Enhanced landslide mobility by basal liquefaction: The 2014 State Route 530 (Oso), Washington, landslide

Size and speed regularly dictate the destructiveness of landslides. Slide mobility can project devastation across large distances, as demonstrated in numerous cases worldwide (see examples in Voight, 1978; Evans and DeGraff, 2002). Such was the case on 22 March 2014, when the State Route 530 (Oso) landslide failed catastrophically, traveled across a flat alluvial valley for more than 1 km, and swept away the Steelhead Haven neighborhood near Oso, Washington, USA (Fig. 1). The 43 people killed and ∼35 homes destroyed (Keaton et al., 2014) were on flat ground, hundreds of meters from the slope where sliding initiated. The entire slide volume exhibited greater mobility than typical rock or debris avalanches (Iverson et al., 2015). Commonly, landslide hazard assessments focus on the origins and frequency of instability on hill or mountain slopes (e.g., Montgomery and Dietrich, 1994; Aleotti and Chowdhury, 1999; Baum et al., 2005a; Guzzetti et al., 2005; Brien and Reid, 2008); however, understanding landslide mobility can be of equal importance when assessing risks.

Landslides exhibit a wide range of mobility, depending on factors such as topographic relief, volume, material properties, topographic confinement, and degree of fluidization (cf. Hsü, 1975; Corominas, 1996; Dade and Huppert, 1998; Legros, 2002; Griswold and Iverson, 2008). The mechanical processes leading to excess mobility or long runout have motivated considerable discussion (see Legros, 2002; Hungr, 2007). Many proposed mechanisms (e.g., Shreve, 1968; Melosh, 1979; Hsü, 1975; Sassa, 1985; Campbell, 1989; Abele, 1997; Davies et al., 1999; Johnson et al., 2016) invoke reduced basal or internal frictional strength to promote mobility. However, field observations to unequivocally support a particular hypothesis commonly remain elusive.

One mechanism to enhance mobility—liquefaction—has found strong support in a variety of field, experimental, and modeling studies (e.g., Sassa, 1985; Voight and Sousa, 1994; Iverson et al., 2000; Hungr and Evans, 2004; Sassa et al., 2010; Iverson et al., 2011; Reid et al., 2011). Highly mobile flowslides (Mitchell and Markell, 1974; Dawson et al., 1998; Hunter and Fell, 2001, 2003; Olivares and Picarelli, 2003) and debris flows typically become liquefied during failure and transport, thereby reducing frictional strength and enhancing mobility (e.g., Iverson, 1997; Iverson et al., 1997; Hungr et al., 2001; Sassa and Wang, 2005). Liquefaction in saturated soil greatly diminishes shear strength due to rapid increases in soil water pressure; liquefaction can occur from a number of causes, such as strong vibrations or undrained loading, and in various components within or underneath a landslide mass. Although the Oso landslide displayed some debris-flow characteristics at its distal margins, it predominantly acted as an unconfined debris avalanche (Iverson et al., 2015), without the pervasive internal deformation typically associated with debris flows. Moreover, although the slide materials contained some lacustrine sediments, these were not sensitive clays that might liquefy upon failure (Stark et al., 2017).

Several previous studies have examined aspects of the 2014 Oso landslide event, including its mobility (Keaton et al., 2014; Iverson et al., 2015; Wartman et al., 2016; Iverson and George, 2016; Stark et al., 2017; Aaron et al., 2017). Nevertheless, two fundamental issues remain unresolved or in dispute: (1) the precise sequence and associated timing of events, and (2) the nature of the mechanism that promoted high mobility. For example, several studies (Keaton et al., 2014; Wartman et al., 2016; Stark et al., 2017) have advocated for two large, distinct failure events separated by ∼4 min, with the first mass traveling across the valley floor. Other studies (Iverson et al., 2015) postulated that the vast majority of landslide activity occurred over the initial minute. These differences have profound implications for interpretation of seismic signals (Hibert et al., 2015; Iverson et al., 2015) and for mechanical models that incorporate momentum transfer to enhance mobility. Although several modeling efforts have incorporated liquefaction as a mechanism to reduce shear strength during the Oso event (Iverson et al., 2015; Iverson and George, 2016; Aaron et al., 2017; Stark et al., 2017), these studies differ in inferred liquefaction processes while focusing on liquefaction of the slide mass itself. Both slide sequencing and the causes of liquefaction bear directly on the destructiveness of the 2014 slide, and on potential future hazards from similar slides.

Here, we used postevent field observations combined with subsurface information and materials testing to decipher the geologic materials and processes of the 2014 Oso landslide, and to clarify how the sequence of events led to the landslide’s enhanced mobility. Our research was aided by fundamental observations obtained via a detailed field mapping effort performed over a 3-yr-period following the landslide. Our investigation focused on understanding the Oso landslide’s mobility, rather than on the causes for its initiation. However, we expect that our findings will provide critical information for subsequent studies that test initiation models.

We begin by providing a brief background on landsliding in the region and at the site of the Oso landslide. Following a description of our methods, we present our data and observations, and then we analyze these data to distinguish the major components of the landslide from one another. We subsequently use these components to interpret and describe a likely sequence of events, thus separating observations from interpretations. We then analyze our proposed mobility mechanisms involving liquefaction processes. Finally, we conclude with a discussion focused on the implications of the mobility of the Oso landslide for other similar settings, and a comparison of the various mechanisms put forth to explain mobility at Oso.

Landsliding in the glacially derived sediments of northwest Washington, typically associated with winter and early spring rainfall, is a well-known geologic hazard (Manson, 1988; Thorsen, 1989; Baum et al., 2007). Landslides can be relatively shallow slope failures (e.g., Baum et al., 2005b; Godt et al., 2006) or larger deep-seated failures (e.g., Gerstel et al., 1997; Brien and Reid, 2008), such as occurred near Oso. Previous geologic mapping of the Oso area (Dragovich et al., 2003) indicated evidence of widespread landslide deposits in the North Fork Stillaguamish River valley, with more-recent mapping (Haugerud, 2014) illustrating relative age relations among larger prehistoric landslides. Absolute dating of two landslides neighboring the 2014 Oso landslide provided calibrated radiometric ages of ca. 500 yr B.P. and 6000 yr B.P. (LaHusen et al., 2016), suggesting that landsliding here is at least partially decoupled from deglaciation processes, which occurred some 16,000 yr ago (Porter and Swanson, 1998).

The Oso landslide occurred at 10:36 a.m. local time (Pacific Daylight Time) on a sunny Saturday morning that followed both anomalously high March and 4 yr cumulative rainfall (Henn et al., 2015; Iverson et al., 2015). The landslide initiated from a 180-m-tall terrace of Pleistocene glacial deposits located on the north side of the North Fork Stillaguamish River (Dragovich et al., 2003). The source area was 760 m in length, but the landslide traveled over 1.5 times this distance (as measured from the toe to the most distal deposit), crossing the entire North Fork Stillaguamish River valley, and inundated 0.87 km² of the relatively flat valley floor including part of State Route 530 (SR530; Fig. 2). As a result, the Steelhead Haven neighborhood (Fig. 2), located on the south side of the North Fork Stillaguamish River, was overrun by debris and destroyed. Remarkably, 11 persons in the neighborhood survived the rapid onslaught of mud and debris and were rescued from distal edges of the deposit (Quistorf, 2014).

The site of the Oso landslide had undergone repeated periods of instability over the past 80 years (Miller and Sias, 1997, 1998; Keaton et al., 2014, and references therein), dating back at least to 1932 (Thorsen, 1969). From this time until 2014, landslide movement was confined to roughly the lower half of the hillslope (herein referred to as the Hazel landslide; Fig. 2), with repeated cycles of landsliding that sometimes required engineering intervention to control sediment discharge and overall stability (Miller and Sias, 1997; Wartman et al., 2016). For example, renewed landslide activity in 2006 prompted construction of a log revetment (Keaton et al., 2014) that subsequently partially failed prior to 2014 (Fig. 2). Although early work (Shannon and Associates, 1952) identified riverbank erosion as the primary cause of landsliding, later work by Thorsen (1969) proposed that surficial springs and groundwater channeling into a large downdropped block formed from a prehistoric landslide event (herein referred to as the Paleo landslide; Fig. 2) likely promoted landslide movement. Subsequent investigations (Miller and Sias, 1997, 1998) identified timber harvest on the relatively flat glacial terrace north of the Hazel landslide area (Whitman Bench; Fig. 2), riverbank incision, and slope surface changes from surface runoff as potential causes of instability for the pre-2014 slope. Importantly, landslide deposits during these previous events traveled only moderate distances (for example, in 2006, material traveled ∼180 m and blocked the North Fork Stillaguamish River) compared to the much larger runout of the 2014 failure, which crossed the entire river valley width. However, a pre-2014 geologic map (Dragovich et al., 2003) indicates that other nearby prehistoric landslides exhibited enhanced mobility with runouts reaching far across the North Fork Stillaguamish River valley, and subsequent light detection and ranging (LiDAR) interpretation (after the 2014 event) revealed more of these large features (Haugerud, 2014).

The 2014 landslide mobilized intact, generally horizontally layered glacial deposits at its northern (proximal) end and reactivated multiple preexisting landslide deposits in the southern part of its source area. The intact glacial deposits at the northern end (i.e., from the Whitman Bench; Fig. 2) are composed of, from top to bottom, recessional outwash (Q_goe), till (Q_gtv), advance outwash (Q_gov), and advance glaciolacustrine deposits (Q_glv; Tabor et al., 2002; Dragovich et al., 2003; Badger, 2015). These are underlain by a (likely) preglacial sand unit (Q_co) mapped on the north side of the North Fork Stillaguamish River (Dragovich et al., 2003). Till and advance outwash deposits are sometimes missing or discontinuous in sections across the Whitman Bench (Badger, 2016), which can result in variations in both site hydrology and deposit stratigraphy. Downslope (to the south) of the Whitman Bench, the 2014 Oso landslide mobilized deposits from multiple episodes of preexisting slope failures (Thorsen, 1969; Miller and Sias, 1997, 1998; Dragovich et al., 2003; Keaton et al., 2014, and references therein). Immediately below the bench, a steep (∼30°) slope formed the headscarp of the prehistoric Paleo landslide complex (Fig. 2). The 100-m-wide, downdropped block of the Paleo landslide (referred to as one of a series of “ancient” landslides in the area by Thorsen [1969] and also mapped by Miller and Sias [1997]) is one of several similar features found on both the north and south sides of the North Fork Stillaguamish River valley. Evidence for similar downdropped blocks of former landslides is also visible in the scarps on both sides of the 2014 Oso failure (Keaton et al., 2014). Further southward and downslope, the Paleo landslide slope led to an even steeper (∼32°) headscarp of the more recent (historic) Hazel landslide complex (Fig. 2). These landslide deposits were composed of the same glacial materials forming the Whitman Bench, but they were variably tilted and disaggregated as a result of previous landsliding. Deposits in the pre-2014 Hazel area were likely highly disaggregated, having been through multiple episodes of landsliding, whereas the glacial stratigraphy near the top (Paleo landslide) was likely more intact, thereby preserving original glacial stratigraphic relations.

The episodes of previous landsliding, combined with ongoing forestry practices (Everett and Casey, 2014), led to a variable forest cover prior to the 2014 landslide. Evergreen conifer timber harvest over most of the Whitman Bench in the 1940s, with subsequent regrowth of Douglas fir (Pseudotsuga menziesii) trees, created a younger forest, with trees generally less than 20 m tall. On the Paleo landslide scarp, much taller coniferous trees (25–40+ m) were present, because recent logging activities in the late 1980s and early 2000s were restricted in this zone directly upslope of historical landsliding (Miller and Sias, 1998; Everett and Casey, 2014). Farther down the slope, in the area of the Hazel landslide, a mix of deciduous and coniferous trees of much shorter height (<15 m) predominated, resulting from (nonplanted) postlandslide regrowth.

To understand the Oso landslide’s mobility, we performed field mapping, geotechnical sampling and testing, and LiDAR-based volumetric, geomorphologic, and vegetation analyses. These analyses, combined with both numerical and empirical analyses of liquefaction mechanisms (presented in the Analysis of Liquefaction Mechanisms section), provide the basis for our interpretations and conclusions.

Detailed field geologic mapping and observations form our primary data. The majority of our field investigations took place between April and September 2014 to take advantage of the relatively undisturbed, postfailure geomorphic conditions, with several weeks of follow-up field work over the next 2 years. We also mapped a 450-m-long section through the west distal lobe (see Fig. 2) made available via the excavation of the SR530 road bed 2 months following the landslide. Timely mapping on high-resolution (1:1200 scale) orthophotos was crucial, as many features underwent rapid disintegration in subsequent months from both anthropogenic (deposit regrading) and meteorological (rainfall) impacts. For some aspects (e.g., mapping of debris-flow limits), our mapping was aided by photographs and video taken from helicopters by emergency responders during the first 48 h following the landslide.

We mapped major scarps, grabens, thrusts, and hummocks (i.e., multisided landslide elements that commonly preserve the uppermost topographic surface) to establish relative senses of motion of the various components of the landslide. The direction, position, and disposition (whether buried, dragged, or fractured) of mobilized trees also provided indications of whether distinct landslide zones experienced extension or compression (or both). We mapped the four major (disrupted) geologic units of the landslide (Q_goe, Q_gtv, Q_gov, Q_glv; see Dragovich et al., 2003) by making on-site observations at ∼1400 locations and carefully correlating these observations with known stratigraphy of the area from previous downhole drilling and coring at the site (Riemer et al., 2015; Badger, 2016), as well as from scarp exposures. In many cases, thin, disrupted surficial materials concealed the primary units forming individual hummocks; we therefore performed shallow digging and trenching to identify these primary units.

Our mapping of features used to infer mobility included the presence and location of evidence of liquefaction such as sand boils and slosh pits, the size and structure of hummocks, and the before- and after-event locations of anthropogenic features such as dislodged pieces of log revetment wall (many with broken steel cables). Sand boils are a classic phenomenon resulting from both cyclic and monotonic (often referred to as “static”) liquefaction (e.g., Seed, 1968; Bardet and Kapuskar, 1993; Ojha et al., 2001) and represent the surface manifestation of eventual post-emplacement pore-water pressure dissipation from depth. They are generated as elevated pore-water pressures eventually move fluid to a free (ground) surface, typically carrying liquefied silt-, sand-, and sometimes gravel-size particles (Kayen et al., 2004) with upwardly flowing water. We readily identified sand boils in the field by their circular, mostly cone-shaped structure and composition of nearly uniform sand-sized particles. Slosh pits are topographic depressions, typically located between hummocks, that contain a mixture of sand and grayish pebbles and cobbles, with a distinctive “bathtub” ring around the margins of the depression. These features likely formed from movement (so-called “sloshing”) of sediment-laden liquid during rapid landslide emplacement. In many cases, sand boils and slosh pit deposits were no longer visible 1 yr following the landslide as a result of ongoing overall erosion at the site resulting from precipitation and bioturbation.

We collected sediment samples from key exposures that provided insights into possible controls on landslide mobility. These included debris-flow deposits (Q_df) sampled at the distal edge of the landslide shortly (weeks) following emplacement, sand boil samples located throughout the hummock field, overridden floodplain alluvial sand (Q_a, as identified and mapped herein) from locations temporarily exposed as the new channel of the North Fork Stillaguamish River cut through the landslide deposit, and recent, post-2014, alluvium deposited along the North Fork Stillaguamish River following the landslide (as a proxy for the deposits forming the floodplain prior to being overrun by the landslide).

Geotechnical testing included soil characterization tests (i.e., grain size, Atterberg limits, specific gravity) on samples to determine their provenance and relation to various components of the landslide. We also measured bulk density and water content, and we calculated dry density (when applicable) to obtain typical average values for subsequent mobility analyses. Finally, we conducted a suite of triaxial geotechnical strength tests to investigate potential failure modes of the underlying alluvium sediments forming the North Fork Stillaguamish River floodplain. These tests were performed at densities, saturation levels, and confining stresses commensurate with the field conditions at the time of the landslide. We present additional details of the geotechnical laboratory testing program in Appendix S1 of the Data Repository.¹

Multiple high-resolution aerial LiDAR data sets acquired before (July 2013) and after (26 March 2014 and 6 April 2014) the 2014 landslide allowed us to analyze three important characteristics: (1) the thickness of the slide deposits, (2) the three-dimensional (3-D) geometry of the slide surface, and (3) the heights of the preslide trees (see Appendix S2 in the Data Repository [^{footnote 1}]).

We calculated the spatial distribution of slide-deposit thickness by first constructing bare-earth digital elevation models (DEMs) for each LiDAR point-cloud data set. This entailed using last-return laser data and performing minor manual editing to remove vegetation and anthropogenic structures from each DEM. We then compared independently georeferenced pre- and postslide pairs of these DEMs to determine elevation differences. Vertical differences between unchanged areas of the models (i.e., away from the landslide) were less than 20 cm for 90% of the data (and less than 100 cm for 99.9% of the data), thereby allowing a spatially robust volumetric comparison of areas affected by the landslide. Areas on the North Fork Stillaguamish River valley plain that increased in elevation following the 2014 landslide indicated slide-deposit thickness (assuming only minor scour and/or compaction of the subsurface). In contrast, elevation changes on the hillslope may have resulted from rearrangement of mass by sliding and were not indicative of deposit thickness. For these areas, we constructed surfaces of the inferred slide plane at depth based on borehole and other data (see next section and Appendix S2 of the Data Repository [^{footnote 1}]).

Combined with borehole data showing 2014 failure surface elevations at five locations within the landslide deposit (Badger, 2016), we used the LiDAR data to reconstruct inferred failure surfaces of the landslide. Using this information in a fully 3-D combination LiDAR- and CAD-based software platform (Maptek I-Site Studio v. 5.1), we constructed both failure and ground surfaces, and we subsequently calculated volumes of both the source and deposit regions. These included the identification and calculation of separate components of the source and deposit areas (e.g., the volume of the Paleo landslide source and the Hazel landslide source). Details of our volume calculations are included in Appendix S2 of the Data Repository (^{footnote 1}).

We used particle-tracking methods focused on readily visible “tall” (>20 m in height) trees in both aerial photographs and LiDAR data to aid the quantification of mobility between source and deposit zones of the landslide. Prior to sliding, taller Douglas fir trees (20–40+ m) were restricted to parts of the landslide source areas, mostly in the upper half of the slope. Other areas lower on the slope were covered with much shorter (typically <15 m in height) mixed deciduous and coniferous forest. After the 2014 event, trees were displaced and distributed throughout various parts of the slide deposit, with most trees toppled over and having subhorizontal orientations. We determined preslide tree heights using the 2013 LiDAR point cloud, where the difference in elevation between first- and last-point LiDAR returns provided us with estimates of individual tree height. Then, using a 14 April 2014 postslide orthophoto, we located and measured the lengths of toppled, subhorizontal “complete” conifer trees, distinguished by visible root systems and pointed, unbroken tips. Numerous other downed “partial” trees existed on the deposit (as noted by Keaton et al., 2014; Wartman et al., 2016), but we focused on complete trees.

We present four maps (Figs. 3–6) depicting the structure, geology, mobility features, and topographic change (deposit thickness), along with a geotechnical borehole–based cross section, as independent data for making our subsequent interpretations of the Oso landslide sequence and its enhanced mobility. Here, we briefly describe our data and observations.

Our structure map (Fig. 3) presents major scarps, grabens, thrusts, and blocks throughout the landslide and identifies ∼900 hummocks in the central to distal end of the deposit. Individual hummocks were identified using weighted and smoothed slope curvatures of the 2014 LiDAR-based DEM at three scales (3–10 m, 25–50 m, and 50–100 m) to isolate flat hummock tops. Geomorphic features transition from large downdropped blocks to laterally continuous block “slices,” hummock fields, and smaller disaggregated hummocks in progression from north to south through the deposit. Nearly all structural features show evidence of extension, with a few notable exceptions described in the “Analysis and Interpretation of Landslide Progression” section.

Our geologic map (Fig. 4) depicts broad swaths of the main stratigraphic units that make up the source and deposit areas. These consist of the four predominant glacial deposit units (Q_goe, Q_gtv, Q_gov, Q_glv; Dragovich et al., 2003) as well as several units that are specific to the landslide’s depositional structure. Units we mapped and informally refer to here include areas where hummock composition is highly mixed (undifferentiated hummocks, Q_gund), the large colluvial infill downslope of the landslide headscarp (Q_c), and debris-flow deposits located at the distal (Q_df) and lateral (Q_dfs) margins of the landslide. Our original mapping data consisting of lithologic observations are presented in Data Sheet S1 of the Data Repository (see ^{footnote 1}).

Our mobility features map (Fig. 5) portrays the tracking of several groups of objects from their preslide to postslide locations; these objects include complete trees, pieces of forest floor, and pieces of a cut-log revetment wall originally located on the north side of the North Fork Stillaguamish River. The locations of these features could be readily identified using pre- and postslide aerial photos, LiDAR-generated tree-height maps, and field observations (see “Methods” section). Although we did not perform a one-to-one correlation between each of these features, patterns of movement are evident. For example, tall trees originally standing on the scarp and bench of the Paleo slide as well as a small section of the Whitman Bench are readily distinguished in their fallen locations on various parts of the landslide deposit (see color-coded tree heights in Figs. 5A and 5B; data provided in Data Sheet S3 of the Data Repository [^{footnote 1}]). Here, we also show areas of disaggregated and stretched forest floor (Fig. 5B) on the surface of the landslide deposit, along with pre- and postslide locations of parts of the log revetment structure.

The deposit thickness and sand boil map (Fig. 6) shows the locations of 107 sand boil sites (composed of one or more of the 379 individual sand boils we mapped; see Data Sheet S2 of the Data Repository for location data of sand boils [^{footnote 1}]) overlain on an elevation change map, where positive values on the valley floor approximate 2014 slide deposit thickness. Most sand boil sites were located south of the former footprint of the northern margin of the North Fork Stillaguamish River, with only two locations mapped north of this boundary. Sand boils were typically several decimeters in diameter and several to ten centimeters tall, although in some cases, they reached over 1 m in diameter and several decimeters tall (Fig. 7). We used sand boils as indicators of liquefaction at depth. We found sand boils in both subaerial and subaqueous environments, indicating that pore-pressure dissipation occurred shortly after emplacement. Some sand boils were located within slosh pit features (Fig. 6); sediment rings were higher on the distal sides of these slosh pits, indicating rapid deceleration.

Our geologic and structural cross section through the landslide (Fig. 8) used data from 12 geotechnical boreholes (Fig. 4; Badger, 2015, 2016) along with our field mapping to portray the disturbed stratigraphy within the landslide deposits. We used a turning point located where major block slices transition into hummocks to present the entire section of the landslide, from source zone to distal deposit. This view provides the subsurface details of the various slide components (i.e., major deposit zones, failure surfaces, etc.), which were subsequently used to interpret the landslide sequence (see next section).

Landslide mobility is commonly quantified using travel paths of particular features. For example, tracking where an anthropogenic (e.g., piece of revetment wall) or natural feature (e.g., tree) was located before and after a landslide can provide a direct indication of surface displacement. At the Oso landslide, strong evidence for travel paths is provided by the three major components of the landslide source area, each with its own stratigraphic package (Hazel, Paleo, and Whitman; Figs. 2 and 4). The Hazel source component was located just north of the North Fork Stillaguamish River and consisted of debris from previous landslide failures. The Paleo source component was located just upslope (north) of the Hazel component and consisted of the remnants of a downdropped block that (prehistorically) failed from the Whitman Bench (see “Study Area Background and Landslide History near Oso” section). Finally, the Whitman source component consisted of the part of the Oso landslide that failed within intact geologic units forming the Whitman Bench. The Whitman component also produced several secondary failures that, although not directly involved in the Oso landslide’s mobility, are important for understanding the deposit geometry. In this section, we use our mapped lithologic relations (Figs. 4 and 8), along with observations of displaced trees, forest floor, and revetment wall (Fig. 5), to identify, interpret, and link the various source (Fig. 9A) and deposit (Fig. 9B) areas of the landslide (Hazel, Paleo, Whitman). Using these linkages, we can then interpret the landslide sequence (presented in a subsequent section), which is in turn necessary for understanding the timing and mobility of the landslide.

Using our geologic mapping of the 2014 deposits, we first examined the distal areas of the Oso landslide deposit south of the North Fork Stillaguamish River, and in particular, south of the compression-extension (C-E) zone (Fig. 3) and north of the distal debris-flow deposits (Fig. 4). The C-E zone forms an important structural boundary between two of the deposit areas and is discussed in more detail in the subsequent section.

Our mapping south of the C-E zone revealed a southward progression of hummocks composed mostly of lithologically distinguishable materials (i.e., recessional outwash [Q_goe], till [Q_gtv], etc.). Toward the southern end of the hummock field, but north of the debris-flow deposits (Fig. 4), we identified a swath of chaotically mixed smaller hummocks (i.e., till, advance outwash, and advance glaciolacustrine deposits all in close, but seemingly random, proximity) juxtaposed against larger more monolithologic hummocks to the north. This mixed zone is labeled “undifferentiated hummocks” (Q_gund) in Figure 4, and we interpret this zone as originating from the Hazel landslide area north of the North Fork Stillaguamish River. Repeated cycles of landsliding at the Hazel site during its 80+ yr history prior to the 2014 landslide likely mixed the glacial units into a mélange, as reflected in the composition of the undifferentiated hummocks. These hummocks, along with the distal debris-flow deposit, compose the Hazel component deposit.

In contrast, the hummocks just north of the inferred Hazel deposits (but still south of the C-E zone; Fig. 3) are primarily stratigraphically conformable (i.e., in sequence of Q_goe–Q_gtv–Q_gov–Q_glv), with the deepest materials from the originally horizontally layered glacial stratigraphy sequence being transported the farthest to the south (Fig. 4). Some unconformities exist locally in the hummock field (e.g., Q_gtv is missing in the far western lobe, and Q_gov is missing in the central distal end); however, these can be explained by source-area stratigraphic relations and material behavior during transport. Large swaths of intact Q_gtv are mostly absent on the western side of the Whitman Bench source area, and thus its absence in parts of the western hummock field is expected. Elsewhere, in the central part of the deposit south of the C-E zone, blocks of stronger, rock-like Q_gtv may have tumbled and overridden weaker (i.e., uncemented sand) units such as Q_gov during transport, providing an explanation for the absence of Q_gov deposits here. This behavior is also visible on the east side of the Whitman component of the 2014 landslide (discussed subsequently), where a large swath of Q_gtv blocks overrode (presumed) areas of Q_gov (Fig. 4, near boreholes EB-07 and EB-18). Overall, the structured nature and distinguishable lithology of the hummocks south of the C-E Zone strongly suggest that they came from a large, and more intact, source area north of the (prelandslide) river. We infer that the Paleo bench area was the source for this part of the 2014 deposit (as opposed to being composed of a relic, undisturbed section of sediments). Importantly, we found no structural evidence of compressional or extensional features separating the primarily stratified (Paleo) from the undifferentiated (Hazel) hummock zones that might indicate separate failure times for these components. Our observations therefore indicate synchronous failure of the combined Hazel-Paleo source for the hummock deposits south of the C-E zone.

Our mapping of the distribution of postsliding revetment logs and vegetation (Fig. 5) serves to further distinguish the provenance of the hummock deposits south of the C-E zone. We mapped revetment logs with cables in numerous locations on the southern border of the hummock field (i.e., primarily among and distal to the undifferentiated hummocks). These logs originally formed part of the erosion-protection revetment structure located at the toe of the pre-2014 Hazel landslide deposit. As such, these dispersed logs denote a Hazel component origin, consistent with our interpretation for the source of the undifferentiated hummocks. Transported remnants of stretched forest floor (Fig. 5B), some with tall (20+ m) trees still intact, also occupy the tops of many hummocks south of the C-E zone and indicate stretching of the deposit during landslide movement. Although a few trees south of the C-E zone remained somewhat upright after sliding, most fell or were snapped. We identified numerous complete trees (see “Data and Observations” section) with heights of 20–40+ m lying within the stratified (inferred Paleo) hummock zone (Fig. 5B). Trees of this height were prevalent in the Paleo bench area but virtually nonexistent in the Hazel slide area prior to the 2014 event, thereby helping to distinguish Paleo from Hazel components in the deposit south of the C-E zone (Fig. 9B).

Finally, additional confirmation of a dual Hazel and Paleo source for the deposits south of the C-E zone is possible through volume-balance calculations. Using 3-D reconstruction of pre- and postslide topographic surfaces and construction of the landslide sliding surface (see “Methods” section), we estimated the combined volumes of the 2014 Hazel and Paleo deposits south of the C-E zone (including debris flow) to be ∼2,875,000 m³. The volumes of the Hazel and Paleo component source areas were 1,930,000 m³ and 766,000 m³, respectively, making for a combined volume of 2,696,000 m³ and resulting in a volumetric expansion ratio of ∼7% (Table 1). This somewhat low expansion ratio (compared to rock avalanches; e.g., Nicoletti and Sorriso-Valvo, 1991) suggests that the majority of the interparticle motion was restricted to the basal slide surface, with large extension between hummocks accounted for in the LiDAR-based volume calculation. Any additional source area volume that might have been part of the Hazel-Paleo failures would result in an even smaller expansion ratio, and assuming that all deposits south of the C-E zone came only from the Hazel source area results in an unrealistically large (∼50%) expansion ratio. Thus, the volumetric calculations reinforce the interpretation that the deposits south of the C-E zone originated from the combined Hazel and Paleo source areas (Figs. 8 and 9).

Most of the overall Oso landslide deposit exhibits extensional features (Fig. 3). However, the northern boundary of the Paleo deposit differs markedly from other features within the deposit. This internal feature represents the contact between deposits from the Hazel-Paleo source area and deposits from the Whitman source area (described subsequently). This contact is readily identifiable by repeated stratigraphy upslope and downslope of the contact zone (Fig. 4) and by vegetation patterns, where upslope Q_glv deposits (inferred Whitman) with few trees are (unconformably) in contact with downslope Q_goe deposits (inferred Paleo) covered with many toppled trees and pieces of stretched forest floor (Fig. 5B). The contact between these two masses is distinctively preserved south of the current North Fork Stillaguamish River (Fig. 10); we made additional detailed field observations to identify interactions in this zone. Whereas we found localized evidence of compression along the length of the contact (isolated snapped and/or buried trees), we also found ample evidence of extension (grabens, trees toppled back onto the Whitman deposit, and trees with drag marks). Grabens are present along ∼60% of the contact between the Whitman and Paleo deposits (Fig. 10; as calculated over the length of contact of the Whitman Q_glv deposit with the Paleo Q_goe deposit; Fig. 4). With the exception of one distinct region of compression on the east containing a high concentration of snapped and buried trees and minor thrusts in the absence of any grabens, few other areas along the contact zone contain exclusively compressional features. Widespread compressional features would be expected if the massive upslope Whitman component impacted a static Hazel-Paleo component. The presence of toppled trees (rooted entirely in Paleo component materials) on top of the Whitman component also indicates that the two components were moving simultaneously. We therefore denote this area of juxtaposed compression and extension as the compression-extension (C-E) zone and interpret its kinematics to infer that the distal edge of the Whitman component and the proximal edge of the Paleo component, although distinct entities, were traveling at similar rates and in close proximity to one another.

Given the C-E zone contact, the surface expressions of source and deposit areas for the Whitman component of the Oso landslide are more distinct than the Hazel and Paleo components. The previously unfailed Whitman source area (Fig. 9A) is characterized by a section of intact and mostly horizontally conformable glacial deposits (i.e., layered Q_goe–Q_gtv–Q_gov–Q_glv from top to bottom) behind the crest of the Paleo landslide headscarp (Fig. 2) and is part of the Whitman Bench topography. The 2014 Whitman deposit, located immediately below (and to the south of) the 2014 Oso headscarp (Fig. 9B), is characterized by large extensional features (i.e., blocks, slices) and exposed scarps and, on flatter ground near the C-E zone, large (10–15-m-tall) hummocks. Although some of the Whitman component traveled out onto the North Fork Stillaguamish River valley floor, most of the mass remained on the slope north of the river valley. The uppermost section of the Whitman component is a block (Fig. 3) composed primarily of recessional outwash that was back-rotated ∼10°–15° and covered with extensive downed trees with tips pointing northward, suggesting rapid downslope acceleration. The interior of the block contains extensional features with transverse scarps and fractures. Downslope of this block, the mass extended and developed multiple slices separated by listric (internal) faults (Figs. 3 and 8), which formed as the slide translated along a slightly inclined (6°) basal failure surface (Fig. 8). These slices created scarps up to 35 m tall, with some (e.g., Fan Lake scarp; Figs. 3 and 8) nearly exposing the Oso basal failure surface. The slice detachment from the Fan Lake scarp (and possible partial collision into the Paleo component ahead of it) may have caused some of the compressional features located on the west side of the C-E zone (Fig. 10). Original stratigraphic relations are mostly preserved in the translated Whitman deposit, and at several locations close to the new alignment of the North Fork Stillaguamish River our mapping (Fig. 4) revealed displaced, but nearly horizontal, stratigraphic contacts for till through glaciolacustrine units (i.e., Q_gtv over Q_gov over Q_glv). Relatively intact blocks of till (which is locally extensive on the east side of the exposed Whitman Bench scarp) were scattered and transported nearly 500 m along the eastern half of the Whitman component (Fig. 4).

The total source volume of the Whitman component (6,311,000 m³) was considerably larger than the combined masses of the Hazel-Paleo slide (Fig. 8; Table 1), and it failed along a slip surface with maximum depth of ∼100 m. The overall shape of the failure surface (as revealed by geotechnical borings and other field observations; see Appendix S2 of the Data Repository [^{footnote 1}]) is arcuate near the headscarp, but translational at a shallow dip over the majority (60%, 330 m) of its length. This shape may be the result of the anisotropy of the dominant (and horizontally bedded) Q_glv sediments, which formed ∼87% (480 m) of the deeper failure surface length (∼550 m; Fig. 8). Anisotropy in clays is known to affect soil shear strength (e.g., Duncan and Seed, 1966; Kirkgard and Lade, 1991) and failure plane shape (e.g., Lo, 1966; Badger and D’Ignazio, 2018).

Two additional aspects of the Oso landslide’s overall structure are notable for their evidence of landslide timing and structure, despite their lack of direct correspondence to overall landslide mobility. These secondary failures were part of the Whitman component, but they occurred subsequent to the major motions of the Hazel, Paleo, and Whitman components. The first consisted of a debris flow along a 400 m length on the east side of the landslide source area (Fig. 4, Q_dfs) that formed soon (<1 h) after the main landslide (see video from search and rescue helicopters, available at https://www.youtube.com/watch?v = UUFByAwcGs0, accessed 23 May 2017). This debris flow mobilized glaciolacustrine clay (Q_glv), likely as a result of groundwater capture from the neighboring drainage basin to the east (Keaton et al., 2014). Notably, this is the only major area of the Whitman component that mobilized into a flow.

The other secondary component of the Oso landslide consisted of an ∼1.1 million m³ section (Table 1) of the intact Whitman Bench glacial deposits that subsequently slid or toppled into the graben formed behind the downdropped back-rotated Whitman block. This deposit (denoted as colluvium, Q_c; Figs. 4 and 8) is composed of juxtaposed outwash and till deposits (i.e., Q_goe, Q_gtv, Q_gov), with modern roots, wood, and other organics (as identified in geotechnical boring EB-04si-15 at 26–38 m depth and within 4 m of the Whitman slide failure surface; Badger, 2016). The presence of recently buried organics at this location suggests that the primary Whitman failure surface intersected the Whitman Bench well in front (to the south) of the final location of the escarpment (see inferred failure surface shown in Fig. 8). The upper part of the colluvial infill consisted of a series of at least three distinct so-called “till” fall events (totaling ∼41,000 m³; Table 1), with the largest event encompassing ∼75% of the total till-fall volume. The largest of these falls, composed of competent blocks of well-cemented glacial till, partially disaggregated upon impact and subsequently spread and buried part of the upper Whitman slide block, splintering back-thrown, 40-cm-diameter trees that had been previously growing on recessional outwash on the top of the Whitman Bench.

The energy associated with the colluvial infilling (sliding and/or falling material, inclusive of the till falls) could have been considerable (estimated to be ∼1150–1950 GJ if the entire mass, with an assumed intact unit weight of ∼2040 kg/m³, experienced a free fall of 90 m [upper value] or if it experienced Coulomb frictional sliding along a roughly 50° backslope with a length of 100 m and a friction angle of 30° [lower value]). Although this infilling may not have contributed directly to the mobility of the 2014 landslide, it may assist interpretation of the recorded 2014 seismic signals from the landslide (see subsequent “Discussion” section). Characteristics of these colluvial materials also provide insight into the composition of the earlier Paleo source area prior to its failure in 2014. For example, the predominantly loose, sandy, forest floor covering recessional outwash deposits (Q_goe) found in hummocks at the back edge of the 2014 Paleo landslide deposit contains seemingly random blocks of till (Q_gtv) and glaciolacustrine clay (Q_glv; mapping locations are provided in Data Sheet S1 of the Data Repository [^{footnote 1}]). These out-of-stratigraphic-sequence blocks may be the result of similar colluvial infilling in the graben of the prehistoric Paleo landslide.

Here, we present our interpretation of the sequence of events of the 2014 Oso landslide. The sequence is important not only for understanding the failure mechanics and kinematics of the landslide, but also because it points to factors that likely enhanced the landslide’s mobility. We base this sequence primarily on our field observations and analysis of the source and deposit components, as presented in previous sections. We organize the sequence into three subsections of motion: (1) slope preconditioning and landslide triggering, (2) initial motion, and (3) landslide transport. The final subsection is further divided into two parts, describing leading-edge debris-flow formation and landslide runout. When discussing landslide runout, we provide detailed information on our proposed mobility mechanism (valley bottom liquefaction), which sets the stage for our subsequent quantitative analyses of this mechanism. Each of the sequence stages references various components of the landslide (Hazel, Paleo, Whitman) as previously described.

Piecing together the sequence of postfailure landslide mobility for the Oso landslide requires some understanding of the reasons for its initiation. Although we deliberately focused on the Oso landslide’s mobility in this study rather than on its initiation, triggering factors can shed light on aspects of mobility. For example, hydrologic conditions are generally important, with fluidization of water-saturated sediment leading to larger mobility (e.g., Legros, 2002; Okura et al., 2002; Iverson et al., 2011). Landslide volume and topography (e.g., Corominas, 1996), as well as failure history, can also affect landslide mobility. For example, the La Conchita, California, landslides in 1995 and 2005 occurred on a slope with evidence of prehistoric landsliding (Jibson, 2005). To this end, the history of landsliding at the Oso site helps to place the 2014 slide’s prefailure state into context.

The site of the Oso landslide consisted of a 180-m-tall terrace of glacially derived sediments on the north side of a valley that had been previously subject to extensive fluvial incision at its base. Although the toe of the 2014 failure surface was above river level at the time of failure (Fig. 8), toe erosion did establish the overall geometric configuration of the slope. Moreover, the stress condition of the slope also likely played a role in its failure, with large slope height (and resultant higher stress state) leading to lower overall stability (Badger and D’Ignazio, 2018).

The effect of multiple episodes of previous landsliding within the same slope as the Oso slide (e.g., Thorsen, 1969; Miller and Sias, 1997) also may have contributed to the 2014 failure (Wartman et al., 2016). The failure planes for the former Hazel and Paleo landslides were likely well established and may have facilitated movement of these parts of the 2014 landslide. Rapid failure may have occurred as a result of significant strength drop upon renewed shearing on these surfaces (e.g., Tika et al., 1996; Schulz and Wang, 2014). Additionally, some research (Badger and D’Ignazio, 2018) indicates that progressive failure of Q_glv silts and clays through the deeper failure surface (Fig. 8) may have played a dominant role in initial triggering, with strain softening of these overconsolidated sediments leading to a loss of shear strength over time (e.g., Terzaghi, 1936; Skempton, 1964; Bjerrum, 1967).

Finally, it has been hypothesized that elevated pore-water pressures in the subsurface, induced by infiltration from significant rainfall in the weeks or years leading up to failure (Henn et al., 2015), may have triggered sliding (Iverson et al., 2015; Stark et al., 2017; Wartman et al., 2016). On the other hand, if a progressive strength-loss mechanism triggered initiation (Badger and D’Ignazio, 2018), elevated pore pressures would not necessarily have been needed. It is possible that some type of hybrid failure mechanism controlled initiation, such as progressive failure of lacustrine clay in the Whitman component combined with pore-pressure–induced failure (with possible loading from the upslope Whitman component) in the Hazel-Paleo component. Although we do not specifically address initial triggering in our analyses, it is clear that the 2014 Oso landslide occurred on a slope that was preconditioned by prior landsliding (i.e., the Hazel and Paleo components) for future instability.

Regardless of the exact mechanism for initiation (precipitation-induced pore pressure changes, progressive strength loss, or other), our analysis of field evidence (see “Relation between Landslide Components” section) indicates that the Hazel and Paleo components failed together as one mass and that they likely initiated just shortly before the Whitman component began to slide. The short time gap in initiation of these two components is evidenced by both compression and extension in the C-E zone (Fig. 10). If a static Hazel-Paleo component was impacted by later failure of the Whitman component, the C-E zone would show predominantly collisional or compressional features. Alternatively, a postfailure static Whitman deposit with secondary motion of the Hazel-Paleo component in front of it would result in solely extensional features in the C-E zone. The presence of both features in this zone indicates jostling between the components and that they were in motion simultaneously.

Although our field evidence indicates that both components were moving closely in time, a precise failure mode is not directly apparent. Geological and structural evidence north of the C-E zone indicates that the Hazel-Paleo component pulled away from the Whitman component by a sufficient distance to allow exposure and subsequent extension of the Q_gov and Q_glv parts of the Whitman mass (Figs. 4 and 8). If the Whitman component failed first (progressive failure in Q_glv sediments), with the Hazel-Paleo component sliding away in response, a large separation of the two masses concurrent with jostling in the C-E zone might be less likely. Thus, our mapping suggests a more retrogressive failure mechanism, with the Hazel-Paleo component moving prior to the Whitman component. Regardless of whether initial motion was progressive or retrogressive, all source components failed closely in time, thereby generating considerable momentum. This has important ramifications for mobility in that the Oso landslide moved out onto the river valley in two closely spaced components, but as one mass.

Given these relations, our interpreted sequence begins with the Hazel-Paleo components sliding (and potentially dropping slightly) from the toe of the basal slide surface located ∼10 m above the existing ground surface at the base of the slope (Fig. 8, section distance 530 m to 560 m). As the Hazel-Paleo component moved out into the valley, it pushed and entrained the North Fork Stillaguamish River immediately in front of it, which was transported across the valley (see subsequent section). The Whitman component, now partially exposed by the backscarp of the newly evacuated Paleo component (Fig. 8), followed closely behind the still-moving Hazel-Paleo mass. As the Whitman component moved southeastward, it rounded a resistant knob (known as the “Pyramid”; Fig. 3) composed of downdropped deposits displaced from prior prehistoric landsliding and collocated with the Devils Mountain fault trace (Johnson et al., 2001; Dragovich et al., 2003). All three components then spread into the unconfined space of the North Fork Stillaguamish River valley floor.

Several events, including the formation of a debris-flow front and transport of the slide mass across the North Fork Stillaguamish River valley, occurred quickly following initial motion of the landslide. Other investigators have referenced these events to seismic records and eye-witness accounts; we compare our findings to these previous versions in the subsequent “Discussion” section. Here, we use and discuss our field observations to support our inferred sequence for landslide transport and mobility.

River entrainment and debris-flow formation, runout, and reflection. Immediately after the 2014 landslide, the most distal parts of the deposit contained fully liquefied materials with a perimeter snout of logs and debris, characteristic of a debris flow. The debris flow represented the leading edge of the landslide mass; as it swept across the floodplain, it entrained nearly all surface features (trees, cars, and structures, including victims) that previously existed on the south side of the river. Pieces of the log revetment formerly located on the north side of the river were also found within some parts of the debris-flow deposit (Fig. 5). Many of these items were transported to the outer margins of the deposit (Fig. 4D), as evidenced by the locations of structural debris and victim recovery (Quistorf, 2014). In the first weeks after the event, we observed liquefied pools of sediment and ponded water at the distal end of the landslide deposit formed from debris-flow emplacement; these were a significant hindrance to rescue workers (Fig. 4D).

Field samples we collected several weeks after the landslide at the distal edges of the debris-flow deposit revealed an average bulk density of 1950 kg/m³ (Table S1 [^{footnote 1}]). With a density roughly twice that of water, the rapidly moving debris-flow slurry had sufficient momentum to sweep houses from their foundations. However, few indications of subsurface entrainment or scour within the Steelhead Haven neighborhood were found during our or other previous investigations (Iverson et al., 2015; Wartman et al., 2016). Our mapping did identify two sections of the SR530 road bed, with a combined length of ∼150 m, that were disrupted by the landslide (Fig. 5B), but they appear to have been part of engineered embankments built 3–4 m above the adjacent topography.

The debris flow likely formed as the toe of the combined Hazel-Paleo component crossed the relatively flat, 200-m-wide area just north of the North Fork Stillaguamish River, struck the 2006 log revetment along the north edge of the North Fork Stillaguamish River, and entrained the river. Prior to the 2014 event, an ∼10-m-thick wedge of colluvium (tapering from 21 m to 1 m in thickness; Fig. 8), derived from 2006 Hazel slide debris, had accumulated behind the roughly 4-m-tall log revetment (Keaton et al., 2014; Wartman et al., 2016). The leading edge of the 2014 slide may have mobilized this flat-lying debris north of the revetment as well. However, given that the failure surface toe emerged from the slope above this older colluvial debris (Fig. 8), the Hazel-Paleo component may have primarily overrun this debris before impacting and entraining the log revetment. Collapse during failure of loose Hazel component materials on the slope may have aided debris-flow formation at the front of the moving mass, although field evidence of hillslope liquefaction is minimal. A significant source of water for debris-flow mobility likely came from entrainment of the North Fork Stillaguamish River, with the remaining being in the previously failed debris. Based on time-series comparison of measured and correlated flows in the river at the time of the 2014 landslide (see Appendix S2 in the Data Repository [^{footnote 1}]), we estimate the entrained river volume to be ∼78,000 m³ or roughly 18% of the debris-flow deposit (424,000 m³; Table 1).

The debris flow was unconfined as it traveled onto the flat North Fork Stillaguamish River valley, and it moved in tandem with the mass of the combined Hazel-Paleo component immediately behind it. Our mapping shows that the debris flow reflected off the southern central wall of the valley, flowed back toward the north, and onlapped onto the hummock field of the Hazel component of the deposit (Fig. 11). This reflection indicates that the debris flow remained close to and in front of the Hazel-Paleo component as the components traveled across the valley. Where the debris flow was not reflected (i.e., in the eastern and western lobes), the debris thinned as it traveled farther, as evidenced by measurements of mud onlap deposits on the base of trees (Fig. 11A). These thinner deposits commonly contained small hummocks (typically <1 m high) of various glacial units that progressively decreased in size with farther travel distance. Average distal tree onlap measurements provide a reasonable indication of the distal deposit thickness (0.8 m and 1.2 m in the west and east lobes, respectively; see Appendix S2 of the Data Repository [^{footnote 1}]), and they likewise help to constrain the volume (424,000 m³) of the debris flow. Although the debris-flow component of the event caused significant damage, it only represented ∼4% of the total landslide deposit volume (Table 1).

Landslide runout and extension via valley bottom liquefaction. After initial motion and subsequent entrainment of the river, the combined Hazel-Paleo component, along with the southern part of the Whitman component, traveled rapidly across the North Fork Stillaguamish River floodplain immediately south of the slide (Fig. 9). The deposit from this part of the landslide mass was primarily a field of hummocks, characteristic of a debris avalanche. The hummocks formed when the mass extended and spread in multiple directions into the accommodation space offered by the open, relatively flat North Fork Stillaguamish River valley bottom. Hummock materials were not liquefied, and most of their cores consisted of one stratigraphic unit (i.e., Q_goe, Q_gtv, etc.), except in distal smaller hummocks, where mixed materials were common. Upright ferns and forest floor also remained intact on hummock tops (Fig. 5B). Proximal hummocks were typically ridgelike, with longer axes oriented perpendicular to the direction of motion. Many distal hummocks displayed nearly equidimensional geometries, indicating that extension occurred similarly in both travel-parallel and travel-perpendicular directions. In some cases, hummocks themselves underwent internal extension with jigsaw-puzzle fabric and vertical cracks extending tens of decimeters downward into the hummocks (Fig. 4C). These features are common in other debris-avalanche hummocks (e.g., Roberti et al., 2017).

As a feature of extensional kinematics, hummocks typically indicate the presence of a weak base underlying the landslide mass (Paguican et al., 2014). The relatively flat North Fork Stillaguamish River valley bottom beneath the Steelhead Haven neighborhood, subsequently buried by hummocks (Fig. 12), consisted of alluvial sediments (primarily silts, sands, and gravels) ranging from <1 to 13 m in thickness (Fiske, 2014; see also Table S5 [^{footnote 1}]). Given the anomalously high precipitation and resultant soil moisture during March 2014 (Henn et al., 2015), near-surface alluvium was likely saturated or nearly saturated with shallow groundwater near the ground surface. Geotechnical borehole drilling conducted 1 month following the landslide (Fiske, 2014) encountered groundwater within 1–2 m of the preslide ground surface over much of the distal edge of the deposit, even after a temporary lake that had formed by the landslide blocking the North Fork Stillaguamish River had drained.

Several lines of field evidence indicate that liquefaction of the North Fork Stillaguamish River valley alluvium was the likely source for this weak base, and for the high mobility of the slide mass. The first is the occurrence of the hummocks themselves. Lateral deformation resulting from liquefaction often results in hummock-like features (e.g., Hansen, 1965; Harty and Lowe, 1995; Davies, 2003), and thus liquefaction of the underlying alluvium offers a compelling explanation for their formation. The second line of evidence is a strong spatial correlation between the distribution of hummocks and the location of the flat, pre-2014 North Fork Stillaguamish River alluvial valley. Our structural map (Fig. 3) shows that nearly ubiquitous extension of hummocks exists only south of the earlier (2003) location of the North Fork Stillaguamish River (near the present-day location of Fan Lake; Fig. 8). On the other hand, the area north of Fan Lake (within the northern part of the Whitman deposit) is composed mainly of large, linear, slice-like blocks with their long axes oriented perpendicular to the direction of landslide motion. The transition from larger scarps in the north to hummocks in the south occurs at the location of the start of the preslide floodplain and indicates that basal liquefaction likely occurred to the south of this transition point.

The third key piece of evidence for basal liquefaction-induced mobility comes from observation and identification of hundreds of liquefaction features (i.e., sand boils; Figs. 6 and 7) throughout the deposit zone south of Steelhead Lake (Fig. 3). Given the occurrence of sand boils within many slosh pit features, slosh pits may also be liquefaction-related features, possibly formed when underlying alluvium was expelled or incorporated into these depressions. Sand boils and slosh pits are primarily present south of the recent (2013) North Fork Stillaguamish River (Fig. 6), with the vast majority located between hummock mounds where thinner landslide deposits allowed a shorter pathway for pore-pressure dissipation from the liquefied base (Fig. 13). The formation of sand boils through and on top of “backwashed” distal debris-flow deposits (Figs. 6 and 13) indicates that the sand boils were a late-stage depressurization response of the underlying alluvium. Our geotechnical characterization of sampled alluvium and sand boils from the field (Fig. 14) revealed pronounced similarities in their grain-size distributions; the alluvium underlying the landslide deposits (gray shade in Fig. 14) appears to have been the source of the sand boils (red lines in Fig. 14). Of the other geologic units forming the slide mass (Q_goe, Q_gtv, Q_gov, Q_glv), only Q_goe has a similar grain-size distribution to that of sand boils; however, it is coarser grained than the sand boils, indicating a poorer match for the provenance of the sand boils.

Liquefaction of the alluvium likely occurred when the quickly moving Hazel-Paleo component and southern part of the Whitman component rapidly loaded, sheared, and possibly vibrated the saturated, porous alluvium of the flat valley (Fig. 12). The dynamic processes of the overriding landslide mass were sufficient to cause increases in pore-water pressure (e.g., Hutchinson and Bhandari, 1971; Sassa and Wang, 2005) and liquefy the underlying sands and gravels. The resulting loss of shear strength in the alluvium would have allowed the landslide mass to continue forward, subsequently loading additional alluvium and causing a wave of progressive liquefaction under the landslide mass in a complex, yet synchronized, sequence of events. We investigate liquefaction mechanisms in the next section.

Our interpretation of the landslide sequence posits that the Oso landslide’s enhanced mobility was caused by liquefaction of sediments (i.e., alluvium) in the North Fork Stillaguamish River floodplain that underlay the overriding landslide materials. This mechanism differs from liquefaction of the entire landslide mass. Liquefaction, when soil effective normal stress decreases to zero as a result of pore-water pressure increases, reduces shear strength (e.g., Poulos et al., 1985; Kramer and Seed, 1988; Yamamuro and Lade, 1997) and can greatly aid mobility (e.g., Sladen et al., 1985; Olson and Stark, 2002). Large surface deformations from basal liquefaction have been well studied and can result from either cyclic shaking (e.g., Bartlett and Youd, 1995; Seid-Karbasi and Byrne, 2007) or monotonic loading (so-called static liquefaction; e.g., Sladen et al., 1985; Fourie and Tshabalala, 2005; Take and Beddoe, 2014). These mechanisms have been replicated in laboratory settings (e.g., Kokusho, 1999), where it has been shown that water films (with zero shear strength) are formed above liquefiable layers and can last hundreds of seconds, allowing for continuous deformation of overlying sediments. Here, we investigated three possible mechanisms that could have liquefied the alluvium during the 2014 Oso event. The first two mechanisms (rapid loading and static shearing) may have acted in progression. The third mechanism (cyclic shearing) could have been a contributing factor to the other two mechanisms.

Upon initial movement of the landslide in 2014, the Hazel-Paleo slide mass dropped ∼10 m over an incised former slope bank of the North Fork Stillaguamish River (see “NFSR (2003)” in Fig. 3 and the section from 530 m to 550 m in Fig. 8), resulting in loading of post-2006 landslide colluvium deposits overlying North Fork Stillaguamish River alluvium. As the mass of the Hazel-Paleo slide, and the Whitman slide behind it, traveled onto the floodplain, it would have rapidly loaded the saturated or nearly saturated alluvial sediments, causing a rapid rise in alluvial pore pressures. If increased pore pressures were generated from rapid loading, then this could have contributed to liquefaction of the alluvial sediments.

It is well known that undrained loading can increase pore-water pressures within underlying sediment (e.g., Lambe and Whitman, 1969). Undrained loading has been invoked to explain both slow and rapid landslide motion (e.g., Hutchinson and Bhandari, 1971; Sassa, 1985; Sassa and Wang, 2005), and in some cases, it has been directly observed using replicable experiments (Iverson et al., 2011; Reid et al., 2011). However, many natural settings are not perfectly undrained. Thus, the degree of fluid pressurization depends on the rate of deformation loading by sliding versus the rate of pore-pressure diffusion as controlled by sediment properties such as compressibility and hydraulic conductivity (e.g., Iverson and LaHusen, 1989; Iverson et al., 2004, 2010). If loading is relatively slow and pore pressures can be quickly dissipated, then little pressure increase will occur. To greatly increase pore pressures, loading must be rapid compared to pressure diffusion.

To examine potential pore-pressure generation from rapid loading during the 2014 Oso event, we used a two-dimensional numerical model of coupled elastic sediment deformation and groundwater flow. The finite-element code, called Biot2 (Hsieh, 1996; Hurwitz et al., 2007), simulates linearly elastic deformation in a porous medium under small strain. For the porous medium, we used parameters determined from North Fork Stillaguamish River alluvium sampled near and under 2014 Oso landslide deposits (Fig. 12; see also Appendix S1 of the Data Repository [^{footnote 1}]); parameters included drained elastic moduli (E = 2.1 × 10⁷ N/m²), saturated hydraulic conductivity (K_sat = 4.8 × 10⁻⁵ m/s), and porosity (n = 0.46; see Appendix S3 in the Data Repository for more information on these simulations [^{footnote 1}]). We used a 10-m-thick domain of saturated alluvial sediment subject to a vertical traction (load) of 100 kPa, representing a 5-m-thick saturated landslide mass. We did not include any transient impact loading or basal stresses induced by velocity. Pore pressures were not allowed to diffuse into the overriding slide mass (a reasonable assumption given the low permeability of much of the overlying deposits), but they could dissipate through the alluvium to the ground surface in front of the sliding mass. To simulate the transient effects of rapid loading by slide motion, we sequentially applied this load piecewise over time at model cells on the ground surface. In our simulations, the speed of the advancing front averaged a constant 10 m/s; this was based on an estimated runout distance of 1000 m for the majority of the landslide materials over a period of 100 s under conditions of strong ground shaking (indicative of landslide motion; Iverson et al., 2015). The 2014 landslide underwent periods of acceleration and deceleration as it crossed the river valley, resulting in nonsteady landslide velocities that are not captured by our model. Dynamic simulations by Iverson and George (2016) produced a maximum slide front speed of ∼30 m/s.

We show pore-pressure distributions in the alluvium at two time steps from these simulations (Fig. 15). Several observations are relevant to possible basal liquefaction of the overridden sediments. Although the compressible alluvium has high hydraulic conductivity, pressures still rapidly increase (in <0.1 s) directly beneath the slide mass. Elevated pressures are concentrated immediately under the mass and remain significantly elevated (>2 m pressure head) to a depth of several meters while the simulated slide passes overhead. For the 5 m (100 kPa) equivalent load, the largest excess (above hydrostatic) basal pressure heads are ∼9 m in this scenario, approximate equivalent to the applied load. Moreover, these elevated basal pore pressures would modify effective stresses in parts of the alluvium, thereby promoting enhanced shearing (see next section). The 1–2 m depth zone of higher pressures correlates well with our field observations of shallow scour (∼1 m) of buried tree roots near the current river location (Fig. 12). The simulation also reveals that elevated pore pressures propagate as a subsurface wave in front of the rapidly moving slide mass, thereby potentially reducing the shear strength of the alluvium in this region and preconditioning it for subsequent failure.

Reasonable variations in sediment parameters affect the simulated pore-pressure responses slightly, but the overall response patterns remain similar. Variations in applied vertical load, however, can greatly affect the responses. For example, doubling the load will roughly double the simulated pore-pressure increase beneath the overriding slide mass. Thus, a larger impact load from dropping of the overriding slide mass in the initial stages of motion could elevate pore pressures beyond the vertical static load imposed in the simulations (Fig. 15). The precise amount of impact loading is difficult to estimate without a dynamic model. During the 2014 Oso event, any significant direct impact loading would have mostly affected sediments at the base of the slope (i.e., near the area of the 2003 North Fork Stillaguamish River; Figs. 3 and 8) as the slide transitioned onto the valley floor. These impact effects would have greatly diminished as the slide mass traversed the alluvial flats. We note that scouring of older slide debris, colluvium, and alluvium present only at the base of the slope during the 2014 event supports this premise (see 5 m elevation difference between pre- and postlandslide ground surface in the area of Fan Lake; Fig. 8A).

Our model simulations show that rapid elastic loading may have generated elevated pore-water pressures nearly sufficient to transiently support the overriding load. In addition, it is possible that the alluvium underwent additional undrained monotonic (static) shearing to further aid liquefaction. In this mechanism, an initially loose soil structure can collapse from the application of normal or shear stresses (e.g., Yamamuro and Lade, 1997; Take and Beddoe, 2014), thereby generating sufficient pore-water pressure to result in liquefaction and a complete loss of effective normal stress and shear strength (Terzaghi et al., 1996). A lack of shear strength in the alluvium would then have allowed flow and mobility of the alluvium and further enhanced the mobility of the overriding landslide material.

To investigate whether the North Fork Stillaguamish River alluvium was capable of this form of liquefaction, we conducted consolidated, undrained (i.e., CU) triaxial tests on both intact and reconstituted alluvium samples taken from the North Fork Stillaguamish River just upstream of the landslide, but unaffected by the overriding slide mass (see Appendix S1 and Figure S2 in the Data Repository [^{footnote 1}]). These are the same types of sediments that were in place throughout the adjacent floodplain prior to the 2014 landslide. We provide full details of the triaxial testing program in Appendix S1 of the Data Repository (see ^{footnote 1}). Triaxial testing allows investigation of both the soil structure collapse potential and the peak loads needed to cause shear. In these tests, a sample is consolidated by an all-around pressure to represent the stress at a particular depth and then vertically loaded until failure. Here, we chose confinement stresses of 20 kPa and 100 kPa, representing 1 m and 5 m depths, respectively, assuming undrained field behavior and a coefficient of earth pressure at rest (K₀) equal to 1. For dense soils, the behavior is normally dilative, and although some positive pore-water pressure can be generated during loading, it usually does not cause a significant loss of shear strength. On the other hand, if a soil is loose, the shearing process may densify the soil, forcing the voids to contract and thereby generating increased pore-water pressure. This results in little resistance to shear loading, with resultant extreme displacements (e.g., Sassa, 1985; Sassa and Wang, 2005).

Our triaxial test results for reconstituted, stratified samples of the North Fork Stillaguamish River alluvium indicated a mixed dilative-contractive response (Fig. 16A), but they also revealed that some layers (Fig. 16C, layer 2) of alluvium underwent a large pore-water pressure response resulting in full liquefaction (i.e., p′ → 0 and q → 0 in Fig. 16A). Consistent with previous work on this subject (e.g., Yamamuro and Lade, 1997), we found that the higher fines content (i.e., percent of soil <0.075 mm) and well-graded size distribution of some layers of the alluvium (e.g., layer 2 in Fig. 16; see also Appendix S1 in the Data Repository [^{footnote 1}]) are likely responsible for the dramatic contractive behavior. Other layers (e.g., layer 3 in Fig. 16) have lower fines content (1.1%) compared to the fines content (6.2%) of liquefiable layers. During shear, these fines can migrate to the void space of the coarser particles and result in an increase in pore-water pressure (Yamamuro and Lade, 1997). Thus, loading the North Fork Stillaguamish River alluvium could have resulted in dramatic strength loss of some sediment layers, which would have aided the landslide’s mobility as it traversed the floodplain.

A final mechanism by which the alluvium may have softened and become mobile is cyclic liquefaction (e.g., Casagrande, 1976; Seed, 1979). Here, ground vibrations from the motion of the landslide itself could have caused cyclic shearing and contraction of the alluvial sediments, leading to increases in sediment pore-water pressure. Whereas earthquakes are a common cause of cyclic shearing, liquefaction has also been documented as a result of other weaker modes of vibration, including seismic exploration (Hryciw et al., 1990), blasting (Charlie et al., 1992; Ashford et al., 2006), train traffic (Pando et al., 2001), pile driving (Hwang et al., 2001), and even ocean waves (Dalrymple, 1979).

Given eye-witness reports and recorded seismic signals of long-duration (∼1 min), low-level vibrations in the early stages of the landslide (see Keaton et al., 2014; Iverson et al., 2015) and the high likelihood that the North Fork Stillaguamish River alluvium met typical conditions for liquefiable soils (they were both loose and saturated), we performed a cyclic liquefaction analysis for the Oso landslide. We used the cyclic stress method of Seed and Idriss (1971) to compare the cyclic stress ratio (CSR), as a mechanism for soil forcing, to the cyclic resistance ratio (CRR), as a measure of soil resistance to liquefaction (see Appendix S4 of the Data Repository [^{footnote 1}]). Using parameters specific to the Oso site (e.g., standard penetration test [SPT] measurements [N_1,60 = 3], in situ effective soil stress ratio [σ_vo/σ_vo′ = 2.08], and peak ground acceleration [a_max = 0.822 m/s²] from the landslide), we calculated a CRR of 0.097 and a CSR of 0.113, resulting in a factor of safety (F_s = CRR/CSR) for liquefaction of 0.9 (where F_s < 1 indicates probable liquefaction). This suggests that cyclic shearing–based liquefaction of the alluvium was possible as a result of landslide-induced ground shaking.

Vibrations from a moving landslide are considerably different from those with which the Seed and Idriss (1971) methodology was developed (i.e., earthquakes), and several assumptions were needed for our analysis. However, our approach is based on both measured sediment forcing characteristics (from seismometers) and sediment resistance parameters (from subsurface investigations), resulting in a robust analysis. Thus, cyclic liquefaction might be a plausible contributing mechanism enhancing liquefaction-induced mobility of the Oso landslide.

Several previous studies have invoked liquefaction as the primary cause of the high mobility of the Oso landslide (Iverson et al., 2015; Iverson and George, 2016; Stark et al., 2017; Aaron et al., 2017); these studies suggested the possibility of rapid compressional undrained loading from collapsing material upslope promoting liquefaction. However, these investigations differ in identifying the nature and location of the materials that liquefied, as well as the processes inducing liquefaction. Understanding these issues is fundamental to evaluating future landslide hazards in similar settings. Aaron et al. (2017) assumed a liquefied material rheology for modeling their downslope failure composed of colluvium from previous landsliding, but not for their larger upslope failure mass. They suggested that widespread colluvial liquefaction might have been caused by strong deformation of previously failed blocks of clay and silt infilled with sediment or by loss of cohesion in lightly cemented, saturated silt. Iverson et al. (2015) presented a variety of mechanisms that could aid liquefaction, but their model focused on dynamic feedbacks among evolving landslide momentum, material porosity, and pore pressure. In this approach, enhanced liquefaction (and landslide mobility) was primarily a product of looser initial materials in the slide mass itself (Iverson and George, 2016).

Our study identifies the valley alluvium (and possibly some colluvium located at the base of the slope from previous sliding) as the primary material that liquefied. This liquefaction occurred at the base of the slide (Fig. 12) with hummocks rafting on top; importantly, the slide mass itself (with the exception of the debris-flow components) did not liquefy. Liquefaction of the entire slide mass would have precluded the formation and preservation of hummocks containing preslide glacial stratigraphy. Evidence for basal liquefaction was notable nearly everywhere that the landslide overran valley bottom alluvial sediments.

Our findings suggest that sediment-filled river valleys can present a prime condition for potential basal liquefaction. Alluvial sands and gravels are common in many river valleys worldwide. In rainy regions, such as the Pacific Northwest, alluvial sediments may be saturated or nearly saturated during the rainy season, when landsliding is more likely from adjacent hillslopes. Near-surface alluvial sands and gravels are typically in a loose state, having not undergone significant consolidation, and are commonly composed of rounded (as opposed to angular) particles. This makes them more susceptible to liquefaction at low (i.e., near-surface) confining pressures when dynamically loaded or sheared when wet (Vaid et al., 1985; Wei and Yang, 2014).

Following the Oso landslide, other large slide deposits were studied in the North Fork Stillaguamish River valley (Haugerud, 2014; LaHusen et al., 2016; Booth et al., 2017; Perkins et al., 2017; Badger and D’Ignazio, 2018), some of which were previously identified by Dragovich et al. (2003). Some of these deposits have similar morphologies to the Oso slide, including extensive hummock fields spread across the valley bottom. As the Oso slide had a nonseismic trigger, it is likely that some previous large-volume events may also have been precipitation triggered, with subsequent rapid loading of saturated valley bottom alluvium. Moreover, large, rapid landslides with failure surfaces in low-permeability materials (e.g., the glaciolacustrine clays present in the North Fork Stillaguamish River glacial terraces) may enhance pore pressures by impeding pressure dissipation. Basal liquefaction of alluvium may have played a prominent role in these prehistoric North Fork Stillaguamish River slides, as reflected in their long-lived hummock morphology.

Basal liquefaction may be a more widespread mechanism than is currently recognized for high-mobility slides that are not more fully liquefied, such as flowslides and debris flows. Large debris-avalanche deposits from volcano flanks commonly exhibit hummock fields, and modern examples at Ontake, Japan (Voight and Sousa, 1994), and Mount St. Helens, Washington, USA (Major et al., 2005), suggest an underlying weak, wet, liquefied layer. Field studies of other landslides also indicate that liquefaction can enhance slide mobility. These include the Nomash River slide (Hungr and Evans, 2004) and the Attachie slide (Fletcher et al., 2002), both in British Columbia, Canada, the Flims landslide in Switzerland (Calhoun et al., 2014), the South Jingyang Platform landslides in China (Peng et al., 2018), and the Gamahara and Kameyama debris flows in Japan (Sassa and Wang, 2005). However, basal liquefaction itself can be difficult to document, as failure surfaces are typically hidden, and surface expressions of deeper liquefaction (e.g., sand boils) are highly transient and subject to disintegration. Some of the sand boils that we documented at Oso could not be found just several months later due to degradation from precipitation, runoff, and bioturbation.

Understanding the behavior of the Oso landslide is crucial to understanding its large mobility and ensuing devastation, as well as enabling accurate assessment of similar landslide hazards elsewhere. Our field observations and measurements provide firm evidence for valley bottom liquefaction as the primary explanation for Oso’s large mobility. Nevertheless, fundamental controls such as failure surface geometry, volume, materials, and failure sequence all contributed to its behavior. Moreover, these controls are essential to analyzing or modeling landslide failure and movement. Several previous investigators have examined these issues (e.g., Keaton et al., 2014; Iverson et al., 2015; Hibert et al., 2015; Wartman et al., 2016; Stark et al., 2017; Aaron et al., 2017; Yerro et al., 2017) and further discussed them in several discussions and closures (Iverson, 2018a; Aaron et al., 2018; Iverson, 2018b; Keaton et al., 2018; Diyaljee, 2018; Stark et al., 2018). However, some of our observations differ in critical ways. Here, we briefly discuss issues that bear directly on how the Oso landslide failed and implications for future hazard assessments.

The geometry and location of landslide failure surfaces are crucial factors for defining initial instability, kinematics, and movement pathways. Based on boring-log information (Badger, 2016), the deeper Whitman slide mass was underlain by a gently sloping basal failure surface that permitted the development of the observed series of back-rotated slices (Hutchinson, 1988) and Toreva block-like structures (Reiche, 1937; Paguican, et al., 2014), as well as allowed material to be transported directly onto the North Fork Stillaguamish River valley bottom. Moreover, because most of the basal failure surface was located in cohesive glaciolacustrine materials, its strength dominantly controlled initial sliding. Our failure surface for the Whitman component differs from the shallower, more circular or log-spiral surfaces previously presented by others (e.g., Iverson et al., 2015; Wartman et al., 2016) prior to borehole data becoming available. Our failure surfaces also differ from both phase I and II composite failure surfaces shown in Stark et al. (2017), which are located at higher elevations. However, our failure surfaces roughly align with those shown in Aaron et al. (2017) and Badger and D’Ignazio (2018).

Material distribution is also crucial to understanding the nature of the basal liquefaction that enhanced mobility. Our mapping of postslide hummock composition illustrated that the distribution of materials was not predominantly chaotic nor was it primarily glaciolacustrine deposits overlain by discontinuous ridges of sand, as might be assumed in the absence of detailed field mapping. Rather, it reflected the distribution of stratigraphic layers present in the pre-2014 hillslope. As the initial (Hazel-Paleo) and subsequent (Whitman) failures traveled across the valley bottom, they were rafted and stretched into hummocks and grabens composed of their original glacial stratigraphic sequences. This pattern was echoed in the distribution of stretched forest floor both north and south of the C-E zone and provides strong evidence for basal sliding, not internal flow, of most of the materials.

Our mapping of sand boils, particle-size analysis, and examination of slide surfaces at the postslide North Fork Stillaguamish River and along the exhumed SR530 roadway indicated that most hillslope materials (including those preserved in the deposit as hummocks) did not liquefy. Although some of the Hazel-related colluvium at the base of the slope (now buried) may have liquefied when loaded during sliding (Aaron et al., 2017; Stark et al., 2017), most of the Hazel-Paleo runout occurred over valley alluvium. As Aaron et al. (2017) noted, a mechanism for direct liquefaction of the Hazel-related colluvium (an amalgam of glacial units) is uncertain. However, virtually all liquefaction features occurred in areas where flat-lying alluvium existed, with very limited observed liquefaction on the hillslope. Moreover, we found glaciolacustrine landslide hummocks directly over alluvium (with no layers of intervening colluvium) in the North Fork Stillaguamish River banks (Fig. 12) and also at the far distal end where the landslide overrode the SR530 road bed. Thus, our observations limit the pushing and flow of fluidized colluvium, previously at the base of the slope (Stark et al., 2017), as the predominant source and cause for widespread liquefaction.

The sequence of landslide failure is also essential for understanding landslide mobility, as it directly affects volume and momentum considerations needed for hazard assessment and dynamic mobility modeling. Our observations indicated two failures very closely spaced in time. The C-E zone defining the contact between the two slide masses showed signs of jostling, compression, and extension (as indicated by grabens and backwardly toppled trees dragged for meters off the upslope mass). Some previous investigators identified this contact zone as primarily collisional (Keaton et al., 2014; Wartman et al., 2016), whereas others noted the absence of large compressional features such as faults or pressure ridges (Aaron et al., 2017). The dynamic jostling recorded in the C-E zone suggests there was at least some momentum transfer between the two masses. Moreover, this momentum transfer may have helped push the initial mass onto the valley bottom, thereby facilitating the development of widespread alluvial liquefaction. Previous landslides at the site may not have been sufficiently large or rapid to provide the momentum to push material onto the floodplain and create alluvial liquefaction. In addition, both major components (Whitman and Hazel-Paleo) experienced basal liquefaction and hummock formation, indicating some behavioral continuum across the components during transport. In this context, our observations do not support liquefaction behavior only in the initial slide and frictional behavior only in the upslope slide (Aaron et al., 2017). Thus, modeling an evolving mass may be a better representation of Oso landslide behavior than two completely noninteracting masses.

The landslide sequence also influences the interpretation of seismic signals generated by sliding. For example, records from a nearby seismic station (11.7 km away) show two high-amplitude signals during the morning of the Oso event, separated by ∼4 min (Hibert et al., 2015; Iverson et al., 2015). The first high-amplitude event represents the onset of major sliding. The source for the second high-amplitude signal has been debated. Some investigators (Keaton et al., 2014; Hibert et al., 2015; Wartman et al., 2016) favor the large second slide (here, termed Whitman) as the second seismic source, whereas others (Iverson et al., 2015; Allstadt, 2015), using different analyses, favor a smaller source such as a secondary failure of the headscarp. In all these analyses, however, the second event was more impulsive and suggested a smaller mass than the initial event. Our volume calculations indicate that the Whitman component was about twice as large as the Hazel-Paleo components (Table 1), thereby making the Whitman component an unlikely candidate for the second event. In addition, our field observations indicate that all slide components (Hazel-Paleo-Whitman) were traveling simultaneously, and therefore failed closely in time (likely in the initial minute+ of strong shaking). Thus, our observations support a secondary failure (perhaps the ∼1,083,000 m³ colluvial infill and till falls from the Whitman headscarp) as the source of the second high-amplitude seismic signal. The source of this secondary failure is more closely aligned with, but larger than, Iverson et al.’s (2015) original hypothesis. The second high-amplitude signal contains multiple high-frequency peaks, which may represent the multiple colluvium and till-fall events deposited below the headscarp of the Whitman slide component. Finally, in addition to constraining the timing of sliding, the seismic signals, combined with eyewitness accounts (Keaton et al., 2014; Iverson et al., 2015), provide strong evidence that rapid failure was integral to the Oso event.

The Oso event was well documented by remote seismic monitoring, pre- and postslide high-resolution aerial LiDAR surveys, and a series of high-resolution aerial orthophotos. As noted above, however, our field-based interpretations of the Oso landslide sequence differ in some key aspects from previously published accounts that relied extensively on these remote-monitoring products. Although these tools were important aids in understanding the landslide event, they were insufficient on their own to fully resolve critical features of the slide behavior. For example, examination of remote imagery alone did not allow full characterization of material distribution in the hummocks; instead, direct excavation of hummocks was needed. Furthermore, field examination was needed to reveal the structural nature of the C-E zone and its implications for sequence timing. The utility of such detailed mapping has been emphasized as a critical component for understanding landslide dynamics (e.g., Coe et al., 2016). Thus, remote imagery and seismic analysis without relevant ground-based observations may lead to spurious interpretations.

Enhanced mobility of landslides poses heightened risk to communities and infrastructure in their paths. The 2014 landslide near Oso, Washington, is a tragic reminder of this risk. Through extensive postevent field mapping (including small-scale features such as ephemeral sand boils and individual hummock lithologies) as well as material testing, we identified likely causes for the Oso landslide’s enhanced mobility. Namely, the landslide moved quickly, and it was sufficiently large to rapidly load and liquefy the saturated alluvial sediments on the valley bottom beyond the base of the slope. Our investigations revealed that slide motion transpired in a rapid sequence and that there was extensive basal liquefaction of the overridden valley alluvium. All of these factors needed to be present to promote large mobility. Had the alluvial sediments underlying the runout zone (including the Steelhead Haven neighborhood) not been contractive and liquefiable, or if the floodplain had not been saturated, the landslide mass may have been arrested after only a few hundred meters of motion (similar to past reactivations of the Hazel landslide), rather than the observed 1000+ m in 2014. Similarly, had only smaller, lower components of the landslide failed (e.g., the Hazel source area), the landslide may not have had sufficient momentum to rapidly push out over the valley bottom and liquefy the underlying alluvial sediments. The fact that mobile landslides have occurred elsewhere in this section of the North Fork Stillaguamish River valley, as well as other similar river valleys around the world, indicates that this combination of factors may promote enhanced mobility of other landslides.

Our observations of liquefaction features throughout the deposit, along with our analyses of liquefaction mechanisms, provide not only key evidence for a basal liquefaction process, but they also provide a starting point for future modeling studies. We identified several possible mechanisms that could have contributed to liquefaction of the alluvial sediments overridden by the Oso landslide. We found that two of these mechanisms, rapid undrained loading and shearing of loose saturated alluvium, likely acted in concert to reduce sediment strength and increase mobility. Additionally, cyclic shearing induced by strong ground vibrations may have contributed to the observed widespread basal liquefaction. We therefore suggest that future landslide investigations and numerical model development explicitly consider basal liquefaction and examine the possibility that multiple mechanisms may act in unison.

We thank the following individuals for assistance with site access during our mapping work: Dale Topham (Snohomish County), Jean Fike, Mark Arneson, and Travis Miranda (Washington Department of Natural Resources), Ken Osborne (Grandy Lake Forest Associates), Ron Burrows (private landowner), and Mark Hammer (Washington State Department of Transportation [WSDOT]). Geotechnical testing was performed under a cooperative agreement with Michael Riemer at the University of California–Berkeley, Department of Civil and Environmental Engineering. Tom Badger (formerly WSDOT) provided us with geotechnical boring logs from the landslide, and we appreciate his assistance in the early stages of this project. We value discussions with, and field observations from, Jeff Jones (Snohomish County) and William Schulz, Jonathan Godt, Richard Iverson, Kevin Schmidt, Jeff Coe, James Vallance, and Jonathan Perkins (all U.S. Geological Survey [USGS]) throughout the course of this project. We thank Dianne Brien and Skye Corbett (USGS) for assistance with geographic information system analysis, Paul Hsieh (USGS) for assistance with poroelastic modeling, and Kate Allstadt (USGS) for assistance with seismologic analyses, and we appreciate reviews on an earlier version of this paper by William Schulz, Kevin Schmidt, and Richard Iverson of the USGS, and two anonymous reviewers.