Sandy trench-fill sediments at accretionary margins are commonly scraped off at the frontal wedge and rarely subducted to the depth of high-pressure (HP) metamorphism. However, some ancient exhumed accretionary complexes are associated with high-pressure–low-temperature (HP-LT) metamorphic rocks, such as psammitic schists, which are derived from sandy trench-fill sediments. This study used sandbox analogue experiments to investigate the role of seafloor topography in the transport of trench-fill sediments to depth during subduction. We conducted two different types of experiments, with or without a rigid topographic high (representing a seamount). We used an undeformable backstop that was unfixed to the side wall of the apparatus to allow a seamount to be subducted beneath the overriding plate. In experiments without a seamount, progressive thickening of the accretionary wedge pushed the backstop down, leading to a stepping down of the décollement, narrowing of the subduction channel, and underplating of the wedge with subducting sediment. In contrast, in experiments with a topographic high, the subduction of the topographic high raised the backstop, leading to a stepping up of the décollement and widening of the subduction channel. These results suggest that the subduction of stiff topographic relief beneath an inflexible overriding plate might enable trench-fill sediments to be deeply subducted and to become the protoliths of HP-LT metamorphic rocks.

High-pressure–low-temperature (HP-LT) metamorphic rocks derived from terrigenous sedimentary rocks are known to occur at subduction margins (e.g., Agard et al., 2009; Guillot et al., 2009). Such metamorphic rocks are exposed alongside low-grade accretionary rocks and forearc basin strata that include coarse-grained sandy deposits with the same depositional ages as the metamorphic rocks. For example, the Sanbagawa metamorphic complex in southwestern Japan contains HP-LT psammitic and even conglomeratic schists, and the depositional ages and geochemical characteristics of the protolith are almost identical to those of sandstone from the low-grade Shimanto accretionary complex (Kiminami et al., 1999; Shibata et al., 2008; Aoki et al., 2012) and submarine fan turbidites deposited in the associated forearc basin (Fig. 1; Noda and Sato, 2018). These observations indicate that terrigenous trench-fill sediments were accreted in a shallow subduction zone and were also subducted to >20 km depth. Other examples of such subduction-accretion–related HP-LT metamorphic rocks can be seen in the Franciscan complex in California (e.g., Hsü, 1974; Ernst, 2011; Jacobson et al., 2011; Dumitru et al., 2015; Raymond, 2018), the Chugach terrane in Alaska (Plafker et al., 1994), the Central Pontides in Turkey (Okay et al., 2006), and the Coastal Cordillera in Chile (Glodny et al., 2005; Willner et al., 2004; Angiboust et al., 2018).

At typical sedimentary accretion zones, such as those in Cascadia (Gulick et al., 1998; Booth-Rea et al., 2008; Calvert et al., 2011), Alaska (Moore et al., 1991; Ye et al., 1997), Java (Kopp et al., 2009), southern Chile (Glodny et al., 2005; Melnick et al., 2006), Sumatra (Singh et al., 2008; Huot and Singh, 2018), and Japan (Park et al., 2002b; Kimura et al., 2010), terrigenous trench-fill sediments are generally scraped off at the frontal wedge, whereas hemipelagic-to-pelagic sediments underplate the base of the accretionary wedge (e.g., Scholl, 2020). This may be because the increased structural thickness of the wedge and progressive dewatering of subducting sediment caused the upper boundary of the subduction channel to step down (e.g., Sample and Moore, 1987; Vannucchi et al., 2012). This suggests that the growth of the accretionary wedge might inhibit the subduction of terrigenous sediment beyond the wedge through the subduction channel. However, occurrences of HP-LT metasandstone at some accretionary margins demonstrate that terrigenous sediment can be subducted beneath the wedge. One hypothesis is that a stiff topographic high enables trench-fill sediments to be subducted under the wedge (Fig. 2). In fact, subducting seamounts followed by subducting material can be observed beneath the wedge along accretionary margins in southwestern Japan (Moore et al., 2014), Alaska (Li et al., 2018), Barbados (Moore et al., 1995), and Hikurangi (Barker et al., 2009; Bell et al., 2010).

The subduction of terrigenous material associated with the rough topography of a subducting oceanic plate has been proposed to explain tectonic erosion of the wedge (e.g., von Huene and Culotta, 1989; Lallemand et al., 1994; von Huene et al., 2004). Sandbox analogue experiments (Lallemand et al., 1992; Dominguez et al., 2000) and numerical simulations (Ruh, 2016; Morgan and Bangs, 2017) have shown the potential for sediment transport below the frontal wedge behind a subducting topographic high. In addition, recent seismic profiles across the accretionary margins of the Nankai Trough (Bangs et al., 2006) and the Hikurangi Trench (Bell et al., 2010) indicate that subducting seamounts or ridges and the surrounding sediment are accommodated by a step up in the décollement, and the surrounding sediment is being transported to depth.

However, the influence of a subducting seamount beneath an accretionary wedge on subduction and accretion fluxes is not well understood. In particular, the role of topographic highs and backstops in modifying the décollement level and in maintaining or rejuvenating the subduction channel as a conduit for sediment subduction needs to be explored. The purpose of this study was (1) to investigate how the topographic roughness of the subducting plate interface influences material fluxes, including the accretion of sediment to the wedge and the subduction of sediment along the subduction channel, and (2) to propose a model that explains how terrigenous trench-fill sediments can be transported to depth. We performed two types of sandbox analogue experiment, one with and one without a topographic high.

Although there are several experiments focusing on seamount subduction (Lallemand et al., 1992; Dominguez et al., 1998, 2000) and material transfer beneath the wedge (Kukowski et al., 1994; Gutscher et al., 1998; Lohrmann et al., 2006; Albert et al., 2018), experiments with a combination of these two factors have not yet been reported to date. In order to reproduce the subduction of a rigid topographic high beneath an accretionary wedge and its associated underplating of sediment, we used an unfixed backstop to the side wall of the apparatus, which is another novelty of this study. We also inserted two weak layers within the sand, to reproduce the situation where the subducting sediment includes several potential slip surfaces. Multiple weak layers are considered to affect the evolution of tectonic underplating and related antiformal stacking (cf. Ruh et al., 2012).

Here, for the sake of clarity of the terminology, we briefly review the term “subduction channel.” A subduction channel model (Shreve and Cloos, 1986; Cloos and Shreve, 1988a, 1988b) was originally introduced from hydrodynamic theory of lubrication in the slip zone between overriding (accretionary wedge or continental crust) and subducting (oceanic) plates (Jischke, 1975; Sorokhtin and Lobkovskiy, 1976). Flow of sediments in this zone is formed by the superposition of a pure shear flow arising as a result of the relative movement of the surfaces bounding the channel on the flow (Sorokhtin and Lobkovskiy, 1976). Because the shear flow is governed by the pressure in this model, material within the subduction channel is modeled by a viscous rheology with constant density, uniform viscosity, and with a permeability depending on porosity (Shreve and Cloos, 1986; Cloos and Shreve, 1988a). Sediment subduction occurs through the channel when the downward-directed shearing overcomes the buoyancy and the opposite-directed pressure gradient.

Cloos and Shreve (1988a) introduced the concept of “inlet,” which means the opening between the deformation front and the toe of the overriding plate or the accretionary wedge, and its capacity controls the subduction channel thickness. When the inlet capacity is lower than the incoming sediment on the subducting oceanic plate, the excess will accrete to the overriding plate. This idea of capacity and the subduction channel model were revised by von Huene et al. (2009) and Vannucchi et al. (2012) to explain tectonic erosion from the bottom of the overriding plate or accretionary wedge based on field and seismic observations of localized shear zones with fault-like slip at both top and bottom boundaries of the subduction channels (cf. Kitamura et al., 2005; Sage et al., 2006; Vannucchi et al., 2008; Collot et al., 2011). In their models, the subduction channel can change its upper and lower boundaries with time and space depending on material supply, fluid content, and heterogeneity of deformation. For example, rheological effects from the progressive dewatering and consolidation of channel material can deactivate the basal décollement and weaken the upper part of the subduction channel with respect to the bottom, leading to upward migration of the roof décollement (cf. von Huene et al., 2004). On the other hand, dewatering from the upper boundary of the subduction channel through fracture networks related to seamount subduction (e.g., Dominguez et al., 2000; Ruh et al., 2016) can reduce the fluid pressure and increase the strength of the sediment below the upper boundary, which results in underplating of the dewatered channel material (Shreve and Cloos, 1986; Cloos and Shreve, 1988a) and downward migration of the upper décollement.

The concept of the subduction channel has been used to explain exhumation of HP and ultrahigh-pressure (UHP) metamorphic rocks due to buoyancy and returning flow within the subduction channel filled with mechanically weak sediments and sedimentary rocks (England and Holland, 1979; Gerya et al., 2002; Jolivet et al., 2003; Agard et al., 2009; Guillot et al., 2009). Such rheological behavior within the channel would be expected for exhumation of HP metamorphic rocks from deep (>15 km) environments.

In this study, we use the term “subduction channel” to mean a zone bounded by the upper bounding fault (décollement) and the top of the oceanic plate. In other words, it represents the capacity of the subduction gap through which sediments are transported beneath the accretionary wedge and the backstop (cf. Cloos and Shreve, 1988a). Both upper and lower boundaries can shift their positions with time and space depending on the topography of the overriding and subducting plates and thicknesses and properties of the incoming sediment. Similar meanings of this term can be found in previous studies of analogue experiments (Kukowski et al., 1994; Gutscher et al., 1998; Lohrmann et al., 2006). The subduction channel here does not presume any effects of buoyancy or returning flow of the sediment within the channel. In this sense, our “subduction channel” is comparable to the “plate interface” of Agard et al. (2018) and different from some interpretative meanings of subduction channel such as the idea representing forced return flow in the channel (e.g., Gerya et al., 2002).

Experimental Apparatus and Materials

A scaled two-dimensional (2-D) analogue model was established for this study so that the results could be compared with naturally occurring geological structures (e.g., Graveleau et al., 2012). A glass-sided rectangular deformation rig with internal dimensions of 100 cm × 30 cm × 20 cm was used (Fig. 3). A steel plate was positioned at one end as a fixed wall with a small open window (1.6 cm in height) at the bottom. A rigid wooden protowedge was placed next to the steel plate but was not fixed to it; it was just loosely lain atop of the sand layer. The wedge was designed to have a higher mechanical strength than the accretionary wedge. The reason we used a rigid wooden backstop was to ensure stability during the experiments and for repeatability. The mobility of the backstop helped to replicate the movable nature of equivalent structures in natural geological systems, and to allow topographic relief to be subducted. The backstop had a surface slope that dipped at 30° and was covered by sandpaper. A plastic (Mylar®) sheet was placed over the rig’s base plate and fixed to a roll that pulled the sheet using a stepper motor (on the left side in Fig. 3). The sheet was pulled beneath the rigid backstop at a rate of 0.5 cm/min, thereby compressing the experimental material above.

Two types of granular material were used for the experiments: Toyoura sand and glass microbeads. Dry granular materials like these are widely used as analogue materials to simulate the nonlinear deformation of crustal rocks under brittle condition, because they display an elastic/frictional plastic behavior with transient strain hardening prior to failure and subsequent strain softening until the onset of stable sliding at constant friction (Lohrmann et al., 2003). Toyoura sand, a standard testing material commonly used by Japanese civil engineers, is a spherical quartz-rich sand with a particle size of 0.14–0.26 mm (D50 = 0.2 mm), a density of ∼1600 kg m−3, an internal coefficient of friction, μ, of 0.59–0.68, and a cohesion, C, of 105–127 Pa (Yamada et al., 2006; Dotare et al., 2016). The glass microbeads are spherical and 0.045–0.063 mm in diameter, have a low internal coefficient of friction (μ = 0.47) and low cohesion (C = 40 Pa), and are considered to be a suitable analogue for weaker layers (Yamada et al., 2006, 2014). The theoretical minimum critical taper angle of the wedge ranges in between 8.7° and 10.6° (cf. Dahlen, 1984).

Layers of sand and glass microbeads with a total thickness of 3.4 cm were used in the experiments. The sand and glass were sprinkled into the rig from a height of ∼30 cm above the rig floor (Fig. 3). Alternating layers of blue, red, and black sand were laid down to help visualize the cross-sectional geometry of the models, without influencing the mechanical homogeneity.

Mechanically, weak layers were created by adding two thin layers of glass microbeads, each 3 mm thick. These glass-bead layers mimicked multiple décollement layers with a lower Mohr-Coulomb failure criterion than the Toyoura sand of the wedge-forming material. The reason we used two weak layers was to reproduce an occurrence of underplating or duplex structures during subduction, which have been investigated by analogue (Konstantinovskaya and Malavieille, 2011; Pla et al., 2019) and numerical studies (Ruh et al., 2012). Such multiple décollements are commonly found within subducting sediments or at the top of the oceanic crust, including at the Nankai (Moore et al., 2001; Park et al., 2002b), Hikurangi (Ghisetti et al., 2016; Plaza-Faverola et al., 2016), and Barbados (Saffer, 2003) accretion zones.

Experimental Setup

Experiment A investigated the subduction of a smooth oceanic plate beneath a static backstop (Fig. 3A). Experiment B investigated the subduction of topographic relief (e.g., a seamount), using a 5-cm-width and 1.6-cm-height wooden block that was attached to the plastic sheet (Fig. 3B). The height of the relief was approximately half of the total thickness of the sediment, which was chosen to avoid drastic deformation of the accretionary wedge. The surface of the topographic relief was covered by a Teflon® sheet. The total amount of horizontal shortening was 30 cm for experiment A and 35 cm for experiment B.

After each 2 cm increment of shortening (every 4 min), we sprinkled dry sand from at least 10 cm above the surface of the accretionary wedge to fill the topographic lows that had developed. The amount of sand input at each supply was varied with repetitions in a cycle of 70, 9, 9, 70, 131, and 131 g in order. Total numbers of sediment supply were 13 and 17 times, meaning repetitions of 2 and 3 cycles for experiment A and experiment B, respectively. When we supplied sand, we first filled the retroside of the wedge while it was underfilled, and then the rest of the sand was sifted in the wedge top. If sand still remained, the rest was spread at the front of the wedge. This sand was used to replicate sedimentation in forearc/slope basins that form on the surfaces of accretionary wedges. A total amount of 910 g of sand was added over the course of experiment A and 1129 g during experiment B. The volumes of sand added during experiments A and B were 569 and 706 cm3, respectively.

In addition to investigating wedge morphology, we studied temporal variations in sediment influx/outflux. The sediment influx and outflux (cm2) were calculated using the thicknesses (cm) of the trench-fill sediments (influx) and the subduction channel underneath the backstop (outflux), which were multiplied by the rig width (30 cm) and divided by the length of shortening (cm). Input and output (cm3) are here defined to be the integrals of influx and outflux, respectively, with respect to shortening length (cm). Time-lapse digital images were taken through the transparent side glass at 5 s intervals using a personal computer–based controller. The images were later analyzed to calculate sediment influx/outflux values and to study the cross-sectional geometry of the wedges. The experiments did not account for the effects of isostatic compensation and erosion, which would have contributed to the differences between our models and natural examples (e.g., Schellart and Strak, 2016).


Models used in laboratory experiments should be properly scaled so that the results can be considered true analogues of geological processes (e.g., Hubbert, 1937). It is assumed that brittle deformation will obey frictional Mohr-Coulomb–type laws. The basic scaling relationship between the physical properties of a model and those in nature, which relates the stress, σ, density, ρ, gravity, g, and length, l (Hubbert, 1937; Schellart, 2000), is
where the subscripts m and g indicate model and geological values, respectively. The cohesion C can substitute for stress, σ (Schellart, 2000; Graveleau et al., 2012), and the experiments were performed under normal gravity (gmgg = 1); consequently, Equation 1 can be modified to give

For mean bulk density values of 2000–2500 kg m−3 and cohesion values of 5–20 MPa, which are typical of sedimentary rocks in accretionary wedges (Schumann et al., 2014), the length scale ratio ranges from ∼3 × 104 to 1.5 × 105. Details of the scaling for the materials used in this study can be found in Noda et al. (2020b).

The 5 cm width and 1.6 cm height of the topographic relief used in experiment B can be scaled to 1.5–7.5 km and 0.5–2.4 km, respectively. The scaled dimensions of the topographic relief are comparable to many seamounts on the Pacific plate. However, the height-to-radius ratio of 0.64 in the model is higher than that of 0.21 for natural seamounts (Jordan et al., 1983; Smith, 1988). This high ratio was used to enhance the effects of topography. The total amount of shortening during the experiments was 30–35 cm, which is equivalent to 9–53 km of displacement. Assuming a plate convergence rate of 5 cm/yr, this in turn corresponds to 0.2–1.1 m.y. The sediment supply to the topographic lows of 910–1129 g for 1 m.y. is equivalent to a sediment budget on the order of 106 t/yr. This calculated sediment budget is the same order of magnitude as the sediment load in many mountainous rivers in Japan and New Zealand (Milliman and Syvitski, 1992), and the sedimentary influx into the Kumano Basin (50 km × 70 km × 2 km) during the last 4 m.y.

Experiment A: Subduction without a Seamount

During the first ∼9 cm of shortening, high-frequency, low-amplitude forethrusts developed in front of the backstop (8 cm of shortening in Figs. 4 and 5A; supporting information1). We defined this term as stage 1 and the following term as stage 2, which was characterized by low-frequency, high-amplitude forethrust activity (Fig. 5A). The rate of wedge widening (wedge width divided by the length of shortening) was nearly constant through stages 1 and 2 (0.21–0.22 cm/cm; Fig. 5B).

The wedge uplifting rate (uplift rate, which was calculated from height of the wedge divided by the length of shortening; Fig. 5D) was large (0.34 cm/cm) in stage 1, and thus the slope increased rapidly, exceeding 12° by the end of stage 1 (Fig. 5C). After the emergence of T6 (stage 2), the frequency of forethrust initiation and the uplift rate of the wedge (0.10 cm/cm) were lower than during stage 1. The slope of the wedge surface ranged from 8.5° to 13° and was 9.5° at the end of the experiment (Fig. 5C).

Deformation was concentrated in the upper layer of glass beads, which acted as a décollement, until 16 cm of shortening (Figs. 4 and 6A). At around 18 cm of shortening, the décollement stepped down to the lower layer of glass beads as the toe of the backstop subsided below the upper layer of glass beads (Figs. 6A and 7). During this stage, the footwall of forethrust T7 underthrust the wedge, and the sand layer between the two layers of glass beads underplated the wedge, creating a duplex structure (18–24 cm of shortening in Figs. 4 and 6A). This underthrusting raised the hanging wall of T7 and created a steep (25°) slope and a wedge-top basin (trench-slope basin) on top of the wedge (22 cm of shortening in Fig. 6A). After the activation of T8, with the lower layer of glass beads acting as a décollement, subducting sediment was accreted to both the frontal and basal parts of the wedge, with an increasing amount of underplating and thickness of the forethrust sheet of T8. The final forethrust, T9, was initiated with the upper layer of glass beads acting as the detachment (30 cm of shortening), but the lower layer of glass beads and even the surface of the plastic sheet were used as the sliding planes below the middle-inner wedges (Fig. 6A). The final wedge was nearly 30 cm in length and had a constant slope of 9.5° (Fig. 5). The toe of the backstop further subsided (Fig. 7) to the lower layer of glass beads (30 cm of shortening in Fig. 4).

Experiment B: Subduction with a Seamount

Stage 1 of experiment B was almost identical to that of experiment A in terms of wedge progradation and the widening and uplift rates of the wedge (Fig. 5; supporting information [footnote 1]). Stage 2 started after the initiation of forethrust T5, earlier than in experiment A. T5 was active for over 12.8 cm of shortening, exceeding that of any other forethrust in either experiment (Fig. 5A). This long activity acted to reduce the width of the wedge and created a wedge-top basin (Fig. 6B). This also steepened the frontal edge up to 30° (Fig. 6B) and the averaged forearc slope to 17.7° (19 cm of shortening in Fig. 5C). The wedge progradation rate during stage 2 was 0.10 cm/cm, i.e., nearly half that of experiment A (Fig. 5A). The uplift rate varied from 0.06 to 0.29 cm/cm, but the mean rate was the same as in experiment A (Fig. 5B).

The wedge deformation process during stage 1 of experiment B was similar to that in experiment A (0–6 cm of shortening in Fig. 8). However, at 7 cm of shortening, the seamount triggered the first forethrust of stage 2 at 10 cm from the toe of the wedge (T5 in Fig. 8). The subduction of the seamount led to an undeformed layer underthrusting the wedge, and then uplifted the hanging wall as a trench-slope basin to create accommodation space (10–16 cm of shortening in Figs. 6B and 8).

A décollement was formed in the upper layer of glass beads on the landward side of the seamount and in T5 on the trenchward side during the period between the initiation of T5 and collision of the seamount with the backstop (8–12 cm of shortening in Fig. 8). Just prior to the collision (12–18 cm of shortening), both the upper and lower layers of glass beads were sliding, and the sand layer between two layers of glass beads was underplated and injected into T5.

The collision of the seamount with the backstop stepped up the décollement to the same height as the top of the backstop, leading to thickening of the subduction channel (>20 cm of shortening in Figs. 6B, 7, and 8). This also uplifted the accretionary wedge and increased the slope up to the angle of repose (36°; Fig. 6B). The subsequent forethrusts, T6 and T7, were rooted in a décollement in the upper layer of glass beads beneath the wedge. After passing through the seamount, a certain range of shear zone was created just below the backstop. Finally, the toe of the backstop subsided slightly (Fig. 7), causing the lower layer of glass beads to act as a décollement (35 cm of shortening in Fig. 8).

Sediment Flux

Experiment A: Subduction without a Seamount

The outflux from the subduction channel (sediment subduction) gradually decreased, but its rate of change increased as the backstop progressively subsided (Figs. 7 and 9A). In particular, after the backstop position was lower than the upper glass bead layer (17 cm of shortening in Figs. 7 and 9A), corresponding to a step down of the décollement, the outflux dropped rapidly. Influx to the accretionary wedge (dashed line in Fig. 9A) increased to balance the total sediment influx. The output-to-input ratio of the experiment was 0.36 (Table 1).

Experiment B: Subduction with a Seamount

Sediment outflux gradually decreased (blue line in Fig. 9B), as it did during experiment A, until the seamount reached the backstop. After the seamount raised the backstop, at around 18–20 cm of shortening (Figs. 7 and 9B), sediment outflux fully recovered and even exceeded its initial rate (Fig. 9B). Outflux soon decreased again as the seamount subducted farther landward and the backstop subsided (Figs. 7 and 9B). The output-to-input ratio of the experiment was 0.46 (Table 1).

Décollement Step Down and Underplating

The gradual decrease of the outflux in experiment A (Fig. 9A) increased the influx to the accretionary wedge, which raised its growth rate. During the time the upper layer of glass beads acted as a décollement, the sediment above it was accreted to the wedge front. As the slip switched to the lower layer of glass beads, the sediment between the two layers of glass beads underplated the wedge, and frontal accretion continued. Similar results have been reported in previous analogue experiments (Bonnet et al., 2007; Konstantinovskaya and Malavieille, 2011) and numerical experiments (Ruh et al., 2012); i.e., underplating becomes significant when the outflux from the subduction channel (sediment subduction) is smaller than the influx (Kukowski et al., 1994). The results of our experiment support the conclusion that a narrowing of the subduction channel and a decrease in outflux can lead to sediment underplating the wedge and faster wedge growth.

If we assume that sand above the upper layer of glass beads is terrigenous sediment, and that sand below this layer is hemipelagic-pelagic sediment, the former can be scraped off at the wedge front, and the latter may be underplated below the wedge. This occurs because terrigenous and hemipelagic sediments tend to be detached as a result of variations in diagenetic alteration (Moore, 1975) or smectite content (Vrolijk, 1990; Deng and Underwood, 2001), or existence of weak smectitic pelagic clay (Moore et al., 2015). This can be observed in the Nankai Trough, where there is a step down in the décollement at 3–5 km depth, in the transitional region between the aseismic and seismic zones (Fig. 10A; Park et al., 2002b; Kimura et al., 2007). This can be ascribed to lithification that concurrently occurred with deformation across the upper boundary of the subduction channel, stepping down the décollement to the lower boundary of the subduction channel (Kimura et al., 2007).

The stepping down of the décollement in this study was associated with subsidence of the backstop, which was probably linked to increased overburden stress caused by thickening of the wedge. Increased overburden stress may inhibit the subduction of terrigenous sediment to great depth. An example from the Ecuador-Colombia margin (Fig. 10B) shows that underplating and progressive seaward migration of duplexes are considered to be related to the splay faults that force the underlying décollement thrust to step down (Collot et al., 2008).

Décollement Step Up and Sediment Subduction

In experiment B, the subduction of the seamount shifted the décollement from the glass bead layers into forethrust T5 along the leading flank of the seamount. While T5 was active as a “top décollement” (cf. Lallemand et al., 1994), incoming undeformed layered sand in the wake of the seamount was underthrust below the accretionary wedge. This is similar to what is seen in seismic profiles in Figure 2, which show a décollement with a step up caused by seamount subduction.

Another effect of seamount subduction in experiment B is that raising the wedge and the backstop widened the subduction channel, allowing thick layers of sediment to subduct below the backstop through the subduction channel (Figs. 10B–10D). In nature, if an oceanic plate with sufficiently large and rigid topographic highs subducts under a static backstop (cf. Tsuji et al., 2015), trench-fill terrigenous sediment accompanying the highs will be transported through the subduction channel to a higher-pressure environment than sediment on a smooth oceanic plate. Experiment B may be analogous to the transport mechanism of the protolith of the ancient Sanbagawa metamorphic and Shimanto accretionary complexes of southwestern Japan.

Where the trench-fill sediments are insufficient to fully cover the topographic relief of the subducting oceanic crust, tectonic erosion may dominate, and the accretionary wedge cannot grow, as seen in northeastern Japan, Costa Rica, and Ecuador (von Huene et al., 2004; Collot et al., 2011) and suggested by other experiments (Dominguez et al., 1998; Ruh, 2016). The field example from southwestern Japan shows that terrigenous sediments can be transported from shallow depths (e.g., the Shimanto accretionary complex) to the depth of HP metamorphism (e.g., the Sanbagawa metamorphic complex) almost coincidentally. The subducting seamount in our experiment, which was totally buried by trench-fill sediments, reproduced frontal accretion and sediment subduction with the seamount at about the same time. Therefore, a sediment-rich subduction zone may be required for terrigenous sediments to accrete and underplate at different depths.

Comparison with Previous Experiments

Previous analogue experiments of seamount subduction showed significant impacts on the geometry and deformation of accretionary wedges (Lallemand et al., 1992; Dominguez et al., 1998, 2000). Those impacts include that seamount subduction developed (1) back thrusts in the landward side of the subducting seamount, (2) dense networking of faults within the overriding wedge, (3) normal faults in the trenchward side, (4) thickening and shortening of the wedge with a slope break, and (5) a shadow zone behind the seamount trailing slope. In particular, item 5, the shadow zone, will be important to transport trench-fill sediments and parts of the wedge front beneath the wedge (Dominguez et al., 2000). Numerical simulations also suggest fracturing or faulting atop of the seamount and the presence of a shadow zone in the wake of the seamount (Ruh, 2016; Ruh et al., 2016). In addition, a subducting seamount can generate a large overpressure above the landward flank and underpressure above the trenchward flank, producing a drop in vertical compressional stress on the trenchward side of the seamount (Baba et al., 2001; Ruh et al., 2016).

The major difference in our study from these previous experiments of seamount subduction is that we used a rigid but unfixed backstop. Dominguez et al. (2000) used a cohesive (deformable) backstop that allowed the décollement to return to its original position just after the seamount passed due to the overburden stress. This indicates that it is difficult for the flexible backstop to maintain the subduction channel, if the overriding plate readily collapses gravitationally along the trenchward flank of the seamount. The rigid backstop in our study kept a certain thickness of the subduction channel during the passage of the seamount beneath the backstop, and this enabled the sediment to be transported more than the deformable backstop. It is also suggested that the strength of the seamount must be high enough to support the weight of the backstop.

Another difference in our study is multiple décollements within the incoming layer. Due to progressive subsidence of the backstop (Fig. 7), sediment underplating with duplex structures was successfully reproduced within the two glass bead layers in experiment A without a seamount (Fig. 6A). The effect of decreasing outflux relative to influx for underplating has been indicated by previous experiments. Kukowski et al. (1994) pointed out that a required condition for underplating is a ratio of incoming to outgoing material that is significantly larger than unity. Our results agree with this point (Fig. 9).

In addition, experiments presented here simultaneously reproduced frontal accretion and underplating by using multiple décollements in experiment A. This suggests that a stepping down of the décollement is a key factor for underplating (cf. Malavieille, 2010). On the other hand, experiment B with a seamount mostly used the upper glass bead layer and the sand layer at the same height as the top of the seamount as the décollement, because the height of the seamount was larger than the height of the upper glass bead layer. Therefore, if the height of the topographic relief is higher than any weak layers within the incoming sediment, the upper boundary of the subduction channel will be determined by the height of the topographic relief.

Subduction of Trench-Fill Sediments to Depth

Here, we propose a schematic model for the subduction of terrigenous sediment to depth (Fig. 11). A progressive thickening of the wedge increases the overburden on the décollement, which develops along weak layers in the cover sediment deposited on the subducting oceanic plate. This overburden results in dewatering and diagenetic alteration of the subducting sediment, which increases its mechanical strength, leading to a step down in the décollement. The reduction of sediment outflux due to narrowing of the subduction channel increases the mass of sediment underplated beneath the wedge and the rate of frontal accretion. When a topographic high (e.g., a seamount or an aseismic ridge) subducts under the wedge, the décollement steps up to the forethrust along the leading flank of the seamount (Figs. 11A and 11B). This likely enables the subduction of terrigenous sediment beneath the wedge. Further subduction of the topographic high would raise the backstop and open the subduction channel for terrigenous sediment to be subducted into a high-pressure environment (Fig. 11C). After the topographic high passes the inner wedge or backstop, the décollement under the accretionary wedge returns to the plate boundary or a weak layer within the trench-fill sediments.

We speculate about the ways in which the trench-fill sediments can be subducted deeper (>15 km of burial) to be protolith of HP-LT metamorphic rocks. As shown in this study (experiment B), the strength of the subducting seamount is crucial to maintain the subduction channel for sediment transfer to the deep interior. Although common isolated seamounts composed of volcaniclastic rocks and hyaloclastite with various degreed of alteration can be shred off during the shallow stage of subduction, some of them can be subducted into the HP environment (Ueda, 2005; John et al., 2010; Bonnet et al., 2019a, 2019b). Alternatively, a large-scale aseismic ridge and horst composed of oceanic crust may have sufficient mechanical strength against the overriding plate and open the subduction channel for sediments. In such a situation, trench-fill sediments will be transported deeper than the updip limit of the seismogenic zone (zone 4, 5–15 km of vertical burial; Vannucchi et al., 2012).

However, at some depth, when the shear force along the décollement has increased and reached the mechanical strength of the topographic highs (i.e., seamount), the topographic highs and the sediments will be sheared off, stacked, and underplated. Although surface erosion is one of the major driving forces for exhumation of HP metamorphic rocks (e.g., Malavieille, 2010; Malavieille and Konstantinovskaya, 2010), the underplating of a seamount may be another trigger. There is an ancient example of stacking of seamounts and clastic sequences under the HP-LT environment in the Lower Cretaceous Iwashimizu accretionary complex in the Kamuikotan HP-LT zone, Hokkaido, Japan (Ueda, 2005). The metavolcanic sequence, composed of hyaloclastite, pillow breccia, and pillow lava (Pirashuke Unit), is considered to have been subducted under a pressure of 0.5–0.6 GPa and temperature of 200 °C, suggesting depths of ∼20 km. This unit is associated with a clastic sequence consisting of sandy turbidite and mudstone, and even conglomerate (Shizunai Unit). Based on the downward progression of underplating accretion and higher pressure in the structurally upper unit as compared to the structurally lower units, Ueda (2005) suggested that structurally upper units were exhumed synchronously with growth of the structurally lower units.


Our experiments did not incorporate effects of pore pressure, but it is important in maintaining subduction channels along the plate interface (e.g., Saffer and Bekins, 2006). As the fluids escape from the subducting material in the subduction channel, a significant part of the overburden load is transferred to the solid fraction of the subducting material, and the mechanical behavior gradually changes to that of a consolidated media with a continuous solid matrix (cf. Nankai and Barbados; Saffer, 2003; Calahorrano et al., 2008). This probably accelerates both the stepping down of the décollement and underplating (Strasser et al., 2009; Kimura et al., 2011). In contrast, numerical simulations predict that the raising of the wedge due to the subduction of a seamount could delay the release of fluid from subducting sediment (Baba et al., 2001; Ruh, 2016). Low-velocity layers observed in the wake of subducting seamounts could provide evidence of undercompacted sediment with potentially high pore pressures (e.g., Sage et al., 2006). Furthermore, the seismic reflection characteristics of the Hikurangi subduction margin also suggest localized reductions in effective stress associated with seamount subduction (Bell et al., 2010). In addition to topographic relief, pore pressure can allow subduction channels to persist for longer than would otherwise be possible. Our experiments cannot currently incorporate such effects of pore pressure; consequently, we need to consider ways to include these effects.

We conducted a series of analogue experiments to investigate how terrigenous sediment is subducted under an accretionary wedge. The results yielded the following conclusions.

  • (1) An increase in overburden stress due to progressive thickening of the accretionary wedge leads the décollement to step down and narrows the subduction channel. This accelerates the growth of the wedge through underplating and frontal accretion.

  • (2) When a topographic high subducts under the wedge, the décollement steps up from a weak detachment layer within the incoming sediment to the forethrust along the landward flank of the seamount. This enables terrigenous sediment in the wake of the seamount to be underthrust beneath the wedge.

  • (3) If a topographic high is rigid enough to uplift the backstop, and the backstop is mechanically stable, it can widen the subduction channel to transport the terrigenous sediment that follows toward a deeper environment.

  • (4) A sufficient sediment supply to the trench and a rough oceanic crust surface are necessary for simultaneous shallow accretion, underplating of the wedge, and transportation of sediment to deeper settings as the protolith of HP-LT metamorphic rocks.

We are grateful to Takato Takemura for useful suggestions regarding experimental material and to Yujiro Ogawa for discussions about sediment subduction and accretion. This work was funded by a Grant-in-Aid from the Japan Society for the Promotion of Science (17K05687). This research was also supported by the Cooperative Program (no. 143, 2017; no. 147, 2018; no. 151, 2019) of the Atmosphere and Ocean Research Institute, The University of Tokyo. Relevant multimedia data files for this study are available on Figshare ( Constructive comments from the associate editor, Philippe Agard, and journal reviewers, Jones Ruh and Stephan Dominguez, are greatly appreciated. Managing Editor Gina Harlow is thanked for facilitating revisions and final publication.

1Supporting information for this paper can be found on Figshare at
Science Editor: Shanaka de Silva
Guest Associate Editor: Philippe Agard
Gold Open Access: This paper is published under the terms of the CC-BY-NC license.