Diverse rupture modes for surface-deforming upper plate earthquakes in the southern Puget Lowland of Washington State

The seismic hazard of active convergent margins derives not only from great earthquakes on the megathrust, but also from shallow faulting in the actively deforming forearc. Along the Cascadia margin, the cities of Seattle and Tacoma occupy the Puget Lowland, a structurally complex, seismically active forearc trough between the Coast Range and the Cascade volcanic arc. Deformation of the lowland is caused by the clockwise motion of western Oregon, pushing northward and compressing western Washington against slower moving Canada (Fig. 1A; Wells and Simpson, 2001; McCaffrey et al., 2007). North-south shortening of the Puget Lowland at ∼4.4 mm/yr (McCaffrey et al., 2007) has produced thrust fault–bounded uplifts that subdivide the lowland into four basins (Fig. 1B; Pratt et al., 1997; Brocher et al., 2001; Blakely et al., 2002). Much of the north- to northeast-directed strain in the southern Puget Lowland is accommodated by deformation in the Seattle fault zone, the Tacoma fault zone, the Saddle Mountain deformation zone, and along the Olympia fault (Figs. 1B, 1C).

The Puget Lowland tectonic setting is similar to that of southwest Japan, where the westward-migrating forearc has deformed the Seto Inland Sea into a series of basins and uplifts bounded by faults. One of these, the Nojima fault, produced the 1995 M_w6.9 Hyogoken Nanbu (Kobe) earthquake, which killed more than 6400 people, destroyed the port of Kobe, and caused $100 billion in damage (Chang, 2010). Unlike the Inland Sea of Japan, the historical record of seismicity in the Puget Lowland is short, and information about past large, surface-deforming earthquakes must come from paleoseismology. In this densely forested, heavily glaciated region, the most successful paleoseismology studies begin with aeromagnetic mapping and seismic reflection surveys to identify potentially active fault zones, move to searches for fault or fold scarps and deformed shorelines on lidar (light detection and ranging; i.e., airborne laser swath mapping) imagery, and follow up with studies of Holocene stratigraphy beneath scarps and shorelines to date prehistoric earthquake deformation (Barnett et al., 2010; e.g., Blakely et al., 2002; Johnson et al., 2004a; Sherrod et al., 2004a, 2008; Kelsey et al., 2012).

Paleoseismology has uncovered several aspects of fault and earthquake behavior in the Seattle-Tacoma-Saddle Mountain fault systems.

1. A very large earthquake is recorded by an uplifted marine terrace first identified at Restoration Point on Bainbridge Island (Fig. 2A). The terrace, uplifted during an earthquake on the master-ramp thrust of the Seattle fault, can be traced over tens of square kilometers on the fault hanging-wall block (Bucknam et al., 1992; Sherrod et al., 2000; Kelsey et al., 2008; Figs. 2A and 3).

2. Surface-rupturing reverse faults are well documented in the hanging-wall blocks of the Seattle and Tacoma faults, as well as along faults of the Saddle Mountain deformation zone (Fig. 1C; Witter et al., 2008; Blakely et al., 2009; this paper). In the central Seattle fault zone, the surface ruptures are on short, en echelon backthrusts (Nelson et al., 2003a), and their relation to slip on the master thrust at depth is problematic (Kelsey et al., 2008).

3. Widespread evidence for surface-deforming earthquakes within ∼100–200 yr of 1000 cal yr B.P. extends over at least 800 km² (Bucknam et al., 1992; Sherrod, 2001) and probably >4000 km² (Blakely et al., 2009; Martin and Bourgeois, 2012) of the southern Puget Lowland. This includes the Seattle fault zone, the Tacoma fault zone and Tacoma basin, and the Saddle Mountain deformation zone (Fig. 1C; Atwater and Moore, 1992; Bucknam et al., 1992; Bucknam and Biasi, 1994; Atwater, 1999; Sherrod et al., 2000; Hughes, 2005; Kelsey et al., 2008; Blakely et al., 2009; Arcos, 2012).

4. Apparent quiescent intervals of 6000–8000 yr have been inferred for 2 faults marked by short scarps of the central Seattle fault zone (Nelson et al., 2003a; this paper). The significance of the apparent clustering of earthquakes on these faults in the late Holocene is unclear, partly because the structural relation of the faults to the master-ramp thrust is not understood.

In this paper we reconstruct the prehistory of large earthquakes in the southern Puget Lowland (Fig. 1B). Our objectives are to interpret new and previously published evidence for surface-deforming earthquakes above the master-ramp thrusts of the Seattle and Tacoma faults, to compare earthquake chronologies among sites, and to assess the timing of large earthquakes on these faults and the kinematically related Saddle Mountain deformation zone. To do this, we describe, date, and interpret new evidence of surface-deforming earthquakes in nine fault scarp trenches in the Seattle and Tacoma fault zones (Nelson et al., 2003b, 2008; Sherrod et al., 2004a) and from wetland stratigraphy adjacent to one of the trenched scarps in the Seattle fault zone. We synthesize these new interpretations with prior studies of uplifted, subsided, or warped shorelines (Bucknam et al., 1992; Sherrod et al., 2000; Sherrod, 2001; Kelsey et al., 2008; Arcos, 2012), fault scarp and fold scarp stratigraphy (Nelson et al., 2002, 2003a; Witter et al., 2008; Blakely et al., 2009), wetland stratigraphy adjacent to scarps (Hughes, 2005; Polenz et al., 2011), strong-shaking–induced liquefaction and landslide deposits (Fig. 1C; Karlin and Abella, 1992, 1996; Pringle et al., 1998; Sherrod, 2001; Karlin et al., 2004; Martin and Bourgeois, 2012; Arcos, 2012), and the deposits of tsunamis generated by fault rupture (Atwater and Moore, 1992; Bourgeois and Johnson, 2001; Arcos, 2012; Martin and Bourgeois, 2012). A key part of our synthesis is an evaluation of the more than 260 ¹⁴C ages obtained during paleoseismic studies in the southern lowland. Using stratigraphic, geomorphic, and age evidence to correlate earthquakes from site to site, we selected ages that most closely limit the times of late Holocene earthquakes (Table 1; Table S1 in the Supplemental File¹) and calculated time intervals for each earthquake.

The subsurface geometry of the Seattle fault zone is the subject of debate. Johnson et al. (1994, 1999), Pratt et al. (1997), and ten Brink et al. (2002) interpreted the Seattle fault zone as a complex, south-dipping (∼40°) master-ramp thrust (Fig. 3A). Short, south-facing scarps in the central Seattle fault zone, such as along the Toe Jam Hill fault on Bainbridge Island, were interpreted as shallow, north-dipping faults antithetic to the master-ramp fault (Figs. 1C, 2A, and 3A). In a variation of this model (adopted by ten Brink et al., 2006), Brocher et al. (2004) proposed a passive-roof duplex in which the shallow faults root into a gently north-dipping detachment fault above the south-dipping master-ramp fault (Fig. 3B). Liberty and Pratt (2008) proposed that the eastern part of the Seattle fault zone is a thrust fault that extends upward into a fault-propagation fold with a breakthrough forelimb fault (Fig. 3C). The tip of the breakthrough fault forms a blind growth fold and may be the fault exposed in a trench across the Vasa Park scarp (Fig. 1C; Sherrod, 2002).

In the central Seattle fault zone, Kelsey et al. (2008) incorporated aspects of the Brocher et al. (2004) and ten Brink et al. (2006) models, but Kelsey et al. (2008) proposed that the upper part of the fault zone is a doubly vergent, wedge-thrust, fault-bend fold. The north-dipping reverse faults marked by scarps on Bainbridge Island are interpreted as shallow flexure slip, bedding-plane thrust faults that root 4–6 km deep in the active axial surface of the wedge-thrust fold (Figs. 2A and 3D). In this model, the shallow bedding-plane faults may slip during moderate earthquakes independent of slip on the deeper master-ramp thrust.

The Tacoma fault zone consists of a series of faults and folds with a north-dipping reverse fault on its western end, a growth fold above a deeper fault beneath Case Inlet, and a fault that intersects a northwest-trending monocline on its eastern end (Figs. 1C, 2B, and 3B; Pratt et al., 1997; Brocher et al., 2001, 2004; Johnson et al., 2004b). Shallow seismic reflection work has confirmed the growth fold and other folds to the west, and identified small faults near the northwest-trending monocline that indicate its postglacial growth. Clement et al. (2010) interpreted the reflection data to suggest Holocene folding and faulting along the Olympia fault (or fold) on the south edge of the Tacoma basin. If north-dipping reverse faults of the Tacoma fault zone sole into a south-dipping master Seattle thrust, they do so at a depth of at least 10 km (Fig. 3B; Brocher et al., 2001, 2004; Johnson et al., 2004b; Clement et al., 2010).

The different structural models affect interpretations of the magnitude of the earthquakes recorded by fault scarps and deformed shorelines in the Seattle and Tacoma fault zones. All of the models are consistent with the largest earthquakes being produced during ruptures tens of kilometers long on the master-ramp thrust of the Seattle fault zone and/or on faults of the Tacoma fault zone that extend deeper than 8–10 km. Kelsey et al. (2008) termed these “regional earthquakes,” based on shorelines uplifted >3 m over an area 5–6 km wide and tens of kilometers long on the hanging wall of the Seattle fault zone during the Restoration Point earthquake. We refer to these largest upper plate earthquakes as “very large,” and follow previous studies in inferring that they were M 7–7.5 (e.g., Bucknam et al., 1992; Sherrod, 2001; Blakely et al., 2009).

As interpreted in Kelsey et al. (2008), earthquakes recorded by short scarps or uplifted terraces extending <1–2 km from faults may have rupture areas limited to the upper 4–6 km of a fault-bend-fold wedge over distances of 8–10 km (Fig. 3D). These events were considered “local earthquakes” by Kelsey et al. (2008) with inferred “moderate” magnitudes of 5.5–6.0. If instead, the ruptures with limited areas of surface deformation sole in a master thrust when it slips during earthquakes (ten Brink et al., 2002; Brocher et al., 2004; Liberty and Pratt, 2008; Pratt and Troost, 2009), rupture areas would be at least an order of magnitude larger, likely generating “large” earthquakes of ∼M 6.5–7.0 (Figs. 3A, 3C). If large, these earthquakes apparently produced less than meters of surface deformation over areas less than kilometers wide and tens of kilometers long.

Farther west, Witter et al. (2008) and Blakely et al. (2009) argued that postglacial movement on faults of the Saddle Mountain deformation zone help accommodate northward-directed slip on the deep thrusts of the Seattle fault zone where it may impinge on the southeast flank of the Olympic Mountains (Fig. 1C; Polenz et al., 2011). Blakely et al. (2009) concluded that an earthquake on the Saddle Mountain fault ca. 1000–1300 cal yr B.P. was M 6.5–7, what we term a “large earthquake.”

Typical 50–200-yr analytical uncertainties and additional stratigraphic and sample context uncertainties in ¹⁴C dating (Sherrod, 2001; Nelson et al., 2003a; Kemp et al., 2013) preclude distinguishing individual earthquakes on the Seattle fault zone, the Tacoma fault zone, and the Saddle Mountain deformation zone that occurred ∼1000 cal yr B.P. Because our paleoseismic methods cannot distinguish between a series of large earthquakes or one very large earthquake ∼1000 cal yr B.P. on these faults, we refer to surface-deforming earthquakes within ∼100–200 yr of 1000 cal yr B.P. as within the “millennial earthquake series.” We term the largest earthquake of the millennial series, first identified from extensive uplifted shorelines on Bainbridge Island (Fig. 2A; Bucknam et al., 1992; Sherrod et al., 2000), the Restoration Point earthquake. Based on less distinct and spatially much more limited stratigraphic and geomorphic earthquake evidence, other, probably smaller, surface-deforming earthquakes in the Seattle fault zone and possibly the Tacoma fault zone occurred before (Nelson et al., 2003a; Kelsey et al., 2008) and after (Sherrod, 2001; Nelson et al., 2003c; Karlin et al., 2004) the Restoration Point earthquake. Although ¹⁴C ages that limit the times of surface faulting or folding help us interpret the evidence of other earthquakes, in most cases the ages are merely consistent with site-to-site correlations of earthquakes based on stratigraphic and geomorphic evidence (Sherrod, 2001; Nelson et al., 2003a; Kelsey et al., 2008). We return to the topic of rupture extent during the millennial earthquake series in the Discussion.

Most evidence of surface deformation during earthquakes of the past few thousand years in the Seattle fault zone have come from studies of fault scarps and deformed shorelines on Bainbridge Island and the Point Glover peninsula to the south (Figs. 1C and 2A; Table 1; Table S1 in the Supplemental File [see ^{footnote 1}]). Other sites with evidence of surface deformation include Alki Point near Seattle, Vasa Park in the eastern part of the fault zone (Sherrod, 2002), and sites along faults of the Saddle Mountain deformation zone. In this section, we summarize earlier investigations in the central Seattle fault zone, describe new evidence for surface-rupturing earthquakes along scarps, and explain how deformed shorelines are related to earthquakes.

Methods used to investigate surface deformation were discussed by Nelson et al. (2003a, 2003b), Sherrod et al. (2004a, 2008), Witter et al. (2008), and Blakely et al. (2009). All basic data for scarp sites, such as complete trench logs, core descriptions, scarp topographic profiles, detailed descriptions and interpretations of stratigraphic units, notes explaining how fault slip was measured, and tables of lithologic, radiocarbon, structural, soil profile, and paleoecologic data can be found in Nelson et al. (2002, 2003b, 2008), Sherrod et al. (2004b), Witter et al. (2008), or in the Supplemental File (see ^{footnote 1}). Shoreline site methods were described in Sherrod et al. (2000), Sherrod (2001), and Kelsey et al. (2008). Our analysis and interpretation of key ¹⁴C ages from sites in the central Seattle fault zone follow the discussion of surface deformation data.

Beginning with the shoreline studies of Bucknam et al. (1992), investigators have attributed most uplifted shorelines in the central Seattle fault zone to uplift of the fault hanging wall during the Restoration Point earthquake. These include uplifted marine platforms located at Restoration Point, 5 km to the east at Alki Point, and along the shores of southern Bainbridge Island and the Point Glover peninsula (Figs. 1C, 2A, and 4; Table 1; Table S1 in the Supplemental File [see ^{footnote 1}]; e.g., Atwater and Moore, 1992; Sherrod et al., 2000; Harding et al., 2002; Ota et al., 2006; Kelsey et al., 2008). Conversely, the marsh at Winslow, 5 km north of Restoration Point on the footwall of the fault (Fig. 2A), probably subsided during the earthquake, as did West Point ∼4 km north of the uplifted platform at Alki Point (Fig. 1C; Atwater and Moore, 1992; Bucknam et al., 1992; Ekblaw and Leopold, 1993). Maximum and minimum limiting ¹⁴C ages for uplift and subsidence in these studies placed the earthquake between 550 and 1750 cal yr B.P. (Table 1; Table S1 in the Supplemental File [see ^{footnote 1}]; Bucknam et al., 1992). Shoreline uplift and a tsunami, caused by slip on faults of the Seattle zone, were dated at Gorst to 1260–560 cal yr B.P. (Fig. 1C; Arcos, 2012).

More precise ages for the Restoration Point earthquake came from matching ¹⁴C ages on different-aged groups of outer ring wood from trees that died about the time of the earthquake with fluctuations in the radiocarbon calibration curve (e.g., Kemp et al., 2013). In an excavation at West Point, Atwater and Moore (1992) identified deposits of the tsunami produced by sudden upper plate uplift during the earthquake. Scirpus sp. stems engulfed by sand exposed in the excavation were flattened by a log carried by the tsunami. By wiggle matching three ages on groups of rings from the log, Atwater (1999) narrowed the time of the earthquake to 1050–1020 cal yr B.P. (Table 1; Fig. 5). Close agreement among ages on Scirpus stems and log ages, and the good preservation of bark encircling the log, showed that log death did not predate the tsunami by more than a few years. Matching tree-ring patterns among the West Point log and the stumps of trees carried into Lake Washington (Fig. 1C) by shaking-induced landslides showed that the log and stumps probably died in the same season of the same year (Jacoby et al., 1992; Karlin and Abella, 1992). South of the Tacoma fault zone at Red Salmon Creek, Sherrod (2001) also used wiggle-matched ages from the rings of a Pseudotsuga menziesii stump, inferred to have been killed by coseismic subsidence, to date subsidence during a large earthquake to 1090–1010 cal yr B.P. (Table 1; Figs. 1C and 5).

Kelsey et al. (2008) inferred, as had previous researchers (e.g., Bucknam et al., 1992; Sherrod et al., 2000; Ota et al., 2006), that the widely distributed, largely continuous, uplifted marine platform in the Seattle fault zone was raised during the Restoration Point earthquake. They interpreted older, higher platforms that merge with the younger continuous platform within hundreds of meters of fault scarps as markers of fault slip during older earthquakes (Fig. 4). Because of the limited extent of the older terraces and their identification at only a few sites, Kelsey et al. (2008) inferred the older terraces to record uplift during moderate magnitude earthquakes on fault-bend-fold faults having lengths of <10 km. At three principal sites, i.e., where the scarp of the Toe Jam Hill fault intersects the west coast of Bainbridge Island (A in Fig. 2A), at the tip of the Point Glover peninsula (Fig. 4), and where a strand of the Seattle fault intersects Puget Sound south of Alki Point (B in Fig. 2A), the uplifted lower shoreline platforms, correlated with the similar platform at Restoration Point, are 2–9 m above Puget Sound. The older platforms on the hanging walls of the faults are ∼3 m (west Bainbridge), ∼4 m (Point Glover), and ∼5 m (Alki Point) higher than the wider, more continuous, lower platforms (terrace maps and profiles in Kelsey et al., 2008, figs. 4–6 therein). If gradual subsidence followed coseismic uplift at many of these sites, as illustrated by stratigraphy and fossils for the Restoration Point earthquake at Gorst (Fig. 1C), the present heights of many shorelines may be minimum values for coseismic uplift (Arcos, 2012).

The scarp of the Toe Jam Hill fault extends east-west 1.5 km across the southern part of Bainbridge Island in the Seattle fault zone (Figs. 1C and 2A). A synthesis of stratigraphy and ¹⁴C ages from 5 trenches across the 1–6-m-high scarp yielded a history of 3, or possibly 4, surface-rupturing earthquakes between 2.5 and 1.0 ka (Nelson et al., 2003a; site maps and complete data in Nelson et al., 2002).

Charcoal ¹⁴C ages from faulted and folded forest soil A horizons date two of the earthquakes, labeled B and C, to ca. 2.5 to 1.9 ka, primarily in the Crane Lake and Mossy Lane trenches. From correlations of earthquakes among trenches (Nelson et al., 2003a), only 3 ages limit the time of the second oldest earthquake (earthquake B), placing it in a 700-yr interval (Fig. 5; Table 1). Charcoal ages from A horizons that most closely constrain the time of earthquake C indicate an age of ca. 1.7–1.2 ka from the Crane Lake trench and ca. 1.3–1.2 ka from the Mossy Lane trench. If earthquakes C and D were correctly correlated by Nelson et al. (2003a) between the two trenches, the youngest maximum age and oldest minimum age from the Mossy Lane trench date earthquake C to 1350–1170 cal yr B.P. (Fig. 5; Table 1). The three other trenches lack firm evidence of earthquake C. The earliest possible earthquake (A), dated between 2.5 ka and 1.9 ka in the Crane Lake trench, may record a toppled tree rather than an earthquake.

Nelson et al. (2003a) correlated the youngest surface-deforming earthquake (earthquake D) in the five trenches with the Restoration Point earthquake. Evidence for the correlations included the youngest detrital ¹⁴C ages in the 1.3–1.0 ka range and the relative amount of fault slip or uplift, which was inferred from stratigraphic relations to be largest for the youngest earthquake in 3 of the 5 trenches. Eight ¹⁴C ages from the 5 trenches were interpreted as maximum ages for the time of earthquake D (Table 1). The two highest quality samples from the trenches, inferred to predate the earthquake by decades, gave maximum ages for earthquake D of 1230–960 cal yr B.P. (cedar leaf at base of fissure in Bear’s Lair trench; Table 1) and 1060–920 cal yr B.P. (sticks of forest-floor charcoal from a fissure in Crane Lake trench; Table 1).

Although Kelsey et al. (2008) acknowledged that older platform uplift at the west end of the Toe Jam Hill fault scarp on the west coast of Bainbridge Island was known only to predate the Restoration Point earthquake and postdate the time that regional Holocene sea level stabilized (ca. 7 ka), Kelsey et al. (2008) suggested that uplift of the older platforms here and at Point Glover occurred during earthquake C of Nelson et al. (2003a; Figs. 2A and 4). Because none of the older platforms have been dated, they could have been uplifted during either or both of earthquakes B and C, or perhaps during an earlier late Holocene earthquake.

Vented sand is a widespread effect of strong shaking during the millennial earthquake series (e.g., Whistler et al., 2002). Martin and Bourgeois (2012) applied stratigraphy and sedimentology to distinguish between sandy beds deposited by tsunamis or through liquefaction venting caused by seismic loading at four of seven previously identified sites in the southern lowland. Thorough work by Martin and Bourgeois (2012) at Lynch Cove (Fig. 1C) led them to reinterpret sandy beds as liquefaction features rather than tsunami deposits, as had been concluded by Hemphill-Haley (1996) and Jovanelly and Moore (2009).

The only long record of strong shaking during Holocene earthquakes in the southern lowland comes from Lake Washington, astride the Seattle fault zone (Figs. 1C and 5). Through study of more than 45 gravity and piston cores, sidescan swath imagery, and high-resolution seismic reflection profiling, Karlin and Abella (1992, 1996) and Karlin et al. (2004) identified 21 widely distributed turbidites, probably induced by strong shaking since 7.6 ka, for an average turbidite recurrence of 300–400 yr.

In the Spotted Frog trench on the IslandWood scarp on southern Bainbridge Island (Fig. 2A), till and other ice-contact deposits on Miocene volcaniclastic deposits were twice thrust over a Holocene forest soil developed on the ice-contact deposits (Figs. 2A and 6; Fig. S1 in the Supplemental File [see ^{footnote 1}]). Based on displacements of the ice-contact deposits and faulted colluvium from the first earthquake, slip during the first earthquake was <1.7 m and during the second earthquake was >0.9 m (measurements explained in Fig. S1 in the Supplemental File [see ^{footnote 1}]). Additional folding during these or possibly earlier earthquakes accounts for the vertical displacement of 3.5 m across the scarp. Two of four ¹⁴C ages on detrital charcoal in overridden soil horizons place the earthquakes in the same time frame as earthquakes C and D in the Toe Jam Hill fault trenches (Table S1 in the Supplemental File [see ^{footnote 1}]). An age of 1.2 ka (1290–1160 cal yr B.P.; Table 1) on charcoal from a thin AB horizon developed on hanging-wall-collapse colluvium deposited following the first earthquake shows that the second earthquake occurred after earthquake C (Figs. 5 and 6). If the ruptures on the IslandWood scarp coincide with those on the subparallel Toe Jam Hill fault scarp 1.3 km to the south, the first earthquake in the Spotted Frog trench correlates with earthquake C and the second earthquake correlates with earthquake D.

Bucknam et al. (1999) inferred that because no scarp was observed where the Toe Jam Hill fault intersects the uplifted platform that continues 2 km east to Restoration Point on the east coast of Bainbridge Island, faulting observed in trenches across the scarp must be older than uplift of the platform (Fig. 2A; Sherrod et al., 2000). However, Nelson et al. (2003a) thought it was unlikely that surface faulting on the Toe Jam Hill scarp would not have occurred during the very large Restoration Point earthquake (which they correlated with earthquake D). The scarp rapidly decreases in height from 6 m to <2 m as it approaches a landslide above the heavily forested platform on the east coast of the island (Nelson et al., 2002; Kelsey et al., 2008). A 30-m-long trench south of the projected intersection of the fault with the platform exposed 1–2.5 m of soft, massive marine sand on its surface, but no trace of a scarp was found in or near the trench.

To test for possible vertical offset of the uplifted platform by the fault we measured the platform’s shoreline angle (back edge) north and south of the fault where the platform intersects the hillside (Figs. S2 and S3 in the Supplemental File [see ^{footnote 1}]). We measured 48 topographic profiles on lidar imagery, parallel with and <200 m north and south of the projection of the Toe Jam Hill scarp, across the platform. We estimated the elevation of the shoreline angle of the platform along each profile by projecting the slope of the platform westward to the hillside. Means of shoreline angle estimates in groups of profiles north and south of the projection of the fault across the platform (n = 24 for each group) differed by 1.2 m. A paired t-test on the means rejects the null hypothesis (p << 0.05) that there is no difference in the elevation of the shoreline angles on either side of the projected trace of the fault. From this result, we infer that surface faulting displaced the platform during earthquake D, but the <1.5-m-high fault scarp in the soft marine sediment on the platform is too broad or too eroded to now identify.

Two east-west–trending fault scarps truncate glacial topography on the Point Glover peninsula ∼2 km southwest of the Toe Jam Hill fault scarp (Fig. 2A). The older of the two, the north-facing Point Glover scarp, is ∼1 km long and as much as 8 m high (Fig. 4; Nelson et al., 2003b; Kelsey et al., 2008). The younger, steeper Waterman Point scarp is 1.6 km long and 4 m high along its western third and tapers to 1 m high on the uplifted platform on the east side of the peninsula.

Three trenches across the Waterman Point scarp exposed weathered and glacially sheared Blakeley Formation sandstone and mudstone of Oligocene and late Eocene age overlain by till and proglacial lake and fluvial deposits (Figs. 4, 7, 8, and 9). We interpret that bedrock in the Madrone Ridge trench was thrust southward over weathered ice-contact deposits along fault F1 during a first earthquake (Fig. 7C). During a second earthquake, bedrock was thrust along fault F2, higher in the hanging wall, over soil A horizons and colluvium deposited in response to the first earthquake. Key evidence that requires a two-earthquake interpretation is a forest A horizon, developed on slope colluvium deposited following the first faulting event, which was sheared and/or overridden during the second faulting event. The faulted A horizon developed on scarp colluvium is particularly distinct in a supplemental exposure 2 m west of the mapped trench wall. A forest A horizon of this thickness (8 cm) takes decades to centuries to develop (e.g., Lukac and Godbold, 2011); therefore, the earthquakes occurred at least decades apart. The total vertical separation during both faulting events is <2.4 m, suggesting that the 4.3 m height of the scarp is the result of slip on faults F2, F3, and F4, or other unmapped faults beneath them, during earlier surface-deforming earthquakes (Fig. 7; slip measurements explained on trench log in Nelson et al., 2003b).

Except for one charcoal age of 0.2 ka in the youngest slope colluvium, which we infer was deposited in a young root cast, the youngest ages (1.2 ka) from the trench are on charcoal in the sheared A horizon beneath the hanging wall of fault F1. These relations show that both surface-rupturing earthquakes are younger than 1290 cal yr B.P. (ages of 1290–1180 cal yr B.P. and 1290–1140 cal yr B.P. in Table 1).

Approximately 100 m west of the Madrone Ridge trench, the Snowberry trench, across a steep, 3.5-m-high scarp, exposed Blakeley Formation that was thrust southward over weathered ice-contact deposits and glacial lake sediment along fault F1 (Figs. 4 and 8). Distinct, overturned blocks of forest soil A, O, and B horizons beneath mudstone and sandstone of the hanging wall suggest that the blocks were plucked from the soil developed on underlying glacial sediments and rotated during thrusting prior to collapse of the thrust tip. The orientations of shear planes within the jumble of rotated blocks of soil, fragments of bedrock, and glacial and root-stirred sediment under the hanging wall suggest rumpling and shearing of the soil and underlying units during a surface-rupturing earthquake. Structural relations in the Snowberry trench do not suggest more than one earthquake.

Four charcoal ages from A and O horizons beneath the hanging wall provide maximum ages for the time of faulting (Fig. 8). The three youngest ages were obtained on samples from a 10-mm-thick seam of charcoal in the O horizon attached to two overturned blocks. Based on excavation and measurement of the charcoal seam in the trench wall, the seam once covered at least 1.7 m² of the forest floor. Because it is continuous, undeformed by tree roots, and well preserved, we infer that the seam’s wood burned <200 yr prior to faulting. Two of the three younger ages (1.2 ka and 1.0 ka; Fig. 8) are consistent with the ages for faulting in the Madrone Ridge trench and the Toe Jam Hill fault trenches, but the third is younger (0.9 ka, age of 980–790 cal yr B.P.; Table 1). In our later analysis of ¹⁴C ages, we explain how the youngest age may be evidence for a second surface-rupturing earthquake in the Snowberry trench.

Approximately 600 m east of the Madrone Ridge trench, the Nettle Grove trench exposed Blakeley Formation thrust southward over sheared and weathered ice-contact deposits along faults F1, F2, and F3 during 1 or 2 earthquakes (Figs. 4 and 9). Faults F1, F2, and F3 may have slipped during the same earthquake or during two earthquakes, the first on F1 and F2, and the second on F1 and F3 (Fig. 9C). Total slip measured by stratigraphic separations along the 3 faults is at least 4.0 m. The vertical separation between the hanging wall and footwall is 3.5–3.7 m, whereas surface displacement measured across the scarp is ∼3.2 m (slip measurements explained on trench log in Nelson et al., 2003b). The youngest sampled charcoal in the trench (1.1 ka; Fig. 9) was found in the sheared A horizon beneath the hanging wall and fault F2. The age (1180–1010 cal yr B.P.; Table 1) is a maximum age for surface rupture at this site.

Thus, based on stratigraphy in the 3 trenches along a 700 m section of the Waterman Point scarp, interpretations range from 1 very large surface-rupturing earthquake to 2 large surface-rupturing earthquakes in the past 1300 yr. We infer the history of surface ruptures to be the same over such a short section of scarp and conclude that the stratigraphy exposed in the Madrone Ridge trench is more complete than that in the other two trenches. The vertical component of slip measured for each of the faulting events in the trenches compared with the height of the scarp also suggests that the scarp records multiple earthquakes.

To test the two-earthquake interpretation, we described and dated wetland stratigraphy at two sites adjacent to the Waterman Point scarp (Fig. 4; scarp map on lidar imagery with 1 m contours in Nelson et al., 2003b). Approximately 70 m east of the Madrone Ridge trench, the stream channel in a 14-m-wide valley cut into the scarp is vertically displaced ∼0.5–1.0 m. A cross section with 15 gouge cores extending south from the base of the fault scarp at the valley mouth shows that sediment of the past 8000 yr is <1 m thick (Madrone East core site, Fig. 10; Fig. S4 in the Supplemental File [see ^{footnote 1}] shows core descriptions with sketch map). Colluvium with fragments of peat derived from the scarp in cores 1 and 4, overlying the younger of two wetland O horizons developed on sediment ponded against the scarp, is bracketed by ages on Prunus-type seeds between 1.2 and 0.7 ka (Table 1). The ages suggest that the colluvium, probably deposited in response to faulting along the scarp, dates from the time of the two earthquakes inferred from the Madrone Ridge trench (and the 1–2 earthquakes recorded in the Snowberry and Nettle Grove trenches; Figs. 7–9).

At the Wataugua Beach core site, ∼140 m east of the Nettle Grove trench site, we found evidence consistent with a two-earthquake interpretation. A composite stratigraphic section with ¹⁴C ages compiled from 6 gouge cores 1–2 m south of the scarp-wetland contact (Figs. 4 and 11; Fig. S5 in the Supplemental File [see ^{footnote 1}] shows composite section description with sketch map) shows two beds of probable colluvium (units 5–7 and 10) separated by 0.24 m of peat and peaty mud (units 8 and 9). Radiocarbon ages on a twig from the top of unit 4 (0.8 ka) and from a fragile herb bract or bud from the upper contact of pond deposits mixed with the colluvium (unit 7; 0.9 ka) are consistent with the colluvium (units 5–7) having been deposited in a pond along the scarp shortly following an earthquake of the millennial series (Fig. 12). Flat-lying Pseudotsuga menziesii needles from the upper 10 mm of peat beneath the upper scarp colluvium 0.25 m higher in the section (unit 9) gave a similar age (0.9 ka), but a Pseudotsuga-type twig (petiole with distinct pegs) at the upper unit 9 contact 20 mm above the needles yielded an age ∼100 yr younger (0.8 ka, Fig. 11; 930–740 cal yr B.P., Table 1). Typical rates of deposition for peat and pond sediments in such a small wetland basin (less than a few mm/yr) suggest that 0.24 m of peat and pond deposits between units 7 and 10 took decades to accumulate. Although bedrock-derived unit 10 may be the disaggregated remains of a block of hanging-wall bedrock that rolled down the scarp decades after the first surface-rupturing earthquake, we infer that it is slope colluvium deposited in response to a second surface-rupturing earthquake. Site proximity and similar ¹⁴C ages suggest that the two possible earthquakes identified in the Wataugua Beach cores are the 2 surface-rupturing earthquakes identified in the Madrone Ridge trench, 140 m to the west.

Remnants of uplifted shoreline platforms north of the Waterman Point fault scarp are additional evidence consistent with a history of at least three late Holocene earthquakes on the Waterman Point and Point Glover faults. The continuous 8–10-m-high platform, which extends around both sides of the Point Glover peninsula, shows >6 m of uplift of this part of the hanging wall of the master Seattle thrust during the Restoration Point earthquake (Fig. 4; Bucknam et al., 1992; Sherrod et al., 2000; Kelsey et al., 2008). Two <50-m-wide remnants of a platform 70–200 m north of the Waterman Point fault and 2–3 m above the continuous platform may mark an earlier episode of uplift of the hanging wall of the fault during a pre–Restoration Point earthquake. These remnants are several meters lower than the platform at the northern tip of the peninsula uplifted during faulting on the Point Glover fault (Fig. 4), which also predates the Restoration Point earthquake (Kelsey et al., 2008). We also mapped remnants of a third, lowest shoreline platform 200 m southeast and 370 m northeast of Waterman Point (Fig. 4) at an elevation of ∼2 m, ∼4 m below the platform of the Restoration Point earthquake. These smaller remnants (<30 m wide and <50 m long) are consistent with 1–2 m of uplift north of the Waterman Point scarp during the younger surface-rupturing earthquake identified in the Madrone Ridge trench. They also match our interpretation of earthquake-induced deposition of the younger colluvium at the Wataugua Beach core site.

To correlate earthquakes from site to site and to make more precise estimates of the times of surface-deforming earthquakes, we reevaluated the stratigraphic context and types of materials dated for ∼260 ¹⁴C ages from paleoseismic sites in the southern Puget Lowland. For each dated sample, we considered the sample stratigraphic and geomorphic context, composition, size, weight, probable source, condition (e.g., decayed, abraded), potential for contamination with younger or older material, potential for being reworked or introduced from a younger horizon (e.g., burned roots), and consistency with other ages at the sample’s site and correlative sites (e.g., Nelson et al., 2003a; Bird, 2007; Kemp et al., 2013). For published studies we mostly made the same site-to-site correlations of earthquakes as made by the original investigators (e.g., Sherrod, 2001; Nelson et al., 2003a; Sherrod et al., 2004a). Of the 214 ages that place some constraint on the times of earthquakes (Table S1 in the Supplemental File [see ^{footnote 1}]), we selected increasingly smaller subsets of ages based on higher sample quality, more stratigraphically consistent ages, and more closely limiting inferred maximum or minimum ages on the times of earthquakes (Table 1). In most cases, we assumed the youngest of detrital ages from faulted or folded deposits to be the most closely limiting maximum age of an earthquake. As explained earlier and in Figure 12, however, we relied on stratigraphic context to infer that six of the detrital charcoal ages are probably minimum ages for earthquakes (Table 1). Following many previous paleoseismic studies (e.g., Fumal et al., 2002; Atwater et al., 2004; Lienkaemper and Williams, 2007; Fraser et al., 2010; DuRoss et al., 2011; Berryman et al., 2012), we then used Bayesian methods (Bronk Ramsey, 1995, 2009; Biasi and Weldon, 2009; Lienkaemper and Bronk Ramsey, 2009) to calculate a probability distribution for the times between our selected youngest maximum ages and oldest minimum ages (ages in bold in Table 1; Figs. 5 and 12; Table S2 in the Supplemental File [see ^{footnote 1}] lists OxCal program code). Although we use the OxCal-modeled probability distributions to correlate and date earthquakes, the error ranges of the distributions may not encompass all uncertainties in the times of earthquakes. For example, the distributions do not include additional types of sample context error, such as the residence time of charcoal on a forest floor (Gavin, 2001; Nelson et al., 2003a), because we have no way of accurately estimating them.

Limiting ages from the Crane Lake and Mossy Lane trenches on the Toe Jam Hill fault place earthquakes B and C of Nelson et al. (2003a) within intervals (2σ ranges of OxCal-modeled probability distributions) of 2650–1940 cal yr B.P. and 1350–1170 cal yr B.P., respectively (Fig. 5). If the youngest charcoal age from the Spotted Frog trench on the IslandWood scarp closely limits the time of the second faulting event in that trench (Fig. 6), then the first faulting event probably records earthquake C. The older shoreline platforms studied by Kelsey et al. (2008) may correlate with earthquakes B and/or C. Sherrod et al. (2000) listed coseismic uplift during several moderate magnitude earthquakes as one of three possible explanations for salinity changes dated to 3.5–2.5 ka in a sparsely sampled section of core from the Restoration Point marsh (Fig. 2A). Kelsey et al. (2008, p. 1595) considered these events as “a cluster of ∼2–3 folding (local) earthquakes.”

We follow Kelsey et al. (2008) in inferring that the Toe Jam Hill and the Waterman Point faults, the east-west trends of which are offset <1 km (Fig. 2A), ruptured during the same earthquake, most likely the Restoration Point earthquake. Our correlation of earthquakes between the two faults suggests, however, that both faults do not always rupture during the same earthquakes. Nelson et al. (2003a) attributed the youngest surface-rupturing event in the Toe Jam Hill fault trenches (earthquake D) to the Restoration Point earthquake, citing ¹⁴C ages and the greater slip exposed in trenches during that earthquake than for earlier earthquakes. In contrast, our dating of two colluviums (units 5–7 and 10; Fig. 11) in wetland cores along the Waterman Point scarp suggests that the older of the two faulting events on that scarp most likely correlates with the Restoration Point earthquake. An age on a conifer twig (930–740 cal yr B.P.; Fig. 12) from the top of the peat (unit 9) that separates the colluviums is ∼100 yr younger than the most precise age for the Restoration Point earthquake (1050–1020 cal yr B.P.; Figs. 5 and 12). Of course, the dating cannot preclude separate rupture of these scarps during other moderate to large earthquakes of the millennial series.

If, as we infer above, the unit 10 colluvium correlates with the second faulting event in the Madrone Ridge trench and with the second of the two possible faulting events in the Nettle Grove trench, then a younger surface rupture (earthquake E) broke the Waterman Point scarp at least decades after earthquake D in the Toe Jam Hill fault trenches. We infer this rupture to record the Restoration Point earthquake (Fig. 5). Trench mapping identified only one large surface-rupturing event in the Snowberry trench 100 m west of the two-event Madrone Ridge trench, but our OxCal analysis of limiting ages for earthquake D from the Toe Jam Hill fault suggests that the youngest age from the Snowberry trench may be a maximum age for earthquake E (Fig. 12) and possibly a minimum age for earthquake D. The age (980–790 cal yr B.P.) on wood charcoal in a layer of charcoal at the top of an overturned block of forest soil in the fault zone (Fig. 8) is too young (at 2σ) to date forest-floor charcoal incorporated into the fault zone during the Restoration Point earthquake (1050–1020 cal yr B.P.; Fig. 12). If the age is a minimum for earthquake D, then part of the slip along fault F1 in the Snowberry trench occurred during a second earthquake that is younger than earthquake D. An alternative statistical possibility is that the true age of the charcoal is within the youngest tail of its 3σ age distribution, in which case it might still provide a close maximum age for earthquake D.

Because we cannot distinguish between these two alternatives, we use the youngest Snowberry trench age with three other ages from the Wataugua Beach cores only to set an older limit for the time of earthquake E (Fig. 12). As we have no minimum ages for earthquake E, our full interval extends to A.D. 1851, when European settlers arrived in southern Puget Sound. Because we interpret our limiting ages as being close maximums, we use the Zero Boundary command in OxCal to skew the probability distribution for earthquake E toward its older end (methods of Bronk Ramsey, 2008, 2009; as explained by DuRoss et al., 2011). The skewed distribution (940–380 cal yr B.P. at 2σ) best represents our interpretation of the time of earthquake E.

In a series of stratigraphic, ¹⁴C dating, and paleoecologic studies, Bucknam et al. (1992), Sherrod (2001), Sherrod et al., (2004a, 2004b), and Arcos (2012) identified evidence of substantial late Holocene uplift and subsidence at 12 sites over much of the region of the Tacoma fault zone (Figs. 1C and 2B). Differences in wetland plant fossils and diatom assemblages across uplift and subsidence contacts indicate >1 m of vertical deformation at most sites. Overall, dating at five of the uplift sites and six of the subsidence sites constrains Tacoma fault zone deformation to the millennial earthquake series (1180–790 cal yr B.P.; Sherrod et al., 2004a). Ages that best constrain the time of shoreline deformation include leaf bases of Triglochin maritima that limit the uplift of the tide flat on the hanging wall of the fault (Lynch Cove; Fig. 2B) to shortly before 1170–960 cal yr B.P., and the ¹⁴C-wiggle-matched age of 1090–1010 cal yr B.P. on rings from a Pseudotsuga menziesii stump killed by earthquake subsidence at the Nisqually delta 25 km south of the fault (Red Salmon Creek; Figs. 2B and 5; Table 1).

The >1–3 m of subsidence measured at Little Skookum Inlet led Sherrod (2001; Fig. 1C) to speculate that this subsidence was the result of 2 earthquakes. Following such reasoning, an age of 970–760 cal yr B.P. on the outermost rings of a Pseudotsuga menziesii stump in growth position on the modern tide flat at Wollochet Bay postdates the most precise age for the Restoration Point earthquake (1050–1020 cal yr B.P.). Perhaps it records subsidence during a later earthquake or a stress-relaxation response of the crust to the Restoration Point earthquake decades after it (Sherrod et al., 2004b; Table 1).

The Catfish Lake fault scarps, which truncate drumlins and outwash channels on the hanging wall of the Tacoma fault (Figs. 1C and 2B), formed in the late Holocene. These linear, 1–6-m-high scarps extend 4.4 km east-west between Hood Canal and Case Inlet (Sherrod et al., 2004a, 2004b; Logan and Walsh, 2007; Nelson et al., 2008, sheet 1) and probably another 2 km to the east beneath Case Inlet (Figs. 1C and 2B; Johnson et al., 2004b; Clement et al., 2010). Seismic reflection profiling east of Allyn shows that the water bottom of Case Inlet on trend with the scarp is warped upward ∼2 m, suggesting Holocene growth of a fold above a blind thrust (Johnson et al., 2004b; Clement et al., 2010). The Blueberry trench across the western scarp showed probable late Holocene faulting of a B/E soil horizon developed on weathered till (Sherrod et al., 2004a, fig. 3D therein, 2004b). Reverse separation along at least 2 of 4 small faults offsets the top of the horizon <0.3 m. Although some folding in the till may have a glacial origin, the sharp truncation of the glacial ridges by the scarp at the trench site shows that the scarp formed after the retreat of the Puget ice lobe (later than 16 ka).

In the more recent Cedars trench, across a 6-m-high scarp 0.8 km east of the Blueberry trench (Fig. 2B), we exposed a 30-m-long section of subglacial, ice-contact, and proglacial deposits (Fig. S6 in the Supplemental File [see ^{footnote 1}]; trench log in Nelson et al., 2008, sheet 2). Bedding contacts along most of the trench are subparallel to the scarp, indicating that scarp folding is postglacial. Several lines of evidence show that a strongly folded syncline of subglacial or proglacial stream and lake deposits near the base of the scarp predates the retreat of the Puget lobe. Tight folds, truncation surfaces, boudins, and flame structures in lake beds of the syncline are evidence for subaqueous soft-sediment loading and flow. The onlap of its successive lake beds also shows that folding of the syncline was contemporaneous with deposition. A thermoluminescence age of 110–119 ka on silt-sized feldspar from one of the lake beds shows that synclinal folding considerably predates the last glacial maximum. The syncline axis is strongly oblique to the scarp trend, suggesting that the scarp is much younger than the syncline.

Near the crest of the scarp in the Cedars trench, parallel-bedded sand, deposited in a stream channel cut into the glacial deposits, contains angular, horizontally bedded, fragments of wood charcoal as young as 4.3 ka (Table S1 in the Supplemental File [see ^{footnote 1}]). The channel position near the crest of the scarp and the tilt of its bedding parallel with the scarp surface show that the stream could only have deposited the sand prior to scarp folding, and therefore that the scarp formed after 4.3 ka.

At least some of the uplift and folding along the Catfish Lake scarps probably occurred during the millennial earthquake series. An uplifted shoreline north of where the trend of the scarp intersects the western shore of Case Inlet is 4 m higher than its modern tide-flat equivalent south of the scarp south of Allyn (Sherrod et al., 2004a, fig. 2B therein). Stratigraphy beneath a marsh south of Allyn shows that a freshwater swamp developed on the tide flat following the most recent uplift. Seeds from the base of the swamp O horizon give a minimum limiting age for the uplift of 1060–830 cal yr B.P. (Sherrod et al., 2004b; Table 1), an age within the 1180–790 cal yr B.P. interval determined for Tacoma fault zone deformation by Sherrod et al., (2004a). In Catfish Lake and Mill Pond to the west of Allyn, Logan and Walsh (2007) dated the stumps of trees probably killed by lake water impounded during uplift of the scarp. Because the outer rings of the stumps have been eroded, the youngest of the stump ages (1180–940 cal yr B.P.; Table 1) is older than the uplift.

Near Stansberry Lake ∼6 km east of Allyn, a 1-m-high, 1.2-km-long, northwest-trending scarp offsets a 1.7-km-long, 4–12-m-high, east-northeast-trending scarp in the hanging wall of the Tacoma fault (Figs. 2B and 13; Sherrod et al., 2004a, 2004b, fig. 3B therein; see lidar figures in Nelson et al., 2008, sheet 2). Only the younger scarp is tectonic. Our thorough study of a 30-m-long trench across the older scarp, which exposed subglacial ice-contact and fluvial deposits, proglacial fluvial deposits, and overlying root-stirred colluvium, showed no sign of faulting or folding. Erosional unconformities cut on the glaciofluvial deposits beneath the upper half of the older scarp suggest that this gently sloping scarp is the margin of a wide proglacial channel, the opposite margin of which has apparently been eroded (Owl trench log in Nelson et al., 2008, sheet 2).

The Micah trench across the younger scarp exposed three zones of mostly subvertical faults that offset proglacial ice-contact and well-stratified fluvial deposits (Fig. 13; Nelson et al., 2008, sheet 2). Many of the faults trend near 330°, and the upper tips of strands in fault zones 2 and 3 extend upward into recently root-stirred sediment in the lower part of the modern forest soil. Strand tips in fault zone 1 did not extend into root-stirred sediment. However, because such strands are unreliable indicators of the relative age of slip on oblique-slip faults, we conclude that the faulting occurred during a single, probably late Holocene, earthquake. The maximum age of the earthquake is constrained by the youngest of six ages (1530–1400 cal yr B.P.; Table 1) on wood charcoal in the B horizon buried by a wedge of colluvium formed from collapse of the 0.4-m-high fault scarp in fault zone 2. The mean of three ages, from a tree root that grew following deposition of the colluvium, dates the earthquake at prior to 1060–960 cal yr B.P., a time interval within the millennial earthquake series.

Right-lateral horizontal separation of sand lenses in proglacial deposits and a flower-structure–like pattern of faults in fault zone 2 (particularly on the shorter of the two trench walls shown in Nelson et al., 2008, sheet 2) suggest significant lateral slip on the faults. However, the only distinct bedding in the proglacial units offset by the faults is horizontal, preventing measurement of the ratio of vertical to lateral slip. Total oblique slip on the faults may be several times the 0.6 m net vertical displacement across the three fault zones in the trench. Because the young scarp trends at a 60° angle to the trend of the surface projection of the main Tacoma fault in this area (170°), it probably records slip on a short, shallow fault in the hanging wall of the main fault probably caused by folding of the hanging wall. Clement et al. (2010) inferred that the small postglacial faults they imaged 7 km southeast of the Stansberry Lake scarps were either shallowly rooted, bending moment faults produced during folding of the northwest-trending monocline along this part of the Tacoma fault zone, or short strike-slip faults accommodating lateral movement parallel with the monocline. The latter origin is consistent with our oblique-slip interpretation of the faults in the Micah trench.

Complex sets of scarps trend northeast for 3.5 km from Hood Canal (Figs. 1C and 2B; lidar figures in Nelson et al., 2008, sheet 3; Polenz et al., 2009), 1–2 km south of the uplift and liquefaction site at Lynch Cove (Bucknam et al., 1992; Sherrod et al., 2004b; Martin and Bourgeois, 2012), and 6–7 km north of the Catfish Lake scarps. Their shore-parallel position and arcuate morphology suggest many of the larger scarps in the steep bluffs along Hood Canal are landslide head scarps. However, most of the scarps on the glaciated upland south of Hood Canal and trending to the northeast away from the landslide complex are probably tectonic rather than landslide scarps. Landslides with headscarps with the orientation and position of those near the northeast end of the upland scarps would require slide planes of exceptionally low angles. From the slight up-to-the-south displacement of reflectors at 100–200 m depth on a seismic reflection profile that crosses the scarps near their northeast end (Nelson et al., 2008, fig. 13 therein), we infer that the upland scarps formed in response to slip on a south-dipping reverse fault in the underlying Tertiary bedrock, or on a larger north-dipping reverse fault ∼1 km to the south. Reflectors suggesting a gently dipping slide plane, as would be expected from a low-angle landslide, are absent on the profiles. Hummocky morphology characteristic of a bulging landslide toe is also not present where it would be expected near the shore of Hood Canal at Lynch Cove (Fig. 1C).

Near the northeast end of the upland scarps, we also dug 2 trenches, ∼12 m long, which exposed steeply dipping normal faults in subglacial ice-contact deposits and overlying proglacial fluvial deposits on either side of a 60–80-m-wide graben (Fig. 14). In each of the two trenches on either side of the graben, the youngest dated fragments of detrital charcoal in the scarp-derived colluvium date the normal faulting to younger than 1290–1080 cal yr B.P. (Table 1; Fig. 14), a limiting age that places the time of faulting within the millennial earthquake series. We infer that the Sunset Beach fault scarps, as well as many of the landslide head scarps along Hood Canal (Sarikhan et al., 2007), formed or increased in height during the millennial earthquake series.

We used OxCal (methods of Fig. 12) with the 12 closest maximum and minimum limiting ages from Tacoma fault zone sites (Table 1) to calculate a fault-wide probability distribution for the time interval of the millennial earthquake series (Fig. 5). This calculation assumes that deformation at all the sites records the same very large earthquake or a cluster of large earthquakes in the span of a few decades (Sherrod et al., 2004a). The interval (1050–980 cal yr B.P.) closely overlaps the wiggle-matched age for a drowned stump inferred to record coseismic subsidence at Red Salmon Creek (1090–1010 cal yr B.P.; Sherrod, 2001) as well as the time of the Restoration Point earthquake (1050–1020 cal yr B.P.).

Recent magnetic, structural, and paleoseismic studies conclude that faults of the Saddle Mountain deformation zone record north-directed uplift of the Seattle uplift area during the Holocene (Fig. 1C; Witter et al., 2008; Blakely et al., 2009; Lamb et al., 2009; Polenz et al., 2010, 2011, fig. 8 therein). Early investigations of the Saddle Mountain deformation zone showed late Holocene slip on the Saddle Mountain East fault and ponding of a swamp against the fault scarp (Fig. 1C; Carson, 1973; Wilson et al., 1979; Table 1; Table S1 in the Supplemental File [see ^{footnote 1}]). Recently, Barnett et al. (2009) and Czajkowski et al. (2009) identified multiple late Holocene faulting events on a fault east of the Saddle Mountain East fault, and Witter et al. (2008) dated two surface-rupturing earthquakes to 17–7.7 ka and after 1.7 ka on a parallel fault to the west, the Saddle Mountain West fault (Fig. 1C). As noted by Jacoby et al. (1992), stumps killed by earthquake-induced flooding in Price Lake and nearby wetlands between the faults near Saddle Mountain give close maximum ages for the most recent movement on the faults (Blakely et al., 2009; Polenz et al., 2011; Table 1). Hughes (2005) used ¹⁴C ages on cones and needles from Price Lake and counts of rings from submerged stumps to limit the time of this earthquake to 1300–1000 cal yr B.P.. Preliminary correlations of tree rings date stumps killed by flooding as a result of surface rupture on Saddle Mountain faults to ∼1300–1200 cal yr B.P. (Snook, 2011). Normal fault displacement (2.5 m) on the Frigid Creek scarps, south of the Saddle Mountain faults, was dated by Blakely et al. (2009) to 5.6–0.3 ka (Fig. 1C; Table S1 in the Supplemental File [see ^{footnote 1}]).

In the Cargill Creek trench on the Saddle Mountain West fault, Witter et al. (2008) inferred that the younger of two charcoal samples in colluvium deposited after the most recent faulting probably postdated the faulting. Following this inference, we used the youngest maximum age from a drowned stump in Price Lake (Wilson et al., 1979; Hughes, 2005; Polenz et al., 2011, Appendix A therein) with this youngest charcoal age as a minimum age (ages of 1270–930 cal yr B.P. and 1170–940 cal yr B.P.; Table 1), to calculate an OxCal modeled probability distribution for the time of the most recent earthquake or earthquakes on the Saddle Mountain West fault (Fig. 5). For the Saddle Mountain East fault, we used the three youngest maximum ages from Price Lake to calculate a skewed probability distribution for its most recent earthquake, as we did for earthquake E in the Seattle fault zone. Both distributions overlap the time of the millennial earthquake series.

To the southeast at the bend in Hood Canal, cores and exposures on the Skokomish River delta reveal the sudden conversion of a tide flat to a forest due to uplift ∼1100 yr ago (Fig. 1C; Martin, 2011). Probable folding of the Skokomish River floodplain above a blind fault may date from about the same time (Polenz et al., 2010). The oblique, left-lateral Canyon River fault, 27 km to the west of the Skokomish River (Fig. 1C), also slipped <2000 yr ago (Fig. 1B; Walsh and Logan, 2007; Polenz et al., 2011).

Terrestrial landslides likely induced by strong shaking, particularly those that dammed stream valleys to form lakes, have been dated in Lake Sammamish near the eastern end of the Seattle fault zone (Fig. 1C), in the Cascade Mountains north and south of Puget Sound, and in the southeast Olympic Mountains (Logan et al., 1998; Pringle et al., 1998). Ages for submerged stumps in Lake Sammamish and four lakes in the Olympic Mountains suggest that lake dams may have coincided with the millennial earthquake series (Schuster et al., 1992; Logan et al., 1998). Other landslides are too imprecisely dated or are too far away from the southern lowland to infer that they record earthquakes in the Seattle or Tacoma fault zones.

Diverse rupture modes characterize the earthquake prehistory of the southern Puget Lowland. However, paleoseismology provides firm evidence for the magnitude of only the largest earthquake, named for Restoration Point. As much as 7 m of coseismic uplift at Restoration Point and >2 m of shoreline uplift along at least 40 km of the central Seattle fault zone imply a magnitude of 7–7.5 for this earthquake (Bucknam et al., 1992; Sherrod et al., 2000; ten Brink et al., 2006; Kelsey et al., 2008). Widespread tsunami, liquefaction, and landslide deposits are consistent with an earthquake of this size (Atwater and Moore, 1992; Blakely et al., 2009; Arcos, 2012; Martin and Bourgeois, 2012). At sites that record thousands of years of relatively continuous sedimentation, evidence attributed to the Restoration Point earthquake is more distinct and/or suggests greater deformation or stronger shaking than evidence for any other earthquake (Sherrod et al., 2000; Nelson et al., 2003a; Karlin et al., 2004; Kelsey et al., 2008).

Surface deformation during several historical earthquakes near M 7 on forearc and intraplate thrust faults is broadly analogous to that inferred for the central Seattle fault zone. Uplift of as much as 2.7 m in the Hikurangi forearc of New Zealand during the 1931 M_w7.8 Hawke’s Bay earthquake extended over an area ∼100 km long and as much as 20 km wide (Hull, 1990). Marine terrace uplift of 1–4 m during 7 prehistoric earthquakes on thrusts in the same region is inferred to record earthquakes of M 7–7.5 (Berryman et al., 2011). Surface ruptures during the 1896 M 7.2 Riku-u earthquake in Japan showed a maximum vertical displacement of 3.5 m along four fault segments (37 km total length), and vertical displacement due to rupture of 4 main fault segments (40 km) during the 1980 M_w6.7 El Asnam, Algeria, earthquake reached 5 m (Rubin, 1996; Wesnousky, 2008). Even the greatest vertical displacements measured across scarps of the largest continental thrust earthquakes (Kaneda et al., 2008a; Xu et al., 2009) do not exceed the uplift measured for the Seattle fault at Restoration Point. These and other historical earthquakes on forearc and intraplate thrusts produced very complex patterns of surface deformation with highly discontinuous fold and fault scarps of variable height and orientation on multiple fault segments (Philip et al., 1992; Rubin, 1996; Kelsey et al., 1998; McCalpin and Carver, 2009).

Widespread evidence for other large to very large Holocene earthquakes in the southern lowland is limited to the millennial series, and the number of earthquakes remains unclear. The spatial extent of evidence for uplift and subsidence (>600 km²) and the >1 m of deformation at 15 sites in the Tacoma fault zone imply a large earthquake or earthquakes within the millennial earthquake series (Fig. 1C; Bucknam et al., 1992; Sherrod, 2001; Sherrod et al., 2004a; Arcos, 2012). If, as assumed in our OxCal analysis of Tacoma fault zones ages (Fig. 5), this deformation was the result of a large to very large earthquake (Sherrod et al., 2004a; Smith and Karlin, 2009), empirical magnitude-rupture area relations suggest that its magnitude was near 7.0 (Wells and Coppersmith, 1994). Because empirical relations for thrust faults are based on a small number of earthquakes with complex, variable rupture patterns, such empirical-relation estimates of magnitude are more uncertain than those for normal or strike-slip faults (Wesnousky, 2008; McCalpin, 2009).

Two sites with long, relatively continuous records of sedimentation show evidence consistent with earlier large, possibly very large, earthquakes in the central Seattle fault zone. In a small marsh at Restoration Point (Fig. 2A), Sherrod et al. (2000) used transfer function statistical methods on fossil diatom floras from a core to infer that an abrupt change in core lithology reflected a sudden change in tidal environments. They suggested that the change was caused either by coseismic uplift or by the accumulation of a beach bar that partially blocked tidal access to the marsh ca. 7.2–6.4 ka. If coseismic, estimated uplift during this event was 1.5 m compared with at least 7 m during the Restoration Point earthquake. Perhaps the 1.5 m of uplift, comparable to that seen along the Toe Jam Hill fault and Waterman Point fault scarps during earthquakes B, C, or E (Fig. 5; Nelson et al., 2003a), records a rupture of a few kilometers or less on a nearby unmapped fault.

Could the Seattle and Tacoma faults have ruptured during the same very large earthquake, perhaps with a rupture extending westward to faults of the Saddle Mountain deformation zone (Blakely et al., 2009)? Although upper plate structure eastward from the Saddle Mountain deformation zone remains unclear, we hesitate to exclude such a possibility, given some unexpected multifault ruptures of the past approximately two decades (e.g., Sieh et al., 1993; Xu et al., 2009). Earlier historical examples of such ruptures include the en echelon folds >60 km long that grew during the 1586 M∼7.7 earthquake on the Yoro blind thrust of central Japan (Ishiyama et al., 2007). Above subduction zone megathrusts, backthrusts tens of kilometers arcward of the megathrust have ruptured during and after great earthquakes (McCalpin and Carver, 2009; Berryman et al., 2011; Wiseman et al., 2011) However, the >20 km distance between the surface projections of the Seattle and Tacoma master-ramp thrusts (Fig. 1C) is greater than the distance between adjacent intraplate faults that have ruptured during the same earthquake (Rubin, 1996; Wesnousky, 2008). How Coulomb stress changes influence the interactions of nearby thrusts depends on many factors in addition to fault proximity, including fault orientations, prior earthquake timing and magnitudes, and whether thrusts are blind (Stein and Yeats, 1989; Lin and Stein, 2004; Ishiyama et al., 2007; Schwartz et al., 2012). Rupture modeling of thrust faults (e.g., Wang and Chen, 2001; Lin and Stein, 2004), and of central Seattle fault deformation during the Restoration Point earthquake (ten Brink et al., 2006), do not directly address simultaneous rupture on thrusts so far apart. A likely rupture scenario is that upper plate stress changes induced by the Restoration Point earthquake on the master-ramp thrust of the Seattle fault triggered slip on adjacent, updip parts of the fault, and perhaps faults of the Tacoma fault zone or Saddle Mountain deformation zone, days to decades after the earthquake (Sherrod, 2001; Sherrod et al., 2004a; Johanson and Bürgmann, 2010; Schwartz et al., 2012). A possible analogous earthquake series is the sequence of four M_w5.8–6.7 thrust earthquakes in the central California Coast Ranges during 1982–1985 that may have been partially triggered by stress changes induced by the 1857 M_w7.9 Fort Tejon earthquake on the San Andreas fault (Lin and Stein, 2004).

Evidence for the magnitude of other Puget Lowland earthquakes during the late Holocene is ambiguous for two reasons: (1) the evidence is spatially more limited than that for the largest earthquakes during the millennial earthquake series, and (2) the typically >50 yr errors on earthquake ages prevent use of routine ¹⁴C dating for site-to-site correlation of closely spaced earthquakes. Paleoseismic studies in a variety of tectonic settings face similar correlation uncertainties (Burbank and Anderson, 2001; Atwater et al., 2004; McCalpin, 2009; e.g., Fraser et al., 2010; DuRoss et al., 2011; Berryman et al., 2012). For example, in a study of scarp geomorphology and stratigraphy in the Hikurangi forearc, Kelsey et al. (1998) concluded that trench stratigraphy could not distinguish among prehistoric ruptures postulated to extend 15, 25, 34, or possibly 130 km along the Poukawa fault zone. In the Seattle fault zone, measured amounts of surface deformation for these other earthquakes are smaller than for deformation that we correlate with the Restoration Point earthquake, suggesting earthquakes of smaller magnitude. Whether these earthquakes were large (M 6.5–7.0) or moderate (M 6.0–6.5), however, depends on which of several fault zone models best applies (Fig. 3). Until structural relations among backthrusts and master-ramp thrusts are clarified, paleoseismic evidence cannot discriminate large from moderate prehistoric earthquakes in the Seattle fault zone.

Limited evidence from the Tacoma fault zone is equally ambiguous about rupture modes, although some evidence suggests multiple earthquakes. For example, Sherrod (2001) ascribed the burial of a marsh soil at Wollochet Bay (Fig. 1C) to subsidence during an earthquake younger than the Restoration Point earthquake. Preliminary tree-ring correlations date stumps killed by flooding, a result of surface rupture on the Saddle Mountain faults, to >50 yr before the Restoration Point earthquake (Hughes, 2005; Snook, 2011). At heights as great as 7–8 m, the Catfish lake scarps within the Tacoma fault zone are too high to have formed entirely during a single large earthquake (e.g., Ishiyama et al., 2007), and the 6-m-high scarp at 1 site (Cedars trench, Nelson et al., 2008, sheet 3) is younger than 4.3 ka (Figs. 1C and 2B; Table 1; Fig. S6 in the Supplemental File [see ^{footnote 1}]).

Two consequences of forearc structure that the Puget Lowland shares with intraplate thrust fault systems cloud estimates of the frequency and degree of clustering of large earthquakes in the lowland: uncertain structural relations among faults and probable blind thrusting. Evidence from 2 short backthrusts in the central Seattle fault zone records a cluster of at least 4 late Holocene earthquakes after 2.7 ka. Earthquakes B and C occurred before and earthquake E occurred after the Restoration Point earthquake (which we correlate with earthquake D), and are separated from prior or subsequent earthquakes by intervals of 100–1500 yr (Fig. 5). Records of wetland sedimentation from these sites make it unlikely that similar earthquakes occurred in the preceding 6000–8000 yr and perhaps not since 15 ka (Nelson et al., 2003a; Figs. 10 and 11). However, because the structural relations of these scarps with the master-ramp thrust of the Seattle fault remain uncertain, we cannot determine to what extent the cluster represents the earthquake history of the master Seattle thrust. The same uncertainties afflict the early and middle Holocene earthquake history of the Tacoma fault, where well-dated earthquake evidence is limited to the millennial earthquake series. An additional complication is that most historical earthquakes on forearc and intraplate thrusts have involved rupture of some shallow secondary faults but not others (McCalpin and Carver, 2009; Philip and Meghraoui, 1983; Hull, 1990; Xu et al., 2009). As inferred by Kelsey et al. (1998) from trenches at three sites on faults of the Poukawa fault zone in the Hirugani forearc, our mapping of trenches in the Seattle fault zone shows that not every late Holocene scarp ruptured during each surface-deforming earthquake.

Because blind or almost blind thrusting is ubiquitous in forearc and intraplate thrust systems, an absence of evidence for surface deformation during Holocene earthquakes in the lowland does not imply an absence of earthquakes. A common theme is that even for well-studied modern earthquakes with tens to hundreds of kilometers of complex fault and fold scarps (e.g., 1988 M_s7 Spitak, Armenia; 2008 M_w7.9 Wenchuan, China), some slip on blind thrusts must be invoked to fully explain patterns of surface deformation (e.g., Hull, 1990; Rubin, 1996; Ishiyama et al., 2007; de Michele et al., 2010). Slip on blind thrusts during earthquakes as large as M_w7.7 has produced too little surface evidence of deformation to be preserved for more than decades (e.g., McCalpin and Thakkar, 2003). More commonly, slip on buried thrusts creates small, discontinuous fold or fault scarps that are so eroded or buried within 50–200 yr that surface evidence of them can no longer be identified (Lettis et al., 1997; McCalpin and Carver, 2009).

However, blind thrusting during successive earthquakes may create subtle, long-wavelength folds that warp marine or fluvial terraces (Burbank and Anderson, 2001; e.g., Philip and Meghraoui, 1983; Scharer et al., 2006; Ishiyama et al., 2007; Kaneda et al., 2008b). Substantial blind thrusting is inferred from evidence of this type along an 8-km-long reach of Big Beef Creek, at the western end of the Seattle uplift about half-way between the surface ruptures at Waterman Point and Saddle Mountain (Fig. 1C). There, recessional glaciofluvial terraces crossing the Seattle fault have been arched upward 23 m since 16 ka (Tabor et al., 2011). The uplift has occurred along a stretch of the fault with no discernible surface rupture, implying either large Restoration Point–style earthquakes or many smaller earthquakes on a blind, south-dipping thrust.

Can records of strong shaking over many thousands of years, such as the Karlin et al. (2004) turbidite history from Lake Washington, help distinguish differing rupture modes or earthquake clustering in the lowland? Karlin and Abella (1996) inferred as many as 21 magnetic susceptibility peaks in Lake Washington cores to mark turbidites that record strong earthquake shaking since 7.6 ka, for an average turbidite recurrence of 300–400 yr. Based on uniform sedimentation rates controlled by the 47 ¹⁴C ages of Karlin and Abella (1996; see Karlin et al., 2004), some of the turbidites recur every 100–300 yr for periods lasting a thousand years. A large peak that could record strong shaking from a large, regional earthquake occurs in all cores ca. 7.4 ka, but at least 4 younger peaks are larger and whether all such peaks record earthquake shaking is uncertain (Karlin and Abella, 1996; Karlin et al., 2004). Conversely, all earthquakes that strongly shook the lowland are not necessarily recorded by Lake Washington turbidites, and some turbidites may have been triggered by earthquakes that left no evidence of surface deformation. The lake’s late Holocene record shows age distributions for shaking-induced turbidites that span the times of as few as 3 of the as many as 9 great subduction zone earthquakes of the past 3500 yr (Karlin et al., 2004; Atwater et al., 2004; Nelson et al., 2006; Goldfinger et al., 2012), and the distributions are too wide to infer which turbidites were generated by strong subduction zone shaking. Shaking from deep intraplate earthquakes, such as the M 7.1 earthquake between Tacoma and Olympia in 1949, may have also generated turbidites. Although we infer that lake turbidites record a greater proportion of Seattle fault zone earthquakes than earthquakes in other upper plate fault zones in the Puget Lowland, without estimates of earthquake frequency and magnitude for other fault zones, the proportions of turbidite-marked earthquakes they produced are unknown. Such uncertainty in the interpretation of earthquake sources is a major issue in studies of shaking-induced deposits in lakes adjacent to Holocene faults (Goldfinger, 2009; e.g., Inouchi et al., 1996; Ken-Tor et al., 2001; Strasser et al., 2013).

Hazard assessment in the Puget Lowland benefits from the additional information our reconstruction of earthquake history provides. The history increases the minimum number of closely spaced, surface-rupturing earthquakes in the central Seattle fault zone from 3 to 4, shows that the most recent of these earthquakes postdates the Restoration Point earthquake, and confirms on a second backthrust scarp the inference of Nelson et al. (2003a) that the cluster of late Holocene earthquakes followed a period of ∼6000–8000 yr of scarp quiescence. The history suggests that faults of the central Seattle fault zone may slip during moderate to large earthquakes as frequently as every few hundred years for periods of ∼1000–2000 yr, and then not slip for periods of at least several thousands of years. Our dating narrows the time interval for the millennial earthquake series, particularly in the Tacoma fault zone, but cannot address the prospect of simultaneous rupture of two or more fault zones in the southern lowland. Despite the uncertainties in actual earthquake history, hazard assessment methods accommodate the diverse rupture modes inferred from the findings described here through weighting of alternative rupture models with a range of probabilities (e.g., Petersen et al., 2008).

Our evaluation of new and previously published paleoseismic data improves the understanding of the earthquake prehistory of the southern Puget Lowland, and illustrates some limitations of using paleoseismic data to reconstruct earthquake history in forearcs with complex upper plate structure. OxCal analysis of 10 of the most closely limiting ¹⁴C ages from fault scarp trenches and cores at 8 sites restrict the age of the M 7–7.5 Restoration Point earthquake in the central Seattle fault zone to 1040–910 cal yr B.P. (2σ; Fig. 5). This age interval overlaps the most precise age for the earthquake’s tsunami (1050–1020 cal yr B.P.) in eastern Puget Sound. Limiting ages date stratigraphic evidence for three smaller earthquakes that occurred during time intervals that do not overlap with 1050–1020 cal yr B.P. Two of these, earthquakes B and C, occurred prior to the Restoration Point earthquake, 2650–1940 cal yr B.P. and 1310–1200 cal yr B.P., respectively, and one (earthquake E) occurred after the Restoration Point earthquake between 940 and 380 cal yr B.P. Whether these earthquakes were of moderate (M 5.5–6.0) or large (M 6.5–7.0) magnitude, the surface-deformation history of the central Seattle fault zone suggests that faults may slip during moderate to large earthquakes as frequently as every few hundred years for periods of ∼1000–2000 yr, and then not slip for periods of at least several thousands of years.

Despite its value as the potentially most complete and longest record of strong shaking in the Puget Lowland, extracting a history of surface-deforming earthquakes on upper plate faults from Lake Washington’s turbidite record is complicated by an abundance of potential earthquakes sources. These include surface-deforming and non-surface-deforming upper plate earthquakes in at least three lowland fault zones (Fig. 1B), Cascadia subduction zone earthquakes, and deep intraplate earthquakes beneath the lowland.

The Tacoma fault zone earthquake history consists of at least one large-magnitude earthquake ca. 1000 cal yr B.P., but evidence is too sparse to determine whether this earthquake was closely predated or postdated by moderate to large earthquakes on faults in the Tacoma basin. Four fault scarp trenches in the Tacoma fault zone described here show the most recent scarp of the Tacoma fault to be late Holocene and date surface rupture on two hanging-wall faults to ca. 1000 cal yr B.P., the time of widespread coseismic subsidence in the Tacoma basin. In the Tacoma fault zone, 12 closely limiting new and previously published ages from 9 sites constrain the earthquake (or a closely spaced series of earthquakes) to 1050–980 cal yr B.P. (Fig. 5). An age from one subsidence site in the Tacoma basin suggests an earthquake slightly younger than this interval. The amount of subsidence measured at another site and the height of the highest fault zone scarps also suggest multiple Holocene earthquakes.

OxCal analysis of previously reported ages for late Holocene earthquakes in the Saddle Mountain deformation zone limit earthquakes on the Saddle Mountain East and Saddle Mountain West faults to 1160–310 cal yr B.P. and 1200–970 cal yr B.P., respectively. Whether these earthquakes were synchronous, with each other or possibly with ruptures farther east in the Seattle fault zone or Tacoma fault zone, remains unknown.

This work was supported by the Earthquake Hazards Program of the U.S. Geological Survey. Jason Buck, Gary Henley II, Liz Schermer, Koji Okumura, Rob Witter, Rich Koehler, Elizabeth Barnett, Eliza Nemser, Trenton Cladouhos, Ian Madin, Vicky McConnell, and Yoko Ota mapped trenches, collected critical ¹⁴C samples or structural data, and helped with interpretation. These and other trench wall mappers are listed in figure captions. Lee Liberty, Bill Stephenson, and many others ran and interpreted geophysical profiles near trenches. We thank many other colleagues for help with trench mapping and for insightful comments during trench reviews. Discussions with Tom Pratt, Kathy Troost, Uri ten Brink, Jonathan Hughes, Rich Briggs, Tim Walsh, Bob Bucknam, and Craig Weaver were particularly helpful. Reviews by Olaf Zielke, Rob Witter, Tom Pratt, Sotiris Kokkalas, and three anonymous reviewers strengthened the paper. Any use of trade, product, or firm names is for descriptive purposes only, and does not imply endorsement by the U.S. Government.