Intraslab Earthquakes: Dehydration of the Cascadia Slab

See allHide authors and affiliations

Science  14 Nov 2003:
Vol. 302, Issue 5648, pp. 1197-1200
DOI: 10.1126/science.1090751


We simultaneously invert travel times of refracted and wide-angle reflected waves for three-dimensional compressional-wave velocity structure, earthquake locations, and reflector geometry in northwest Washington state. The reflector, interpreted to be the crust-mantle boundary (Moho) of the subducting Juan de Fuca plate, separates intraslab earthquakes into two groups, permitting a new understanding of the origins of intraslab earthquakes in Cascadia. Earthquakes up-dip of the Moho's 45-kilometer depth contour occur below the reflector, in the subducted oceanic mantle, consistent with serpentinite dehydration; earthquakes located down-dip occur primarily within the subducted crust, consistent with the basalt-to-eclogite transformation.

The most damaging earthquakes in western Washington have been intraslab events, also known as Wadati-Benioff earthquakes. These include earthquakes in 1949, 1965, and, most recently, the 2001 moment magnitude (Mw) 6.8 Nisqually event. Although megathrust earthquakes are typically larger in magnitude, intraslab events can be more damaging because they often occur directly beneath population centers, may have shorter recurrence intervals (as in Washington), and tend to have larger seismic energy–to–moment ratios (1) than megathrust events have.

The physical mechanisms responsible for intraslab earthquakes in the depth range 30 to 200 km have been debated for decades. A prominent theory, dehydration embrittlement, asserts that volatiles released during metamorphic dehydration reactions reduce the effective normal stresses across faults, allowing slip (2, 3). In light of the damaging 2001 Nisqually (Mw 6.8), 2001 Geiyo, Japan (Mw 6.7), and 2001 El Salvador (Mw 7.7) intraslab earthquakes, gaining physical insight into the mechanics controlling these earthquakes is important for earthquake hazard estimates and mitigation efforts. The 1998 Seismic Hazards Investigation in Puget Sound (Wet SHIPS) experiment (4) provided an opportunity to investigate the high-resolution structure of the subducting plate in northwest Washington and southwest British Columbia (5).

The Juan de Fuca plate is relatively young and warm (maximum age ∼10 million years), subducting obliquely at about 40 mm/year northeastward under Washington and Oregon. Active intraslab seismicity extends to 60-km depth, with some small events reaching depths as great as 100 km (6).

The data for the structural inversion consist of 90,000 first-arrival travel times from the Wet SHIPS, Dry SHIPS, western Cascades, and southwest Washington experiments (fig. S1) (7, 8); 27,000 first-arrival times from 1400 local earthquakes [200 of which are intraslab events (9)]; and 1200 secondary arrivals from the Wet SHIPS experiment that are consistent in slowness and travel time with reflections from the Juan de Fuca slab. We developed a nonlinear iterative inversion scheme that simultaneously inverts these travel times for earthquake locations, three-dimensional (3D) velocity structure, and reflector geometry (10). A wellknown trade-off exists between reflector depth and velocity structure. By including independent first-arrival information, we reduce this trade-off and extract reliable reflector depths. A smooth velocity model is regularized by minimizing second-order spatial derivatives of the velocity structure and reflector surface. Travel times of first arrivals are calculated using the Vidale-Hole (11, 12) 3D finite-difference code. Theoretical reflected bounce points and travel times are determined by summing calculated travel times from the source and receiver to points on the reflector surface and determining the position and time corresponding to the minimum summed time according to Fermat's principle. Reflected rays are then independently traced from the bounce point to the source and to the receiver. Reflector geometry and the 3D velocity model are each adjusted to fit the times of reflected waves. This nonlinear procedure converges stably after 10 iterations. The final model gives root mean square travel-time residuals of 0.09, 0.12, and 0.08 s for the active-source, earthquake, and reflection data, respectively, amounting to variance reductions of 98 and 91% for the active-source and earthquake travel times, respectively, relative to the standard regional 1D velocity model (13).

We interpret the reflector as the Moho of the subducting Juan de Fuca plate on the basis of two observations. First, the wide-angle reflections are often larger in amplitude than direct arrivals beyond 100-km source-receiver distance and are not observed closer than 55 km, indicative of postcritical reflections associated with an increase in velocities with depth across the reflector. Second, the 3D model demonstrates increases in velocities with depth in the vicinity of the reflector, leading to typical mantle velocities of 8 km/s just below the reflector.

We separate the intraslab earthquakes into two groups: those up-dip (west) of the 45-km reflector contour and those down-dip of this contour. The up-dip events generally occur at or below the reflector (Figs. 1 and 2). None of these events unambiguously occur above the reflector, considering the combined estimated 2-km uncertainty (14) in the earthquake locations and reflector position. In contrast, the down-dip events generally occur at or above the reflector, defining an 8-km-thick zone (Fig. 2).

Fig. 1.

Depth of reflector surface (colored area) and relocated intraslab earthquakes relative to the reflector [inverted blue triangles: more than 2 km beneath the reflector; maroon circles: within 2 km of the reflector; red triangles: more than 2 km above the reflector; black circles: reflector depth unknown; stars: Mw 5.8 Satsop (left) and Mw 6.8 Nisqually (right) earthquakes]. The dashed box corresponds to the cross section shown in Fig. 2 and is parallel to the relative plate motion direction.

Fig. 2.

Interpreted cross section (see dashed box in Fig. 1) showing compressional velocities (contoured at 0.5-km/s intervals), relocated seismicity (black circles: continental crustal events; colored symbols: intraslab events coded as described in Fig. 1), and Moho reflector (blue line). Interpreted top of subducting plate (red line) is drawn 7 km above reflector. The region between these lines is interpreted to be the subducting oceanic crust, composed of basalt above 40-km depth (horizontal green line) and beginning to transform to eclogite below. Subducting mantle is below the blue line. Low velocities in the mantle wedge imply the presence of serpentinite. There is no vertical exaggeration.

We propose that the up-dip events occur within the subducted oceanic mantle. In the laboratory, serpentinite, a hydrated mantle rock expected to exist in the uppermost oceanic mantle (15), becomes brittle under dehydration conditions, forming a visibly wet, clearly defined fault in laboratory samples (16, 17). The pressure-temperature (P-T) conditions experimentally determined for this reaction coincide with the P-T conditions predicted for Cascadia in the vicinity of the up-dip events (16, 18) (Fig. 3). Thus, we interpret the up-dip events as earthquakes induced by dehydration of serpentinite in the mantle. These mantle earthquakes are confined to a roughly 5-km-thick zone with a dip that is slightly shallower than that of the slab Moho, but which is parallel to predicted isotherms (18). Dehydration of serpentinite occurs nearly isothermally at these depths (Fig. 3), and thus this behavior is expected (19). In contrast, we propose that the down-dip events generally occur in the subducted oceanic crust as a result of embrittlement associated with progressive dehydration (Fig. 3).

Fig. 3.

Simplified phase diagram for the basalt (white)-to-eclogite (gray) transformation and serpentinite dehydration reaction (gray line) (19) overlaid with the calculated pressure-temperature path for the Cascadia slab (hashed). Note that, in the vicinity of the Cascadia geotherm, the basalt-to-eclogite transformation occurs at nearly constant pressure, independent of temperature, whereas the serpentinite dehydration reaction occurs at nearly constant temperature, independent of pressure. A geotherm just below the Moho would lie to the right (hotter) of the dashed line and, thus, would intersect the serpentinite dehydration line at a shallower depth. [Modified from (18)]

The velocity of the rocks within which the earthquakes occur is consistent with this proposed spatial change in earthquake driving mechanism. The up-dip events nucleate in rocks with a velocity of 7.5 to 8.1 km/s, whereas the down-dip events nucleate in rocks with a velocity of 6.8 to 7.5 km/s (20). Deserpentinization of partially serpentinized mantle rocks should result in an upper-mantle velocity of ∼8 km/s, whereas eclogitization of lower oceanic crust should result in a progressive velocity increase from ∼6.8 to 8 km/s (19).

Our interpretation requires knowledge of the relative locations of hypocenters, the reflector, and wave-speed contours to within about 2 km. We have performed specific resolution and error-analysis tests to determine our ability to resolve these parameters. Velocity checkerboard tests indicate the necessary resolvability, i.e., little smearing, strong pattern matching, and more than 50% amplitude return along the cross section shown in Fig. 2 using 30-km horizontal and 15-km vertical length scales, especially in the top 20 km of the model, but also within the subducting slab where earthquakes occur (fig. S2). To test whether we would be able to see a low-velocity zone (LVZ) associated with the subducted crust, we perturbed our model by placing an 8-km-thick LVZ above and parallel to the reflector, calculated travel times for this model, added random noise to these times, and then inverted them (fig. S3). The inverted model shows only slight smearing of structure and demonstrates returns of more than 50% perturbed velocity amplitude throughout the region with earthquakes, and upward of 75% within the seismic region. Thus, we have the ability to resolve the velocity structure within the slab and at depths shallower than 50 km, and can state that the subducted crust is not a low-velocity wave guide.

In warm subduction zones, such as Cascadia and southwest Japan, circumstantial evidence has pointed to intraslab earthquakes occurring generally within the crust. Warm subduction zones lack the lower plane of intraslab seismicity, which is predominant in cold zones where it is clearly occurring within the subducting mantle. Thermal modeling showed that the seismicity cut-off of intraslab seismicity in southwest Japan (65 km) and in the cold subduction zone of northeast Japan (160 km) is consistent with the respective depths at which dehydration of the oceanic crust should be complete (21). However, intraslab seismicity in southwest and northeast Japan occur at much shallower depths than would be predicted for the basalt-to-eclogite transformation and indicate that basalt-to-eclogite reactions cannot explain all the seismicity. In Cascadia, which has a thermal structure similar to that of southwest Japan, the bulk of intraslab seismicity deeper than ∼50 km was interpreted to occur within the subducting oceanic crust under southwest British Columbia (22). Farther to the south, under southwestern Washington, however, the bulk of intraslab seismicity (generally shallower than 40 km) was interpreted as occurring within the subducted mantle (8). Our results reconcile these apparently contradictory observations in Cascadia: The down-dip events occur within the subducted oceanic crust, and up-dip events occur primarily in the subducted oceanic mantle.

If intraslab earthquakes were purely caused by the basalt-to-eclogite transformation in the oceanic crust, the intraslab seismicity would be confined to the subducting crust. This geometrically constrains the maximum expected magnitude of an event to about Mw 7 (23). Indeed, the three largest historical intraslab events in 1949, 1965, and 2001 are near this limit. Although we do not observe reflections in the vicinity of these events and, thus, are uncertain whether these events occur within the oceanic crust or mantle, the wave speeds are well constrained, suggesting that these events nucleate near the slab Moho. In Cascadia, earthquakes occur within the oceanic crust and mantle (Fig. 2), geometrically allowing the possibility of a larger event, such as the 2001 Mw 7.7 El Salvador intraslab earthquake.

Fluids released from the down-going plate by the basalt-to-eclogite or other transformations can have additional consequences. Slow slip events producing 2 cm of thrust-type slip may propagate a few hundred km along strike over a period of a few weeks (24, 25). These faults appear to coincide with the plate interface and extend down from the down-dip edge of the megathrust locked zone. Coincident in both space and time are recently discovered deep tremor events: nonimpulsive sources that are detected at 2 to 6 Hz (26) and may be caused by a fluid-driven process (27). Seven episodic slip events, with a repeat time of about 14 months, have been detected from the Olympic Mountains into southern or central Vancouver Island. The colocation of these deep-creep events with the region interpreted in our model to be undergoing transformation of basalt to eclogite corroborates the hypothesis that these events are controlled by fluid processes. Geophysical evidence suggests the existence of a serpentinized continental mantle wedge in Cascadia (28). This would be expected because fluids released from the subducting plate infiltrate the overlying continental mantle. The lack of a well-defined continental reflector west of the Cascades is consistent with a serpentinized low-velocity mantle wedge. Indeed, wave speeds in our model at depths of 35 to 45 km in the mantle wedge are less than 7 km/s (Fig. 2), consistent with high concentrations of serpentinite.

Supporting Online Material

Materials and Methods

Figs. S1 to S5

Table S1


References and Notes

View Abstract

Navigate This Article