Seismic reflection data show that the eastern Aleutian Arc is characterized by reflectors that extend continuously from the lower arc crust to >50 km depth, which is considerably deeper than the crustal thickness of 27–35 km previously inferred from coincident wide-angle seismic surveys. Because the upper mantle is commonly homogeneous, and therefore nonreflective, relative to the overlying crust, we interpret these reflectors to be gabbro, garnet gabbro, and pyroxenite intrusions within two 50-km-wide roots that represent a >25-km-thick heterogeneous transition from mafic lower crustal rocks to ultramafic mantle rocks. We suggest that the reflectivity is linked to repeated differentiation and intrusion of mantle-derived melts into the subarc lithosphere, and that the depth of these roots shows that fractionation of arc crust can extend well below the seismically determined Moho. Because these deep roots are not evident beneath the central Aleutian Arc, either the roots form sporadically, perhaps as a consequence of an elevated magmatic supply, or such roots ultimately founder into the underlying mantle due to their relatively high mass density.


Intraoceanic island arcs are one of the fundamental building blocks of Earth’s continental crust (Stern, 2010). The bulk composition of the continental crust is andesitic (Taylor and McLennan, 1985), implying that island arcs should contain significant volumes of andesitic crust, that such crust is produced during arc-continent collision, or both. Petrologic modeling of the volume of andesitic crust requires a thick layer of underlying mafic restite, which may extend as much as 15 km below the seismically determined Moho (Tatsumi et al., 2008). Field observations of exhumed arc sections (Ringuette et al., 1999) and experimental studies of high-pressure melt crystallization (Müntener and Ulmer, 2006) indicate that cumulate rocks can crystallize beneath arcs at depths as great as 50 km. Isolated reflectors identified by wide-angle seismic surveys provide some direct evidence of sub-Moho structure in island arcs (Kodaira et al., 2007; Takahashi et al., 2009), with reflectors at ∼42 km depth interpreted to be the downward transition to upper mantle peridotite (Sato et al., 2009). To better image the deepest island arc crust and uppermost mantle, we have reprocessed higher resolution multichannel seismic reflection profiles from the Aleutian Arc, recorded where the seafloor is shallow, thus limiting interference from water-layer multiple reflections. Rather than isolated reflectors, we observe in some areas seismic reflectors that are continuously distributed from the lower crust to depths >50 km.


At ca. 60 Ma, the oceanic Kula plate was subducting beneath the margins of Beringia and southwestern Alaska, which includes the Peninsula and Chugach terranes, corresponding respectively to the arc and forearc of the accreted Jurassic Talkeetna arc complex (Plafker et al., 1994) (Fig. 1). The present-day Aleutian Arc developed from ca. 56–42 Ma when a series of southward and westward jumps in the location of subduction beneath North America trapped a large remnant of the Kula plate against the North American plate, creating an intraoceanic subduction zone where the Pacific plate now subducts beneath lithosphere of the former Kula plate (Scholl, 2007). The southwestern end of the Alaska Peninsula marks the westward transition from subduction of the Pacific plate beneath the accreted terranes and continental crust of North America to subduction beneath oceanic lithosphere attached to the North America plate. Unlike most intraoceanic arcs, the Aleutian Arc has not been subject to arc rifting and backarc basin development, implying that the entire >42 m.y. history of the arc is preserved in a relatively unmodified fashion. The lack of significant tectonic extension has resulted in the thickest island arc crust on Earth with a crustal thickness variably estimated to be 27–35 km (Holbrook et al., 1999; Fliedner and Klemperer, 1999; Lizarralde et al., 2002; Shillington et al., 2004).


We present here two reprocessed normal-incidence reflection images from the eastern arc: line A2 is located along the arc massif in the inner forearc region, and line A3 crosses the arc at the southwest end of the Alaska Peninsula (Fig. 1). The two multichannel seismic reflection lines were shot by the R/V Ewing in 1994 using an airgun array with a total volume of 8400 in3 (138 L) and recorded by a 160 channel hydrophone streamer with a far offset of 4.23 km. The seismic data were conventionally processed to stack, and coherent arrivals were enhanced using a semblance-weighted coherency filter. Migrated sections were computed using a segment migration algorithm that does not generate wave-equation artifacts, and converted to depth using seismic velocity models derived previously from the coincident wide-angle seismic surveys (Lizarralde et al., 2002; Shillington et al., 2004).

On both seismic lines, the deepest reflections, which are only clear beneath the shelf where the seafloor is shallow, originate from the top of the subducting Pacific plate (Figs. 2 and 3). On line A2, the apparent depth of these reflections increases as the distance from the trench increases, and they arrive too late to be recorded between common depth points (CDP) 52000 and 57000 (Fig. 2A). Because the Pacific plate dips to the NNW, these reflections originate out of the plane of the WSW-oriented seismic line, being reflected from the shallower updip part of the interplate boundary to the south. Although shallower crustal reflections, which likely correspond to a shear zone at the base of the accreted Chugach terrane, can be identified at CDP 66500–71500 (14–19 km depth) at the ENE end of line A2, the upper and middle arc crust above 20 km depth is mostly nonreflective. In contrast, two thick sets of reflections, which arrive at recording times from 7 to 15 s (23–55 km depth for reflectors directly below the seismic line) and earlier than reflections from the top of the subducting plate, are observed at CDP 46000–49000 and CDP 53000–57000; these two reflective regions are identified as root 1 and root 2 in Figure 2. Cross-arc line A3 intersects along-arc line A2 at the position of root 1, and this same package of subhorizontal reflections can be seen on line A3 at 8–13 s (Fig. 3A). Two thinner sets of deep reflections above the subducting plate are also observed on line A2 at CDP 39000–44000 and 11.5–13.5 s (MR1 at 40–47 km depth in Fig. 2) and at CDP 52000–54000 and 14.5–15.5 s (MR2 at 52–56 km depth)

In most island arcs, there does not appear to be a sharp, seismically well-defined Moho. Rather, where resolved, there is a change over ∼4–8 km from high crustal seismic velocities of ∼7.2–7.4 km s−1 to low upper mantle velocities of 7.6–7.8 km s−1 (Kodaira et al., 2007; Sato et al., 2009). Wide-angle seismic surveys typically locate the Moho from PmP reflections, which are mostly generated at the top of this transition zone (Takahashi et al., 2009). At the intersection of lines A2 and A3, the wide-angle seismic surveys locate the Moho at depths of 35 km and 30 km, respectively (Lizarralde et al., 2002; Shillington et al., 2004); this discrepancy indicates uncertainty due to reflector identification and inversion methodology in these relatively sparsely sampled surveys. It is possible that some deep normal incidence reflections in Figure 2 originate out of the plane of seismic profile A2; however, at the intersection of reflection profiles A2 and A3 a reflection that originates significantly out of plane, for example in the upper crust, will have a steep apparent dip on one of the two orthogonal profiles; this is not observed (Fig. DR1 in the GSA Data Repository1). Thus, even with an uncertainty of ∼5 km in the depth of the Moho inferred from wide-angle seismic data, subhorizontal reflections at the line intersection clearly extend from ∼10 km above the Moho to >12 km below this boundary, and it is likely that the majority of the reflections that comprise roots 1 and 2 are distributed as much as 15 km below the inferred Moho.

Hypocenters of local earthquakes within 15 km of the seismic lines have been superimposed on the migrated sections. There are almost no earthquakes within root 2 and only a few earthquakes in root 1, in contrast to the adjacent crust of the North America plate and the subducting Pacific plate. The maximum depth of seismicity in the upper plate decreases toward the relatively warm arc (Fig. 3B), and may indicate that this depth is thermally controlled.


The eastern Aleutian Arc is underlain by two seismically reflective mostly aseismic roots, each of which extends from ∼25 km to 50–55 km depth, and ∼50 km along the arc. Although some reflectors within these roots could be partially molten sills related to rising melt, this explanation seems to be unlikely in general, because few reflectors exhibit the high amplitudes typical of known melt zones and most reflections appear weaker than those from the interplate boundary. Furthermore, similar reflective roots are not observed farther west, where the seismic line also approaches the arc and rising melt should also be evident if commonly reflective (Fig. DR2). Seismic reflections within the forearc mantle of the Hikurangi subduction zone have been tentatively linked to water released by dehydration of partially serpentinized peridotite (Davey and Ristau, 2011). These reflectors, however, correlate closely with a cluster of seismicity associated with dehydration embrittlement, and no similar seismicity is known under the eastern Aleutian Arc. Therefore the absence of seismicity coincident with the roots suggests that fluids, whether melt or water, are unlikely to be the predominant cause of the observed deep seismic reflections. In modern and fossil subduction zones, 4–8-km-thick aseismic zones of seismic reflectivity that are laterally continuous over tens of kilometers in the upper mantle resemble those from the top of the subducting Pacific plate, and are interpreted as shear zones (Warner and McGeary, 1987; Calvert et al., 2006); however, these zones differ from the >25-km-thick regions of reflectivity that characterize the eastern Aleutian Arc roots.

Island arcs are the locus of long-lived extraction of largely basaltic melts from the upper mantle. Following establishment of the initial arc, these melts pond beneath and repeatedly intrude the thickening crust. In exhumed arcs, the crust-mantle transition, the petrologic Moho, corresponds to the change from plagioclase-bearing to plagioclase-free lithologies, e.g., crystallized garnet gabbro to cumulate pyroxenite, and occurs over a distance of ∼1 km (Müntener and Ulmer, 2006; DeBari and Greene, 2011). The in situ P wave velocity of garnet gabbro varies from 7.1 to 7.6 km s−1 as the proportion of garnet increases (Kono et al., 2009), implying that garnet gabbro could constitute much of the region beneath some seismically determined Moho discontinuities, and that the petrologic Moho could be considerably deeper than the seismic Moho. If lithological variation over the interval between the seismic and petrologic Moho is small, then this depth range will not be strongly reflective in normal incidence seismic images. Alternatively, if basaltic melts with varying water content intrude the arc over a broad depth range, and each intrusion package differentiates into layered gabbro and pyroxenite (with or without garnet depending on the depth), then there will be a broad petrologic crust-mantle transition zone, which will be vertically heterogeneous due to the numerous lithological contrasts (Kelemen et al., 2003; Müntener and Ulmer, 2006; Jagoutz et al., 2011). In a hot arc at 30 km depth, the gabbro to pyroxenite and garnet-gabbro to garnet-pyroxenite reflection coefficients are 0.13 and 0.14, respectively (Table DR1 in the Data Repository), sufficient to produce observable reflections (Warner and McGeary, 1987), especially if amplitudes are further increased by constructive interference. In the eastern Aleutian Arc the reflective roots extend across the seismic Moho, indicating that while the bulk composition in the reflective roots increases downward from mafic at 25 km depth to ultramafic at >40 km depth, significant vertical lithological variation is present within the root. Although the maximum depth of 55 km at which mafic rocks are inferred to be present is similar to that of the exhumed Kohistan arc, the latter exhibits interlayered gabbros, garnet gabbros, and pyroxenites over only a few kilometers (DeBari and Greene, 2011).

For typical island arc geotherms, pyroxenites are denser than upper mantle peridotites over a broad depth range (Behn and Kelemen, 2006; Tatsumi et al., 2008). Gabbroic rocks can also become denser than typical upper mantle rocks at depths >∼35 km as the arc cools and garnet forms. This density distribution may result in foundering into the mantle of the deepest sections of arc crust if it is still sufficiently warm to allow density-driven descent. This process, which may recur on time scales as short as 1–10 m.y. (Behn et al., 2007), has been inferred in the exhumed Talkeetna arc from the absence of a high Mg# lower crust and sharp crust-mantle transition (Kelemen et al., 2003; DeBari and Greene, 2011). The large-scale removal of dense heterogeneous roots, as inferred in the Talkeetna arc, would result in their replacement by a relatively homogeneous and therefore nonreflective upper mantle beneath the arc, which is characteristic of the Aleutian Arc further west. If arc roots persist until arc-continent collision, they are likely to subduct into the mantle with much of the oceanic plate due to their high density and depth.

We thank Donna Shillington and Dan Lizarralde for generously providing their seismic velocity models. Sue DeBari, Othmar Müntener, and two anonymous reviewers made valuable comments that improved the paper. The seismic reflection data were acquired by the R/V Ewing, facilitated by Nathan Bangs, Steve Holbrook, and Simon Klemperer and funded by the U.S. National Science Foundation. The reprocessing of these data was funded by the Natural Sciences and Engineering Research Council of Canada.

1GSA Data Repository item 2013050, Figures DR1 and DR2 (additional displays of the seismic data) and Table DR1 (calculated reflection coefficients), is available online at www.geosociety.org/pubs/ft2013.htm, or on request from editing@geosociety.org or Documents Secretary, GSA, P.O. Box 9140, Boulder, CO 80301, USA.