Volume 48, Issue 9 e2020GL091433
Research Letter
Free Access

Volcano Clustering Promoted by the Cessation of Back-Arc Spreading and Ensuing Nascent Lithospheric Drips

Changyeol Lee

Changyeol Lee

Department of Earth System Sciences, Yonsei University, Seoul, Korea

Search for more papers by this author
Ikuko Wada

Corresponding Author

Ikuko Wada

Department of Earth and Environmental Sciences, University of Minnesota, Minneapolis, MN, USA

Correspondence to:

I. Wada,

[email protected]

Search for more papers by this author
First published: 11 April 2021
Citations: 3


In Northeast Japan and Izu-Bonin, arc volcanoes form in clusters or as cross-arc chains. Their occurrence emphasizes the non-uniform distributions of sub-arc temperature and fluids that control the spacing of arc volcanoes. Here, using 3-D numerical models, we show that the cessation of back-arc spreading promotes volcano clustering by triggering the formation of nascent lithospheric drips – downward protrusions of cold and dense lithosphere-adjacent to the thinned back-arc lithosphere. The nascent drips interfere with the flow of the hot asthenospheric mantle from the back-arc toward the arc, leading to gradual development of alternating hot and cold regions beneath the arc. The results indicate that along-arc variation in the sub-arc mantle temperature is largest not during back-arc spreading but after its cessation, explaining the time offset by several million years between back-arc spreading and volcano clustering in Northeast Japan and Izu-Bonin.

Key Points

  • Cessation of back-arc spreading triggers the formation of nascent lithospheric drips adjacent to the thinned back-arc lithosphere

  • Nascent lithospheric drips interfere with the mantle wedge flow, leading to alternating hot and cold regions beneath the arc

  • Clustering and cross-arc chains of volcanoes tend to occur not during but after back-arc spreading

Plain Language Summary

Arc volcanoes form in subduction zones because the mantle beneath them experiences partial melting due to high temperature and the presence of water. Arc volcanoes in Northeast Japan and Izu-Bonin, cluster together, indicating that temperature and water are not distributed uniformly in the underlying mantle wedge. Behind these two arcs, the back-arc basin formed through rifting and seafloor spreading in the past. We use 3-D numerical models to study how the back-arc spreading affects the temperature and flow conditions beneath the arc. We found that alternating hot and cold regions develop in the mantle beneath the arc after the back-arc spreading stops. This occurs because some parts of the back-arc lithosphere next to the newly formed thin back-arc lithosphere become unstable and protrude into the underlying mantle. These protrusions interfere with moving mantle that transfers heat from back-arc region to the arc, causing cold regions beneath the arc.

1 Introduction

Arc volcanism is a surface expression of subduction, which is an essential process in plate tectonics and mantle convection. It forms a chain of volcanoes parallel to the trench, and volcanic activities are spatially and temporally more discontinuous than at mid-ocean ridges, indicating different magmatic processes beneath the surface. Although it is widely accepted that the addition of slab-derived aqueous fluids to the overriding hot mantle lowers the mantle solidus and causes flux melting (e.g., Gaetani & Grove, 1998; Mysen & Boettcher, 1975), what controls the location of melt generation in the mantle wedge and the distribution of volcanoes is unclear.

Volcano clusters in Northeast (NE) Japan and cross-arc volcano chains in Izu-Bonin likely reflect along-arc variation in the distributions of temperature and fluids in the mantle wedge (Figure 1a). In NE Japan, volcano clusters had formed by 8 Ma (Kondo et al., 1998) and are spaced at 79 ± 16 km on average (Tamura et al., 2002). Seismic tomography indicates a strong spatial correlation between low-velocity anomalies and volcano clusters, and the anomalies have been interpreted as high-temperature regions (Tamura et al., 2002). In Izu-Bonin, cross-arc chains of what are presently seamounts are 80–100 km apart (Tamura et al., 2009) and are thought to have largely formed during 10–5 Ma (Sato et al., 2020). Geodetic and geochemical observations in Izu-Bonin also indicate spatial correlation between the cross-arc volcano chains and high-temperature anomalies (Honda et al., 2007; Tamura et al., 2009).

Details are in the caption following the image

(a) Map showing NE Japan and Izu-Bonin and (b) a schematic diagram of the model geometry with boundary conditions. In (a), red triangles indicate Quaternary volcanoes, thin dashed lines indicate the depth contours of the subducting slab surface at every 40 km based on Slab2.0 (Hayes, 2018), color indicates bathymetry based on ETOPO1 (Amante & Eakins, 2009), and blue rectangles indicate volcano clustering in NE Japan and cross-arc seamount chains in Izu-Bonin. In (b), Vcv, Vsd and Vsp correspond to the plate convergence rate, subduction rate, and back-arc spreading rate, respectively.

Several mechanisms have been proposed for the genesis of hot regions in the mantle wedge beneath the arc. Reduced viscosity of the mantle wedge due to the addition of slab-derived aqueous fluids has been invoked to allow small-scale convection in the mantle wedge, leading to high-temperature anomalies in the regions of mantle upwelling (Honda & Yoshida, 2005; Wirth & Korenaga, 2012). A thin low-viscosity layer along the top of the subducting slab is also proposed to cause localized return flow and along-arc temperature anomalies beneath the arc (Morishige, 2015). However, these mechanisms do not address why volcano clustering occurs only in some subduction zones even though slab-derived fluids are available at sub-arc and post-arc depths in most subduction zones except warm-slab subduction zones (e.g., Lee et al., 2021; van Keken et al., 2011). Other studies show that the heterogeneity in the thermal structure of the back-arc causes along-arc variation in sub-arc mantle temperature that are amplified by the development of complex three-dimensional mantle flow pattern (Davies et al., 2016; Lee & Wada, 2017), but the cause of such thermal heterogeneity is unclear.

One commonality between NE Japan and Izu-Bonin, where clusters or cross-arc chains of volcanoes exist, is the past occurrence and timing of back-arc spreading; spreading occurred during 21–17 Ma behind NE Japan (Yoshida et al., 2014) and during 25–15 Ma behind Izu-Bonin (Tamura et al., 2009). Previous modeling studies have shown that the thinning of the back-arc lithosphere due to spreading increases the vigor of small-scale convection (e.g., Davies et al., 2016; Honda et al., 2002), leading to high-temperature anomalies in the sub-arc mantle. In this study, we investigate the causal relationship between back-arc spreading and the development of high-temperature anomalies in the sub-arc mantle through a series of three-dimensional numerical simulations.

2 Methods

We use a three-dimensional coupled kinematic-dynamic subduction model (Figure 1b), following the approach of Lee and Wada (2017). The subducting slab is kinematically driven and induces corner flow in the mantle wedge. Our model differs from that of Lee and Wada (2017) in two key aspects: our model incorporates the effects of (1) back-arc spreading and (2) thermal buoyancy to simulate small-scale convection in the back-arc mantle (Supporting Information Text S1). Effects of the radiogenic heat production and viscous dissipation on the thermomechanical behavior in the subduction zones are relatively small (Hall, 2012; Lee & King, 2009) and are excluded in our model (see Table S1 for model parameters).

The model domain is 800-km long along the trench (x-axis), 800-km wide (y-axis), and 200-km deep (z-axis) (Figure 1b). The geometry of the subducting slab is defined by an arc of a 629-km radius with its center at 629-km depth under the trench and does not change along the trench or with time. This geometry is similar to that of the subducting slab in NE Japan and Izu-Bonin (Syracuse & Abers, 2006). The corner of the mantle wedge down to 70-km depth is assumed to be decoupled from the subducting slab (Wada & Wang, 2009) and is non-deforming.

To approximate the effect of back-arc spreading, we move the back-arc spreading ridge away from the trench (Figure 1b). Slab roll-back is assumed to accommodate the back-arc extension during spreading by increasing the subduction rate by the full spreading rate (Vsp) (e.g., Conder et al., 2002). Because the spreading is assumed to be symmetric, the ridge axis and the far side of the back-arc basin closer to the back-arc vertical boundary migrate away from the trench at 1/2Vsp and Vsp, respectively. To accommodate the back-arc-ward mantle flow due to spreading, the horizontal velocity component along the back-arc-side vertical boundary linearly decreases from Vsp at the surface to 0 at 200-km depth during spreading. However, the no-slip boundary condition is applied when spreading is not occurring. For the entire model run, an insulated temperature boundary condition is applied to the back-arc-side vertical boundary. A geotherm calculated for a 100-Ma plate using the half-space cooling model with the mantle potential temperature (Tp) of 1,350°C is used for the subducting slab at the trench-side vertical boundary (e.g., Lee & Wada, 2017). A stress-free boundary condition is applied to the bottom boundary of the model domain. The mantle potential temperature is assigned at the bottom of the model domain from y = 540–800 km. Insulated and free-slip boundary conditions are applied to both trench-normal side boundaries. A geotherm calculated for a 50-Ma plate using the half-spacing cooling model is used for the initial temperature distribution. The adiabatic temperature gradient of 0.35 °C/km is added a posteriori after the simulation is completed to provide the distribution of real temperature as opposed to potential temperature that is being used in the energy equation (Equation S3), as commonly practiced (e.g., King et al., 2010; van Keken et al., 2002).

We approximated the mantle rheology using the rheological parameters for diffusion creep of wet olivine in Karato and Wu (1993) (Supporting Information Text S1). Addition of aqueous fluids and partial melts in the mantle wedge is known to reduce mantle viscosity (Mei et al., 2002). In our model, this effect is approximated by multiplying the viscosity with a reduction factor (f):
where y1 and y2 are the trench-side (y1 = 288.4 km) and the back-arc-side (y2 = 460 km) of the reduced-viscosity zone, respectively. The viscosity of the mantle wedge increases from 5% to 100% of the wet olivine viscosity (Equation S4) at y1 and y2, respectively. The reference viscosity at 200-km depth is 1.0821 × 1019 (Pa s), corresponding to a Rayleigh number of 1.1215 × 106.

We refer to the simulation with the following intermediate parameter values as the reference experiment: 6-cm/yr convergence rate, 4-cm/yr full spreading rate, 2-cm/yr ridge migration rate, and initial ridge location at 400 km from the trench (Figures 2-4). All the experiments were first run without back-arc spreading for 52.1 Myr (t = −52.1 to 0 Myr) to minimize the effects of initial conditions on the temperature and flow in the mantle, and then with back-arc spreading for 7.92 Myr (t = 0–7.92 Myr) followed by a period without back-arc spreading for 19.14 Myr (t = 7.92–27.06 Myr). During spreading, the subduction rate linearly increases from 6 cm/yr at t = 0–10 cm/yr at t = 1.32 Myr, remains at 6 cm/yr for 5.28 Myr, and decreases linearly from 10 cm/yr at t = 6.6 Myr to 6 cm/yr at t = 7.92 Myr, minimizing the model instability. With this timeframe, the back-arc grows by 264 km in width with the final location of the ridge axis at 532 km from the trench.

Details are in the caption following the image

Evolution of the thermal structure of the reference experiment, expressed by the 1,275°C isothermal surfaces (red plane) in the mantle wedge. Vertical cross-sections shown in the panels are plotted in Figure 3 with mantle flow velocities. Green dashed line indicates the initial spreading ridge location.

Details are in the caption following the image

Mantle flow and temperature fields along the vertical cross-sections are shown in Figure 2. For clarity, the flow velocities within and below the subducting slab are excluded.

Details are in the caption following the image

Evolution of the temperature field (a–h) along the slab surface and (i–j) along a dipping plane through the mantle wedge. Vertical dashed lines indicate the locations of vertical cross-sections in Figure 3, and horizontal lines indicate 100-km depth on the slab surface.

3 Modeling Results

Prior to and shortly after the initiation of back-arc spreading (Panels a and b in Figures 2-4), the mantle flow is driven largely by the motion of the subducting slab with little along-arc variation in the sub-arc mantle and slab surface temperatures. As the back-arc spreading continues and the ridge migrates away from the trench, strong upwelling develops beneath the ridge (Panels c and d in Figures 2 and 3). In our experiment, the growth of the back-arc basin is assumed to be accommodated by slab roll-back, and the subduction rate is increased by the spreading rate from 6 to 10 cm/yr. The faster subduction results in more vigorous corner flow and the upwelling in the back-arc provides hotter mantle to the sub-arc region, increasing the mean mantle temperature by 56.9°C at 70-km depth (y = 340.3 km) (Figure 3a vs. Figure 3c). With the faster subduction, the mean slab surface temperature at a given depth initially decreases (e.g., 769.3 vs. 744.3°C at 150-km depth; Figure 4a vs. Figure 4c), but despite this cooling effect, the mean slab surface temperature eventually increases due to the heating effect of the hotter overriding mantle (e.g., 769.3°C vs. 781.7°C at 150-km depth; Figure 4a vs. Figure 4d).

Ramping down the spreading rate reduces the vigor of mantle flow beneath the back-arc basin and in the mantle wedge corner (Figure 2e). The reduced vigor of mantle flow leads to a slight decrease (<5°C at 150-km depth) in the mean slab surface temperature (Figure 4d vs. Figure 4e). After the cessation of the back-arc spreading, mantle flow beneath the back-arc basin slowly wanes down (Figure 3f), and at the transition between regions of the slab-driven corner flow and the waning back-arc mantle flow (y = ∼400 km), the difference in the lithosphere thickness triggers the formation of nascent lithospheric drips (Figures 3g and 3h). Cold nascent drips not only interfere with the inflowing corner flow but also gradually become entrained and advected toward the sub-arc without completely detaching and sinking as drips, reducing the sub-arc mantle temperatures to below the pre-spreading temperature in the cold regions (Figure 3a vs. Figure 3h2; Figure 4a vs. Figure 4h). Between these cold regions, the mantle flow toward the arc is more vigorous than prior to the back-arc spreading (Figure 3a vs. Figure 3h1), owing to the hotter back-arc mantle caused by the mantle upwelling during the spreading, resulting in alternating cold and hot regions beneath the arc (Figures 2h and 4h).

At 70-km depth beneath the arc (y = 340.3 km in Figure 4j) in the middle 400 km of the model (i.e., x = 200–600 km) where boundary effects are small, the mean peak temperature in the hot regions is 1,357.3°C with a standard deviation (urn:x-wiley:00948276:media:grl62253:grl62253-math-0002) of 2.2°C, and the mean minimum temperature in the cold regions 1,184.7°C with urn:x-wiley:00948276:media:grl62253:grl62253-math-0003 = 10.1°C, yielding a difference (hereafter referred to as the max-to-min difference) of 172.6°C (Figure 3j). The hot sub-arc mantle heats up the subducting slab more efficiently than cold sub-arc mantle, resulting in along-arc variation in the slab temperature. The mean peak and minimum slab surface temperatures beneath the arc (z = 100 km) are 747.9 ± 3.2 and 687.0 ± 4.7°C, respectively, yielding a max-to-min difference of 60.9°C (Figure 3h).

We approximate the weakening effect of aqueous fluids and melts by reducing the effective viscosity of the mantle as discussed above. However, if we assume no viscosity reduction in the mantle wedge, nascent lithospheric drips do not form in the model after the cessation of back-arc spreading. The low-viscosity mantle wedge, therefore, is one of the critical factors that facilitate the development of hot and cold sub-arc regions as previously reported (e.g., Honda & Yoshida, 2005). With the low-viscosity mantle, small-scale convection occurs without back-arc spreading (Figure S2). However, the max-to-min differences in the sub-arc mantle and the sub-arc slab-surface temperatures are 33.7 and 23.1°C, respectively, much smaller than those of the reference experiment (172.6 and 60.9°C, respectively), and the nascent lithospheric drips after the cessation of back-arc spreading amplify along-arc variation in the sub-arc mantle and slab temperatures, promoting volcano clustering.

The spacing of the hot regions depends on the spacing of the nascent lithospheric drips, which are manifestation of small-scale convection, corresponding to the downwelling part of the convective pattern. The cell width of small-scale convection depends on the vertical dimension of the viscous mantle (e.g., Honda​ ​et al., 2002), and thus the spacing of the nascent lithospheric drips also depends on the vertical dimension of the viscous mantle wedge beneath where they form. With the slab geometry that resembles that in NE Japan, the predicted spacing of the hot regions beneath the arc ranges from 59.3 to 75.0 km (mean spacing of 66.9 ± 5.0 km) (between x = 200 and 600 km at 27.06 Myr) (Figures 2h and 4h), comparable to the spacing of the hot mantle anomalies (79 ± 16 km) in NE Japan.

We tested the effects of the initial location of the spreading ridge, convergence rate, spreading rate, and mantle potential temperature (Supporting ​Information Text S2; Figures S3–S9). The results indicate that the max-to-min difference in the sub-arc mantle temperature become smaller with increasing trench-ridge distance (e.g., the max-to-min difference of 75.0°C for the initial ridge location of 500 km) (Figures S3), indicating that the relative location of initial spreading plays an important role in the thermal evolution of the sub-arc region. Further, nascent lithospheric drips and the alternating hot and cold sub-arc regions develop upon the cessation of back-arc spreading over a wide range of convergence and spreading rates. The high-temperature anomalies in the sub-arc mantle are less pronounced for higher subduction rate (e.g., the max-to-min difference of 169.6°C for Vsd = 9 cm/yr) (Figures S4 and S5). This occurs because the fast convergence rate increases the vigor of the mantle inflow, which hinders the development of nascent lithospheric drips and also competes with the cooling effects of the drips by a faster turnaround of hot mantle. By contrast, the high-temperature anomalies become much more pronounced for higher back-arc spreading rate (e.g., the max-to-min difference of 197.9°C for Vsp = 6 cm/yr) (Figures S6 and S7). This occurs because the faster spreading rate increases the back-arc mantle temperature and thus the peak temperature while the thermal impact of the nascent lithospheric drips remains relatively unchanged. Lower mantle potential temperatures result in smaller along-arc variation in the sub-arc temperature since the higher mantle viscosity (smaller Ra number) weakens development of the lithospheric drips and small-scale convection (Figure S8). By contrast, higher mantle potential temperatures promote nascent lithospheric drips even without back-arc spreading by lowering the mantle viscosity (e.g., with 1,450°C mantle potential temperature; Figures S9). However, such high mantle potential temperature is unlikely because it would result in partial melting of the subducted oceanic crust and thus the formation of adakites in the arc, which have not been observed in the subduction zones after the back-arc spreading ceased (Kimura & Yoshida, 2006). Further, the predicted mantle flow pattern and temperature field are similar regardless of whether the growth of the back-arc basin is accommodated by slab roll-back or by backward migration of the overlying lithosphere away from the trench (Figure S10). The only notable difference is that in the reference experiment, the subduction rate increases during back-arc spreading, resulting in a cooler subducting slab (e.g., by ∼11.4°C at 100-km depth).

4 Discussion and Conclusion

The max-to-min difference in the sub-arc mantle temperature (172.6°C) between the cold and the hot regions is likely large enough to affect the degree and the location of partial melting. In addition, the heating effect of the hot mantle can lead to efficient dehydration of the underlying subducting slab, which can in turn lead to addition of more aqueous fluids into the overlying hot region and thus a higher degree of flux melting (Figure S11). A previous modeling study that incorporates the impact of slab-derived fluids on melting in the mantle wedge beneath NE Japan indicates that that the degree of partial melting in the hot region can be as high as 20% whereas that in the cold region is negligibly small (Yoo & Lee, 2020). The higher degree of partial melting is attributed to the hotter temperature and more efficient dehydration of the underlying subducting slab beneath the hot region. The higher degree of partial melting increases the vigor of mantle flow by reducing the mantle viscosity (Mei et al., 2002), and it is likely to cause positive feedback on temperature and melting, resulting in a greater max-to-min difference in the temperature than those predicted by our experiment.

The mechanism that is proposed here differs from previous studies in that the underlying cause of the alternating hot and cold regions in the mantle wedge is the change in the lithospheric thickness across the transition from the arc lithosphere to the back-arc basin lithosphere. The thickness change exists during back-arc spreading. However, the nascent lithospheric drips do not form until the back-arc spreading ceases because the mantle flow beneath the back-arc is dominated by larger-scale convection that is compatible with the kinematics of the back-arc spreading, discouraging small-scale convection. Further, in our simplified model, the mantle upwelling that feeds into the back-arc spreading ridge is skewed toward the arc and flows by the transition between the arc lithosphere and the thinner back-arc lithosphere, imposing shear drag on the base of the lithosphere and further discouraging the formation of nascent lithospheric drips during back-arc spreading.

The gradual growth of nascent lithospheric drips after the cessation of back-arc spreading is consistent with previous studies that indicate that it can take significantly longer than 10 Myr for drips to form without the presence of a weak layer, along which drips can separate from the lithosphere (e.g., Beall et al., 2017), particularly when the lithospheric thinning as the initial perturbation is relatively small as in this study. As a result, the drips never fully develop as thermal diffusion reduces the instability over time. However, the drips affect mantle flow and advective heat transfer beneath the arc, which should manifest in magmatic processes. The delayed development of the volcano clustering by ∼10 Ma in NE Japan is consistent with the time scale predicted by our experiment. In Izu, the lithologies and geochemistry of the basement rocks near rear-arc seamounts indicate that clustered volcanism began several million years after the cessation of back-arc spreading (Sato et al., 2020). Continued volcanism with slab roll-back over the following ∼9 Myr resulted in the formation of cross-arc volcano chains. In Izu, the hot regions developed more rapidly than predicted by our experiment. The slab roll-back likely had some impact on the development of small-scale convection but is incorporated only through an increase in subduction rate in our model. Further, there were two episodes of back-arc spreading in Izu (Arculus et al., 2015), the first of which is not included in our experiment. Lack of spatial correlation between the crust that formed during the first and second episodes (Kodaira et al., 2008) indicates that the distribution of hot regions likely changed between two episodes, possibly affecting the development of small-scale convection during the second episode.

The geometry of the slab beneath NE Japan and Izu-Bonin is relatively simple in that it does not change significantly along the arc, and the effect of slab geometry on the mantle flow pattern beneath these arcs is likely small (e.g., Wada et al., 2015). However, complex slab geometries that are observed at other subduction zones (e.g., the along-arc wavy slab beneath SW Japan in Figure 1a) and their time evolution may interfere either constructively or destructively with the development of hot and cold regions beneath the arc after the cessation of back-arc spreading and are needed to be examined further.

The cessation of back-arc spreading is likely one of several mechanisms that can lead to volcano clustering. For example, cross-arc volcano chains have been forming behind southern Kermadec, where the back-arc spreading is currently occurring (Todd et al., 2011). The spreading started within or near the arc volcanic zone, which likely has a significant impact on the magmatic processes and the distribution of volcanism. A spatial correlation between the volcano distribution and the seismologically constrained thermal heterogeneity in the underlying mantle has also been reported for Cascadia (Gao & Shen, 2014). However, the thermal heterogeneity is at a much larger scale than NE Japan or Izu-Bonin, and the dynamics that is responsible for the mantle heterogeneity in Cascadia likely differs from that in NE Japan and Izu-Bonin. Further, the dehydration of the subducting slab in the deeper part of the mantle can lead to volcanism in a wider region behind the arc, such as the intraplate volcanism that extends from NE Japan to north-central China (e.g., Yang & Faccenda, 2020). Thus, any mechanisms that cause strong lateral heterogeneity in temperature or fluid content in the mantle wedge can trigger nonlinear feedback through their effects on mantle viscosity and lead to isolated populations of volcanoes. The variation of the distribution of arc volcanoes among different subduction systems, therefore, likely reflects the variety in the spatial scales and extent at which the mantle dynamics and fluid transport impact melt generation.

The above modeling results indicate a transition in the back-arc mantle flow pattern from large-scale mantle convection during back-arc spreading to small-scale convection after the spreading. The small-scale convection is inhibited during back-arc spreading but facilitated post-spreading by the high temperature that is achieved by the preceding large-scale convection. The development of lateral temperature variation in the back-arc mantle and nascent lithospheric drips due to the small-scale convection lead to the development of alternating hot and cold regions beneath the arc, explaining the occurrence of volcano clustering and cross-arc volcano chains not during but after the cessation of back-arc spreading.


We thank two anonymous reviewers for useful comments, which helped to improve the manuscript. C.L. acknowledges the financial support from the National Research Foundation of Korea (Grant #: 2017R1A6A1A07015374 and 2019R1A2C1002517) and the Yonsei University (2019-22-0010). I.W. acknowledges the financial support from the University of Minnesota in the form of start-up funds.

    Data Availability Statement

    The model parameter values and additional modeling results that are discussed in this article are found in Supporting Information, and the modeling results are made available on Zenodo (https://sandbox.zenodo.org/record/746569#.YE5_LDqg_-h).