Realignments of the Colorado River by ∼2 m.y. of rotational bedrock landsliding: The Surprise Valley landslide complex, Grand Canyon, Arizona

The greater Colorado River drainage (southwestern United States) has many examples of cut-off bedrock meanders (e.g., Harden, 1990) and isolated bedrock channel segments (Howard et al., 2008), but there are relatively few examples of major river-diverting landslide dams (cf. Unaweep Canyon, western Colorado; Aslan et al., 2014). The purpose of this paper is to present a dynamic solution to a persistent geologic enigma in Grand Canyon geology, the origin and history of the Surprise Valley landslide complex, Arizona. This is done in the context of a summary of prior research and examination of the interaction of bedrock landslides, cut-off meanders, and canyon carving. In particular, we explore a landslide-river interaction that has been termed “epigenetic gorges” (Ouimet et al., 2007), which occurs when channels get laterally displaced due to river blockage and then the river incises a new bedrock gorge rather than re-incising the buried paleovalley. We present new geologic mapping, area-balanced cross-sections, and paleo–river channel heights to document the structural character and variable timing of a series of large landslides, and we document how those landslides altered the planimetric organization of the Colorado River and its tributaries through the study area. These data highlight the importance of landslide-caused river realignments in bedrock canyons as an important geologic process in canyon country.

The Grand Canyon region of the western Colorado Plateau has been and is being eroded by the Colorado River. This region is characterized by largely horizontal Paleozoic and Mesozoic sedimentary strata (Fig. 1). Most of the Mesozoic and younger strata have been stripped northward from the canyon rims to form the stepped topography of the Grand Staircase. The cliff retreat process involves landsliding where resistant sandstone cliffs are undermined at their basal contacts by erosion of weak shale units. The main weak units of the region are shown in Figure 1. The generally flat-lying Paleozoic strata in the Grand Canyon region are warped by broad northerly trending low-amplitude anticlines and synclines and east-dipping Laramide monoclines and are cut by west-down Tertiary faults that commonly are spaced tens of kilometers apart.

The Surprise Valley landslide complex includes the most prominent landslides in Grand Canyon which also appear to have had the greatest effects on the path of the river during canyon incision. They are located in central Grand Canyon in and adjacent to Surprise Valley between river miles 131 and 138(1) (river km 212–224). This area is between the Kaibab anticline to the east and Kanab-Cataract syncline to the west (Fig. 1). The nearest significant fault zones are the West Kaibab system (Crazy Jug and Big Springs faults), which consists of west-up Laramide monoclines with almost an equal amount of west-down Neogene faulting, and the Sinyala fault, a northeast-trending west-down Neogene normal fault (Huntoon et al., 1996).

The main weak strata of the region (shales and mudstones) that might be expected to host landslides are the Triassic Chinle Formation, Permian Hermit Formation, Cambrian Bright Angel Formation, and Neoproterozoic shales of the Chuar Group. Questions posed by this paper are: (1) What localized and triggered the inner-canyon Surprise Valley landslides? (2) When did they occur? and (3) How did bedrock landsliding interact with incision of Grand Canyon?

Bedrock landslides were first documented in Grand Canyon by Reiche (1937). This author defined the term “Toreva block” for landslides near Toreva, Arizona: “A Toreva-block is a landslide consisting essentially of a single large mass of unjostled material which, during descent, has undergone a backward rotation toward the parent cliff about a horizontal axis which roughly parallels it” (Reiche, 1937, p. 538). The author of this paper mentioned the back-rotated blocks in Surprise Valley as classic examples. Descriptions of the Surprise Valley landslides were improved by Ford et al. (1974), Huntoon (1975), Rogers and Pyles (1980), Savage (2002), and Savage et al. (2003). Elston et al. (1989, p. 28) noted that deposits of the Surprise Valley complex near river mile 136 involved cross-canyon runup landsliding based on the observation that a brecciated Rampart Cave Member of the Bright Angel Formation (McKee and Resser, 1945; Karlstrom et al., 2020) dolomite slab was thrust up onto its intact equivalent on the canyon's south wall. This feature is referred to here as Poncho's runup as shortened from “Poncho's Radical Runup” of Warme (2004, 2010). Figure 1 shows that in the eastern Grand Canyon region, landslides formed in Chuar Valley on the Galeros Formation of the Chuar Group (Ford and Breed, 1970) and at the base of the Vermillion Cliffs on slopes of Chinle Formation, but these did not interact with the river. In western Grand Canyon, other 1000-m-thick degraded landslides have been documented along the Hurricane fault corridor between RM206 and RM224 (Billingsley et al., 2006; Rogers, 1979).

As noted by Huntoon (2003), two distinct types of landslides occur in the Surprise Valley complex. The primary slides are those involving thick sections of Paleozoic bedrock strata that collapsed along listric detachments in shales of the Bright Angel Formation that soled to near river level at the time of landsliding. Failures caused back-rotated slide masses (Toreva blocks) to move into the free space of a paleocanyon. Where more free space existed, slides broke up into smaller back-rotated domino-style blocks. The bedrock slides, unlike simple rock falls, involved listric normal faults of the 700-m-thick Paleozoic strata dominated by carbonate rocks that rest on the Cambrian Bright Angel Formation. Break-away structures as high as the Esplanade Sandstone soled into the Cambrian Bright Angel Formation at or immediately above the floors of the canyons into which the slides collapsed. The second type consists of parasitic talus landslides that originated within and involved debris from the primary bedrock slides. Huntoon (1975) recognized that the Bright Angel Formation possesses insufficient structural strength to support the load of the overburden where the shale crops out at the base of sheer-walled canyons. He suggested that this type of landsliding represents a primary process in the widening of Grand Canyon and the development of Grand Canyon's characteristic cliff-slope profiles where competent cliff-forming strata get undermined by weaker slope-forming shales (Huntoon, 2003).

The river-subparallel east-west trend of Surprise Valley and the presence of Cogswell Butte, a bedrock island surrounded by landslides (Fig. 2), have long-been recognized as enigmatic features in Grand Canyon. Various geologists (see summary in Savage, 2002) have speculated on how landslides of the Surprise Valley complex altered the drainage patterns in this area. Three main hypotheses have been advanced to explain these features (Fig. 3).

Ford et al. (1974) and Savage (2002) assumed that the valley north of Cogswell Butte represented the coalescence of two opposing minor drainages that were eroding in a headward direction toward each other, respectively westward from Tapeats Creek canyon and eastward from Deer Creek canyon (Fig. 3A). In this hypothesis, before the slide occurred, Tapeats Creek canyon was in about its present configuration including the dogleg turn to the southwest, east of Cogswell Butte, before joining the Colorado River. All that the Surprise Valley landslide accomplished was to bury much of the topography northeast of Cogswell Butte and cause the lower reach of Tapeats Creek canyon to be displaced eastward to its current position. A primary drawback of this model is that it does not adequately account for the fact that Surprise Valley, on the north side of Cogswell Butte, has an uncharacteristic broad U-shaped cross-section and gentle longitudinal sub-landslide profile. In contrast, headward-eroding gullies in Grand Canyon have V-shaped cross-sections with very steep gradients.

Rogers and Pyles (1980) proposed that prior to the slide, the valley north of Cogswell Butte was the westward extension of Tapeats Creek canyon, which joined Deer Creek canyon at a right angle (Fig. 3B). In this hypothesis, the Surprise Valley landslide fell into and blocked paleo–Tapeats Creek, causing it to overtop a divide east of Cogswell Butte and subsequently incise the downstream southwest-flowing reach of modern Tapeats Creek canyon. A disadvantage of this hypothesis is that the west wall of Deer Creek canyon lies athwart the trend of the postulated westward reach of Tapeats Creek canyon. Consequently, Tapeats Creek canyon had to have joined Deer Creek canyon at a peculiar T intersection, which is uncharacteristic of tributary intersections within Grand Canyon and occurs only when the intersecting tributary is localized along a fault. However, there are no fault trends along the valley on the north side of Cogswell Butte.

The hypothesis of this paper (Fig. 3C) is that Surprise Valley formed as a Colorado River meander that looped north of Cogswell Butte. Catastrophic landsliding filled the valley and rerouted the Colorado River to its present position south of Cogswell Butte. This model is supported by restored cross-sections that suggest the landslide filled a large Colorado River paleocanyon. Furthermore, the area buried by slide debris east of Cogswell Butte and in the domino blocks in Bonita Creek canyon is more than twice as wide as Tapeats Creek canyon, so there are inherent volume problems with the other two hypotheses.

Fieldwork for this study included mapping of landslide stratigraphy and fluvial deposits, identifying and measuring strath terrace heights, and geochronologic analyses. Colorado River terraces are differentiated by the presence of far-traveled Colorado River gravels (e.g., Precambrian quartzite from the Needle Mountains of Colorado, San Juan volcanic field volcanics from southwestern Colorado, and Henry Mountain plagioclase-hornblende porphyry from Utah), which are easily distinguished from tributary terraces composed exclusively of locally derived Proterozoic and Paleozoic clasts. Strath terrace heights were measured by laser rangefinder, relative to nearby modern river water level or tributary water level, and from topographic maps. Heights are reported as either above modern river level (ARL), above the bedrock strath beneath the modern river, above modern tributary level, or as elevations above sea level (ASL).

Incision rate of Grand Canyon by the Colorado River has been reasonably steady along its profile, with variations attributed mainly to faulting and differential uplift (Crow et al., 2014). East of the Toroweap fault (Fig. 1), this rate has been semi-steady at ∼100–160 m/m.y. over the past several million years. The result is that the profile of the river has not changed appreciably while being lowered steadily during the period under consideration. Consequently, the modern profile of the river provides a datum from which we estimate the relative ages of features that owe their origin to past river levels. Specifically, the relative age of such features is directly proportional to the height of their bases above modern river level (Karlstrom et al., 2007, 2008; Crow et al., 2014). The features that are tied to height above river level are listed in Table 1 and include: (1) the bases of the landslide deposits that fill paleochannels, indicating a channel-damming landslide; (2) the elevation of far-traveled river gravels atop landslide material, indicative of dam-overtopping events; (3) the bases of lake sediments behind landslide dams along the Colorado River; and (4) dam heights (and lake-pool depths). To estimate lake extent upstream of dams, we take the elevation of river gravels atop the landslide dams and project that elevation upstream to its intersection with the paleo-profile of the river at the time the landslide occurred. The reliability of this age-height relationship assumes that the river profile readjusted to the regional gradient relatively rapidly after any landslide or lava damming events, for example, over time scales of several thousands of years as is suggested from lava-dam studies in Grand Canyon (Hamblin, 1994; Karlstrom et al., 2007; Crow et al., 2018, 2015) and by studies of landslide dams in major river systems such as Owyhee River of Oregon (Ely et al., 2012) and Tibetan rivers (Ouimet et al., 2007).

Bedrock incision (defined as canyon deepening by river incision) is measured from the base of a paleoriver gravel (terrace strath) to the modern bedrock beneath the river. The latter is estimated using maximum pool depth determined from river bathymetry, as informed by drill holes that penetrated sub-river gravels in a few places (Hanks and Webb, 2006; Karlstrom et al., 2007). The reach of the river in Figure 2 includes the Granite Narrows at RM 136, which is one of the deepest and narrowest reaches of the entire Colorado River. Here the mean pool depth is ∼23 m (Karlstrom et al., 2007; Crow et al., 2014). Bedrock incision rates in this reach of the Colorado River are best documented just upstream near Elves Chasm, where they have been estimated both by dates on individual straths relative to the assumed modern sub-river bedrock and by strath-to-strath age-height regressions that do not assume a bedrock depth but derive it. In central Grand Canyon, the strath-to-strath regression projects toward a bedrock depth of 13 ± 11 m, closer to mean rather than maximum pool depth (Karlstrom et al., 2007, their figure 8). Both methods yielded rates of ∼100 m/m.y. in the central Grand Canyon over the past 1–2 m.y. (Crow et al., 2014). New incision rates from detrital sanidine dating of the Piano slide, presented below, are 138 m/m.y., and this rate is used for estimating relative ages of other landslide elements.

Strath age-height relationships along the Colorado River also generally hold in tributaries, as suggested by a lack of hanging valleys along the Colorado River, but may not hold in small tributaries that cannot adjust as rapidly to the lowering base level of the Colorado River. In the area of Figure 2, Deer Creek and the ephemeral Bonita Creek tributaries form hanging valleys perched well above the Colorado River. The spring-fed perennial Deer Creek currently pours over a 55 m (180 ft) waterfall directly to the river's edge at Deer Creek Falls, a spectacular and rare tributary hanging-valley knickpoint along the river. In contrast, Tapeats Creek is graded to the Colorado River such that the heights of the bases of landslides in this tributary are used here to estimate their relative age.

The absolute ages of two of the landslides, Piano and Poncho's runup slides, were previously dated via cosmogenic burial age dating on amalgamated sands or quartzite cobbles done at Purdue University's PRIME Lab and reported by Crow et al. (2014). Cosmogenic burial age dating is based on the principle that the cosmogenic nuclides ¹⁰Be and ²⁶Al accumulate in quartz because of collisions with secondary cosmogenic particles near the Earth's surface and evolve toward a ¹⁰Be/²⁶Al ratio of 6.75/1 (Granger, 2006). If a well-dosed quartz sample is buried and effectively shielded from cosmogenic particles, the production of cosmogenic ¹⁰Be and ²⁶Al ceases, and these radioactive nuclides begin to decay at two different rates. By measuring the ¹⁰Be/²⁶Al of buried river sand and gravel, it is possible to calculate when the sample became effectively shielded, thus determining the age of the landslide that buried the sample. All cosmogenic ages are calculated assuming instantaneous burial, effective and continuous shielding by at least 10 m of cover, and recent exposure.

In this study, detrital sanidine was dated from paleoriver sands at the base of the Piano slide. Dating a high number (n = 186 in this case) of sanidine detrital grains gives an age distribution. The youngest detrital grains yield a maximum depositional age of 0.674 Ma, hence a minimum incision rate of >138 m/m.y., for the paleoriver deposits that were buried by this portion of the landslide. Analyses were done at the New Mexico Geochronology Research Laboratory (Socorro, New Mexico). Sanidine grains were irradiated in the Oregon State University reactor (Corvallis, Oregon) along with standard Fish Canyon Tuff sanidine (FC-2) with an assigned age of 28.201 Ma. The ⁴⁰K decay constant used for age calculation is 5.463 × 10⁻¹⁰/a. Argon gas was extracted by single-crystal laser fusion using a CO₂ laser, and isotopes were measured using a Thermo Scientific ARGUS VI Multicollector noble gas mass spectrometer. Complete analytical details are provided in Supplemental Table S1¹.

Area-balanced cross-sections were used to reconstruct pre-landslide paleotopography, and slide volumes were reconstructed by using multiple cross-sections and the stratigraphic thicknesses that Billingsley and Beus (1989) reported for RM 140 (Fig. 4). Cross-section restoration is made possible by identification and restoration of key stratigraphic marker beds within the landslides. The most useful marker beds include the rusty brown several-meters-thick pair of dolostone ledges encased within the Bright Angel Formation shales: the Rampart Cave Member and Sanup Plateau Member dolostones. Other distinctive units include a gray to white cliff-forming cherty limestone within the Watahomigi Formation of the Supai Group and basal conglomeratic beds of the Wescogame Formation and Esplanade Sandstone of the Supai Group (McKee, 1982). Portions of the Supai Group are essentially intact within landslide blocks, as shown by the correct sequence and approximately true stratigraphic thickness; these have been rotated and translated and somewhat thinned by layer-parallel slip within weak horizons. Similarly, where the Watahomigi Formation's white cliff-forming cherty limestone crops out near exposures of Redwall Limestone within landslide blocks, the Supai Group stratigraphy is conformable and considered to be within a generally intact Toreva slide block. In most of the landslides, stratigraphic thickness is reduced, drastically so for the weak Bright Angel shales, and most major landslide detachments sole into shales near the upper and lower contacts of the Sanup Plateau and Rampart Cave Member dolostones.

The preferred terminology used in this paper for landslide elements is that of structural geology, mainly because the listric and domino-style structures that we mapped within the landslide masses and the detachment zones at their bases have similar structural styles to those observed within the same rock strata in the nearby Lake Mead extensional domain (e.g., Karlstrom et al., 2010). This terminology helps us emphasize controls exerted by stratal heterogeneity of weak and strong layers, restored cross-section structural methodologies that reveal pre-existing topography involving paleocanyons rather than simple slopes, plus a long time frame of development of different landslide processes and elements. Conventional (engineering-based) geomorphology landslide terminology (Varnes, 1978; Cruden and Varnes, 1996; Hungr et al., 2014) that highlights type of material and movement type can also be applied to individual elements of the Surprise Valley landslide complex but, if applied overall, would lead to terms of limited usefulness such as “complex rock slide—rock flow” (Hungr et al., 2014) to refer to slides that involved a combination of rotational sliding, translation, spreading, and flow. To help bridge terminology differences, our terms (first) and approximately equivalent geomorphic terms (second) would be: “breakaway” for “crown” and “main” scarp; “detachment” for “surface of rupture”; “runup” for “zone of accumulation”; and “thrust” for “surface of separation” (cf. Cruden and Varnes, 1996, their figure 3.3, p. 406).

This section describes the main landslide elements and a proposed relative age sequence of landslides (keyed to Table 1) based on height in the landscape, maps and restored cross-sections of landslide elements, and available numeric dating.

The Surprise Valley landslide is a primary bedrock landslide, the highest and oldest landslide in the complex (Fig. 5; Supplemental Plate S1 [^{footnote 1}]) (Robertson, 2015). The breakaway was along the south-facing Cambrian through Permian cliffs that extend westward a distance of 2.4 km from a point ∼0.5 km east of Thunder Spring (Fig. 5). Cross-sections (Figs. 6 and 7; locations in Fig. 2) show that many of the landslide Toreva blocks have recognizable and restorable stratigraphy. There are two main parts of the landslide. Western parts of the slide, just north of Cogswell Butte, consist of several Toreva blocks with semi-intact stratigraphy (Fig. 6). The visually prominent 0.4-km-wide tier of back-rotated Toreva blocks seen from the Bill Hall Trail as one drops down to Surprise Valley from the north rim is displaced vertically ∼300 m (Fig. 6B). Disaggregated material lapped up against the flank of Cogswell Butte and also flowed west as far as Deer Creek canyon. A well-exposed slab of the 240-m-thick Permian Esplanade Sandstone measuring 1 km along strike by 200 m thick dips ∼50°NE toward the center of Surprise Valley where it is partially buried by alluvium. The highest part of the modern slab occurs on its western end at 1143 m ASL, an elevation that marks the modern divide between Surprise Valley and Deer Creek canyon and that has been little reduced by erosion since the slab was emplaced. Restoration of the western cross-section (Fig. 6B) shows that the tier of rocks that collapsed ranges in width from 0.8 to 1.3 km and involved an 800-m-thick section from the top of the Permian Esplanade Sandstone to the Rampart Cave Member. The volume of material in this slide mass was ∼2 km³.

In contrast, the parts of the landslide east of Cogswell Butte exposed in Bonita Creek (Fig. 7A) consist of a series of back-rotated domino blocks. The leading blocks moved on a mass of disaggregated rock pushed in front of the slide. The minimum vertical displacement at the Toreva-block breakaway headwall is 365 m (Fig. 7B).

In both cross-sections A-A′ and B-B′ (see Fig. 2 for locations), restoration shows a U-shaped paleo–Surprise Valley that exceeded 500 m in depth and 800 m in width. The detachment for the collapse of the Surprise Valley slide occurred along a listric normal fault that curved from near vertical against the breakaway headwall to near horizontal along the top of the Rampart Cave Member and at the free surface. Where material ran up the north side of Cogswell Butte, it did so along an upward-curving thrust fault that was a continuation of the listric fault (Fig. 6A). The lowest part of the listric fault beneath the Surprise Valley slide is well exposed near Thunder River Spring (Plate S1). At Thunder Spring, it curves and flattens to follow the Rampart Cave dolostone. At this location, however, there is a section where the Rampart Cave unit is missing, implying the presence of a prior paleocanyon likely occupied by paleo–Tapeats Creek. The detachment and the base of the slide can be traced along the west sides of modern Tapeats and Bonita Creek canyons (Plate S1). On the ridge between Tapeats and Bonita Creek canyons, ∼1 km north of the mouth of Tapeats Creek canyon, the detachment crops out at 866 m ASL and landslide material fills a paleocanyon (Fig. 8A). The lowest part of the paleocanyon at the west end of the landslide is 854 m ASL and is filled with brecciated carbonates (Fig. 8B). In summary, cross-section reconstructions show that when landslide blocks are restored to their original positions on Grand Canyon's North Rim, the major paleo–Surprise Valley and parts of a paleo–Tapeats Creek canyon are revealed, which had been carved to the level of the Bright Angel Formation at ∼860 m ASL.

Cogswell West is a secondary slide and is shown in Figure 9 in an oblique aerial view looking northeast at the west summit of Cogswell Butte. The saddle between the west and east summits of Cogswell Butte is occupied by a north-south–trending graben with down-dropped Watahomigi and Manakacha Formations against Redwall Limestone. These faults strike at high angle to the east-west part of the breakaway in western Surprise Valley but are parallel to tilted beds and hence may form part of the breakaway zone for the Cogswell West landslide (Plate S1 [^{footnote 1}]). The Cogswell West landslide lobe extends southwest from Cogswell Butte and is bounded to the west by Deer Creek canyon. This landslide contains mainly jumbled debris rather than map-scale Toreva blocks. A detachment beneath the Cogswell West landslide is present on top of the Rampart Cave Member upstream from Deer Creek Falls at a height of ∼226 m ARL. This contact can be traced semi-continuously between Deer and Bonita Creeks along the top of the Rampart Cave Member, but no analogous detachment was observed in the eastern wall of the Deer Creek canyon. The low point of the Cogswell West landslide toe fills paleotopography ∼110 m ARL between two intact Rampart Cave outcrops (Fig. 9). Rounded tributary alluvium was found perched within a pothole-like outcrop at 90 m above the Colorado River at river mile 135.8 (Fig. 9).

A physically distinct secondary landslide is found upstream on the south side of Cogswell Butte, at river mile 135. The Piano slide filled a paleo–Colorado River channel that had been incised into the Tapeats Sandstone at the head of the Granite Narrows (Fig. 10A). The slide consists of rotated blocks of Muav Limestone through Redwall Limestone. Colorado River gravels are found beneath the Piano slide toe at 70 m ARL at RM 135. A second sub-landslide river gravel crops out at 65 m ARL near RM 135.5R; at this location, sand and river gravel fill centimeter-scale clastic dikes that were injected into the basal landslide bedrock blocks, suggesting rapid emplacement of the landslide into the paleoriver (Fig. 10B). The paleochannel (Fig. 11A) was sampled for cosmogenic burial age dating; the sample consisted of several quartzite cobbles from a 2-m-thick basal part of the channel where well-rounded river cobbles are mixed with Colorado River sand and landslide-derived sand. The pebbles were crushed and analyzed together such that the age is an average of the Al/Be ratio of five to eight far-traveled quartzite clasts. The age was 880 ± 440 ka (2σ) (Crow et al., 2014). Variable inheritance of cosmogenic nuclides among the different clasts and imperfect shielding of this cliff outcrop, which is exposed on its east side, would both tend to make this age a minimum age.

Potassium-feldspar grains (microcline, orthoclase, sanidine) were separated from paleoriver deposits at the base of the Piano landslide, just above a bedrock strath 70 m ARL (Fig. 11A). From the K-feldspar population, sanidine was hand-picked based on microtextural characteristics (optical clarity, lack of perthite, twinning, and microporosity). Analysis of 186 grains yielded ages between ca. 670 ka and ca. 1325 Ma, with 165 grains being Phanerozoic and likely sanidine. The Precambrian grains are dominated by basement-derived orthoclase, which can also display minimal microtextures. The age data are shown on a probability distribution plot along with K/Ca values (derived from measured ³⁹Ar/³⁷Ar ratio) and radiogenic yield (Fig. 11B). Dates show an age distribution with a small mode at ca. 175–225 Ma (Fig. 11B), and grains younger than 40 Ma (Fig. 11C) are represented by a large number of grains between ca. 20 and 38 Ma. The majority of dates range between 25 and 35 Ma, which was a period of voluminous sanidine-producing eruptions in the San Juan volcanic field. Two grains are distinctly younger and give a weighted mean age of 0.674 ± 0.022 Ma (1σ) (Fig. 11C). This maximum depositional age is distinctly younger than the age of the ca. 765 ka Bishop Tuff eruption (Anderson et al., 2017) but nearly within 2σ of the ca. 631 ka Lava Creek B eruption (i.e., Matthews et al., 2015). Considering systematic errors associated with analytical details of very small and young crystals, we cannot unambiguously distinguish our 674 ka result from the age of the Lava Creek or some other similar-aged eruption perhaps from Yellowstone (cf. Troch et al., 2017).

The Poncho's runup slide, west of Deer Creek, is a primary bedrock slide, the second largest of the bedrock landslides in the Surprise Valley landslide complex. Elston et al. (1989) made the observation that deposits on the south canyon wall at RM 136.5 in what is now known to be the base of the Poncho's Runup consisted of blocks of the Rampart Cave Member that originated from the north side of the river. Warme (2004, 2010, 2012), Warme et al. (2012), and Watkins et al. (2007a) documented that Poncho's runup overran the Colorado River, buried the canyon floor, and was emplaced on the south canyon wall.

Figure 12 (and Plate S1 [^{footnote 1}]) shows several isolated erosional remnants with similar stratigraphy of a basal detachment, brecciated, attenuated, and welded-together Rampart Cave and Sanup Plateau dolostones, and brecciated remnants of Supai Group and Redwall Limestone overlying higher detachments. They are thrust above in-place Rampart Cave and Sanup Plateau Member dolostones (Fig. 12B). Most of the stratigraphic thickness of Bright Angel Formation and Supai Group was thinned and removed during sliding. One intact block of the Esplanade Sandstone rode on top of the moving debris and was deposited ∼100 m above the river on the south flank of the canyon. The direction of the movement as indicated by slickenlines and back-rotated blocks was to the southwest, parallel to Deer Creek canyon. Decimeter- to meter-scale blocks of rounded Precambrian rock are found near the highest parts of the runup (black stars in Fig. 12A), indicating that the landslide acquired rounded granite river boulders in the river and pushed them to the top of the runup; these lithologies are similar to granites exposed at Deer Creek Falls today and in a granite monadnock ∼100 m upstream.

On the north side of the river, the absence of in-place Rampart Cave Member west of Deer Creek canyon suggests that the bedrock slide came from the north side of the river west of Deer Creek canyon. A prominent breakaway headwall extends from the west wall of Deer Creek canyon for several kilometers downstream, and a major north-dipping Toreva block forms a valley just below the headwall. One topographically closed 0.5-km-long half-graben playa occupies the surface of the debris field between the highest and a second Toreva block. It is slowly filling with locally derived silt. This nominal degree of erosion contrasts sharply with the Surprise Valley slide where the entire surface of the slide mass and associated lake deposits are well drained and deeply eroded at their distal limits.

The detachment zone at the base of Poncho's runup (Fig. 13A) consists of brecciated Rampart Cave dolostone with convoluted shale around the lowest breccia blocks. A 5-m-thick section of river sand and gravel is exposed just beneath the slide at a level 61 m ARL (star in Fig. 13B). The age of the Poncho's runup is constrained by a cosmogenic burial date obtained from sand in this paleochannel that is capped by landslide debris, found in a cave perched 61 m ARL above the river near the top of the Tapeats cliff near river mile 138 (at star in Fig. 13B). At the sampled location within the cave, there are slickensides on the Tapeats Sandstone and on sheared river cobbles that were caught between the sandstone and the runup debris. The dated sample consisted of quartz grains collected from the ceiling of an overhang that was overlain by >10 m of bedrock landslide material, providing sufficient post-burial shielding. Thus, the sample represents an average age derived from the nuclide inventory of thousands of quartz grains. Uncertainties reflect measurement error only; post-burial production by muons would make the burial age older than reported. The burial age is 0.98 ± 0.42 Ma (2σ) (Crow et al., 2014).

A restored cross-section (Fig. 13D) shows that the bedrock slide was a >1.6-km-long section of the north wall measuring 700 m high by 500–600 m thick and involved 0.8 km³ of rock. It collapsed along a listric surface that soled into shales of the Bright Angel Formation below the Rampart Cave Member and at the top of the Tapeats Sandstone near the floor of the river canyon. The detached rock involved the stratigraphic section from the base of the Bright Angel Formation to the Redwall Limestone. Rock closest to the headwall fell a minimum of 460 m.

The western part of the Tapeats Creek canyon tributary contains secondary landslide debris and brecciated landslide blocks of the Rampart Cave Member that flowed in from the east end of the Surprise Valley landslide (Fig. 14). These deposits include two major outcrops of brecciated rusty-brown Rampart Cave Member carbonates that flowed over and are inset below the subhorizontal Surprise Valley landslide detachment that extends from Thunder Spring to near the mouth of Tapeats Creek. Rounded tributary gravels were buried beneath Rampart Cave breccia and fill Tapeats Creek paleochannels carved into the Shinumo Sandstone (Timmons et al., 2005) and upper Hakatai Shale at 83 and 63 m above the modern tributary, respectively. Additional tributary gravels are present at 78 m above Tapeats Creek in a paleochannel carved into the lower Hakatai Formation of the Mesoproterozoic Unkar Group.

Figure 15 shows the 133 Mile landslide (RM 131.5–133.5), a primary bedrock landslide involving the uppermost Bass Formation as well as Cambrian Tapeats Sandstone through Muav Limestone, which were derived from the south canyon wall and were translated on a detachment in the upper Mesoproterozoic Bass Formation. Rotated Toreva blocks of Tapeats Sandstone through Rampart Cave Member are steeply dipping (Fig. 15B). The detachment beneath the 133 Mile landslide strikes parallel to the modern river course and is exposed at 75–86 m ARL. No Colorado River gravels have been found along this basal detachment. The detachment contact between disturbed Tapeats Sandstone and intact Bass Formation is shown in Figure 15C. The detachment zone apparently followed weak ash-rich shales of the lower Bass Formation. This outcrop reveals classic domino-style deformation due to bedrock extension above the detachment. Colorado River gravels are found in a synclinal structure on top of the main slide mass at a height as high as 164 m ARL. Additional lags of river gravel mantle colluvial slopes both upstream and downstream of the main slide mass at a height of 136 m ARL (Fig. 15A).

Figure 16 shows younger secondary slides that affected Deer Creek's confluence with the Colorado River between RM 136.5 and RM 137.5R. Brecciated landslide debris is observed against a Tapeats Sandstone bedrock wall in upper Deer Creek in an area referred to by river runners as “the patio” (Fig. 16A), and the toe of the slide reaches down to the Colorado River just downstream of the Deer Creek Falls spring. The Patio slide detachment (Fig. 16B) is outlined by four minor head-scarp basins within the debris of the Poncho's slide (Plate S1 [^{footnote 1}]).

The youngest secondary slide, the Backeddy landslide, fills a paleochannel (Fig. 17A) located between RM 137.0R and RM 137.2R. The base of the channel is below the modern river level, and both the north and south walls of the paleochannel in the Tapeats Sandstone are exposed on the north side of the modern river. The material that filled this paleochannel is mostly brecciated debris from the Muav Limestone through the Temple Butte Formation that was reworked from the Poncho's slide.

Figure 18 summarizes each landslide element, our interpreted sequence of landslides, and the planimetric reconstructions of the changing course of the Colorado River and its tributaries through the development of the Surprise Valley landslide complex.

Of the three prior models for Surprise Valley shown in Figure 3, the presence of landslide blocks filling a broad paleovalley argues against the interpretation of Figure 3A. The gap in Rampart Cave Member just south of Thunder River is interpreted to represent paleo–Tapeats Creek canyon prior to the Surprise Valley slide. This interpretation is compatible with Surprise Valley as either a paleo–Tapeats Creek (Fig. 3B) or a paleo–Colorado River (Fig. 3C). However, the geometry and similarity in elevation of basal landslide deposits at the east and west ends of Surprise Valley support the model that Surprise Valley is an ancient path of the Colorado River (Fig. 3C) and that the paleo–Tapeats Creek intersected it near Thunder Spring.

The far-traveled eastern landslide dominos are explained as debris having flowed in an upstream direction within the paleocanyon of the Colorado River more than 2 km from the breakaway. The paleochannel was floored by Bright Angel Formation and was cut to just below the Rampart Cave Member (Fig. 8A), with its base at 866 m ASL. Landslide material likely proceeded still further south (upstream at the time) but has been removed in the area of the mouth of modern Tapeats Creek. Just north of Cogswell Butte, the landslide is different in character because it rode virtually intact above the disintegrating substrate to where it docked against Cogswell Butte. The large Esplanade Sandstone slab described previously is interpreted to have filled the Colorado River channel and formed an effective dam with a height of 1143 m.

Interpolating between the downstream and upstream paleochannel elevations of Figures 8A and 8B yields an average elevation of 860 m ASL for the ancient Colorado River channel and base of the Surprise Valley landslide dam. Therefore, the height of the dam can be computed from the paleochannel bottom (860 m) to the top of the Surprise Valley bedrock divide (1143 m), hence it was 283 m tall. Based on an initial dam elevation of 1143 m, the Surprise Valley landslide dam may have backed up water more than 135 river miles to Lees Ferry, at the base of the Vermillion Cliffs (Fig. 1) (950 m ASL). A single clast of far-traveled Colorado River gravel was found along the southeast wall of Deer Creek canyon; this is consistent with the Surprise Valley paleochannel having been cut by the Colorado River. We cannot rule out that it could have been transported there by humans, thus additional evidence for Colorado River gravels or channel sands within the basal parts of the landslide is needed.

A lake within and behind the landslide dam is inferred from a ∼40-m-thick sequence of fine-grained red siltstones that contain palynomorphs, including diverse pollens and spores, freshwater algae, and fungal hyphae (Watkins et al., 2007b). This lake is interpreted to have spilled over an existing divide on the south side of Cogswell Butte that was similar in character to those between other buttes and the plateaus from which they are separated in Grand Canyon. In this case, the divide within the 4-km-wide neck of the meander loop separated opposing, steep, headward-eroding gullies that respectively drained eastward and westward to the Colorado River, herein referred to as East and West Cogswell gulches. Implicit is that the elevation of the divide was lower than the top of the landslide dam (<1143 m ASL). Following lake spillover, the rerouted Colorado River would have eroded the divide, carved a new bedrock channel, and simultaneously excavated the sediments that had accumulated upstream from the dam.

Once the river was established in approximately its current path, we infer that it cut down to its current depth at the Granite Narrows south of Cogswell Butte at a semi-steady rate of 138 m/m.y., the rate determined from the detrital sanidine maximum depositional age of the Piano slide. Under these assumptions, the age of the Surprise Valley landslide is ca. 2.1 Ma based on the average height (260 m) of the landslide base above the Colorado River (Fig. 8), which translates to 283 m of bedrock incision and a bedrock incision rate of 138 m/m.y.

Cogswell West landslide is interpreted to be a separate landslide element of the complex because of its geometry (Fig. 19). Alternatively, it has been proposed that the same basal detachment, at the top of the Rampart Cave Member, underlies all of Surprise Valley (Watkins et al., 2007a) and Cogswell Butte, and that Cogswell Butte itself has moved. If so, the detachment is concealed by colluvium along the eastern margin of Deer Creek. However, the north-south graben structure on top of Cogswell Butte suggests east-west extension, which is opposed to the north-south extensional breakaway of Surprise Valley. The Cogswell West slide detached from the oversteepened east wall of lower Deer Creek canyon and a large back-rotated block of Bright Angel and younger Paleozoic debris plugged the lowest Deer Creek canyon, which had its confluence with the river ∼0.75 km upstream of modern-day Deer Creek Falls. The landslide filled paleo–Deer Creek, which was positioned 110 m ARL (133 m above the river's bedrock strath) or 90 m ARL (113 m above bedrock) if the side-stream gravels are part of the same paleo–Deer Creek that carved the saddle. The estimated magnitude of bedrock incision suggests an age of ca. 0.96–0.82 Ma for this slide, which displaced Deer Creek ∼0.75 km to the west.

By ca. 2 Ma, the Colorado River had been established along the Granite Narrows section with a channel carved into Paleozoic strata. Between 2 Ma and the time of the Piano slide at ≤0.674 ± 0.022 Ma, the Colorado River is interpreted to have incised from ∼600 m ARL to the ∼70 m paleochannel height preserved beneath the Piano slide at an average incision rate of 400 m/m.y. This is consistent with observations that the rate of bedrock incision immediately following formation of an epigenetic gorge can temporarily be very high (>1 cm/yr) before returning to long-term average incision rates (Ouimet et al., 2007). The Piano slide was derived from strata near the top cliffs on the south side of Cogswell Butte (Fig. 18).

The new detrital sanidine age of 0.674 Ma is interpreted to be from ash fall from either the Yellowstone region or an unknown eruption. This age provides the maximum depositional age for this paleoriver deposit and a minimum incision rate of 138 m/m.y. This much higher-precision age is within error of but now supersedes the lower-precision cosmogenic burial date of 0.88 ± 0.44 Ma (Crow et al., 2014). The Piano slide rerouted the river south of its prior course and explains the Granite Narrows as a young (<0.674 Ma), narrow, and deep bedrock reach of the Colorado River.

At the time of the Poncho's event, the Colorado River had incised to within 84 m of its present bedrock depth (Table 1). The river was just beginning to incise into the top of the Tapeats Sandstone at this locality and into Precambrian granitic basement ∼1 km upstream. The front of what may have been a single bedrock slide bulldozed bedload from the channel of the Colorado River as it rafted up the south flank of the canyon. The excavated material included river cobbles and subrounded river-fluted Precambrian boulders having dimensions as large as 1 m³. Most of the bedload material became overrun and mixed into the basal slide mass, but some of it was pushed intact in front of the landslide debris to the highest reaches of the runup. The leading edge of the runup material reached the base of the Peach Springs Canyon Member of the Muav Limestone ∼200 m above its point of origin on the floor of canyon.

The movement of the Poncho slide debris into the Colorado River and lower Deer Creek was constrained by the respective opposing canyon walls. The appreciable runup that has been documented on the south wall of the Colorado River segment may have been accompanied by lesser runup on the east wall of lower Deer Creek canyon. Following the slide, the topographic low across the slide debris in both the Colorado River and Deer Creek coincided with the trend of the underlying buried channels. Consequently, when both streams reincised, they reexcavated their former channels before incising deeper into and through the Tapeats Sandstone. Significantly, the north wall of the new Colorado River and west wall of the new Deer Creek channels positioned above the top of the Tapeats Sandstone comprised oversteepened relatively unstable Poncho's slide debris that rose to a height of 300 m above the sandstone.

Far-traveled river gravels occur as lags and remnants of a 10-m-thick fill terrace atop the Poncho's slide; these indicate the Colorado River became established atop the landslide after the landslide lake had filled with sediments. The aggradation of far-traveled river gravels atop the slide indicates that the river had overtopped and perhaps begun to erode through the dam; hence, we take the height of the gravels as an approximate spillway height of 719 m ASL (133 m ARL). The height of this dam would have backed water up to Elves Chasm (RM 117) and possibly triggered upstream landslides such as the 133 Mile landslide, the only landslide to have originated on the south side of the Colorado River.

The cosmogenic burial age of sands just above the strath from a cave perched on the rim of the Tapeats cliff (Fig. 13B) is 0.98 ± 0.42 Ma, within error of the cosmogenic age of 0.88 ± 0.44 Ma age of the Piano slide. Crow et al. (2014) merged the two ages and got 932 ± 304 Ma (2σ) for the combined samples. These ages have large uncertainties, and we use the detrital sanidine age of <0.674 Ma from the Piano slide to refine both age estimates. Although precise age control is lacking, the relative heights of the straths above the river's modern bedrock strath for the Piano slide (93 m) and Poncho's Runup (84 m) suggests that the Piano slide is older. An alternative interpretation, given their similarity in heights and known variations in depth to bedrock below the river, is that the downstream Poncho's slide and resulting lake may have destabilized upstream areas and triggered the Piano slide. Also, as discussed further below, these slides formed during a time of major lava-dam activity to the west (RM 177) that could have backed up water into this area (Hamblin, 1994; Crow et al., 2015).

Figure 18 shows that landslides pushed Tapeats Creek to the east several hundred meters. Tapeats Creek presumably reestablished its course along a topographic low across the landslide debris, but now drained to the southwest. Incision of the creek along this trend accounts for the more pronounced dogleg bend that now characterizes the lower reach of Tapeats Creek canyon below the Thunder River confluence. Using the 83–63 m heights of side-stream terrace straths above the modern tributary and an average incision rate of 138 m/m.y., much of this realignment had taken place by 0.61–0.46 Ma. Tapeats Creek slides may also have been triggered or affected by the Poncho's and/or Piano slide lake(s) backing up into Tapeats Creek.

We interpret the detachment at 75 m above the river to represent a Colorado River paleochannel, and the gravels on top of the 133 Mile landslide to represent post–133 Mile river levels established on top of a landslide dam (Fig. 18). This landslide is on the south side of the river, and its height above the river would suggest an age older than Poncho's slide. However, we envision that a lake, backed up from downstream, may have destabilized the south-dipping and weak ash layers in the lower Bass Formation and Bright Angel Formation, causing bedrock landsliding. If the Poncho's slide dam were the cause, its spillway elevation of 720 m ASL would have backed water to about RM 117 and a lake would have inundated the 133 Mile detachment to depths of 30 m, well into the Bright Angel Formation. In turn, gravels atop the 133 Mile landslide indicate that a lake behind it filled with fluvial sediments and overtopped the landslide dam such that a spillway of the dam, as represented by the highest gravels, was initially at 164 m ARL. River gravels at 135 m ARL may record the river cutting down to reestablish its previous grade in a new bedrock channel located north of its paleochannel.

The Patio slide involved collapse of oversteepened older landslide debris piles of the Poncho's landslide. Material flowed both toward the river and into the west part of Deer Creek. This event occurred after the Deer Creek channel had incised to within 30 m of modern river level as suggested by the elevation of springs formed at a bedrock strath located immediately west of Deer Creek Falls (Fig. 19). These springs are interpreted as groundwater derived from Deer Creek that flows through debris now filling the buried paleochannel. The Patio slide resulted from a single parasitic slump that emanated from the eastern parts of the Poncho slide mass. The landslide blocked Deer Creek and displaced it to a new location ∼150 m to the east. The modern stream currently occupies this new position, where it has eroded a narrow slot canyon through Tapeats Sandstone (Deer Creek narrows) before it pours over a basement knickpoint at Deer Creek Falls. Assuming the spring area was near the paleo–Colorado River–Deer Creek confluence, its 32 m height ARL and 55 m height above bedrock beneath the river implies that this landslide took place ca. 400 ka. This was also a time of major downstream lava-dam activity that could have backed up water into this area (Crow et al., 2015). This hypothesis for a young, ca. 400 ka hanging valley occupied by Deer Creek Falls and its narrows may explain why these falls are unique among Grand Canyon side canyons. In contrast, Tapeats Creek is graded to the Colorado River without a similar waterfall for a combination of reasons: its pathway was pre-established (going the other direction) by the paleo–Colorado River; it has higher spring discharge and larger drainage basin area and can keep up with mainstem incision; it likely underwent rapid incision during lake drawdown of the Piano and Poncho's landslide lakes; and its basement rock substrate was likely more easily eroded than the granite monadnock encountered by the rerouted Deer Creek.

The inset nature and below-river-level position of the strath suggests the Backeddy slide resulted from recent readjustments to oversteepening of unstable Poncho's landslide debris. This parasitic slide (Fig. 18) blocked a Colorado River paleochannel, which at the time of the slide had incised to near the level of the modern river. The slide debris forced the Colorado River into a new mile-long alignment and its modern bedrock channel, which lies 0.4 km south of the buried channel. Assuming the base of this landslide is near river level, that bedrock is ∼23 m below river level, and that bedrock incision averaged 138 m/m.y., this landslide is estimated to be ca. 170 ka. This is older than the optically stimulated luminescence age of 21 ka from silty lake sediments dated by Webb et al. (2005) from the mouth of Owl Eyes Canyon (RM 133) about 6 km upstream (Fig. 17B), suggesting they may be unrelated.

Past studies have discussed landslide triggering mechanisms and landslide localization in Grand Canyon as due to the presence of shales of the Cambrian Bright Angel Formation. However, this unit is exposed along the Colorado River in both eastern and western Grand Canyon without producing widespread landslide complexes (Fig. 1). Hereford and Huntoon (1990) proposed that eastward coarsening of the Bright Angel Formation inhibited this type of mass wasting along the Marble Canyon section of eastern Grand Canyon. Western Grand Canyon has many locations that expose shales of the Bright Angel Formation near river level, including along the Toroweap and Hurricane faults which have Quaternary slip, suggesting that the presence of Bright Angel Formation is just one factor affecting landslide susceptibility. This observation poses the question about why the Surprise Valley area hosts Grand Canyon's most prominent landslides.

Landslide triggering by seismicity is well known globally (e.g., Hadley, 1994; Ouimet, 2010) but is not a ready explanation for the Surprise Valley complex. The nearest fault, the Sinyala fault, crosses the Colorado River near RM 138; its offset at river level is ∼12 m and decreases to 3–4 m where exposed on the Esplanade rim (Billingsley and Hampton, 2000), higher in the canyon's walls. This and similar faults have remained active into the Quaternary, as evidenced by microseismicity (Brumbaugh, 2005; Karlstrom and Timmons, 2012), and Fishtail Canyon landslides just downstream (RM 139) may be related to this fault (Watkins and Rogers, 2004). But the Surprise Valley complex landslide volumes decrease approaching the Sinyala fault, and hence seismicity is not considered the primary triggering or localization mechanism for Grand Canyon's largest bedrock landslides.

Groundwater is a possible mechanism for cliff destabilization given that increased pore pressure reduces effective normal stress and allows detachment and slip (Rogers and Pyles, 1980). The West Kaibab fault zone and an extensive cave network on the west side of the Kaibab uplift direct high-volume groundwater flow from snowmelt at higher elevations into the study area, currently discharging at Tapeats, Thunder, Dutton, and Deer Springs out of the Rampart Cave Member of the Bright Angel Formation. The shear strength of shales can decrease over time in the presence of water, and alternating shale-dolostone lithologies above and below the Rampart Cave Member may have allowed groundwater flow to amplify the strength contrast between lithologies as demonstrated by detachments just above (Surprise Valley) and just below (Poncho's) the Rampart Cave Member. Thus, our favored localization model for the initial Surprise Valley landslide involves weakening of the Bright Angel Formation shales by north-rim groundwater prior to tributary incision through the Muav–Bright Angel contact, where abundant groundwater in the Redwall-Muav aquifer was perched, as it is today.

For the subsequent landslides, we call on instability due to oversteepening of the earlier landslide debris caused by incision of the Colorado River into and through the slide and the underlying Cambrian Bright Angel Formation. A combination of hydrologic influences on favorable structural and/or oversteepening configurations is most likely. The dip of the Paleozoic strata through the Surprise Valley area averages ∼2° or 32 m/km toward the west. In contrast, the gradient of the west-flowing Colorado River is only 1.3 m/km. This disparity may have influenced the location of the primary bedrock landslides (Surprise Valley and Poncho's) along the Colorado River because they developed shortly after the river incised into the Bright Angel Formation. These bedrock slides lie just to the east of where the Bright Angel shales dip below the Muav Formation at river level immediately downstream of the Surprise Valley landslide complex (Huntoon and Elston, 1980), and the progression of younger slides to the west is consistent with ongoing river incision across this contact.

Surface water could have a similar weakening effect, involving either the river itself or, as Hamblin (1994) postulated, lakes created by downstream lava dams (RM 178–188), or landslide dams. Hamblin suggested that landslide activity in the Surprise Valley area of central Grand Canyon was facilitated by lava-dammed lakes that might have destabilized upstream canyon walls. A summary of timing data (Fig. 20; Table 1) shows that the lava-dam activity is younger than the ca. 2.1 Ma Surprise Valley slide, but that several major lava dams formed in the time frame of the <0.674 Ma Piano and ca. 0.61 Ma Poncho's landslides, and we interpret rapid drawdown by dam failure to be the likely trigger for the Poncho's, Piano, Tapeats, and 133 Mile landslides. Numerous downstream lava dams and cascades in the 400–100 ka time frame also provide a plausible mechanism for the ca. 400 ka Patio and ca. 200 ka Backeddy slides, although dating is not precise enough to link discrete lava-dam and landsliding events. Upstream lake deposits were proposed by Hamblin (1994) in eastern Grand Canyon. For many of these, their identity as lake deposits was questioned (Kaufman et al., 2002) such that verification and dating of upstream lake deposits remains an important test for landslide- and lava-dammed lake models.

A contribution of this paper is the documentation of far-traveled river gravels on top of several of the landslides. Figure 21 shows how we interpret these in terms of a sequence of landslide-dammed lakes. Poncho's landslide dam has mainstem river gravels atop it that extend to a height of 133 m ARL. We interpret this to record a time when the dam was overtopped, with the river established on top of it. Hence this is considered as an approximate spillway elevation of a dam that would have backed water to RM 117, inundating the 133 Mile area by 30 m of water and potentially triggering sliding. Gravels atop the 133 Mile slide extend to 164 m ARL (766 m ASL) which, by similar reasoning, may record a dam spillway elevation that would have backed water to RM 100.5.

From the downstream end, there are several lava dams that might have backed water up and destabilized the Surprise Valley area. Figure 20 shows the dated basalt flows, many of which formed lava dams, with a peak in ⁴⁰Ar-³⁹Ar age distribution ca. 611 ka. As listed in Table 1, older flows include the High Remnant flows near Lava Falls (mean age of 617 ka), lower Black Ledge flows (605–615 ka), and the Buried Canyon flows (mean age of 524 ka, but individual flows as old as 609 ka; Crow et al., 2015). Lakes behind these lava dams may have triggered the Poncho's or Piano slides which, in turn, may have triggered the Tapeats and 133 Mile landslides. Dating is not precise enough to link individual lava flows to landsliding events, but the similarity in age of the Piano landslide (≤674 ka) to that of some of the early and most voluminous basalt lava dams (650–600 ka), which likely created lava dams extending to heights >200 m ARL, seems unlikely to be a coincidence. Some of the downstream lava dams, for example, the 524 ka Buried Canyon lava dam, also have gravels on top, indicating that they backed water well upstream of the Surprise Valley area, to RM 80 (Fig. 21). Younger lava flows that may have backed up lava dam lakes that could have triggered the Patio slide include the ca. 448 ka Toroweap flows near Lava Falls. Backeddy slide may have been triggered by lakes formed behind the 243 ka upper Whitmore flows or the 209 ka lower Gray Ledge flow (Crow et al., 2015, their table 1).

Longevity of landslide dams and lava dams is highly variable but very short compared to the accuracy of available geochronology. Historical discharge records for the Colorado River and the history of filling of Lake Mead (Nevada and Arizona) and Lake Powell (Utah and Arizona) suggest that dams would have been overtopped in years to decades depending on how leaky they were (Crow et al., 2015). The presence of gravels atop the landslide (and lava) dams indicates that the lakes filled with sediment, which allowed far-traveled gravels to be transported across the dam. This would have likely occurred in <100–1000 yr based on projections for filling of modern reservoirs with sediment, leading to overtopping of current dams. Dam failure and erosional removal begins quickly once the dam is overtopped, and the return to steady incision rates after lava-dam emplacement was estimated by Crow et al. (2015) to have been <10,000 yr. Landslide dams may have been somewhat longer lived such that the multiple landsliding events in the Surprise Valley complex might have reduced incision rates in this reach by tens of meters per million years. But the long-term average of 138 m/m.y. here compared to steady rates of ∼100 m/m.y. at Elves Chasm (RM 116) and 160 m/m.y. in eastern Grand Canyon (Crow et al., 2014) supports the general concept that semi-steady bedrock incision took place in spite of short-term damming events.

A primary geomorphic interpretation is that Cogswell Butte, which originated south of the Colorado River, ended up north of the river. This happened when the principal and oldest landslide in the complex, the ca. 2.1 Ma Surprise Valley landslide, dammed the main stem of the Colorado River, thereby causing the meander loop surrounding Cogswell Butte to be cut off after the resulting lake overtopped a low divide south of the butte. In this hypothesis, the Colorado River occupied a large meander loop, similar in size and geometry to the modern Elves Chasm meander loop (RM 116). The river wrapped around the north side of Cogswell Butte and separated the butte from the North Rim of the Grand Canyon, similar to numerous modern-day buttes. East and West Cogswell gulches extended to a saddle at the time the Surprise Valley slide occurred. The collapse of the north wall of the meander loop and flow of landslide domino blocks a significant distance upstream into the river's paleovalley effectively dammed the meander. This caused the resulting lake to overtop a saddle between the East and West Cogswell gulches south of Cogswell Butte, leaving the meander loop high and dry. Tapeats Creek became aligned with the former north-flowing reach of the Colorado River on the east side of Cogswell Butte and began incision of what is now the lower reach of Tapeats Creek canyon.

The proposed sequence of subsequent landslides is based on heights of the base of the subsequent slides above the modern river (Table 1), and an assumed steady incision rate of 138 m/m.y. The Cogswell West slide (0.96 Ma) was a secondary debris landslide from the eroding Cogswell Butte that dammed an ancient Deer Creek course, diverting it ∼500 m west. Similarly, the Piano slide (ca. 0.67 Ma) was a secondary debris landslide that dammed and diverted the river and rerouted the river to its present position at the Granite Narrows. The second major bedrock landslide, Poncho's slide (0.61 Ma), may have also been driven by groundwater saturation that was a precursor to modern day Deer and Dutton Springs; this slide drove a northerly derived bedrock runup across the Colorado River and up the south canyon wall to 823 m ARL and dammed the river to 719 m elevation ASL. The upstream 133 Mile slide (ca. 0.6 Ma), the only slide originating on the south side of the canyon, was possibly triggered by landslide lakes formed behind Poncho's slide which dammed and then pushed the river to the north. This 133 Mile slide has landslide-mantling gravels that suggest a 766 m ASL spillway elevation (Fig. 21). The Tapeats slides (ca. 0.53 Ma) fill tributary terraces at 63 and 83 m above Tapeats Creek and diverted the tributary to the east. The Patio slide (ca. 0.40 Ma) was a smaller reactivation of Poncho's slide that dammed and diverted both the Colorado River and Deer Creek and instigated the Deer Creek Falls knickpoint. The Backeddy slide (0.17 Ma) is another small reactivation of the Poncho's slide mass that dammed and diverted the river to the south and buried a well-preserved paleochannel about river mile 138. These subsequent slides overlapped in timing with downstream lava dams and were possibly triggered by inundation by water from lakes behind lava dams.

Overall, the unique character of the Surprise Valley landslide complex reflects multiple triggering mechanisms during >2 m.y. in which river incision interacted with landslides. The main instability formed when the Colorado River incised through the weak Bright Angel Formation in a meander loop where this shale aquitard at the base of the Rampart Cave Member karst aquifer horizon was already saturated with groundwater. This triggered failure of Toreva blocks of the Surprise Valley landslide, which filled the meander bend of the river and rerouted the river to the south of Cogswell Butte to incise the Granite Narrows. Subsequent landslide instability in this new reach of the river was triggered by combinations of oversteepened slopes (Cogswell West slide), continued saturation by springs and groundwater (Poncho's runup), and surface-water saturation due to both downstream lava dams (Piano slide) and landslide lakes (133 Mile slide). The rerouting of Deer Creek by the youngest landslides provides an explanation for Grand Canyon's unique hanging-valley tributary at Deer Creek Falls.

This study of the Surprise Valley landslide complex of Grand Canyon has various broader implications. Its unique nature within Grand Canyon itself is explained in this paper by the triggering potential of the voluminous North Rim–derived springs that became focused near the top of the upper Bright Angel Formation aquitard as the Colorado River incised through the overlying Redwall-Muav aquifer. The triggering of Toreva-block bedrock landslides was favored by the layered nature of the strata with strong strength and hydrologic-property contrasts between the Rampart Cave Member dolostone and the shales of the Bright Angel Formation. The geometry of resulting detachments resembles the ramp-and-flat geometry of thrust belts or extensional detachments, and displacement was catalyzed by elevated pore-fluid pressure which reduced the effective normal stress across weak detachment horizons. The lesson from the 2 m.y. long-lived nature of landsliding in this region is that major bedrock landslides have legacies in terms of subsequent slides such that once started, long-lived landslide complexes are expected.

Comparisons up and down the length of the greater Colorado River system of features formed over the past 10 m.y. are also instructive. Surprise Valley is arguably the second-most dramatic known example where bedrock landslides displaced the Colorado River. Another singular event is that of Unaweep Canyon in Colorado (Aslan et al., 2014). In that case, a landslide caused a major piracy event that was poised to take place because of regional bedrock heterogeneity. It resulted in an entirely new course for the Colorado River being established in soft rocks around the northwest margin of the Uncompahgre uplift. Abandonment of the kilometer-deep Unaweep Canyon basement gorge that had been carved directly across the uplift occurred when a landslide dammed the canyon and lake spillover found a more favorable path through softer rocks.

A comparison to epigenetic gorges and landslides of major rivers of the eastern Tibetan Plateau (Ouimet et al., 2007, 2008) offers insights into feedbacks between landslides and river incision and how large landslides influence river profiles. Our results suggest that large landslides in Grand Canyon were probably infrequent enough that they did not provide significant reduction of long-term river incision rates. Fast rates of 400 m/m.y. between 2 Ma and 674 ka are suggested here, compatible with the concept that carving of a new bedrock channel was facilitated by development of a steep channel gradient and narrow channel width, both generating the increased stream power necessary to erode the bedrock rapidly (Finnegan et al., 2005; Ouimet et al., 2008), which resulted in formation of the Granite Narrows. Average incision rates at Surprise Valley of 138 m/m.y. over the past 674 k.y. are ∼15% slower than the 160 m/m.y. steady rates documented upstream and downstream in eastern and western Grand Canyon over the same interval (Crow et al., 2014). The Surprise Valley landslide complex is a probable explanation for slower but still steady incision rates of 100 m/m.y. over the past 600 k.y. near Elves Chasm at river mile 117, about 48 km upstream (Crow et al., 2014). Thus, the landslide dams had more of an effect on long-term incision rates than lava dams, which were quickly removed and hence did not dampen long-term incision rates. In general, the Colorado River had sufficient stream power to remove both lava and landslide dams efficiently between damming events. Once Grand Canyon became deeply incised within laterally continuous layered strata, and in the absence of repeated fault triggering as is seen in Tibetan examples, it required an unusual Toreva-block failure event to cut off a meander and displace the channel several kilometers to carve a new bedrock gorge. Subsequent landslides have caused 100-m-scale rerouting of the Colorado River within its newer gorge several times in the past 2 m.y. that are similar in scale to the Tibetan epigenetic gorges described by Ouimet et al. (2008).

Research was supported in part by U.S. National Science Foundation grants from the Tectonics Program (EAR-1119629, EAR-1242028, and EAR-1348007) and Sedimentary Geology and Paleobiology Program (EAR-1545986). This paper builds on the 2015 M.S. thesis by Jesse Robertson and mapping done by the University of New Mexico collaborative team in 2012–2014. We dedicate this paper to Peter Huntoon for his collaboration in shaping many of the concepts in this paper, in particular the role of landslides as an ongoing mechanism in the widening Grand Canyon and that the relative ages of the slides correlate to the depth of the canyon when they formed. We thank Grand Canyon National Park for research and collecting permits. Ryan Crow provided insight and assistance throughout the project. Discussions and insights from John Warme improved the paper, especially with respect to Poncho's runup. We thank Alan Herring for use of oblique aerial photographs. Jordan Anderson helped with drafting of final figures. Cosmogenic burial dating was done by Darryl Granger. Reviews by Kyle House and Andres Aslan helped improve the paper.