Amphitheater-headed canyons formed by megaflooding at Malad Gorge, Idaho

Significance

The shapes of bedrock canyons offer clues to the history of surface water on Earth and Mars. Using field examples in Idaho, we found that canyons with amphitheater-shaped heads were likely carved rapidly by outburst flooding about 46,000 y ago and that canyons with more pointed heads evolved progressively by river erosion over tens of thousands of years. Our study suggests that the many amphitheater-headed canyons in fractured basalt on Mars, long inferred to be carved by groundwater seepage, may owe their origins instead to megafloods.

Abstract

Many bedrock canyons on Earth and Mars were eroded by upstream propagating headwalls, and a prominent goal in geomorphology and planetary science is to determine formation processes from canyon morphology. A diagnostic link between process and form remains highly controversial, however, and field investigations that isolate controls on canyon morphology are needed. Here we investigate the origin of Malad Gorge, Idaho, a canyon system cut into basalt with three remarkably distinct heads: two with amphitheater headwalls and the third housing the active Wood River and ending in a 7% grade knickzone. Scoured rims of the headwalls, relict plunge pools, sediment-transport constraints, and cosmogenic (³He) exposure ages indicate formation of the amphitheater-headed canyons by large-scale flooding ∼46 ka, coeval with formation of Box Canyon 18 km to the south as well as the eruption of McKinney Butte Basalt, suggesting widespread canyon formation following lava-flow diversion of the paleo-Wood River. Exposure ages within the knickzone-headed canyon indicate progressive upstream younging of strath terraces and a knickzone propagation rate of 2.5 cm/y over at least the past 33 ka. Results point to a potential diagnostic link between vertical amphitheater headwalls in basalt and rapid erosion during megaflooding due to the onset of block toppling, rather than previous interpretations of seepage erosion, with implications for quantifying the early hydrosphere of Mars.

Landscapes adjust to perturbations in tectonics and base level through upstream propagation of steepened river reaches, or knickzones, thereby communicating environmental signals throughout a drainage basin (e.g., ref. 1). Nowhere are knickzones more important and apparent than in landscapes where canyon heads actively cut into plateaus, such as tributaries of the Grand Canyon, United States, and the basaltic plains of Mars (e.g., refs. 2–4). Here the stark topographic contrast between low-relief uplands and deeply incised canyons sharply delineates canyon rims and planform morphology. Canyon heads can have varied shapes from amphitheaters with vertical headwalls to more pointed planform shapes with lower gradients, and a prominent goal in geomorphology and planetary science is to link canyon morphology to formation processes (e.g., refs. 4–8), with implications for understanding the history of water on Mars.

Amphitheater-headed canyons on Mars are most likely cut into layered basalt (9, 10), and canyon-formation interpretations have ranged widely from slow seepage erosion to catastrophic megafloods (4–6, 11, 12). Few studies have been conducted on the formation of amphitheater-headed canyons in basalt on Earth, however, and instead, terrestrial canyons in other substrates are often used as Martian analogs. For example, groundwater sapping is a key process in forming amphitheater-headed canyons in unconsolidated sand (e.g., refs. 8, 13, 14), but its importance is controversial in rock (5, 12, 15). Amphitheater-headed canyons are also common to plateaus with strong-over-weak sedimentary rocks (3, 16); however, here the tendency for undercutting is so strong that canyon-head morphology may bear little information about erosional processes, whether driven by groundwater or overland flow (e.g., refs. 3, 5, 17). Canyons in some basaltic landscapes lack strong-over-weak stratigraphy, contain large boulders that require transport, and show potential for headwall retreat by block toppling (18–21), all of which make extension of process–form relationships in sand and sedimentary rocks to basalt and Mars uncertain.

To test the hypothesis of a link between canyon formation and canyon morphology in basalt, we need field measurements that can constrain formation processes for canyons with distinct morphologies, but carved into the same rock type. Here we report on the origin of Malad Gorge, a canyon complex eroded into columnar basalt with markedly different shaped canyon heads. Results point to a potential diagnostic link between canyon-head morphology and formative process by megaflood erosion in basalt.

Malad Gorge is a tributary to the Snake River Canyon, Idaho, within the Snake River Plain, a broad depression filled by volcanic flows that erupted between ∼15 Ma and ∼2 ka (22, 23). The gorge sits at the northern extent of Hagerman Valley, a particularly wide (∼7 km) part of the Snake River Canyon (Fig. 1). Malad Gorge is eroded into the Gooding Butte Basalt [⁴⁰Ar/³⁹Ar eruption age: 373 ± 12 ka (25)] which is composed of stacked lava beds, each several meters thick with similar well-defined columns bounded by cooling joints and no apparent differences in strength between beds. The Wood (or Malad) River, a major drainage system from the Sawtooth Range to the north, drains through Malad Gorge before joining the Snake River. The Wood River is thought to have been diverted from an ancestral, now pillow lava-filled canyon into Malad Gorge by McKinney Butte basalt flows (24) [⁴⁰Ar/³⁹Ar eruption age: 52 ± 24 ka (25) (Fig. 1).

Fig. 1.

Shaded relief map of the study region (50-m contour interval) showing basalt flows (23), their exposure age sample locations, and the path of the ancestral Wood River following Malde (24) (US Geological Survey).

Malad Gorge contains three distinct canyon heads herein referred to as Woody’s Cove, Stubby Canyon, and Pointed Canyon (Fig. 2A). Woody’s Cove and Stubby have amphitheater heads with ∼50-m-high vertical headwalls (Fig. 2C), and talus accumulation at headwall bases indicates long-lived inactive fluvial transport (Fig. 3 A and B). Woody’s Cove, the shortest of the three canyons, lacks major spring flows and has minor, intermittent overland flow partially fed by irrigation runoff that spills over the canyon rim. Stubby has no modern-day overland flow entering the canyon, and springs emanate from a pool near its headwall (Fig. 3B). In contrast, Pointed Canyon is distinctly more acute in planform morphology, contains a 7% grade knickzone composed of multiple steps rather than a vertical headwall (Figs. 2C and 3C), and extends the farthest upstream.

Fig. 2.

Malad Gorge topography (10-m contour interval) and aerial orthophotography (US Geological Survey). (A) Overview map and (B) close-up for Stubby and Pointed canyons showing mapped bedrock scours (white arrows), exposure age sample locations (red circles) with age results, location of the uppermost active knickpoint (black circle), abandoned bedrock channels (blue dashed lines), and grain-size analysis sites (blue squares). The blue star shows the reconstructed location of the headwall of Pointed Canyon at 46 ka (see Discussion and Fig. 5). (C) Longitudinal profile along Stubby and Pointed canyons from their confluence (shown as white lines in B) with local slope, S, averaged over regions demarked by dashed lines (Fig. 4A shows close-up of profile in Stubby Canyon).

Fig. 3.

Photographs of (A) headwall of Woody’s Cove (person for scale, circled), (B) ∼50-m-high headwall of Stubby Canyon, (C) downstream-most waterfall at Pointed Canyon knickzone (12-m-high waterfall with overcrossing highway for scale), (D) fluted and polished notch at the rim of Stubby Canyon (notch relief is 10 m), (E) upstream-most waterfall at Pointed Canyon knickzone (within the southern anabranch of Fig. S2), and (F) upstream-most abandoned channel in Fig. 2B and Fig. S2 (channel relief is ∼10 m). White coloring on the headwalls in A and B is likely residue from irrigation runoff.

Early work attributed the amphitheater-headed canyons in this region—Malad Gorge, Box Canyon, located 18 km south of Malad Gorge (Fig. 1), and Blue Lakes Canyon located 42 km to the SE—to formation by seepage erosion because of no modern overland flow and the occurrence of some of the largest springs in the United States in this region (7). Because spring flows (e.g., ∼10 m³/s in Box Canyon; US Geological Survey gauge 13095500) are far deficient to move the boulders that line the canyon floors, Stearns (7) reasoned that the boulders must chemically erode in place. This explanation is improbable, however, given the young age of the Quaternary basalt (25), spring water saturated in dissolved solids (19), and no evidence of rapid chemical weathering (e.g., talus blocks are angular and have little to no weathering rinds). Instead of groundwater sapping, Box Canyon was likely carved by a large-scale flood event that occurred ∼45 ka based on ³He cosmogenic exposure age dating of the scoured rim of the canyon headwall (19, 26). In addition, Blue Lakes Canyon was formed during the Bonneville Flood [∼18–22 ka (27, 28)], one of the world’s largest outburst floods that occurred as a result of catastrophic draining of glacial lake Bonneville (21). In both cases, canyon formation was inferred to have occurred through upstream headwall propagation by waterfall erosion.

Herein we aim to test whether the amphitheater-headed canyons at Malad Gorge also owe their origin to catastrophic flooding, whether Pointed Canyon has a different origin, and whether canyon morphology is diagnostic of formation process. To this end we present field observations, sediment-size measurements, hydraulic modeling, and cosmogenic exposure ages of water-scoured rock surfaces and basalt-flow surfaces (Methods and Tables S1 and S2).

Results

We inspected the canyon rims and the escarpment that separates Woody’s Cove from the rest of Malad Gorge and mapped scoured rock as indications of overland flow. Scours consist of linear abrasion marks (flutes), often millimeters in depth and centimeters long, that fan outward in the inferred downstream flow direction (Fig. S1). The amphitheater-headed canyons at Malad Gorge have scoured rock upstream indicating overland flow in the past (Fig. 2). Both amphitheater-headed canyons have notches cut into their headwalls showing evidence for plucking (missing blocks) and abrasion (polished and fluted rock surfaces). For example, the notch at the head of Stubby Canyon is ∼10 m deep with respect to the neighboring basaltic plain and cuts basalt-flow stratigraphy (Fig. 3D). In addition, both amphitheater-headed canyons have large pools at their heads (e.g., Fig. 3B), similar to plunge pools at the base of waterfalls. The pool at Stubby Canyon, for example, is ∼120 m in diameter (Fig. 4A) and partially filled with sediment. Canyon headwalls are vertical and not undercut (Fig. 3 A and B), and the bedrock is blocky and jointed suggesting canyon-head erosion by block plucking and toppling (18, 29). The bedrock scours upstream of the canyon heads, scoured notches at the canyon-head rims, vertical and blocky canyon headwalls, and large relict plunge pools at Malad Gorge canyons are similar to features found at Box Canyon and Blue Lakes Canyon, evidence used at those sites and elsewhere to support canyon formation by large-scale flooding (e.g., refs. 19–21, 30).

Fig. 4.

(A) Thalweg long profile of the head of Stubby Canyon showing the modern (spring-driven) water level. The region of the river bed downstream of the plunge pool (S = 0.0043) was used for hydraulic calculations. (B) Sediment-size distributions within the survey location of Stubby Canyon (GS1 in Fig. 2B; 113 counts) and of a boulder bar upstream of the canyons (GS2 in Fig. 2B; 32 counts). The median sediment size (D₅₀) at location GS1 was used for calculations (Methods). (C) Surveyed canyon cross-section of Stubby Canyon at GS1 (Fig. 2B) showing the flood depth calculated to move the bouldery bed (Methods) and the modern flow depth. The break in slope at 958-m elevation represents the transition from bedrock walls above to talus below.

We collected samples of polished bedrock surfaces from the notches of the amphitheater-headed canyons for cosmogenic ³He exposure age dating (Methods and Tables S1 and S2). In both cases, eroded notches are sufficiently deep (10 m for Stubby Canyon and 2.6 m for Woody’s Cove) that any inherited cosmogenic exposure before erosion is negligible. The exposure age at the rim of Woody’s Cove is 47 ± 3 ka, which is within error of the oldest of three samples taken from the rim of Stubby (46 ± 3 ka). These ages are also within error of the age of the notch cut into the rim of Box Canyon [45 ± 5 ka (26)]. The other four samples of scoured rock within Stubby Canyon cluster with an average of 20.6 ± 2.6 ka (Table S2), coincident with upper age constraints for the Bonneville Flood (27, 28). Together, these exposure ages indicate the formation of all three amphitheater-headed canyons in this region (Box Canyon, Woody’s Cove, and Stubby Canyon) may have been coeval, ceasing ∼46 ka, except for later reworking during backwater inundation by the Bonneville Flood.

To constrain the discharge necessary to mobilize the boulders that line the canyon floors, which is a necessary condition for canyon formation, we measured sediment sizes and surveyed channel dimensions within Stubby Canyon (at GS1 in Fig. 2B; Methods). Median particle diameters are 0.58 m with the largest boulders exceeding 3 m (Fig. 4B). The river-bed gradient downstream of the plunge pool has a near-constant slope of 0.0043 (Fig. 4A), and modern spring-fed water depths average 1.0 m (Fig. 4C). These data were used as inputs into hydraulic resistance and incipient sediment motion formulas (Methods) to find a modern spring discharge of 11 m³/s [similar to measurements within Box Canyon (19)] and a paleoflood necessary to mobilize the boulder bed that has a calculated minimum discharge of 1,250 m³/s and a minimum water depth of 9 m (Fig. 4C). Large paleodischarges are also inferred from boulder bars (median grain diameter of 0.32 m; Fig. 4B) upstream of the canyon (GS2 in Fig. 2B) where the flood was largely unconfined. The calculated minimum canyon-forming discharge is approximately sevenfold the largest historic discharge of the Wood River (181 m³/s over a 98-y record; US Geological Survey gauge 13152500) and more than 100-fold the calculated modern spring discharge, indicating that the flood(s) that carved the amphitheater-headed canyons of Malad Gorge were extraordinary. Our discharge estimate is consistent with calculated flow depths of 2–6 m needed to exceed the threshold for megaflood erosion by block toppling in this region (18). Moreover, it is similar to the discharge calculated for the canyon-carving flood at Box Canyon [800–2,800 m³/s (19)] and far smaller than discharge calculated for the Bonneville Flood [10⁶ m³/s (30)].

In contrast to the two amphitheater-headed canyons at Malad Gorge, the third canyon ends in an active knickzone that houses the Wood River (Fig. 3C) and contains abundant markers of active fluvial abrasion (rather than block toppling), most notably, nested potholes (Fig. 3E). At the upstream extent of the knickzone, the active channel narrows from ∼20 to ∼1 m in width as water plunges into the gorge and abandons the pothole-laden river bed at its margins as a strath terrace (Fig. 3E). Similar strath terraces abut both sides of Pointed Canyon along its upper ∼1 km. Upstream of Malad Gorge, the Wood River is anabranching, and the propagation of the uppermost knickpoint appears to have pirated water that once flowed into a neighboring channel (Fig. S2). The abandoned channel contains abundant potholes and a now-dry waterfall (Fig. 3F). Additional abandoned channels exist farther downstream on both sides of Pointed Canyon (Fig. 2B), and abrasion marks within the channels indicate flow into Pointed Canyon in some cases and out of Pointed Canyon in other cases, supporting progressive abandonment of anabranches due to knickzone propagation.

To test whether Pointed Canyon was formed by progressive knickzone retreat, we collected samples for exposure age dating from strath terraces and abandoned channels (Fig. 2B, Methods, and Tables S1 and S2). Three of the resulting dates fall within the age range of the Bonneville Flood, and because the Bonneville Flood likely inundated Malad Gorge in full (30), reworking and erosion of even the uppermost terraces cannot be ruled out. The other exposure ages show progressive younging in the upstream direction (Fig. 5) with a ∼33-ka abandoned channel at x = 1,300 m (where x is the distance from the confluence of Pointed and Stubby canyons), to an 11-ka abandoned channel at x = 1,900 m, to the most upstream extent of the knickzone where terrace abandonment is active (Fig. S2 and Fig. 3E). These ages suggest, over at least the past 33 ka, a constant rate of knickzone retreat of 2.5 cm/y (by linear regression, r² = 0.99), and a constant rate is expected given that 1 km of retreat has not affected the 7,800 km² drainage area to the knickpoint (e.g., refs. 1, 2). It is possible that the Bonneville Flood partially affected all of the exposure ages in Pointed Canyon, but this cannot explain the 11 ka age that is younger than the flood, nor the linear, upstream-younging trend in ages that intersects the modern knickpoint location at ∼0 ka (Fig. 5). Moreover, flood erosion, where it occurred, likely completely reset ages owing to plucking of meter-scale blocks.

Fig. 5.

Exposure age versus distance projected along the long profile (Fig. 2C) of Pointed Canyon. Three samples (open symbols) yielded ages consistent with reworking by the Bonneville Flood and are not included in the fit (note one of the three falls nearly on the trend line). Error bars represent analytical uncertainty (Table S2), which is often smaller than the symbol size. The most upstream age of zero (filled circle) is assumed based on the observed location of active knickzone retreat (Fig. 3E and Fig. S2). Data suggest Pointed Canyon was 1.05 km long (blue star in Fig. 2B) when formation of the other canyons ceased (∼46 ka). The Bonneville Flood age range shown extends a few thousand years older than is typically reported based on revised chronology of paleoshorelines (27) and canyon exposure ages (28).

Discussion

Scoured rock, eroded notches into canyon-head rims, plunge pools, sediment-transport constraints, and cosmogenic exposure ages suggest that the amphitheater-headed canyons of Malad Gorge were carved by large-scale flooding that ceased ∼46 ka, and similar data from Box Canyon (19, 26) suggest a common origin. Exposure ages of the canyon heads are similar to the eruption age of the McKinney Butte Basalt [⁴⁰Ar/³⁹Ar age of 52 ± 24 ka (25)], and independent evidence that the Wood River was diverted to its present course at this time (24) suggests a causal connection between diversion of the Wood River and the incision of Malad Gorge and Box Canyon. This hypothesis is consistent with the onset of a regional loess deposit [∼40 ka (31)] that does not show flood scour and must postdate flooding, in addition to the surface exposure age we measured for Notch Butte Basalt (³He age of 52.8 ± 3.4 ka; Methods and Tables S1 and S2), which is largely loess free, crosscut by the Wood River just upstream of the knickzone, and must predate river diversion (Fig. 1). To improve age constraints for McKinney Butte Basalt, we measured an ³He exposure age of the flow surface of 31.9 ± 1.9 ka (Fig. 1, Methods, and Tables S1 and S2); however, this age is necessarily a minimum eruption age due to abundance of loess cover here (23) and potential for erosion of the original flow surface (32). Still, we cannot rule out the possibility that the Wood River diversion postdates formation of the amphitheater-headed canyons, in which case an alternative flood source such as a glacial lake outburst (e.g., refs. 19, 33, 34) from the Sawtooth Range must be invoked.

The similar timing of the McKinney Butte basalt flows and the cessation of flooding recorded by exposure ages on the canyon rims suggests that flooding was short-lived and possibly a single event. Given that the necessary discharge for canyon cutting far exceeds historical floods of the Wood River, we suggest lava flows must have dammed the Wood River resulting in outburst flooding where sheets of flood water focused locally to erode distinct amphitheater-headed canyons. These floods may also have eroded the eastern wall of the Snake River canyon between Malad Gorge and Box Canyon resulting in the anomalously wide Hagerman Valley (Fig. 1). Furthermore, a short-duration flood event explains why there is little landscape dissection upstream of the canyon heads; in the absence of an unbuttressed escarpment, block toppling is not possible (18, 29), and the floods must have been limited to the comparatively slow processes of fluvial abrasion and localized plucking, which require thousands of years or more for channelization (1).

Once volcanism at McKinney Butte ceased, we infer that the Wood River established its modern path to Pointed Canyon and abandoned the amphitheater-headed canyons. Given the measured knickzone retreat rates of 2.5 cm/y in Pointed Canyon, our interpretation implies that Pointed Canyon was ∼1 km long at the cessation of megaflooding (Fig. 5), placing the ancestral canyon head near the location where the modern-day canyon begins to taper in width (blue star in Fig. 2B). Therefore, Pointed Canyon may have been partly carved by megaflooding and may once have had an amphitheater head, perhaps similar to the branching amphitheater-headed canyons on Mars (4). Since the cessation of megaflooding, the comparatively low discharges of the Wood River do not exceed the threshold for block toppling, leaving only the steady upstream propagation of the pointed headwall by fluvial abrasion.

Our study suggests that amphitheater-headed canyons with vertical walls in columnar basalt may be a diagnostic indicator of rapid megaflood erosion rather than persistent fluvial abrasion or seepage erosion. The rationale for this hypothesis is that i) basalt with vertical cooling joints and horizontal bedding planes tends to break down to large, meter-sized boulders that cannot be transported by seepage alone; ii) vertically jointed rock promotes persistent vertical headwalls and rapid erosion if a threshold discharge for block toppling is surpassed, which in turn requires high-magnitude overland flows (18); and iii) the overland flow events must be of short duration, otherwise fluvial abrasion will tend to flatten the headwall and dissect the upstream landscape, as shown in Pointed Canyon. There are a host of canyons cut into columnar basalt by megafloods that support our hypothesis: Asbyrgi Canyon in Iceland (35), canyons cut by the Missoula Floods (20), Blue Lakes Canyon and Devil’s Corral cut by the Bonneville Flood (21, 28, 30), and Box Canyon (19).

Many of these terrestrial amphitheater-headed canyons in basalt appear morphologically similar to putative cataracts in outflow channels and valley networks with amphitheater heads on Mars (e.g., refs. 4, 6, 12). Given confirmation of columnar basalt (e.g., ref. 10) and canyon heads that lack well-developed undercuts, Martian canyons with amphitheater heads may also owe their origin to short-lived, high-magnitude flood events, possibly sourced from subsurface water eruptions, dam-burst floods, or lake spillover (e.g., ref. 36). Most landscape evolution models applied to Earth and Mars do not include jointed rock or erosion by toppling and consequently only produce amphitheater-headed canyons by using rules for seepage erosion in sand (e.g., refs. 8, 15, 37). In contrast, in basalt the onset of toppling may be the tipping point to form amphitheater-headed canyons, which implies drastically different water discharges and flow durations for canyon formation than groundwater sapping.

Methods

Fourteen rock samples were taken for ³He cosmogenic exposure-age dating: five from Stubby Canyon, six from Pointed Canyon, one from Woody’s Cove, one from McKinney Butte Basalt, and one from Notch Butte Basalt (Figs. 1 and 2 and Tables S1 and S2). A cosmic ray-shielded sample used to correct all canyon erosion samples (BG0 in Table S1) was taken from 5.8 m deep within a crack located ∼4 m from the canyon sidewall of Pointed Canyon. The crack splits a viewing platform indicating it has opened historically. Shielded samples for McKinney Butte and Notch Butte basalts were taken from road cuts (MB0 and NB0 in Table S2).

Samples were collected by chipping off the upper 4 cm of rock on near-horizontal surfaces. We separated 250- to 450-μm-diameter olivine and pyroxene grains from the crushed host rock by standard magnetic, heavy liquid, and hand picking techniques. Adhering groundmass was removed by sonicating samples in 5% (vol/vol) 2:1 HF:HNO3 acid for ∼1 h. Samples were ground and wet-sieved to <37 microns which largely removes mantle-derived ³He trapped in melt and/or fluid inclusions (38). The remaining matrix-sited ³He and ⁴He was measured on a MAP 215-50 noble gas mass spectrometer following heating to 1,300 °C in vacuum to release the gas (39) (Table S1). Shielded samples were used to correct for remaining mantle-derived and nucleogenic ³He (40), which only averaged 6% of the total measured ³He (Table S1). Accumulation ages and production rates (Tables S1 and S2) were calculated using the Lifton/Sato scaling scheme (41, 42) on the CRONOS ³He calculator (43, 44).

To calculate minimum paleodischarge, we surveyed a cross-section and long profile using a total station (Fig. 4 A and C). Sediment-size measurements of the intermediate particle axis (Fig. 4B) were made on a regular-spaced grid (Fig. 2A). Thalweg flow depths required for transport of the median particle size were calculated using ref. 45. Using calculated water depth and surveyed canyon cross-sectional area (Fig. 4C), we calculated the minimum paleoflood discharge following (46), where the bed-roughness length scale was 1.3 D₈₄ (where D₈₄ = 1.3 m is the sediment diameter in which 84% of the bed is finer; Fig. 4B) following a calibration in similar Box Canyon (19).