Rethinking turbidite paleoseismology along the Cascadia subduction zone

Twenty-first century earthquakes are projected to take millions of lives (Holzer and Savage, 2013). Efforts to reduce the losses employ estimates of earthquake sizes and recurrence intervals. Paleoseismology has improved such estimates by extending earthquake histories thousands of years into the past.

Turbidite paleoseismology is based on observations that earthquakes are among the triggers for sediment gravity flows that evolve downslope into turbidity currents, and that shaking may cause slopes to fail in the same few minutes at sites widely separated along fault strike, a coincidence not expected of other potential triggering mechanisms (Talling, 2014). If turbidites indicative of earthquakes can be positively identified and correlated, their geographic extent would provide a measure of earthquake size, and their repetition would represent earthquake recurrence intervals. Settings shown to be conducive to turbidite paleoseismology include lakes in Switzerland and Chile (Strasser et al., 2013; Moernaut et al., 2014). Less favorable settings include parts of the continental slope off Sumatra, where great earthquakes rarely trigger turbidity currents (Sumner et al., 2013).

Interpretations of deep-sea turbidites are guiding hazard assessment at the Cascadia subduction zone, which extends 1100 km along the North American Pacific coast (Fig. 1). The study of turbidite paleoseismology began there with a confluence test for seismic triggering of turbidity currents (Fig. 2A). Holocene sequences of deep-sea turbidites were found to be similar above and below a confluence of deep-sea channels, and this similarity was ascribed to abyssal merger of turbidity currents that had been triggered simultaneously in different submarine canyons (Adams, 1990). Today, deep-sea turbidites provide the main basis for a proposed 10,000 yr Cascadia history that specifies 19 full-length ruptures and a similar number of shorter southern ruptures (Goldfinger et al., 2012) (Fig. 1B). This inferred history underpins tsunami scenarios (Priest et al., 2010), fault-energy budgets (Goldfinger et al., 2013b), and giant earthquake probabilities (Kulkarni et al., 2013).

In this paper, we reappraise deep-sea turbidites as guides to Cascadia earthquake hazards. We bring together data from legacy cores off the Washington coast that previous paleoseismological studies overlooked. We draw on physiography, stratigraphy, and sedimentology in disputing the confluence test and in questioning the turbidite basis for influential estimates of earthquake size and recurrence.

Cascadia’s largest network of deep-sea channels originates in submarine canyons that notch Washington’s continental shelf (Figs. 1–3^,1). The upper canyons debouch onto a lower continental slope ribbed with Pleistocene anticlines (Adam et al., 2004). The northern thalwegs from Nitinat, Juan de Fuca, and Quillayute Canyons drain into Juan de Fuca Channel, which heads at the apex of Nitinat Fan and descends its eastern edge (Fig. 3A). Southern thalwegs emanating from Quinault, Grays, and Willapa Canyons merge on the lower slope. Their shared outlet reaches an abyssal confluence with Juan de Fuca Channel at the head of Cascadia Channel, which drains southward between Nitinat and Astoria Fans and continues westward through the Blanco Fracture Zone. The network is a palimpsest, shaped during Pleistocene glaciations and selectively overwritten in the Holocene (Barnard, 1978, p. 110).

Turbidity currents were not always confined to the network’s thalwegs. Spillover paths to lower Juan de Fuca Channel are marked by abyssal plunge pools and sediment waves west of the Columbia River (Fig. 3A), a source of Pleistocene hyperpycnal flows (Normark and Reid, 2003). An upper reach of Cascadia Channel transmitted Holocene flows that overtopped its banks (Griggs and Kulm, 1970).

Cascadia turbidite paleoseismology has roots in sedimentological studies that were unconcerned with earthquakes. Three of these studies defined deep-sea stratigraphy off the Washington coast, in the south near the head of Cascadia Channel (Griggs, 1969), in the north on upper Nitinat Fan (Carson, 1971), and along the continental slope (Barnard, 1973). We draw on this coverage more fully than did Adams (1990) and Goldfinger et al. (2012), who used data from one-quarter of the legacy cores in Figure 3 and overlooked the legacy cores off northern Washington.

Supporting information for Figure 3 includes an extended explanation (Table DR1 in the GSA Data Repository²), a core list (Table DR2), volcanic-ash data (Table DR3), notes on inferred flow paths (Table DR4), and logs of cores on Nitinat Fan and the Washington slope (Figs. DR1 and DR2). An index map to cores keys them by source (Fig. DR3). Figure DR4, showing plunge pools west of Willapa Bay, includes EM 122 multibeam bathymetry and acoustic backscatter data derived from Holbrook et al. (2012). Figure DR5 schematically relates findings in Figure 3 to the confluence test of Adams (1990).

Holocene stratigraphy off Washington differs between north and south, both on the continental slope and on the adjacent abyssal plain. The differences involve numbers of successive turbidites and abundance of reworked Mazama ash. Erupted sometime between 7500 and 7800 yr ago, Mazama ash was delivered to the sea by the Columbia River (Peterson et al., 2012) and to the deep sea by turbidity currents (Nelson et al., 1968). Offshore, the ash serves as a tracer of Columbia River sediment (Fig. 3D), and its lowest stratigraphic occurrence provides a widely used stratigraphic datum (Figs. 3A, 3B).

Off northern Washington, post-Mazama turbidites are few and volcanic ash contents are low (Figs. 3A, 3D). Core logs and supporting analyses (Carson, 1971; Barnard, 1973; Goldfinger et al., 2012) are available for 11 sites within or near Juan de Fuca Canyon and upper Nitinat Fan: 63-18, 63-19, 63-20, 39-8, 39-9, 29-23, 39-26, 39-27, M9907-5, 29-25, and 39-7. The cores at all these sites show six or fewer post-Mazama turbidites. At M9907-5, where the likely number is between 3 and 6, Goldfinger et al. (2012, p. 42–43) counted 13 by positing, implausibly, that turbidites high in the piston core overlie the uppermost turbidites of the associated gravity core (details in Table DR4, path N1). No reported ash content exceeds 10% at any of the 11 sites. Only from site 63-17, between Quillayute and Quinault Canyons, did a sample from the northern Washington slope yield an ash content of 10%–35%.

Off southern Washington, post-Mazama turbidites are comparatively numerous, and reworked Mazama ash is widely abundant. These characteristics prevail on the slope west and south of Quinault Canyon, and along channels on the adjoining abyssal plain: the lower reaches of Juan de Fuca Channel, the nearby outlet from southern canyons, and Cascadia Channel (Fig. 3). Some of the sequences contain a dozen or more post-Mazama turbidites (yellow numerals, Fig. 3A), and in many sequences the maximum volcanic ash contents exceed 35% (Figs. 3A and 3B; individual analyses from five slope cores, Fig. 3D). Superimposed on these shared properties is pronounced variability in turbidite thickness, grain size, and bedding, and in the thickness of intervening hemipelagic clay, near the head of Cascadia Channel (Fig. 3C).

These north-south contrasts in deep-sea stratigraphy parallel trends in Holocene sediment supply on the Washington continental shelf. In the north, the Holocene transgression isolated Juan de Fuca Canyon from sediment of the Fraser River (Williams and Roberts, 1989). Reworking of glacial drift on the ocean floor accounts for much of the Holocene deposition in the Strait of Juan de Fuca and on the adjoining continental shelf (Herzer and Bornhold, 1982; Hewitt and Mosher, 2001). From the south, by contrast, Columbia River detritus has produced a mid-shelf silt of Holocene age that exceeds 20 m in thickness west of Willapa Bay and thins to the northwest (Fig. 3A). Upper reaches of Willapa and Quinault Canyons indent this mid-shelf silt (Nittrouer et al., 1979; Wolf et al., 1999), and Quinault Canyon is known to receive modern shelf sediment during storms (Carson et al., 1986).

These congruent patterns in turbidite stratigraphy and sediment supply cast doubt on a longstanding assumption about the provenance of Holocene turbidites in lower Juan de Fuca Channel, on the abyssal plain off southern Washington (location in Fig. 2C). Adams (1990) and Goldfinger et al. (2012) assumed that these turbidites represent flows that originated in canyons off northern Washington (paths N1 and N2; Fig. 2A; Table DR4). This routing requires flows that deposited few turbidites and little Mazama ash on upper Nitinat Fan to have produced numerous turbidites and abundant Mazama ash farther downstream (Fig. 3A).

It is unlikely that numerous flows from northern canyons would have bypassed all core sites off northern Washington. Juan de Fuca Canyon is lined with dozens of slumps (Carson and McManus, 1969) that subsequent Holocene flows have not removed. In contrast, on the southern Washington slope, erosive flows are evidenced by a scarcity of slumps and by the presence of unconformities in canyon thalwegs (at 53-19, M9907-13, and 63-08), and these features coexist with Holocene turbidites about as numerous and ash-rich as those in lower Juan de Fuca Channel (Figs. 3A and 3B).

We therefore propose that Holocene flows from canyons off the southern Washington slope spilled into lower Juan de Fuca Channel (Fig. 2B, paths S1–S3). The most important of the proposed paths would enable flows from upper Quinault Canyon to reach the abyssal plain as far north as site 48-8, the northernmost known occurrence of numerous ash-rich turbidites in Juan de Fuca Channel. This path, S1, is evidenced by stratigraphy and appears consistent with heavy-mineral assemblages.

Holocene flows using path S1 are evidenced by 8 post-Mazama turbidites and abundant ash at 53-14, on the lower slope midway between upper Quinault Canyon and Juan de Fuca Channel. To account for these turbidites, westward discharge from the upper canyon would continue directly across an outside bend at the canyon outlet and would rise ∼200 m from there to pass through a saddle in an anticlinal ridge (Fig. 3A). The upper portions of turbidity currents are known to diverge from canyons or channels at outside bends (Piper and Normark, 1983). Off California, split flows have exited Monterey Canyon 150 m above its floor more than 12 times in the past 3000 yr (Fildani et al., 2006), and flows have overtopped a wall 280 m high at an outside bend of Eel Canyon (Lamb et al., 2008).

Holocene turbidite sand of lower Juan de Fuca Channel contains heavy-mineral assemblages consistent with sources on upper Nitinat Fan or Vancouver Island (Goldfinger et al., 2013a). However, such assemblages are also known from volcanic rocks of the Olympic Mountains (Carson, 1971, p. 81) and from surficial sand on the inner shelf north of Grays Harbor (Venkatarathnam and McManus, 1973). Furthermore, assemblages derived from Vancouver Island may be available for reworking, both along path S1 in Pleistocene turbidites derived from the Cordilleran ice sheet and along lower Juan de Fuca Channel in slide debris (Fig. 3A; Fig. DR4).

Facies variations impede stratigraphic correlation among cores near the head of Cascadia Channel (Fig. 3C). The uniformly thin, sand-poor turbidites in Juan de Fuca Channel at site 6705-2 contrast with the thick, sandy turbidites along the southern thalweg at 6705-4 and 6705-5. The core sites nearest the head of Cascadia Channel, 6705-6 and 6508-K1, show elements of both.

It has been proposed that sandy units several centimeters thick correlate among cores taken hundreds of kilometers apart in mostly unconnected channels and basins along the Cascadia subduction zone (Goldfinger et al., 2012). In Pleistocene deposits, turbidite correlations this fine have been made along a North Atlantic levee (Hesse, 1995), but are statistically precluded by variable facies on Nitinat Fan (Carson and McManus, 1971). Facies variability also impedes fine correlations among Holocene turbidites near the head of Cascadia Channel (Fig. 3C) and between lower Juan de Fuca Channel and lower Cascadia Channel (Atwater and Griggs, 2012, their figure 5).

The confluence test of Adams (1990) asks whether turbidity currents off Washington were triggered regionally by earthquakes or locally by other means (Fig. 2A; Fig. DR5A). The regional earthquake hypothesis fails if the numbers of post-Mazama turbidites differ greatly among thalweg sites above and below the head of Cascadia Channel. In the test design, flows from northern canyons aggrade lower Juan de Fuca Channel sites 6705-2 and 6705-6; flows from southern canyons register on the slope at 53-18 and downstream at 6705-5; and flows from both north and south reach the confluence site 6508-K1 and are capable of reaching Cascadia Channel sites 6509-15 and 6609-24 (Figs. 1A, 3A, and 3B). Citing similarities in turbidite counts among these 7 sites, Adams (1990) concluded that each of 13 post-Mazama earthquakes shook the entire region, north and south.

The previously overlooked cores off northern Washington confound this venerable result (Fig. 2B; Fig. DR5B). Regional seismic triggering fails a rerun confluence test in which post-Mazama turbidites are few off northern Washington (range of 0–6) but numerous off southern Washington (range of 8–14; Fig. 3). The test design also fails if Holocene flows from northern canyons rarely reached lower Juan de Fuca Channel.

Geologic dating usually lacks the time resolution to show whether a long fault broke as a whole during the seconds or minutes of a single earthquake, or whether it ruptured piecemeal in a series of lesser earthquakes distributed across days, years, or even decades (Nelson et al., 1995). The 10,000 yr earthquake history proposed by Goldfinger et al. (2012) seemingly overcomes this ambiguity, because it includes 19 full-length ruptures but no instance of serial rupture (Fig. 1B). This inference of long ruptures hinges, however, on centimeter-scale turbidite-bed correlations in which each correlated bed represents a pulse of seismic shaking that was felt during the same few minutes along hundreds of kilometers of the subduction zone (Goldfinger et al., 2012, p. 135–136). Correlation this exact does not accord with the variable turbidite facies noted above, and it would not be expected of earthquake motions that vary along fault strike, of slope failures that lag the shaking, or of tributaries of unequal length that issue flows at different times (Atwater and Griggs, 2012, p. 13–17).

As interpreted by Goldfinger et al. (2012), Holocene turbidites of the lower Juan de Fuca Channel help define rupture length by representing megathrust rupture off northern Washington. If, however, these turbidites reflect spillover from canyons off southern Washington, they do not reliably show whether the fault broke north of those canyons.

Submarine mass movements need not correspond dependably with earthquakes (Völker et al., 2011; Goldfinger et al., 2012, p. 138; Sumner et al., 2013). Off northern Washington, Holocene turbidites from Juan de Fuca Canyon compare to less than one-third of the 19 Holocene ruptures posited to have extended beneath it (Figs. 1B and 3A). Conversely, off southern Oregon, turbidites outnumber megathrust earthquakes at a locality where the region’s most recent megathrust earthquake (A.D. 1700) predates three turbidites of unknown cause (Goldfinger et al., 2013c, p. 2141).

In the earthquake history of Goldfinger et al. (2012), the Cascadia plate boundary breaks twice as often off northern California and southern Oregon as off Washington and British Columbia (Fig. 1B). This estimate requires one-to-one correspondence of turbidites and earthquakes off Oregon and Washington (Goldfinger et al., 2012, p. 92, 112), in contrast with the likely turbidite deficit off northern Washington and the observed surplus off southern Oregon.

Holocene turbidity currents have partly overwritten a Pleistocene channel network off Washington. The resulting turbidite stratigraphy reflects latitudinal differences in Holocene sediment supply to canyon heads and probable Holocene spillover from canyons on the southern Washington slope. The stratigraphy has provided important estimates of earthquake magnitudes and recurrence intervals, but these estimates hinge on simplifying assumptions about flow paths, stratigraphic correlation, and record completeness. The way forward requires renewed attention to sediment supply, flow initiation, downstream pathways, and uncertainty in turbidite correlations.

We are particularly indebted to John Adams, Danny Brothers, Jake Covault, Audrey Dallimore, Chris Goldfinger, Joan Gomberg, Ross Heath, Sam Johnson, Harvey Kelsey, Katherine Maier, Patricia McCrory, Alan Nelson, Hans Nelson, David Piper, Brian Romans, David Scholl, Esther Sumner, Peter Talling, Rob Witter, Ivan Wong, and an anonymous referee.