Geochemical and geophysical evidence indicate that splay faults cutting subduction zone forearcs are a key hydraulic connection between the plate boundary at depth and the seafloor. Existing modeling studies have generally not included these structures, and therefore a quantitative understanding of their role in overall fluid budgets, the distribution of fluid egress at the seafloor, and advection of heat and solutes has been lacking. Here, we use a two-dimensional numerical model to address these questions at non-accretionary subduction zones, using the well-studied Costa Rican margin as an example. We find that for a range of splay fault permeabilities from 10−16 m2 to 10−13 m2, they capture between 6 and 35% of the total dewatering flux. Simulated flow rates of 0.1–17 cm/yr are highly consistent with those reported at seafloor seeps and along the décollement near the trench. Our results provide a quantitative link between permeability architecture, fluid budgets, and flow rates, and illustrate that these features play a fundamental role in forearc dewatering, and in efficiently channeling heat and solutes from depth.
 Major splay faults that cut the forearc are ubiquitous at convergent margins, yet a systematic, quantitative understanding of their impact on overall fluid budgets, rates and distribution of seafloor venting, and transport of heat and volatiles has been lacking. Significant focused fluid flow along such faults has been inferred from mapping of seafloor seeps [e.g., McAdoo et al., 1996; Suess et al., 1998], and local shoaling of the BSR [e.g., Bohrmann et al., 2002]. Geochemical observations at seep sites, including low-Cl, elevated B, and ∂18O values, imply that these faults provide a hydraulic connection from deep within the subduction zone to the seafloor, and therefore play an important role in the flux of volatiles from the subducting slab and sediments at or beneath the plate boundary décollement [e.g., Ranero et al., 2008; Hensen et al., 2004; Sahling et al., 2008]. However, previous modeling studies of subduction zone hydrogeology have generally considered only the décollement zone as a permeable pathway [e.g., Bekins et al., 1995; Spinelli et al., 2006]. Some studies have considered the effects of splay faults in local sediment dewatering [Shi et al., 1989] and in transporting heat [Cutillo et al., 2003] but have not systematically explored their role in forearc hydrogeology.
 Here, we quantitatively evaluate the role of permeable splay faults on forearc fluid budgets and flow rates at non-accretionary subduction zones. We focus on a case study of the Costa Rican margin, where geochemical and thermal data and seafloor mapping show that these structures transport fluids from depth [e.g., Hensen et al., 2004; Sahling et al., 2008], and where the relevant rock and sediment physical properties, thermal regime, and geology are well-constrained by drilling and geophysical studies [Kimura et al., 1997; Ranero et al., 2008]. We use a 2D model of fluid flow in a cross-section oriented perpendicular to the trench, in order to (1) quantify the budget and partitioning of fluid escape between permeable splay faults, diffuse flow through the overriding plate, and the décollement, (2) calculate flow rates at the seafloor for comparison with values inferred from field measurements of geochemistry and heat flow, and (3) explore the sensitivity of these fluxes to fault and décollement permeability (kf and kd) in order to provide a robust and quantitative framework for future interpretation of seafloor and subsurface data.
2. Geologic and Hydrogeologic Setting
 The Middle America Trench (MAT) is formed by subduction of the Cocos Plate beneath the Caribbean Plate at ∼8.5 km Myr−1 (Figure 1a) [DeMets, 2001]. In our study area offshore the Nicoya Peninsula, the age of the subducting Cocos plate is 24 Ma. Anomalously low heat flow of 20–40 mW m−2 has been documented in this region, and is attributed to vigorous low-temperature hydrothermal circulation in the ocean crust [e.g., Langseth and Silver, 1996]. The sediments on the Cocos plate include ∼250 m of pelagic carbonate ooze overlain by ∼150 m of clay-rich diatomaceous mud [Kimura et al., 1997]. The margin is non-accretionary; the incoming sediment section is completely subducted, and the overriding margin wedge is composed of the Nicoya Complex, an extension of the onshore basement [Kimura et al., 1997]. The margin wedge is overlain by up to ∼2 km of slope sediment, and a narrow (5–10 km) deforming prism at the toe of the margin wedge is formed by offscraping and reworking of these sediments (Figure 1).
 The Costa Rican margin has been the focus of numerous geological, geophysical, and hydrologic investigations, including seismic reflection surveys, heat flow campaigns, Ocean Drilling and Integrated Ocean Drilling Program (ODP and IODP) drilling, and mapping and sampling of vent sites on the continental slope [e.g., Shipley et al., 1992; Kimura et al., 1997; Hensen et al., 2004]. Seafloor mapping, shallow heat flow, and geochemical data from seep sites all show that splay faults form an important flow pathway for fluids derived from subducting and dehydrating sediments at depth [e.g., Hensen et al., 2004].
3. Modeling Methods
 We use the two-dimensional finite element code SUTRA [Voss, 1984] to simulate steady state fluid flow along an 70-km long transect oriented perpendicular to the trench and approximately parallel to the plate convergence direction [e.g., Spinelli et al., 2006] (Figures 1 and 2). These steady-state models represent the time-averaged fluid pressures and flow rates relevant to the first-order fluid budget of the forearc [e.g., Saffer and Bekins, 1998]. In our models, we prescribe fluid sources to emulate tectonic loading and compaction of the sediments on the incoming plate [e.g., Bekins and Dreiss, 1992]. These fluid sources drive pore pressure and fluid flow in the model, mediated by the spatial distribution of permeability. Although our model does not explicitly couple pore fluid pressure, effective stress, and sediment compaction, the prescribed fluid source terms are consistent with modestly overpressured conditions [e.g., Spinelli et al., 2006]. Moreover, previous studies have shown that to first order, sediment dewatering and the resulting fluid pressures and flow patterns are not highly sensitive to the detailed distribution of fluid source terms when varied over a realistic range [Bekins and Dreiss, 1992; Saffer and Bekins, 1998].
 Our model domain extends from 10-km seaward to 60-km landward of the trench, and includes the upper plate, décollement zone, and subducting sediments. In order to explore the role of splay faults on flow rates and the partitioning of fluid expulsion, we include seven faults that cut the upper plate margin wedge and connect the décollement to the seafloor (Figure 2). Based on constraints from seismic reflection data and drilling, we include a 2.5 km-thick drape of slope sediments overlying the margin wedge, and a 5 km-wide accretionary prism near the trench composed of reworked slope sediment (Figures 1 and 2). We define the model geometry and a fault spacing of 5-km based on seismic reflection data [Ranero et al., 2008; Shipley et al., 1992]. The top and seaward boundaries of the model domain are prescribed as hydrostatic constant head. The bottom and landward boundaries of the model are set to a no-flow condition (Figure 2), under the assumptions that (1) there is little hydrologic communication between the sediments and oceanic crust, as documented by inferred pore fluid pressure distribution within the subducting sediment section [e.g., Saffer, 2003]; and (2) fluid sources and permeabilities >60-km from the trench are sufficiently low that fluid flow there is negligible [e.g., Bekins et al., 1995]. Previous modeling studies have shown that the inclusion of a permeable upper oceanic crust has little effect on fluid fluxes and pore pressures unless the sediments themselves are highly permeable [e.g., Spinelli et al., 2006], so relaxing the first assumption should not significantly impact our results. The primary model inputs are the fluid source terms and hydraulic conductivity structure; each of these is described in detail below.
3.1. Porosity and Fluid Sources
 We calculate fluid source terms from porosity loss as sediments are transported into the subduction zone and buried [e.g., Bekins and Dreiss, 1992]. We define porosity (ϕ) as a function of depth (z, km below seafloor), constrained by data from both drilling and exhumed subduction zone sediments [Bray and Karig, 1985; Spinelli et al., 2006]:
Porosities range from 0.7 at the trench to ∼0.03 at 12 km depth. For the case of subducting sediments subjected to vertical (uniaxial) consolidation, the dewatering rate (Γ, units of Vfluid/Vsed s−1) is then defined by
where vp is the plate convergence rate (m s−1), and α and β are the surface slope and décollement dip angle, respectively. We use a total taper angle of 12-degrees (α = 3.8°; β = 7.6°) [Shipley et al., 1992]. In computing fluid sources, we do not include the contribution of fluids from mineral dehydration, because they are generally 10–100 times smaller than those derived from sediment compaction [e.g., Bekins et al., 1995; Spinelli et al., 2006]. On the basis of porosity data at ODP drillsites. The total inventory of pore fluids entering the subduction zone is 21.7 m3 yr−1 per m along-strike.
3.2. Permeability and Hydraulic Conductivity
 We define the permeability of the incoming and subducted sediments using a permeability-porosity relationship defined by laboratory measurements on core samples from ODP Sites 1039, 1040, 1254, and 1255 [Saffer and McKiernan, 2005]. We assign a permeability of 2 × 10−17 m2 to the slope sediments, which represents the effective vertical permeability for a 2.5 km-thick section, using a permeability-porosity relationship for similar slope sediments offshore Peru [Marsters and Christian, 1990]. In a subset of models, we do not include the slope sediments, in order to simulate a scenario in which the slope apron is breached by faulting or mud volcanoes [e.g., Hensen et al., 2004]. Based on lithologic similarities identified through drilling [Kimura et al., 1997], we assign the narrow frontal prism the same permeability as the slope sediments.
 To evaluate the impact of splay faults on the fluid budget and flow, we establish a baseline model with Nicoya Complex (kn) and décollement (kd) permeabilities of 10−19 m2 and 10−14 m2, respectively, and without including any other faults. In subsequent model runs, we include permeable splay faults, and consider a range of splay fault (kf) and décollement permeabilities from 10−16 to 10−12 m2. These values are consistent with the range of subduction fault zone permeabilities measured by drill stem testing and inferred from flow rates, geochemical signals, estimated pore pressures, and numerical modeling studies [Saffer and Tobin, 2011, and references therein]. A previous study of this margin [Spinelli et al., 2006] established that the fluid budget is insensitive to the permeability of the intact upper plate; therefore we assign the Nicoya Complex a permeability of 10−19 m2 for all simulations. In defining hydraulic conductivity, we account for the effects of temperature on fluid properties, using temperatures from the thermal model of Spinelli et al. .
4. Results and Implications for Forearc Hydrogeology
4.1. Forearc Fluid Budget
 Our results show that splay faults play a major role in the fluid budget of convergent margins that has, until recently, been overlooked [e.g., Ranero et al., 2008]. Their effects on the partitioning of fluid escape scale with their permeability and that of the décollement zone. In our baseline model (no permeable splay faults), 64% of the incoming fluids exit the forearc along the décollement, 36% via diffuse flow through the overriding plate, and <2% are subducted beyond the arcward model boundary with the downgoing plate. This is consistent with results of previous modeling studies at this and other subduction margins, which indicate that 60–70% of fluids exit along a permeable décollement and most of the remaining fluids escape by diffuse flow across the seafloor of the forearc [Spinelli et al., 2006; Screaton et al., 1990; Saffer and Bekins, 1998]. Although not the focus of our work, it is important to note that pore pressures in all of our simulations are modestly elevated, but do not exceed lithostatic values, consistent with both large-scale mechanical constraints and pore pressures inferred from drilling data [e.g., Saffer, 2003; Spinelli et al., 2006].
 In models that include permeable splay faults, the percentage of fluids captured by the splay faults varies systematically with both kd and kf (Figure 3). For simulations with kd = 10−14 m2, the proportion of incoming fluid exiting along splay faults ranges from 6% to 30% as kf is increased from 10−16 m2 to 10−13 m2 (Figure 3). For simulations with lower décollement permeability (kd = 10−15m2), the splay faults capture between 17% and 35% of the incoming fluid. The budget is generally not sensitive to further increases in splay fault permeability for values of kf > kd (Figure 3), because for sufficiently high conduit permeabilities, the flux of fluids along the décollement and splay faults is limited by upward flow of fluids from the low-permeability subducting sediments. We anticipate that for higher underthrusting sediment permeability, the same patterns would persist, but the faults could tap a larger proportion of the total fluids.
 In general, with increased splay fault permeability, the proportion of fluid exiting the forearc along the décollement and via diffuse flow both decrease, because the splay faults capture a larger fraction of fluid from the dewatering subducting sediments. This effect is also partly mediated by the décollement permeability; with higher kd, splay faults access and intercept more fluid from the décollement. This decreases flow rates along the décollement, but results in only a small impact on diffuse flow. In contrast, for lower values of kd, the splay faults access less fluid directly from the décollement, because flow to the splays along the décollement is the limiting factor (Figure 3). In this case, the splay faults tap fluids that would otherwise exit the forearc via diffuse flow.
4.2. Fluid Flow Rates
 Modeled seafloor flow rates demonstrate the efficiency of faults in channeling fluid, heat and solutes away from the plate boundary (Figure 4). For cases with kd = 10−14m2, simulated seafloor seepage rates at the splay faults range from 0.08 cm yr−1 (kfaults = 10−16 m2) to 2.6 cm yr−1 (kfaults = 10−12 m2). For lower décollement permeability (kd = 10−15 m2) the flow rates range from ∼0.2 to 4.4 cm yr−1. In models that do not include slope sediment, simulated peak flow rates are considerably higher (up to 25 cm yr−1), because flow at the seafloor remains focused along the structures rather than being diffused by the less permeable slope sediment. The simulated flow rates are in good agreement with rates of 0.3 – 20 cm yr−1 inferred from geochemical profiles and heat flow measurements at seep sites (cf. Figure 1c) [Hensen et al., 2004; Ranero et al., 2008]; only the most active site of fluid expulsion exhibited a local flow rate significantly higher than our simulations (300 cm yr−1 [Hensen et al., 2004]). It is important to note that the flow rates reported at seep sites reflect significant focusing of fluid flow along-strike (i.e., these rates are not sustained at all locations along a given fault outcrop) [e.g., Ranero et al., 2008]. As such, the reported rates should be considered maxima, whereas those we simulate in our 2-D model, which does not account for any out-of-plane effects, represent mean flow rates per unit length along-strike (Figure 4). Modeled flow rates within the décollement near the trench are also consistent with rates of 1.4–45 cm yr−1 and 0.02–5 cm yr−1 at Sites 1043/1255 estimated from drilling data [Saffer and Screaton, 2003] and measured by a borehole flowmeter [Solomon et al., 2009], respectively (Figure 4). Taken together, our results suggest that splay fault permeabilities are likely in the range ∼10−12 to 10−14 m2. In all of our models, rates of diffuse flow at the seafloor are <0.01 cm/yr over the Nicoya Complex, and range from 0.02 to 0.2 cm/yr over the frontal prism.
 Simulated flow rates along the splay faults are sufficiently high to advect solutes from the plate boundary at depth to the seafloor, and are up to ∼600 times higher than in adjacent areas (Figure 4). Geochemical signals associated with deeply sourced reactions should therefore be effectively channelized, with strong and highly localized expression where splay faults intersect the seafloor, or near structures with an underlying hydraulic connection to the plate boundary (e.g. mud mounds, mud volcanoes). Our results are consistent with patterns of focused flow through the upper plate inferred from geochemical data [e.g., Teichert et al., 2005; Ranero et al., 2008], which document an increase in the proportion of chemically distinct deeply sourced fluids exiting the forearc with distance landward from the trench. This pattern is readily explained by splay faults that capture fluids originating at or below the plate interface.
 The fate of fluids released by dewatering of subducted sediments has been assessed previously using numerical models, but generally assuming a simplified permeability architecture in which fluids reach the seafloor via diffuse flow through the overriding plate or along a highly permeable décollement. Recent observations of geochemical and thermal anomalies on the continental slope have underscored the importance of splay faults that cut the upper plate as key dewatering pathways that transport deeply sourced fluids to the seafloor across the forearc, especially at non-accretionary margins [e.g., Hensen et al., 2004; Ranero et al., 2008].
 We use a numerical model of fluid flow at the Costa Rica convergent margin to systematically investigate the role of splay faults in forearc hydrogeology. Our results show quantitatively that splay faults exert a primary control on the partitioning of fluid expulsion. Simulated flow rates are consistent with flow rates determined from field measurements, and are sufficiently high that the faults should efficiently advect heat and solutes through the forearc to sites of focused expulsion at the seafloor. One key implication is that these structures play a fundamental role in fluid, heat, and chemical transport through the forearc, with implications for non-accretionary convergent margins globally. Our results also provide a quantitative framework for interpreting observed flow rates and geochemical anomalies, by linking surface observations with the underlying physical hydrogeology.
 This work was supported by NSF grants OCE-1100599, 0752114, and 0623633 to D.S. We thank the two anonymous reviewers, whose comments and suggestions have greatly improved the manuscript.
 The Editor thanks two anonymous reviewers for their assistance in evaluating this paper.