Here we report on deep long-period earthquakes (DLPs) newly observed in four places in western Oregon. The DLPs are noteworthy for their location within the subduction fore arc: 40–80 km west of the volcanic arc, well above the slab, and near the Moho. These “offset DLPs” occur near the top of the inferred stagnant mantle wedge, which is likely to be serpentinized and cold. The lack of fore-arc DLPs elsewhere along the arc suggests that localized heating may be dehydrating the serpentinized mantle wedge at these latitudes and causing DLPs by dehydration embrittlement. Higher heat flow in this region could be introduced by anomalously hot mantle, associated with the western migration of volcanism across the High Lava Plains of eastern Oregon, entrained in the corner flow proximal to the mantle wedge. Alternatively, fluids rising from the subducting slab through the mantle wedge may be the source of offset DLPs. As far as we know, these are among the first DLPs to be observed in the fore arc of a subduction-zone system.
Deep long-period earthquakes (DLPs), commonly observed under volcanoes, are thought to represent the movement of fluids [e.g., Chouet and Matoza, 2013; Nichols et al., 2011; Pitt and Hill, 1994; Power et al., 2004; White, 1996]. DLPs occur deeper than long-period earthquakes within volcanic edifices [e.g., Chouet, 1996; Minakami, 1960], most often at epicentral distances from volcanoes less than the hypocentral depth (<45 km depth). DLPs have been interpreted to mark the zone where magma is rising toward volcanoes [e.g., Chouet and Matoza, 2013; Zhao et al., 2011], and they often fill a volume in the shape of a vertical cylinder or cone [Hasegawa and Yamamoto, 1994; Nichols et al., 2011; Power et al., 2004]. The mechanism of seismic radiation may be choked flow, faulting, or resonating cracks, perhaps abetted by dehydration embrittlement or fluid lubrication [Chouet and Matoza, 2013]. Well-studied areas with DLPs include Japan [Hasegawa and Yamamoto, 1994], Alaska [Power et al., 2004], Hawaii [Okubo and Wolfe, 2008], Long Valley [Hill et al., 2002], and Cascadia [Nichols et al., 2011; Pitt et al., 2002].
DLPs are marked by less high frequency seismic energy than regular earthquakes and often have extended codas. The excitation often begins gradually, sometimes with high-frequency content similar to that of a regular earthquake. DLP activity is sometimes observed in clusters of events and accompanied by tremor-like signals. An example from this data set of a DLP event with related tremor is shown in Figure 1. This event is clearly deficient in energy above 10 Hz and even depleted in energy above 2 Hz. In the Pacific Northwest Seismic Network (PNSN) catalog, DLPs are defined as depleted in energy at higher frequencies and often having an extended coda, making them clearly anomalous compared to regular earthquakes. Figure 2 illustrates the seismograms and spectrograms from several DLP events used in this study at a single station contrasted with two typical regular earthquakes at a similar depth and distance.
Table 1. Offset DLPs Deeper Than 35 km
aDates are formatted as month/day/year.
38.0 ± 32
41.6 ± 1
41.3 ± 1
41.3 ± 1
41.6 ± 1
44.1 ± 32
43.9 ± 32
42.1 ± 2
42.1 ± 2
42.5 ± 1
41.9 ± 1
41.0 ± 1
39.6 ± 1
42.1 ± 2
42.7 ± 1
42.4 ± 1
41.8 ± 1
41.4 ± 1
41.7 ± 1
44.5 ± 2
43.8 ± 1
36.3 ± 1
43.9 ± 1
36.4 ± 1
43.9 ± 1
39.2 ± 2
39.6 ± 2
37.1 ± 1
41.9 ± 1
DLPs are also sometimes called “volcanic” low-frequency earthquakes (LFE) to contrast them with “tectonic” LFEs, one form of the slow-slip phenomenon (also called Episodic Tremor and Slip, ETS) recently discovered to occur deep on major faults [e.g., Ide et al., 2007; Vidale and Houston, 2012]. “Volcanic” DLPs tend to occur throughout a volume and may have a compensated linear vector dipole focal mechanism [Aso et al., 2013; Nakamichi et al., 2003], whereas “tectonic” LFEs tend to fall on a fault plane and have the focal mechanism of the dominant motion on that plane [e.g., Vidale and Houston, 2012]. While neither process is well understood, the process by which seismic energy is generated conceivably could be similar.
In this paper, we identify DLPs that are neither directly associated with volcanoes nor with the ETS-associated slow-slip LFEs but rather are located just above the stagnant tip of the mantle wedge at depths near the continental Moho. The presence of a stagnant wedge has been inferred from geodynamic simulations that match heat flow patterns and the expected properties of mantle cooled by proximity to cold subducting crust [Wada et al., 2008]. Herein, we explore the proposal that these DLPs provide evidence of active dehydration in the stagnant mantle wedge.
2 Catalog of Regular and DLP Earthquakes
The PNSN has catalogued long-period earthquakes for many years. Nichols et al.  reviewed these DLPs and found patterns under many volcanoes in the Cascades similar to those seen under Japan [Hasegawa and Yamamoto, 1994] and Alaska [Power et al., 2004]. With the observation of a large offset DLP on 24 July 2013, we reviewed the deep seismicity east of the subducting Juan de Fuca slab to reexamine the pattern of DLPs, particularly away from volcanoes.
We visually reviewed and adjusted or added picks for every earthquake in the PNSN catalog since 1990 that was deeper than 25 km and in the western three fourths of Washington and Oregon to improve picking consistency across events. We recalculated locations for earthquakes with revised picks, rejected events with poorly constrained depths, and assigned earthquake types of “regular” or “long-period” (if significantly deficient in energy above 10 Hz). More DLPs were seen in more recent years, probably because of improving station coverage. The increase in three component seismometers over the last decade allowed us to better detect S waves and greatly improve the depth resolution. We could not determine focal mechanisms owing to unclear first motions and limited coverage of the focal sphere, and full waveform inversion is beyond the scope of this paper.
We observe DLPs under most volcanoes along the arc, but also around 40 km depth significantly to the west of volcanoes in west-central Oregon (Table 1 and Figure 3). Station coverage in Washington is better than Oregon, so absence of offset DLPs in Washington is likely real and not due to missed detection. Seismicity within the subducting Juan de Fuca slab is visible down to 100 km depth in the Puget Sound region and, much more sparsely, to 50 km depth in Oregon. The slab seismicity is consistently below the plate boundary location estimates [Preston et al., 2003; McCrory et al., 2012]. There are both tectonic and low-frequency events near the plate boundary associated with tremor [Vidale and Houston, 2012], but these are not locatable with normal network data processing and are not the subject of this paper.
Figure 4 plots all the DLPs in Oregon in cross section against seismicity, heat flow, and volcanoes in Oregon. A clear pattern emerges of (1) 40 km deep DLPs offset significantly to the west of the volcanoes, and (2) DLPs at a range of depths under volcanoes. Regular crustal seismicity is widespread in many places down to 25 km depth, reaching 30 km in a few locales including Puget Sound in Washington. Although we have detected several additional DLPs relative to those reported in Nichols et al. , the distribution of DLPs shallower than 30 km depth is unchanged, with DLPs dominantly occurring around most volcanoes, particularly those in Washington (Mount Baker, Glacier Peak, Mount Rainier, and Mount St. Helens). However, the additional detected DLPs outline a new pattern in the 35–45 km depth range west of the volcanic arc in Oregon.
As shown in Figure 4, four locales that are 40–80 km west of the volcanic arc have DLPs near 40 km depth, all in Oregon. Our analysis revealed no offset DLPs in Washington. The largest cluster of 40 km deep DLPs is approximately 70 km southeast of Eugene, over 70 km west (trenchward) of the nearest major volcanic center. Intraslab seismicity is well west (trenchward) and deeper than these DLPs, as are very small low-frequency earthquakes located very near the plate interface as detected by focused studies with denser instrumentation further north [e.g., La Rocca et al., 2009] and tectonic tremor throughout Oregon [Boyarko and Brudzinski, 2010; Wech and Creager, 2011]. Analysis by the PNSN has revealed no DLPs near the slab.
Given that tremor is often modulated by transient stresses, we checked the stresses from Earth and ocean tidal loading at the times and locations of each of the DLPs to see if there was a systematic relationship. In particular, the volumetric stress (related to dilatation) and the North-South normal stress (considered because it is oriented similarly to the background tectonic stress) did not correlate with DLP timing. This is consistent with prior results [Aso et al., 2011] for one patch of similar nontectonic DLPs beneath Osaka Bay, Japan, about 50 km from a volcano. The lack of tidal correlation is another indication that the DLPs differ from the slow slip nearby on the subduction zone, which is known to correlate with tides [e.g., Rubinstein et al., 2008].
3 DLPs Suggest Serpentine Dehydration
As far as we know, these are the first DLPs to be observed deep in the fore arc of a subduction-zone system, with the possible exception of Aso's study  of one location in Japan discussed below, and some less offset locations farther south near Yosemite [Pitt et al., 2002]. Figure 4 illustrates the remote position of these new DLPs relative to Cascade Range volcanoes and previously reported volcanic DLPs. If DLPs near volcanoes are caused by fluid or magma flow in the deep crust or upper mantle, then how best to explain these new DLPs that are so offset from volcanic centers and separated from the slab by the stagnant mantle wedge?
There have been a few other observations of DLPs away from volcanoes. In the review of deep seismicity in Japan [Hasegawa and Yamamoto, 1994], a number of DLP clusters that were near Moho depth and not under volcanoes were noted to underlie midcrustal reflectors that could be lenses of fluid, although they were mostly between volcanoes or in the back arc. One of the clusters, a nest of DLPs under Osaka Bay, was near the Moho at 30 km depth and interpreted as indicating the presence of fluids [Aso et al., 2011].
We interpret these offset DLPs in western Oregon to represent the embrittlement of serpentine within the stagnant mantle wedge. The position of the DLPs within geodynamical models for Cascadia [Wada et al., 2008] indicates that they reside within a temperature range of 450–600°C. Petrological studies predict, for the pressure range of 1.1–1.3 GPa at this depth, that antigorite would undergo dehydration either due to raising the temperature or lowering the pressure [Ulmer and Trommsdorff, 1995]. Acoustic emissions in antigorite samples have been documented at pressure-temperature conditions (580°C and 1.5GPa) [Dobson et al., 2002], near the predicted ranges of the offset DLP locations. Antigorite acoustic emissions were characterized by long-duration waveforms interpreted to represent the superposition of multiple small-slip events, analogous to tremor. The weakening or embrittlement of the rock is hypothesized to be caused by a drop in effective normal stress associated with dehydration [Murrell and Ismail, 1976; Raleigh and Paterson, 1965]; however, the mechanical details of the weakening mechanism continue to be studied [i.e., Chernak and Hirth, 2010].
The limited north-south extent of the offset DLPs is consistent with the hypothesis that the Oregon portion of the stagnant mantle wedge has undergone dehydration induced by a recent increase in temperature. This theory is corroborated by high heat flow and dispersed volcanism in the Oregon back arc at this latitude. The westward progression of rhyolitic volcanism across the High Lava Plains over the past 10 Ma is thought to be the consequence of anomalously hot mantle that has upwelled beneath eastern Oregon and has been entrained in the corner flow [Long et al., 2012]. The clustering of rhyolitic activity on South Sister in the late Quaternary is evidence that the anomalous mantle material has reached the arc [Fierstein et al., 2011]. We propose that the offset DLPs record the arrival of this anomalously hot mantle into the mantle corner, thereby conducting heat into the stagnant wedge. Our hypothesis can be tested through future geochemical studies that infer the variation in the temperature of melts along arc.
An alternative explanation is that the offset DLPs are a result of fluids rising from the subducting slab's eclogitizing crust through the mantle wedge to near the base of the continental crust. It is known that this type of fluid circulation causes hydration of mantle rocks, or serpentinization, reducing their seismic velocities [Bostock et al., 2002]. This reduced seismic velocity in the uppermost mantle leads to a “missing” Moho velocity contrast under the Puget Sound-Willamette trough, a feature of structural studies in the region since the early 1970s [Brocher et al., 2003; Crosson, 1972]. Note that the “missing Moho” area in Oregon, inferred from long baseline refraction studies in the Cascadia fore arc [Brocher et al., 2003], is very close to or includes DLP locations (Figure 3). The DLPs thus may be direct evidence for slab-derived fluids arriving near the Moho. If only 10–20 Myr may be required to saturate the mantle wedge with serpentine [Hyndman and Peacock, 2003], then the wedge may no longer be able to accept more fluid. But present models of the rheology of the mantle wedge [e.g., Abers et al., 2006] offer several possibilities concerning the salient question of the rate at which fluid can penetrate the stagnant wedge. One shortcoming of rising fluids as the cause of the offset DLPs is why they are restricted to Oregon and are not observed Cascadia wide.
We observe DLPs that locate in the Oregon fore arc and hypothesize, based on their location and long-period character, they are the result of dehydration embrittlement caused by recent heating. If our interpretation of a dehydration mechanism is correct, then the offset DLPs would fall along an antigorite stability contour within the mantle and demark the eastern edge of the serpentinized stagnant mantle wedge in Oregon. The DLPs would presumably sit below the Moho within the wedge. Given the specific temperature-pressure conditions required for dehydration, these observations can be used to calibrate future geophysical models. The introduction of fluids to the base of the crust provides an opportunity to induce partial melting; however, temperatures there would be too low. Nevertheless, the production of melt above the stagnant mantle wedge may explain enigmatic cases of volcanism observed above the fore arc, such as the Boring Volcanic Field [Evarts et al., 2009]. This new pattern of DLPs revealed by our analysis shows that fluids are more mobile than expected and offers new clues to the structure and dynamics of subduction zones.
The authors thank Naofumi Aso and Victor Tsai for enlightening discussions on DLP mechanics. This study was funded in part by the U.S. Geological Survey.
The Editor thanks Jacqueline Caplan-Auerbach and David Hill for their assistance in evaluating this paper.