hydrothermal cooling of a subduction zone is maximized when the subducting slab is young, slowly converging, steeply dipping, and the crustal aquifer is ventilated near the trench. Incoming plate age is one of the primary controls on the effectiveness of advective heat transfer in the aquifer, and the greatest temperature effects occur with an incoming plate <50 Ma. The thermal effects of fluid circulation decrease dramatically with increasing age of the incoming plate. Temperatures in the Cascadia, Nankai, southern Chile, Colombia/Ecuador, Mexico, and Solomon Islands subduction zones are likely strongly affected by fluid circulation; for these systems, only thermal models of Cascadia and Nankai have included fluid flow in subducting crust.

1. Introduction

[2] Accurate estimates of temperatures are important for seismic and petrologic interpretations in subduction zones [e.g., Peacock, 2009]. Thermally controlled changes in mineralogy, fluid generation/pressure, and frictional behavior may limit the area of a fault capable of nucleating earthquakes (i.e., the seismogenic zone) to temperatures of ∼150–350°C [Blanpied et al., 1995; Moore et al., 2007]. In subduction zones, temperatures are used to: estimate the extent of potential earthquake rupture areas [Hyndman and Wang, 1993; Oleskevich et al., 1999; Moore et al., 2007], infer the physical and chemical conditions for episodic tremor and slow slip [Peacock, 2009], and constrain the location and distribution of slab alteration and dehydration, which are important for understanding volatile cycling in subduction zones [Hacker et al., 2003; Wada et al., 2012]. Thus, thermal models have been generated to estimate temperature distributions at subduction zones worldwide [e.g., Wada and Wang, 2009; Syracuse et al., 2010].

[3] Rapid fluid flow through the oceanic crust can extract heat from the lithosphere, affecting its thermal evolution [Davis and Lister, 1977; Parsons and Sclater, 1977; Stein and Stein, 1994], transport large quantities of heat laterally for tens to hundreds of km [Davis et al., 1997; Becker and Davis, 2004; Elderfield et al., 2004], and affect subduction zone temperatures [e.g., Spinelli and Wang, 2008]. The basalt comprising the upper ∼600 m of basement rock in the oceanic crust [e.g., Christeson et al., 1992; Canales et al., 2005] is typically viewed as a high-permeability aquifer between overlying low-permeability sediment and underlying low-permeability intrusive rocks (Figure 1) [Fisher, 1998; Becker and Davis, 2004]. The permeability of the basaltic basement aquifer is a key parameter in controlling the vigor of fluid circulation and the efficiency of heat transport within the oceanic crust. Permeability estimates for this basement aquifer are based on laboratory testing of core samples, field measurements of thermal, chemical, and pressure conditions during active and passive testing in boreholes, near-seafloor measurements of thermal and chemical conditions, and modeling of coupled heat and fluid flow [e.g., Davis et al., 1999, 2000; Wheat et al., 2003; Becker and Davis, 2004; Spinelli and Fisher, 2004; Becker and Fisher, 2008; Fisher et al., 2008; Davis et al., 2010]. Measurements and models of permeability in heterogeneous aquifers (particularly fractured systems, like the basaltic oceanic crustal aquifer) yield widely different values depending on the scale of measurement [e.g., Gelhar et al., 1992; Person et al., 1996; Butler and Healy, 1998; Fisher, 1998; Becker and Davis, 2004]. Studies of heat transport or the formation response to loading (both of which interrogate the aquifer over tens of km) indicate permeabilities ∼1000× higher than measurements limited to sampling a few meters of aquifer around a borehole [Becker and Davis, 2004]. The scale dependence of permeability in the oceanic crustal aquifer is consistent with most of the fluid flow occurring within a relatively small fraction of the rock [Fisher and Becker, 2000].

Figure 1.

Cross section of the shallow (<40 km) part of a subduction zone, showing the major geologic components of our thermal models. The basaltic aquifer is 600 m thick. For clarity, the layer of subducting sediment on top of the aquifer is not shown.

[4] For the regional scale (tens of km) examined in this study, the permeability of the basement aquifer of the oceanic crust is typically ≥10−10 m2 [Becker and Davis, 2004]. Studies of oceanic lithosphere <25 Ma offshore Cascadia, Costa Rica, and southern Japan indicate regional-scale permeability of the crustal aquifer ranging from 10−11 to 10−8 m2 [Davis et al., 1997, 2000, 2001; Stein and Fisher, 2003; Spinelli and Fisher, 2004; Hutnak et al., 2008, 2007; G. A. Spinelli, Long-distance fluid and heat transport in the oceanic crust entering the Nankai subduction zone, NanTroSEIZE transect, 2013 submitted to Earth and Planetary Science Letters]. In the subduction zones for these three regions, surface heat flux anomalies and the distributions of thermally controlled alteration of the subducting slabs are most consistent with a 600 m thick aquifer with permeability of ∼10−9 m2 prior to subduction, then gradually decreasing with depth as the oceanic crust is subducted [Spinelli and Wang, 2008, 2009; Harris et al., 2010a, 2010b]. In the only study to date examining regional-scale fluid transport in oceanic lithosphere >25 Ma, Fisher and Von Herzen [2005] estimate aquifer permeability of 10−12 to 10−10 m2 for 106 Ma ocean lithosphere in the eastern Atlantic.

[5] The amount of heat redistribution in an aquifer is controlled by both the aquifer thickness and the efficiency of advective heat transport. Decreasing the aquifer thickness would require an increase in efficiency (via an increase in aquifer permeability) in order to achieve the same lateral heat transport [Davis et al., 1997; Stein and Fisher, 2003; Fisher and Von Herzen, 2005]. For example, observed thermal anomalies on the flank of the Juan de Fuca Ridge are consistent with either a 600 m thick aquifer with permeability of 3 × 10−11 m2 or a 300 m thick aquifer with permeability of 10−10 m2 [Stein and Fisher, 2003]. In addition to restricting the area through which heat can be transported laterally, thinner aquifers tend to support smaller convection cells, which are less efficient at moving heat horizontally. Fisher and Von Herzen [2005] found that lateral heat transport in a 100 m thick aquifer is unable to explain observed thermal anomalies on the 106 Ma eastern Atlantic lithosphere, even with permeability of 10−10 m2, because the convection cells are small.

[6] In the oceanic crustal aquifer at subduction zones, wide convection cells can extend from the incoming plate into the shallow subduction zone, allowing for significant lateral heat redistribution [Kummer and Spinelli, 2009]. Thermal constraints at the Nankai, Costa Rica, and Cascadia subduction zones are most consistent with fluid flow in the oceanic crustal aquifer moving heat updip, warming the subducting plate, and plate interface at shallow depths and cooling them at depths relevant to large (Mw > 8) earthquakes [Spinelli and Wang, 2008; Harris et al., 2010a, 2010b; Cozzens and Spinelli, 2012]. This may shift thermally controlled seismogenic limits landward and increase the width of the seismogenic zone [Spinelli and Wang, 2008; Cozzens and Spinelli, 2012]. In addition, this heat redistribution can generate anomalously high surface heat flux in and near the trench [Spinelli and Wang, 2008]. Although isolated heat flux observations have been used to infer thermal anomalies at subduction zones [e.g., Watanabe et al., 1970; Burch and Langseth, 1981; Yamano et al., 1984], detailed surface heat flux transects (measurement spacing ≤1 km) extending tens of km across subduction zones are rare [e.g., Davis et al., 1990; Hyndman and Wang, 1995; Langseth and Silver, 1996; Yamano et al., 2003; Harris et al., 2010a, 2010b]. The paucity of detailed heat flux transects across subduction zones limits the ability to constrain margin-specific thermal models, but allows synoptic modeling studies to provide a useful role.

[7] Fluid circulation tends to homogenize temperatures in the oceanic crustal aquifer [e.g., Davis et al., 1997; Harris and Chapman, 2004]. Therefore, its thermal effects are most important where temperature gradients in the absence of fluid flow would be large. For example, fluid circulation in the oceanic crust moves heat from more deeply buried (i.e., warmer) regions of the aquifer to areas where the aquifer is closer to the seafloor (i.e., cooler). In addition to redistributing heat vertically in the ∼600 m thick aquifer, fluid circulation can homogenize temperatures laterally over tens of km, mining heat from warm areas and transporting it to cooler regions [e.g., Davis et al., 1997; Spinelli and Fisher, 2004]. Systems with a large lateral temperature gradient in the oceanic crustal aquifer (ΔT/ΔL) are susceptible to considerable heat redistribution by hydrothermal circulation. Because ΔT/ΔL in a subducting aquifer may be large, subduction zone temperatures may be substantially influenced by fluid circulation. Prior to subduction, the oceanic crustal aquifer is relatively cool due to its proximity to the seafloor; upon subduction beneath the margin wedge, temperatures in the aquifer increase. In subduction zones, ΔT/ΔL is largest in systems with young (i.e., hot) subducting lithosphere and slow convergence (Figure 2). Young oceanic lithosphere and slow convergence lead to hot subduction zones, in which the subducting crust warms close to the trench. These hot systems are characterized by a small subduction zone thermal parameter (ϕ):

display math(1)

where A is incoming plate age, v is convergence rate, and θ is subducting slab dip [Kirby et al., 1991]. A small subduction zone thermal parameter provides some indication of the propensity for hydrothermal heat redistribution, but there is not a simple inverse relationship between ϕ and ΔT/ΔL in the aquifer. Steep slab dip results in high ΔT/ΔL in the subducting aquifer, as the upper oceanic crust is more deeply buried at a small distance from the trench (Figure 2), but it also increases ϕ.

Figure 2.

Temperatures in the upper 600 m of oceanic crust within the subduction zone (Tsub) relative to temperatures at the trench (T0), for models with no fluid flow. Ages are for crust at the seaward boundary of the model, 150 km from the deformation front. The 100, 15, and 5 Ma average dip cases are for a slab dip of 25° and a convergence rate of 7 cm yr−1. The 5 Ma average dip, steep, and shallow cases all have the same thermal parameter; changes in convergence rate compensate for changes in dip (Table 1). Distance from trench (x axis) is measured along the center of the aquifer.

[8] Here, we identify the values of key subduction zone parameters that maximize the thermal effects of fluid circulation in subducting oceanic crust. We highlight margins for which hydrothermal circulation should be considered in thermal models. We indicate some prominent margins for which hydrothermal circulation likely has minimal influence on the thermal state.

2. Methods

[9] We use a 2-D finite element model to simulate subduction zone temperatures [Hyndman and Wang, 1993; Peacock and Wang, 1999]. The thermal model accounts for heat production by radioactive decay and heat transport by conduction, advection of the subducting slab, mantle wedge flow, and vigorous fluid circulation in a highly permeable upper oceanic crustal aquifer [Spinelli and Wang, 2008]. The governing equation for our model is:

display math(2)

where K is thermal conductivity, T is temperature, ρ is density, c is specific heat, ν is velocity of the subducting plate, and H represents sources of heat. Each element has eight nodes on its boundaries; the elements have uniform thermal properties. The numerical approach is described in Wang et al. [1995] and Peacock and Wang [1999]. The subducting slab heats up as it is advected under the overriding plate and into contact with warmer surrounding material [McKenzie, 1969; Sleep, 1973].

[10] We run simulations for three different subduction zone geometries; the dip of the subducting slab at 40 km depth equal to 15°, 25°, or 40°. The models extend from 150 km seaward of the trench to 400 km landward of the trench; both side boundaries are far from the region of interest (the shallow, seismogenic portion of the subduction zone) to avoid boundary effects. For all three geometries, we include 1.5 km of sediment in the trench; a global compilation seismic reflection data in subduction trenches yields an average sediment thickness of 1.44 km [Heuret et al., 2012]. We demonstrate the effects of varying the sediment thickness in the supporting information provided. 1 Based on the typical stratigraphy and hydrogeology of oceanic crust, we use an aquifer thickness of 600 m (Figure 1) [e.g., Christeson et al., 1992; Fisher, 1998; Becker and Davis, 2004; Canales et al., 2005]. The base of the model is 100 km below the top of the oceanic lithosphere. At the landward end of the model (i.e., where the base of the slab is deepest), the base of the model is at 213 km depth for the case with the shallowest slab dip; it is at 763 km depth for the case with the steepest slab dip. The crust of the overriding plate is 32 km thick. Thermal properties throughout the system (Table 1) are consistent with previous subduction zone thermal models [e.g., Hyndman et al., 1995; Wada and Wang, 2009; Marcaillou et al., 2012]. Changing the distributions of thermal conductivity and heat production to that for a thinner (i.e., oceanic) overriding plate has a small (∼10°C) effect on modeled temperatures. Element dimensions range from ∼10 m to ∼15 km, typical for models using a similar approach [e.g., Currie et al., 2002; Wada and Wang, 2009].

Table 1. Thermal Properties of Stratigraphic Layers in the Numerical Model
 Thermal Conductivity, K (W m−1 °C−1)Heat Production, H (μW m−3)Thermal Capacity, ρc (MJ m−3 °C−1)
  1. a

    Nu is Nusselt number in aquifer (values determined from equations (3) and (4)).

  2. b

    K increases 0.02 W m−1 °C−1 per km landward throughout margin wedge.

  3. c

    H decreases from 0.8 μW m−3 at the land surface to 0.2 μW m−3 at 10 km depth; below 10 km depth, H in the continental crust is 0.2 μW m−3.

Seafloor sediment1.750.82.5
Basaltic aquifer2.9 Nua0.013.3
Lower oceanic lithosphere2.90.013.3
Margin wedge2–2.9b0.8 
Continental crust2.90.2–1.0c 
Continental mantle3.10.01 

[11] We assign constant temperatures at the boundaries of the model. The top of the model is fixed at 2°C; the base of the model is fixed at 1400°C. At the landward boundary, we use a geotherm consistent with a back-arc setting. We calculate the geotherm from the global average back-arc heat flux at the land surface, 80 mW m−2 [Currie and Hyndman, 2006], and the thermal conductivity and heat production in the continental lithosphere. At the seaward boundary, we use a geotherm for conductively cooled oceanic lithosphere; this varies based on the age of the incoming lithosphere considered in each simulation (Figure 3). We run simulations with insulated or ventilated hydrothermal circulation. Insulated hydrothermal circulation redistributes heat within the aquifer, but all heat transport to the ocean is via conduction through low-permeability seafloor sediments. In this case, heat flux on the incoming plate, particularly in the trench, is anomalously high [e.g., Spinelli and Wang, 2008; Harris et al., 2010a, 2010b]. In regions with prominent plate bending normal faults and/or where the oceanic crustal aquifer crops out at the seafloor, fluid exchange between the aquifer and the ocean (i.e., ventilated hydrothermal circulation) extracts heat from the crust [e.g., Hutnak et al., 2008]. In such cases, heat transported updip in the subducted oceanic crustal aquifer may vent to the ocean [Harris et al., 2010a, 2010b]. Ventilated hydrothermal circulation causes anomalously low heat flux on the incoming plate [e.g., Lister, 1972; Davis et al., 1999; Fisher et al., 2003; Grevemeyer et al., 2005; Hutnak et al., 2007]. In simulations with ventilated hydrothermal circulation, we fix the geotherm through the oceanic lithosphere at the trench (Figure 3b); thus, heat advected updip from the subducted crust does not warm the incoming plate, rather it is removed from the plate (to the overlying ocean). The fluid flow associated with mantle hydration and dehydration along the deep extensions of the faults that may allow ventilated circulation likely occurs at thermally insignificant rates, ∼0.01–0.03 m yr−1 [Faccenda et al., 2009]; thermally important regional-scale fluid flow in the oceanic crustal aquifer occurs at rates ∼1–30 m yr−1 [Stein and Fisher, 2003; Spinelli and Fisher, 2004; Fisher and Von Herzen, 2005; Hutnak et al., 2007]. The models are steady state; the results are not sensitive to the initial conditions.

Figure 3.

Examples of seaward boundary temperatures. (a) For simulations with no fluid flow or insulated hydrothermal circulation, temperatures 150 km seaward of the trench are based on conductive cooling of oceanic lithosphere. We show geotherms for lithosphere 5, 10, 25, 50, and 100 Ma in age. Simulations are also run with 15, 20, 35, 70, 120, and 160 Ma oceanic lithosphere; seaward boundary geotherms for those ages are omitted from this figure for clarity. (b) For simulations with ventilated hydrothermal circulation, temperatures are fixed in the trench based on conductive cooling of oceanic lithosphere below 1.5 km of sediment with a heat flux of 20 mW m−2 and a 600 m thick isothermal basement aquifer. Insets show geotherms through the entire oceanic lithosphere; the main figures show the upper 20 km of the same geotherms.

[12] We use high thermal conductivity in the oceanic crustal aquifer to simulate the thermal effects of vigorous fluid circulation [e.g., Davis et al., 1997]. In an aquifer, the Rayleigh number (Ra) quantifies the tendency for convection; the Nusselt number (Nu) measures the efficiency of advective heat transfer. In an aquifer with vigorous fluid convection (i.e., Ra > 350), advection transports at least 20 times more heat than would be transported by conduction alone (i.e., Nu ≥ 20) and fluid convection is nonstable (i.e., periodic or chaotic) [Wang, 2004]. The nonstable convection homogenizes temperatures within the aquifer; the thermal effects of this vigorous convection are accurately approximated by multiplying the intrinsic thermal conductivity of the aquifer by Nu [Davis et al., 1997]. This high-conductivity proxy for coupled fluid and heat transport is advantageous because it precludes the need to know the detailed permeability structure in the aquifer (e.g., distribution of high-permeability conduits within a lower permeability matrix), which is not well defined on the regional scale of this study but can affect fluid circulation patterns [e.g., Spinelli and Fisher, 2004]. Because we use the high-conductivity proxy (i.e., model only heat transport, not coupled fluid and heat flow), we do not use the model results to infer fluid flow patterns in the aquifer; rather we focus on the thermal processes and results in the subduction zones.

[13] In our simulations, we calculate Nu for each aquifer element by first determining Ra:

display math(3)

where α is fluid thermal expansivity, g is gravity, k is permeability, L is aquifer thickness, ρƒ is fluid density, q is conductive heat flux into the base of the aquifer, µ is fluid viscosity, and κ is thermal diffusivity [Bessler et al., 1994; Spinelli and Wang, 2008]. Fluid thermal expansivity, density, and viscosity are determined for the pressure-temperature conditions in each aquifer element [Parry et al., 2000]. We use the thermal conductivity of the lower oceanic lithosphere and the vertical temperature gradient at the base of each aquifer element to calculate q. In our simulations, aquifer permeability decreases with depth:

display math(4)

where z is depth (km) below seafloor, based on the heat redistribution necessary in a 600 m thick aquifer to explain thermal anomalies in the well-constrained southern Japan subduction zone [Spinelli and Wang, 2009]. Because there are no well-defined trends showing the dependence of regional-scale oceanic crustal aquifer permeability on lithospheric age, we use this permeability trend for all lithospheric ages in most of our simulations. In one suite of simulations, we explore the potential effects of aquifer permeability decreasing with lithospheric age. In this set of simulations, we decrease the logarithm of aquifer permeability by 0.025 per Ma for lithosphere >25 Ma [Fisher and Von Herzen, 2005]. The value of Nu for each element is determined with an empirical equation derived from comparison of model results with coupled fluid and heat transport to results from the same model using a conductive proxy for hydrothermal circulation [Kummer and Spinelli, 2008; Spinelli and Wang, 2008]:

display math(5)

[14] Adjacent aquifer elements with elevated thermal conductivity simulate the wide convection cells that redistribute heat in the oceanic crust [Davis et al., 1997; Spinelli and Fisher, 2004; Kummer and Spinelli, 2009]. Because there is a nonlinear feedback between temperature and Nu, we iteratively converge on a solution. We begin by using results from a simulation without the effects of fluid flow to calculate Ra and Nu for each aquifer element. Then, we increase the thermal conductivity of each aquifer element by a factor of Nu. We rerun the thermal model and recalculate Ra and Nu for each aquifer element. We iterate until temperatures between successive models change by <1°C.

[15] We apply no frictional heating on the plate boundary fault. This minimizes ΔT/ΔL for the no fluid flow cases. Thus, our modeled effects of hydrothermal circulation are a conservative estimate. For a subduction zone that is substantially warmed by frictional heating, the lateral temperature gradient in the subducting aquifer would be larger; therefore, the thermal influence of fluid circulation would be greater. Mechanical studies indicate that the temporally and spatially averaged effective friction coefficient on the plate boundary fault in subduction zones is usually <0.05 [Wang et al., 1995; Wang and He, 1999]. Thermal observations at the Cascadia and Chile subduction zones have been interpreted to indicate no frictional heating on the plate boundary fault [Hyndman and Wang, 1993; Grevemeyer et al., 2003].

[16] The focus area of our study is the shallow portion of the subduction zone where the thermal effects of fluid circulation are largest and the plate boundary fault may be seismogenic; this is tens of km seaward of the region of mantle wedge flow. Temperatures in this forearc region are primarily controlled by advection of the subducting slab, heat conduction, and hydrothermal circulation; the influence of mantle wedge flow is minimal in this region [e.g., Currie et al., 2004; Volker et al., 2011]. Thus, we do not consider mantle wedge flow. We do not use our results to examine subarc or back-arc processes, where the thermal effects of mantle wedge flow are substantial [Currie et al., 2004].

[17] For each of the three geometries, we run simulations with eight different convergence rates and 5–11 incoming plate ages. We examine the effect of hydrothermal circulation on the thermal state of subduction zones for convergence rates from 1.6 to 21.2 cm yr−1 and incoming plate ages from 5 to 160 Ma, spanning conditions at most convergent margins (Tables 2 and 3) [McCaffrey, 2008; Müller et al., 2008; DeMets et al., 2010; Hayes et al., 2012]. We isolate the influence of slab dip by varying the dip and adjusting the convergence rate to maintain a constant thermal parameter. Convergence rates used for each geometry are summarized in Table 2. For all geometries (shallow, average, and steep slab dips), we run simulations with incoming plate ages of 5, 15, 25, 50, 100, and 160 Ma. For the cases with the average slab dip, we run additional simulations with incoming plate ages of 10, 20, 35, 70, and 120 Ma. We quantify the potential thermal effects of fluid circulation in the oceanic crustal aquifer by finding the maximum difference between modeled subduction zone temperatures for simulations without and with hydrothermal circulation.

Table 2. Convergence Rates Used in Models with Various Slab Dips
 Shallowa θ = 15°Average θ = 25°Steep θ = 40°
  1. a

    The slab dip in all column headers is the value of θ at 40 km depth.

  2. b

    Dip-velocity combinations used in Figures 2 and 7.

Slab velocities (cm yr-1)
Table 3. Properties of Subduction Zones shown in Figures 5 and 6
Subduction zonePlate age (Ma)aConvergence rate, v (cm yr-1)bSlab dip at 40 km depth, θcv sin(θ) (cm yr-1)Age uncertainty (Ma)aConvergence rate uncertainty (cm yr-1)bThermal parameter (km)Peak hydrothermal cooling (°C)d
  1. a

    Müller et al. [2008].

  2. b

    DeMets et al. [2010].

  3. c

    Hayes et al. [2012].

  4. d

    Preferred value is for a given thermal parameter, insulated circulation, and most appropriate generic slab dip (“shallow” for <20°; “intermediate” for 20–32°; “steep” for >32°). Range provided in parentheses is for insulated circulation over all generic geometries at a given thermal parameter. Ventilated circulation increases the peak hydrothermal cooling by 5-40 °C over the insulated values, with the largest increases for the hottest subduction zones.

  5. e

    McCrory et al. [2012].

  6. f

    Moreno et al. [2009].

  7. g

    Seno et al. [1993].

  8. h

    Eberhart-Phillips et al. [2006].

Cascadia (Vancouver Is.)53.522e1.320.165155 (89-219)
Solomon Islands52.8652.550.2125138 (60-138)
Cascadia (Oregon)83.521e1.210.193120 (72-170)
south Chile (∼45°S)56.527f2.910.314886 (54-123)
Mexico85.7242.330.317477 (49-110)
Colombia/Ecuador145.4222.070.228354 (36-78)
Nankai (Muroto)154.0b, g181.231.0b, g18048 (48-107)
Nankai (Kumano)204.0b, g261.831.0b, g36046 (31-66)
Alaska (Prince William Sound)405.57h0.720.126837 (37-81)
Middle America208.0384.950.398533 (17-33)
Vanautu/New Hebrides1204.7493.5250.5106431 (16-31)
Philippines444.5332.5110.8134926 (14-26)
Ryukyu238.4263.740.284725 (19-36)
Andaman485.2211.910.389424 (18-35)
Alaska (Kodiak Island)425.525h2.320.297622 (17-33)
central Chile348.0192.620.272020 (20-41)
Alaska Peninsula485.8171.770.181319 (19-37)
north Peru437.5151.920.183519 (19-37)
north Marianas1501.9260.8250.2125919 (15-27)
north Vanautu1205.0403.2250.6213719 (10-19)
Aleutians595.8171.740.2100117 (17-32)
central Marianas1502.5281.2250.3176115 (12-22)
Sumatra765.2191.730.3128714 (14-27)
Hikurangi608.0314.151.0205513 (11-19)
Izu1434.5332.5130.3350513 (8-13)
Kamchatka1048.5253.6240.237369 (7-13)
Japan1439.0162.5130.335478 (8-13)
Java-Timor1456.8252.970.241438 (7-12)
Tonga/Kermadec8418.5329.8163.284795 (5-7)

3. Results

[18] Hydrothermal circulation in the subducting crustal aquifer extracts heat from the subduction zone and transports it to the trench and incoming plate (Figure 4). The maximum hydrothermal cooling occurs in the upper portion of the subducting plate at ∼40–50 km depth. With insulated hydrothermal circulation, the trench is warmer than in comparable cases with no fluid flow. With the fixed trench temperatures for ventilated hydrothermal circulation, heat extracted from the subduction zone is removed from the lithosphere to the ocean rather than warming the trench. All simulations with insulated hydrothermal circulation share similar patterns of heat redistribution, as do all simulations with ventilated hydrothermal circulation. The amount of heat redistributed by hydrothermal circulation varies widely depending on the incoming plate age and convergence rate.

Figure 4.

Examples of modeled temperatures and distribution of hydrothermal cooling. (a) Modeled temperatures are for a simulation with insulated hydrothermal circulation, 5 Ma incoming lithosphere, a convergence rate of 7 cm yr−1, and the intermediate slab dip (25°). The black line is the plate interface. (b) The effect of hydrothermal circulation is quantified by the difference between simulations with and without fluid flow. These figures are focused on the trench and shallow subduction zone where heat redistribution by hydrothermal circulation is greatest; the full model extends from 150 km seaward to 400 km landward of the trench. Hydrothermal cooling is greatest in the uppermost portion of the subducting plate, ∼125 km landward of the trench. Heat is moved from this part of the system to the trench.

Figure 5.

Maximum hydrothermal cooling for (a) insulated and (b) ventilated scenarios with average slab dip. Incoming plate age is at the seaward boundary of the model, 150 km from the deformation front. Data for all subduction zones plotted are summarized in Table 3. Labeled subduction zones are those for which hydrothermal circulation has a large influence on temperatures (MA = Middle America; SC = South Chile; SI = Solomon Islands; MX = Mexico; NN = Nankai; CE = Colombia/Ecuador; CA = Cascadia) or those that have hosted Mw ≥ 9 earthquakes, but whose temperatures are not dramatically affected by hydrothermal circulation (CC = central Chile; AK = Alaska (Kodiak Island); SU = Sumatra; KK = Kamchatka; JA = Japan). Italicized labels indicate those margins are documented as having trench sediment <0.8 km [e.g., Clift and Vannucchi, 2004] and may have ventilated hydrothermal circulation.

Figure 6.

Maximum hydrothermal cooling for insulated (i.e., not ventilated) scenarios for (a) shallowly dipping and (b) steeply dipping slabs. Incoming plate age is at the seaward boundary of the model, 150 km from the deformation front. Plotted and labeled subduction zones are the same as in Figure 5.

Figure 7.

Maximum hydrothermal cooling versus thermal parameter (ϕ) for all model scenarios. Shallow, average, and steep refer to different slab dips. Ventilated circulation allows for advection of heat from the aquifer to the ocean (see text for details). For ϕ ≥ 800 km, maximum hydrothermal cooling is <50°C.

[19] We find that among the parameters tested, incoming plate age (A) exerts the greatest control on the efficacy of hydrothermal circulation in influencing subduction zone temperatures. Where A is <50 Ma, fluid circulation in the oceanic crustal aquifer can have a large effect on subduction zone temperatures; where A is >50 Ma, the thermal effect of fluid circulation is <50°C (Figure 5). We highlight scenarios with >50°C hydrothermal cooling because perturbations of this magnitude exceed the uncertainty commonly reported by varying thermal model parameters (e.g., convergence rate, geometry, material properties, boundary conditions) [e.g., Oleskevich et al., 1999; Harris et al., 2010a, 2010b]. The downward velocity of the subducting plate (vsin(θ)) is a secondary control on the influence of hydrothermal circulation on subduction zone temperatures. For example, reducing vsin(θ) from 5.5 to 1.0 cm yr−1 (i.e., across the entire range examined) effects a 75°C increase in the amount of hydrothermal cooling; whereas, reducing A from 160 to 5 Ma increases the amount of hydrothermal cooling by 130°C, with most of the increase focused between 20 and 5 Ma (Figure 5). Where heat in the oceanic crustal aquifer seaward of the trench vents to the ocean, hydrothermal cooling of the subduction zone is greater than for cases with insulated circulation (Figure 5b). For these cases with an average slab dip, the maximum hydrothermal cooling (at A = 5 Ma; vsin(θ) = 1.0 cm yr−1) is 160°C with ventilated circulation and 145°C with insulated circulation.

[20] Estimates of the thermal effects of fluid circulation from our generic models compare favorably with results of the few existing site-specific thermal models examining hydrothermal circulation. Plate ages, convergence rates, and slab dips for subduction zones indicated in Figures 5 and 6 are summarized in Table 3 [McCaffrey, 2008; Müller et al., 2008; DeMets et al., 2010; Hayes et al., 2012]. While previous detailed models for the Middle America, Cascadia, and Nankai margins showed that hydrothermal circulation confined in the oceanic crustal aquifer under a blanket of low-permeability sediment (i.e., insulated hydrothermal circulation) reduces slab surface temperatures by up to 25, 95, and 100°C, respectively [Spinelli and Wang, 2008; Harris et al., 2010a, 2010b; Cozzens and Spinelli, 2012], for the same margins our generic models with average slab dip and insulated circulation yield maximum hydrothermal cooling of approximately 20, 125, and 75°C (Figure 5a). Hydrothermal circulation has a larger effect on temperatures in the warmer Cascadia and Nankai margins than in the cooler Middle America margin. Geometric differences (e.g., sediment distribution and thickness in trench; slab dip) between the specific sites and the generic model used here cause the deviations between the site specific and generic estimates for temperature reductions due to hydrothermal circulation. Detailed models for Middle America showed that ventilated hydrothermal circulation reduces subduction zone temperatures by up to 30°C [Harris et al., 2010a, 2010b]; our generic models show the maximum temperature reduction by ventilated hydrothermal circulation for Middle America's plate age and convergence rate is approximately 35°C (Figure 5b).

[21] For all scenarios examined, models of steeper slab dip result in larger hydrothermal cooling (Figure 6). Although steeper slab dip (without invoking fluid circulation) tends to cool a subduction zone (i.e., increase the subduction zone thermal parameter) due to the increased downward velocity of cool subducting lithosphere, the steeper dip generates a larger lateral temperature gradient between the incoming and subducted aquifer due to the deep burial of the aquifer close to the trench (Figure 2). Thus, steeper slab dip enhances the heat redistribution by hydrothermal circulation. For a given subduction zone thermal parameter, simulations with the steep slab dip show up to ∼50°C more hydrothermal cooling than simulations with a shallow slab dip (Figure 7). The cooling of a subduction zone by hydrothermal circulation in the oceanic crust is maximized by ventilated circulation from a steeply dipping slab with young incoming lithosphere and slow convergence. Predicted peak hydrothermal cooling for subduction zones indicated in Figures 5 and 6 is given in Table 3.

[22] If the permeability of the oceanic crustal aquifer decreases with lithospheric age, then systems with old subducting lithosphere show ∼4°C less hydrothermal cooling relative to cases where permeability does not decrease with age (Figure 8). As a result, age-dependent permeability reduction slightly accentuates the primary trend found in this study—hydrothermal cooling is most important in systems with young subducting lithosphere; its influence in systems with old subducting lithosphere is small.

Figure 8.

Maximum hydrothermal cooling versus incoming plate age for simulations without and with age-dependent aquifer permeability. Decreasing the presubduction aquifer permeability with lithospheric age reduces the maximum amount of hydrothermal cooling in systems with older subducting lithosphere (dashed line).

4. Discussion and Conclusions

[23] Because the age of the incoming lithosphere is the strongest control on the magnitude of heat redistribution, temperatures in the young-slab Cascadia, Nankai, Mexico, Solomon Islands, Colombia/Ecuador, and southern Chile subduction zones are likely reduced substantially due to fluid flow in the oceanic crust (Figure 5). Of these subduction zones, the effects of fluid circulation have only been considered in thermal models of Cascadia [Cozzens and Spinelli, 2012] and Nankai [Spinelli and Wang, 2008]. Thermal models (and inferences drawn on them) for the Mexico [Currie et al., 2002; Manea et al., 2004], Colombia/Ecuador [Marcaillou et al., 2008], southern Chile [Grevemeyer et al., 2003; Volker et al., 2011], and Solomon Islands [Syracuse et al., 2010] subduction zones should include the effects of fluid circulation. Our generic models suggest hydrothermal cooling of up to ∼140°C for the Solomon Islands, ∼85°C for southern Chile, ∼75°C for Mexico, and ∼55°C for Columbia/Ecuador.

[24] Hydrothermal cooling can affect the rates and distributions of diagenetic and metamorphic reactions in subduction zones. For example, hydrothermal cooling in the Nankai and Cascadia margins likely shifts the location of basalt-to-eclogite transition in the subducting slab farther landward than it would occur without the cooling due to fluid flow [Spinelli and Wang, 2009; Cozzens and Spinelli, 2012]. At lower temperatures, sediment dewatering reactions (e.g., opal-to-quartz or smectite-to-illite) may be delayed and shifted farther into subduction zones [Spinelli and Saffer, 2004]. Shifting the location of dehydration reactions may affect the fluid pressure distribution in subduction zones [Spinelli et al., 2006], and therefore the mechanical behavior of faults [Hubbert and Rubey, 1959; Saffer and Tobin, 2011]. Estimates for the distribution of dewatering reactions (or other alteration) in hot subduction zones should include the effects of hydrothermal cooling.

[25] Another broad-reaching implication of our results is the potential influence on the landward rupture extent of large earthquakes at subduction zones. Hydrothermal circulation shifts both the thermally defined updip and downdip limits of seismicity (i.e., 150°C and 350°C isotherms on the plate boundary fault) landward. However, the effect of hydrothermal circulation on plate boundary fault temperatures is small at the trench and increases to a maximum at ∼40 km depth (Figures 4 and 9). Thus, the landward shift of the updip limit is smaller than that for the downdip limit, and the width of the thermally defined seismogenic zone is increased. Temperatures in hot subduction zones are affected most dramatically by hydrothermal circulation. Hot subduction zones are also most likely to have the thermally defined downdip limit of seismicity (350°C) occur seaward of the intersection of the plate boundary and the continental Moho. Thus, hydrothermal cooling of a hot subduction zone may shift 350°C on the fault landward of the intersection with the continental Moho, maximizing the width of the potential seismogenic zone [Moore et al., 2007]. For the subduction zones known or suspected to generate earthquakes Mw ≥ 9.0, the extent of earthquake rupture is likely controlled by the intersection of the plate boundary fault with the Moho in the relatively cool Alaska-Aleutians, Sumatra-Andaman, Japan, Kamchatka, Peru, and central Chile margins [e.g., Oleskevich et al., 1999]. In the warm Cascadia and southern Chile margins, earthquake rupture extent is more likely to be thermally controlled and affected by hydrothermal cooling of the plate boundary fault (Figure 5). Seismic risk for hydrothermally cooled subduction zones may need to be reassessed.

Figure 9.

Plate boundary fault temperature versus depth for 5 Ma incoming crust and different slab dip and fluid flow conditions.


[26] We thank I. Wada and two anonymous reviewers for suggestions that improved this manuscript. This work was funded by National Science Foundation grant EAR-0943994.


  1. 1

    Additional supporting information may be found in the online version of this article.