Main

On human timescales, mountain peaks seem eternal. Yet, on geological timescales, mountain peaks are ephemeral: their shape and altitude are constantly evolving in response to the competition between tectonic uplift and erosion. However, very few direct observations exist of how Earth’s highest mountain peaks evolve. Whereas their growth can be considered progressive over time, we do not know if their erosion is also progressive or episodic, or whether summit erosion proceeds at the same rates as valley-floor lowering5.

It has been postulated that a glacial buzzsaw mechanism places an upper limit on mountain elevation and relief irrespective of the tectonic uplift rate3,6, confining mountain heights to within about 1.5 km of the local equilibrium-line altitude7 (ELA). In this concept, efficient periglacial weathering, frost cracking and rockfalls along steep hillslopes produce rapid retreat of glacier cirque headwalls, limiting peak elevation at rates commensurate with rapid valley-floor downcutting by glaciers8. In various settings, however, this concept has been challenged9,10,11,12. In the Himalayas, one of the most active mountain ranges on Earth, the >8,000-m peaks culminate about 3,000 m above the ELA and seem to contradict the notion of a comprehensively operating glacial buzzsaw1,12. At these extreme elevations, local measurements of cirque headwall retreat show rates mostly lower than regional erosion rates13,14,15,16. Frozen-based ice cover, as well as a sharp decrease in frost-cracking efficiency for increasing altitudes, owing to the absence of common freeze–thaw cycles1,2, are suspected to strongly limit the erosion of the glaciated peaks protruding far above the ELA1. If high-elevation summit erosion is unable to match rapid valley-floor downcutting rates, this mismatch could lead—in theory—to disproportionate growth of the Himalayan peaks.

In steep and active mountains, it is conjectured, however, that mechanical strength of the mountain rock mass sets an upper limit to peak elevation and the height of oversteepened slopes4, such that slope adjustment might occur through landslides at a rate paced by river downcutting. This equilibrium has been corroborated in non-glaciated and vegetated landscapes, in which landslides can be tracked from satellite-image inventory17,18,19. The 2021 rock-and-ice avalanche of the Ronti peak (India)20 illustrated that landslides can also contribute markedly to the erosion of high-elevation glaciated Himalayan peaks. Nevertheless, the lack of systematic landslide inventories precludes any reliable assessment of the relative contribution of this erosive process in long-term evolution of high-elevation mountainous areas. Instead, the few coseismic landslide inventories show an apparent sharp decrease in landslide occurrence at high elevations21. This observation indicates either an erosion mechanism distinct from landsliding or a size-frequency distribution of landslides17 that would be clearly skewed in favour of very large landslides, remaining infrequent enough not to be seen by remote-sensing techniques.

In this study, we document evidence of a recent giant rockslide in the Annapurna region (central Himalayas, Nepal; Fig. 1), which illustrates the topographic evolution of high Himalayan peaks, their abrupt erosion by large rockslides and the disproportionate consequences of such collapses along downstream drainage systems.

Fig. 1: Location maps of the Sabche breccia deposits.
figure 1

a, Map of Nepal and Narayani basin (in red) with the >8,000-m summits (triangles). G, drill cores GR1 and GR2 (purple symbols) in Dudahi in India; k, Kathmandu; P, Pokhara; S, Sabche. Scale bar, 100 km. b, Topographic map of central Nepal focusing on the Seti river with the rock-avalanche breccia deposit (in yellow) in the Sabche cirque (headwaters of the Seti river) and along the upper Seti valley owing to the rockslide of the Annapurna IV southwest flank. Subsequent erosion of this brecciated material led to the downstream sediment aggradation with massive conglomeratic terraces (in pink) in Pokhara basin and as far as south of the Main Boundary Thrust (MBT), near Narayangad (N). K, Karuwa; MCT, Main Central Thrust; P, Pokhara; S, Sabche; STD, South Tibetan Detachment. c, Zoomed-in map of the glaciated Sabche cirque showing contrasting breccia facies and sample locations. d, Aerial view of the Sabche cirque; its breccia filling (outlined by the dashed line and ‘br’) is exposed by large whitish penitents, hundreds of metres high, left by the erosion of the rock-avalanche deposits.

Full size image

Uncovering a giant rockslide deposit

The southern flank of the Annapurna massif, culminating at 8,091 m above sea level at present, shows a 6-km drop in elevation from the high summits to the Lesser Himalayas hills. The area encompasses steep slopes, active microseismicity22 and strong ground shaking by large earthquakes23,24 (Mw ≥ 8). Heavy monsoon precipitation prevails over its southern slopes. This tectonic and climatic setting induces some of the highest regional erosion rates in the Himalayan range, with values of 3–5 mm year−1 (refs. 25,26,27).

The Annapurna massif, dominated by calcareous metasediments of the Tethyan Sedimentary Series28,29 (TSS) (Extended Data Fig. 1), is punctuated by several large glacial cirques, separated by >7,500-m glaciated peaks (Fig. 1). Between Annapurna III and IV, the Sabche cirque is a deep and wide depression bordered by extremely steep cliffs (Fig. 1). The cirque bottom is filled with light-coloured sedimentary deposits, which have been eroded into large and tapering penitents several hundred metres high (Fig. 1d). Inaccessibility has so far prevented precise field description of these deposits, which have been interpreted from afar as subglacial tills or glaciolacustrine sediments30. Up close, these massive deposits are well consolidated, especially in the southern, most basal part, comprising finely crushed matrix-supported breccia with centimetric to decimetric angular clasts from various TSS carbonate lithologies (Extended Data Fig. 2 and Supplementary Information SI-1). The coarse elements are more abundant in the upper part of the deposit and reach sizes of several metres. The breccia colour is uniform (white to greyish), except locally with yellowish units reminiscent of the yellow limestones of the Annapurnas28 (Fig. 1c). The breccia sediment infill seems relatively continuous without any apparent unconformity but shows incomplete mixing, internal shear zones and rare jigsaw facies (Extended Data Fig. 2). These sedimentary features31, the intense rock fragmentation, the exceptional thickness (>400 m; Fig. 1d) and the exclusive presence of these breccia among all cirques of the Annapurna massif unequivocally indicate that these breccia sediments were generated by a single rockslide granular avalanche.

The elevation of the breccia remnants along the southern side of the cirque and of a few preserved top surfaces in its northern part (Fig. 1) suggests that the cirque was completely filled by rockslide debris (Supplementary Information SI-4), with thickness locally >1 km. Furthermore, a debris avalanche overflowed the cirque and spilled into the upper Seti valley, in which scattered outcrops of massive breccia deposits can be observed for roughly 10 km downstream along the valley walls or as isolated >200-m-high massive spur remnants (Extended Data Fig. 3 and Supplementary Information SI-2). These remnants are dominated by angular TSS limestone clasts in a finely fragmented and pulverized carbonate-rich matrix. Little mixing appears at the metric scale (as observed for Sturzströme in motion32), dyke injections are deformed by brittle deformation and the basal shearing contact with the gneissic-bedrock valley flanks is overlain by a mixing zone incorporating fragmented vegetal debris.

Volume and age of the rockslide deposit

Subtraction of the restored basal topography of the Sabche cirque from the initial breccia deposit surface provides an estimated volume of the rock-avalanche deposits of 23.5 +4/−3 km3. A further volume of about 3.5 km3 is similarly inferred along the upper Seti valley (Methods, Extended Data Fig. 4 and Supplementary Information SI-4). Accounting for an average breccia porosity of 15 ± 5% (Supplementary Information SI-5), this reconstructed deposit would correspond to a total volume of collapsed rock of 23 +3.5/−3 km3, constituting the largest rockslide described in the Himalayas so far33,34.

Four 2–5-m high blocks of Silurian marly limestone were sampled on a large blocky and chaotic remnant of the original deposit surface (Extended Data Fig. 5) for 36Cl surface-exposure dating (Supplementary Table SI-11-2). They show consistent 36Cl concentrations with an average surface-exposure age of 1196 ± 75(2σ) ad (Fig. 2a).

Fig. 2: Chronological constraints on deposits related to the Sabche rockslide.
figure 2

a, Dating of the rock-avalanche deposit and Pokhara conglomeratic formation along the Seti river from the Sabche cirque down to Narayangad (site locations according to letters in Fig. 1). From top to bottom: the probability density function (PDF) of 36Cl exposure ages on blocks located on an upper-surface relic of the rockslide deposit; zoom-in on the young tail of the feldspar IRSL age PDF showing partial signal reset (see complete PDF and individual aliquot ages in Extended Data Fig. 6) of a shearing band towards the base of the rockslide deposit in Sabche cirque; calibrated 14C age PDF (black or grey solid lines) of wood debris sampled in the rock-avalanche deposit in Karuwa (site K in Fig. 1b) at the bedrock contact of the Seti valley; calibrated 14C age PDF of organic fragments taken in the most basal conglomeratic units in Pokhara basin36,38; calibrated 14C age PDF of a charcoal sampled in fine conglomerates north of Narayangad. The red curve represents the composite PDF, combining all available dating at each site. The numbers in parentheses indicate the number of dated samples. b, 14C ages of numerous organic-debris samples from the central part of Pokhara basin (published and new dating), represented according to their stratigraphic relative elevation. Left axis: gives the elevation of a sample above the bottom (it is a relative position, with 0 for a position close to the base of the conglomerate filling and 1 for a position close to the top of the filling), counted from bottom to top of axis. Right axis: gives the depth below the surface, counted from top to bottom of axis. Points and error bars correspond to the median value and the 95% confidence interval of the calibrated ages PDF, respectively. Analyses restricted to leaves, wood fragments and tree trunks to minimize the effect of suspected reworking, in particular for charcoals. A notably rapid sedimentation rate of about 1 m year−1 is documented in the vicinity of the centre of the Pokhara basin. Note that ‘Charcoal in buried palaeosols’ refers to charcoal collected from former soils on the flanks of the Seti valley that were buried by and predate the Pokhara conglomerate.

Full size image

In parallel, infrared stimulated luminescence (IRSL) dating on feldspar was conducted on three breccia samples from an internal shear zone, assuming that increased temperature and/or shear heating35 during the rock avalanche did partially reset the IRSL signal (Extended Data Fig. 6 and Supplementary Information SI-12). The observed partial reset is moderate for two samples with average apparent IRSL ages of several thousand years, but is highly pronounced for one sample taken in the most-deformed levels. For the latter, the IRSL age distribution is highly scattered with an apparent modal age of approximately 1,200 years before present and an estimated minimum age of about 900 years before present (Fig. 2a), consistent with surface-exposure dating of the top relic surface.

Finally, eight 14C analyses on plant fragments of stripped vegetation in the contact zone between rock-avalanche material and bedrock in the upper Seti valley (site K in Fig. 1) provided reproducible calibrated ages (Fig. 2a and Supplementary Table SI-10-1). The most probable age of 1190 ± 26(2σ) ad is highly consistent with the 36Cl surface-exposure ages of the rockslide deposit surface, thereby confirming the monogenic nature of the rock avalanche between the Sabche cirque and the upper Seti valley.

Reconstruction of the collapsed palaeo-peak

To understand the origin of the collapsed rock mass, we identified the failure surface in the northeast cirque part below Annapurna IV (Figs. 1 and 3). There, the morphology of the rock cliffs stands out (Methods and Extended Data Fig. 4d), with fresh planar or prismatic structural surfaces, devoid of any substantial glacier or runoff-related erosion. The proposed giant rockslide, collapsing along this uneroded failure surface, would have beheaded the palaeo-summit of Annapurna IV.

Fig. 3: Reconstruction of the Annapurna IV palaeo-summit before its collapse.
figure 3

a, Thickness of the rock-avalanche deposit estimated from breccia relics in Sabche cirque and along the upper Seti valley. The orange arrow (oriented at 245° N) corresponds to the optimal direction of collapse (Extended Data Fig. 9e). Circled letters correspond to sites described in Fig. 1b. Scale bar, 5 km. MCT, Main Central Thrust; STD, South Tibetan Detachment. b, Average shape of the reconstructed palaeotopographies that satisfy rock-volume and hillslope-stability constraints (greyed palaeo-summit on top of the present topography in slight transparency). c, Cross-section (location in a) through the middle of the Sabche cirque, with averaged palaeotopography before collapse and approximate geometry of the post-collapse breccia deposit; ‘rsb’ and ‘bdb’ indicate the projection on the section plane of the centres of mass of the rockslide and breccia deposit volumes, respectively, and whose elevation difference is used to calculate the released potential gravitational energy. Fm, formation. d, Probability densities (presented as a histogram and resulting from 2,000 random palaeotopographies) of the summit-point elevation of the palaeo-Annapurna IV satisfying increasing levels of requirement: first, the collapsed volume VRS is in the range 20–26.5 km3 (grey); second, the summit slopes were stable with regard to small-sized to medium-sized landslides (pink); third, the safety factor along the failure surface was minimal (red) (Methods and Extended Data Fig. 9). The probability that the summit point could have exceeded 8,000 m (thick line) is 65% and 76% if the palaeotopographies satisfy the first two and three conditions, respectively.

Full size image

To reconstruct the initial geometry of the palaeo-Annapurna IV from the restored rock volume of 23 +3.5/−3 km3, we suggested that the stability of the palaeo-summit hillslopes follows mechanical characteristics that prevail for the whole Annapurna massif, for similar altitudes and lithologies. Stability of 82–100% of the high-elevation slopes requires an apparent rock mass cohesion of 1–2 MPa, assuming a friction angle of 35° (Methods and Extended Data Figs. 7 and 8). For these geomechanical properties, we generated thousands of random possible topographies for the Annapurna IV palaeo-summit and considered as acceptable only those fitting both the collapsed rock volume and the regional hillslope stability (Methods and Extended Data Fig. 9). Through this Bayesian approach, we estimated that the palaeo-Annapurna IV may have peaked at an average elevation of about 8,100 m (about 600 m above the modern summit and comparable with Annapurna I) and exceeded 8,000 m with a probability of 0.65–0.76 (Fig. 3d). The eroded volume corresponds to an average 1 km of bedrock lowering and involves lithologies ranging from Silurian to Cambrian (Fig. 3c), in good agreement with the diversity of colour and carbonate content of the breccia material (Fig. 4).

Fig. 4: Far-downstream signature of the reworked rockslide deposits evidenced by sediment carbonate content.
figure 4

Comparison of the carbonate content in brecciated rock-avalanche material, in TSS units at the source of the rockslide, in Pokhara conglomeratic formation and in modern sediment along the Seti river (see also Supplementary Table SI-13-1). Unlike rockslide-derived sediments, which show a very high and relatively constant proportion of carbonate, modern sediments show downstream increasing dilution of the carbonate fraction, in particular after the confluence of the Seti with the Madi or Trisuli rivers. Rivers: Ka, Kali Gandaki; M, Madi; Ma, Mardi; Tr, Trisuli; numbers in parentheses, ratio of drainage areas, with the Seti as the denominator. Circled letters and acronyms correspond to the sites and faults mentioned in Fig. 1a,b.

Full size image

Sediment export and downstream consequences

The collapse released a huge amount of potential gravitational energy of approximately 1.2 × 1018 J (Fig. 3). Substantial energy release was consumed in rock-material fragmentation, which, aided by the schistose nature of the marly-calcareous bedrock, probably explains the extreme degree of disintegration observed in the breccia material. After the palaeo-summit collapse about 830 years ago, the erosion of the fine-grained breccia by rainwash, landsliding, proglacial streams or glaciers, and their export by the upper Seti river, were extremely efficient and rapid: of the initial 23.5 km3 of rockslide debris, only roughly 10% can still be observed today in the Sabche cirque (Extended Data Fig. 4c). Assuming a long-term denudation rate of about 3 mm year−1 in the High Himalaya26, the return time of a giant rockslide of similar volume would be >100 kyr at the scale of the entire Sabche cirque headwalls. The response time of the glaciofluvial system to evacuate the Sabche breccia material therefore seems very short in comparison.

Farther downstream of the Seti river, the Pokhara basin (Fig. 1) is known to have been filled very rapidly 500–1,000 years ago30,36 by about 5 km3 of mostly conglomeratic sediment with calcareous clasts, whose only possible origin is the Sabche cirque. Sedimentary fill is characterized by the stratigraphic continuity of a succession of planar beds with varied facies (Supplementary Information SI-9) that reflect rapid aggradation by fluvial, hyperconcentrated or turbulent, sediment-laden flows30,36.

Extensive 14C dating of organic fragments found in the fine-grained units of the Pokhara sediments36,37,38, collected at different burial depths, and complemented by eight new dates (Supplementary Table SI-10-1), provide robust constraints on the timing of aggradation (Fig. 2b). They indicate onset of aggradation around 1200 ad at an average rate of 1 m year−1 until about 1300 ad. Together, the similarity in carbonate content (Fig. 4) and the age of the basal units in close agreement with the giant rockslide age, but stratigraphically postdating the breccia (Extended Data Fig. 3a), strongly suggest a cogenetic link between these two sedimentary units. The Pokhara conglomerates are thus best explained by the active erosion of the Sabche rock-avalanche breccia and by the rapid sediment transfer by Seti river down to Pokhara valley.

Fine conglomerates with calcareous pebbles irregularly crop out far downstream of Pokhara, along the Seti, the Trisuli and even the Narayani rivers (Fig. 4) in the foreland (site N in Fig. 1). There, a fine, carbonate-rich conglomerate unit, dated 1150–1220 ad (14C dating; Fig. 2a), shows how the Narayani was temporarily overwhelmed by sediments from the palaeo-Annapurna IV collapse. Such sediment pulse may even have reached the mid-Ganga plain, as suggested by a >3-m-thick unit at the top of a core drilled into an elevated Narayani megafan lobe (site G in Fig. 1). This unit is anomalously rich in TSS carbonate (30–60%) as compared with the average proportion of about 10% in Late Pleistocene to modern Narayani sediments (Fig. 4 and Extended Data Fig. 10).

The amalgamated sediment volume in the Pokhara basin, middle Seti valley and Narayani megafan lobe may have reached 15 km3 (Supplementary Information SI-9-C), thereby representing ≤50% of the initial Sabche breccia volume (assuming 30% fluvial sediment porosity). The other half was presumably partly dissolved and partly exported farther downstream as suspended load, and may have ended up in the Bengal fan, the largest sediment sink for most Himalayan sediment39.

Sediment aggradation in Pokhara valley suggests that the most intense phase of breccia removal in the Sabche cirque and associated sediment supply downstream lasted for a century (Fig. 2b) and affected the whole downstream sediment-routing system. During that period, the rate of sediment yield (0.1–0.2 km3 year−1 exported from a 60-km2 area) was comparable with the highest measured values of post-volcanic-eruption sediment transport40 and led to an overwhelming content of TSS material in river sediments >300 km downstream. Compared with the Narayani basin (30,000 km2), which exports approximately 0.05 km3 of sediment per year41,42, post-collapse erosion of the Sabche breccias (only 0.2% of the Narayani total basin area) would have increased this annual flux by a factor of 3 over 100–150 years (see also carbonate-based budget in Supplementary Information SI-9-D).

The collapse has paradoxically positive and negative downstream consequences. In the long term, the creation of a large flat surface in the Pokhara valley has undoubtedly been beneficial, allowing the development of extensive agricultural areas and, above all, the recent development of the country’s second largest city with all of its infrastructures. By contrast, in the short term, if the palaeo-Annapurna IV collapse occurred today, the Pokhara metropolitan area would be decimated, with massive human and economic costs. Not only would the upstream parts of the valley be swept away by the granular avalanche but numerous sediment-laden flows would rapidly bury the valley under tens of metres of sediment, leaving a vast gravel plain in the Seti valley. Even the Indian side of the Ganga plain, a densely populated region, might be affected by increased probability of river-channel avulsions and rapid sedimentation along the Narayani megafan. Although such events are rare, understanding the causes of the palaeo-Annapurna IV collapse and its downstream consequences thus seem fundamental to comprehend the notable threat that potential giant rockslides can pose to the local populations and infrastructures in these high-mountain regions.

Possible causes for the Sabche rockslide

Several factors make the High Himalaya region clearly prone to large-scale slope failure. Glacial valley deepening owing to subglacial carving, combined with intense frost cracking close to the ELA2, represents one of the main preconditioning factors that produce steep, unstable headwalls surrounding glacier cirques. Schistosity orientation and structural planes also probably guide slope failure of the palaeo-Annapurna IV (Extended Data Fig. 4d).

By contrast, causes triggering the Sabche rockslide remain speculative. It has been suggested previously36 that the filling of the Pokhara basin arose from various pulses of debris flows following numerous landslides activated by three large earthquakes, at 1100 ± 50 ad (ref. 23), 1255 ad and 1344 ad (ref. 24). According to 14C dating of breccia at Karuwa (site K in Fig. 1), the Sabche rockslide occurred after 1150 ad and before 1255 ad at >99% confidence. Seismic triggering resulting from one of these three large earthquakes therefore seems to be excluded. Nevertheless, because the current catalogue of palaeo-earthquakes is probably incomplete for medieval times, we cannot exclude the impact of an undocumented earthquake in the region.

Independently, climate might also be involved as a preparatory or/and triggering factor for the Sabche collapse. The Medieval Climate Anomaly43, from 900 to 1200 ad, was possibly characterized by >1 °C warmer temperatures in the High Himalaya. This could have thawed part of the failure surface (see estimate in Supplementary Information SI-8) and favoured high-elevation rockslide(s)20. Although the cohesion loss from thawing seems to have been probably insufficient to trigger the palaeo-Annapurna IV collapse (unless the rock mass was already close to a critical state), such climatic factor on high-elevation bedrock destabilization calls for closer attention on the effects of current global warming when assessing future gravitational hazards related to the high Himalayan peaks.

The fate of the high Himalayan peaks

The identification of a cataclysmic rockslide around 1190 ad, which beheaded what may have been the Himalaya’s 15th 8,000-m peak, provides key insights into the probable evolution of the highest Himalayan peaks. First, this is not an isolated case. Several giant rockslides have been documented33,34,44 in central Nepal, with deposits reaching up to 10–15 km3. All are located in the high Himalayas and originated from summits/ridges between 4,500 and 6,000 m elevation. Although the current catalogue is incomplete and precludes definition of a precise erosion budget, massive summit collapses resulting in a sudden reduction of their height by several hundred metres could be the primary denudation mode of high, glaciated Himalayan peaks.

Notably, high summits commonly show steeper slopes, with median values rising from 35° below the ELA to 50° above it2,45. Steep slopes surrounding the glacial cirques of the Annapurnas are maintained by higher bulk material strength. The regional cohesion values (C ≈ 1–2 MPa) calculated for the glaciated relief cannot be considered as absolute values, because the transient effects of pore pressure or seismic acceleration have not been taken into account46. Nevertheless, they seem to be, for similar internal friction angles, about four times higher than the cohesion values (C ≈ 0.2–0.7 MPa) for the fluvially dominated high Himalayan landforms (below 4,000 m, south of the glaciated Annapurna ridges; Supplementary Information SI-6) and an order of magnitude higher than in other non-glaciated high mountains47 (C ≤ 0.1 MPa but using a distinct approach to derive cohesion values). In the Himalayas, this higher rock strength is not related to a lithological effect12, given that the steeper slopes at high altitudes in central Himalaya have values independent of the rock type2. Instead, we surmise that the higher rock-mass strength is favoured by permanently frozen conditions at high elevations. Such climatic conditions hold rocks in place by ice-rock interlocking48,49, while preventing regolith formation, penetrative chemical weathering and limiting groundwater recharge and high pore pressures at depth. If permafrost strengthens the first tens or hundreds of metres below the ground surface (Supplementary Information SI-8) and/or if weakened bedrock is absent near the surface, then the initiation of small and medium-sized landslides becomes more difficult. This could explain their very rare occurrence at high altitudes and the need to erode the high summits through rare but large deep-seated rockslides.

The above considerations on the evolution of glaciated high peaks allow us to revisit the concept of glacial buzzsaw. Although topographic data suggest that the mean elevation of mountain ranges may be directly limited by active glacial and periglacial erosion processes3,7, the glacial buzzsaw mechanism is not sufficiently effective at preventing, in areas undergoing sustained uplift rate, the emergence of high peaks >3,000 m above the snowline12, as observed in the Annapurnas massif. By contrast, once a mountain peak passes over the zone in which rock mass is permanently frozen and the intensity of frost cracking decreases substantially, the frozen conditions provide enhanced protection against surface erosion. This may even initiate a positive feedback. The increased mechanical strength of the frozen, unweathered rock mass promotes headwall stability and an increase in both their length and steepness46. In turn, the increased steepness of the headwalls that prevents the accumulation of ice and glacial erosion11, combined with headwall lengthening12, allows the growth of high peaks and their further protrusion into the realm of permanently frozen and muted periglacial erosion conditions (Fig. 5). This positive feedback has a limit, however, set by the cohesion of the deep unweathered rock mass, and our observations suggest that deep-seated rockslides are, ultimately, the controlling mechanism that limits the persistent growth of the high Himalayan glaciated summits (Fig. 5).

Fig. 5: Evolutionary sketch of the high peak emergence well above the ELA in a glaciated landscape undergoing sustained rock uplift.
figure 5

a–c, Starting from a fluvially dominated landscape (a) raised above the ELA (dotted white line), glacier development and subglacial erosion leads to lengthening and steepening of headwalls (b and c). c,d, Whereas subglacial erosion keeps much of the relief below the ELA (that is, buzzsaw effect3) and frost-cracking erosion above the upper glaciated rim steepens the headwalls, further rock uplift drives the summit into the zone in which the rock mass becomes permanently frozen and undergoes minimal surface erosion. d,e, The continued growth of the peak elevation ends in a marked gravitational instability (d and e), that is, a giant rockslide that beheads the summit and results in a thick rock-avalanche deposit (yellow shading in e).

Full size image

Such a conceptual scenario (Fig. 5) might operate on both regional (central Himalaya) as well as global scales. Many large-scale tectonic zones undergoing local sustained uplift rate present high peaks that protrude well above the snowline (that is, by 2,500–4,000 m) in the Himalayan syntaxes (Nanga Parbat and Namcha Barwa), the Karakoram, the Pamir, the western Kunlun Cordillera, the Gongga Shan, the central Tian Shan (Jengish Chokusu and Khan Tengri massifs) or far from the India–Asia collision, for example, in the Denali or Saint Elias massifs in Alaska. Future work should assess the erosive contribution of giant rockslides on the long-term topographic evolution of these regions and also quantify a probable relationship between the rock uplift rate and the amplitude of relief above the snowline50.

Methods

14C dating

Sixteen organic fragments were collected from three distinct areas for 14C dating: seven wood-fragments samples taken in the breccia deposit at Karuwa (K site in Fig. 1), eight samples of charcoal and wood fragments in the southern part of the Pokhara basin (P site) and one microcharcoal sample found in Pokhara-type sediments outcropping at the range outlet (site N in Fig. 1b).

Organic fragment preparation includes standard cleaning and ABA treatment and combustion at 850 °C, as described in ref. 51. The CO2 was converted to graphite and analysed on the LMC14 AMS facility52 ‘ARTEMIS’, except for one sample prepared and analysed at the National Ocean Sciences Accelerator Mass Spectrometry (NOSAMS) facility at the Woods Hole Oceanographic Institution.

Radiocarbon ages are calculated53 by correcting the fractionation with the δ13C calculated from the 13C/12C ratio measured on ARTEMIS. We used the IntCal13 curve54 for calibrating our 14C ages to calendar years using the OxCal online calculator (https://c14.arch.ox.ac.uk/oxcal/OxCal.html) (Supplementary Table SI-10-1).

36Cl exposure age dating

We sampled plurimetric blocks of Siluro-Devonian black shale and calcschists, deposited on top of the main northern remnant of the original Sabche breccia surface. Samples were prepared at the Centre Européen de Recherche et d’Enseignement en Géosciences de l’Envi (CEREGE, Aix-en-Provence, France), following standard procedures55,56. Cl and 36Cl concentrations were obtained on each sample by isotope dilution accelerator mass spectrometry at ASTER-CEREGE. Major and trace elements were measured at the French National facility for geochemical analysis (SARM-CRPG) (Supplementary Table SI-11-3).

36Cl production rate was calculated according to concentration in target elements (Supplementary Table SI-11-2) and Lal/Stone time-dependent scaling using the CRONUS online calculator (http://hess.ess.washington.edu/math/index_dev.html). Several blocks made of finely schistosed rocks show evidence of frost cracking and physical weathering, but the shape of the original surface is generally still identifiable and indicates a maximum surface erosion of roughly 10 cm (Supplementary Table SI-11-1). A further correction for snow shielding was considered from snow-cover measurements on the Himalayan southern flanks 25 km to the east57. Snow and erosion corrections represent approximately 25 and 40 years, respectively.

IRSL dating on sheared breccia deposits

TSS carbonate units may contain small amounts of quartz and feldspar grains. The luminescence signal can be reset (called triboluminescence) by heating, mechanical crushing or shearing35 processes in intense sheared bands such as frictionnites. In the southern part of the rockslide deposit, we collected three samples from two sites approximately 50 m apart within an internal basal shear zone (CA-13-273, CA-13-274 and CA-13-283; Extended Data Fig. 6). Collected samples measured 15 × 12 × 7 cm and 7 × 7 × 5 cm, respectively. Sample preparation followed the procedure described in ref. 58 to isolate K-feldspar fraction. Bulk samples were sent to Actlabs or SARM-CRPG for major-element and trace-element analysis, used to estimate (see Supplementary Information SI-12 for full details) the total dose rate with the online Dose Rate and Age Calculator59.

Luminescence measurements have been performed at the University of Lausanne (Switzerland) using a post-infrared IRSL protocol60 at 50 °C (IR50) and 225 °C (pIR225), as described in Supplementary Information SI-12. Dose-equivalent distributions were statistically analysed61 using the Central Age Model (CAM) and also the Minimum Age Model (MAM) for sample CA-13-283, which best represents luminescence reset during shearing (Extended Data Fig. 6). Final ages were calculated with individual fading corrections62.

Volume of the rockslide deposit

To calculate the total volume of the initial rock-avalanche deposit in the Sabche cirque and the upper Seti (Fig. 3), we consider the difference between the initial surface of the breccia deposit and the original topographies of the cirque bottom and upper Seti valley before the collapse (Extended Data Fig. 4).

The original cirque surface is based on the recognition from air views and satellite images of the outcropping substrate areas beneath the deposits (Fig. 1 and Extended Data Fig. 4a). These outcrops spread outwards from near the cirque centre (and after removing the ice cover in the northeast part of the cirque) and permit a rough interpolation of a smoothed initial surface. It is difficult to assess the second-order relief (mostly small ridges separating glacier streams, considering by equivalence the cirque of the Annapurna Sanctuary located just west of that of Sabche). Introducing an a priori moderate relief carved by glaciers suggests an uncertainty on the volume reconstruction on the order of +3/−2 km3. Also on the basis of observations made in the Annapurna Sanctuary cirque, the volume of moraines and ice that could have been entrained by the rock avalanche was estimated to be <0.1 and 0.5 km3, respectively.

The initial top surface of the rockslide deposit has been estimated on the basis of a few remnants in the northern and northwestern parts of the cirque (Extended Data Figs. 4b and 5a) and on top of the ridge along its southern part (Extended Data Fig. 2f). These surfaces are all located at an altitude of about 4,600 m but are not sufficient to constrain the geometry of the deposit surface in its central part, in which the breccia penitents left by erosion provide only a minimum elevation (Supplementary Fig. SI-4-1c). Nevertheless, literature evidence from the geometry of large collapses characterized by frontal confinement, such as the Sabche one, suggest that this central part was probably at an elevation above 4,600 m (Supplementary Information SI-4). The present course of the glacier descending from Annapurna III, which has been diverted westward, also requires the presence of an initially convex surface of the rock-avalanche deposit (Fig. 1). Despite uncertainties related to secondary hummocky features, a central convexity or a downward transition between the cirque and the upper Seti valley, the mean rock-avalanche surface can presumably be approximated by a nearly horizontal deposit at an elevation of about 4,600 m (Extended Data Fig. 4b). We estimate that these uncertainties affect the reconstruction of the breccia volume by an error of 2–3 km3 (Supplementary Fig. SI-4-1).

Along the upper Seti, a similar approach was used to estimate the breccia volume, assuming, for the original surface of the rock-avalanche deposits, a progressive thickness decrease along the upper Seti valley down to the last breccia occurrence (Supplementary Fig. SI-1-1) and relying on the minimum thickness indicated by the relic breccia spurs.

The combination (mean squared error) of the different sources of error on the basal and summit surfaces of the breccia deposit results in a final uncertainty in the volume reconstruction of +4/−3 km3.

Finally, to derive the volume of solid rock that collapsed from the Annapurna IV summit, the breccia deposit volume was corrected for breccia porosity, measured from various rockslide breccia sampled in the Annapurna region (Supplementary Information SI-5). An average value of 15 ± 5% was found, consistent with previous estimates on rock-avalanche material63.

Original failure surface of the Sabche rockslide

The entire northeast and north-northeast area of the Sabche cirque shows numerous structural surfaces related to stratification or schistosity, and oriented at 110° N (Extended Data Fig. 1). The eastern part of the ridge joining Annapurna III and Annapurna IV (Fig. 3) shows large structural slabs with dips of about 45° to the south, that is, well oriented to represent slip planes. Below Annapurna IV, the cliffs are characterized by geometric breaks (prisms, dihedrals) that connect these structural surfaces to irregular surfaces oriented at approximately 0° N. In contrast with slopes that dominate the western or eastern sides of the cirque, these cliffs do not show any obvious trace of glacial or avalanche erosion, that is, any trace of channelizing forms. This absence suggests that their surfaces were created by recent fracturing and most probably correspond to the rockslide failure surface.

To corroborate these observations in a more objective way and to draw the a priori extension of the failure surface, a morphological criterion is defined by looking at the transverse curvature (Extended Data Fig. 4d) of the high-resolution topography (2-m Pleiades tri-stereo-derived digital terrain model; see realization details in Supplementary Information SI-3), excluding ice and snow-pack areas. The structural surface appears with minimal curvature, whereas the slopes affected by erosion show converging thalwegs underlined by negative curvature. Using these indicators, we identified a zone of ‘fresh’ planar or prismatic surfaces just south and west of the Annapurna IV summit, distinct from the areas beyond which erosion has begun to carve out channelized or converging relief. This first area (the upper and lateral limits of which are represented by the pink line in Fig. 1c and Extended Data Figs. 1–4) is considered as delimiting the original failure surface of the rockslide.

Even if we cannot exclude the possibility that small/secondary rockslides or rockfalls would have partly rejuvenated this surface, analogous to the 5 May 2012 rockslide event64 at the southeast end of Annapurna IV, the absence of deposits on top of the northern remnant of the original surface (Extended Data Fig. 5) suggests that no notable reshaping event has affected the initial failure surface.

Palaeo-summit reconstruction

To reconstruct the palaeo-summit, three constraints must be verified: first, the restored rock volume must equal the collapsed one, that is 23 +3.5/−3 km3; second, the palaeotopography has to be on the verge of collapse; and third, the slopes and cliffs below the palaeo-summit must be mechanically stable, otherwise they would have failed before the giant mass through smaller rockslides. The failure surface follows large portions of structural surfaces associated with the local schistosity developed in marly limestone and possibly presenting lower cohesion and friction angles. Defining a safety factor (SF) associated with this failure plane is difficult in the absence of specific measurements on these planes and because of uncertainties on the exact geometry of the most basal part of the failure surface (below modern glaciers and debris). Other than mass conservation, the restoration therefore relies only on cliff average stability.

Mechanical characteristics of the Annapurna range substratum

To check the stability of headwalls below the palaeo-summit, we explored the average mechanical rock properties that ensure the stability of the regional relief in the entire high Annapurna range. Estimating the average material properties of a region can be done by considering the best-fitting strength parameters that reproduce the frequency-area distribution of landslides47. Here, in the absence of such a landslide distribution in the Annapurna region, we instead estimate the stability of the Annapurna summits by exploring the maximum value for the cohesion that allows the stability of all regional slopes. At this scale, testing slope stability through complex numerical approaches is computationally too demanding, and simplified limit equilibrium analysis, despite its limitations, is usually preferred65,66.

To calculate the spatial distribution of the SF in the Annapurna region, we have developed a MATLAB code largely adapted from previous approaches66. The slope stability analysis is based on the limit equilibrium concept for a 3D model with rotational movement on a potential ellipsoidal slip surface.

At each topographic grid point, our model evaluates the slope stability conditions for a large number of randomly selected ellipsoidal slip surfaces (Extended Data Fig. 7). These are defined by the geographic coordinates of the centre, the length of their three half-axes a, b and c, by their orientation α in the direction of the steepest topographic slope β0 and inclination β that follows the steepest slope plus or minus 15°, and by an offset of the ellipsoid centre above the terrain zc. For each failure surface, the SF is computed using the 3D sliding surface Hovland’s model67:

$$\begin{array}{c}{\rm{SF}}=\frac{{\sum }_{{\rm{slip}}{\rm{surface}}}C\,.{A}_{(x,y)}+\rho g{D}_{(x,y)}\cos ({\beta }_{{\rm{c}}})\tan ({\phi }_{{\rm{c}}})}{{\sum }_{{\rm{slip}}{\rm{surface}}}\rho g{D}_{(x,y)}\sin ({\beta }_{{\rm{m}}})}\\ {\rm{with}}\,{\beta }_{{\rm{m}}}{=\tan }^{-1}(\tan ({\beta }_{{\rm{c}}})\,.\,\cos ({\alpha }_{{\rm{c}}}-\alpha ))\,{\rm{and}}\,A(x,y)=\frac{1}{\cos ({\beta }_{{\rm{c}}})}\end{array}$$
(1)

in which the summation sign corresponds to the pixel summation of the resisting and driving forces (projected on each surface element) applied on each vertical rock column of thickness D and A is the 3D area of the slip surface for the considered pixel, ρ the rock density, C the apparent cohesion of the substrate, φ its internal friction angle, βc and αc the local slope and aspect, respectively, of the surface of rupture at the base of the considered column. The driving force and sliding act in the direction of α. βm is the apparent dip of the slip surface at the considered column in the direction of sliding. No intercolumn forces or external forces, such as seismic loading, are considered in the model.

In contrast with a previous model66, and because soil is absent above 4,500 m, we do not consider potential slip at the soil–substratum interface but assume a homogeneous substratum. Because we are not interested in exact friction and cohesion values, but only by apparent regional values that could be applied to the slopes of Annapurna IV, we do not consider either seepage forces or pore-pressure effects.

Practically (Extended Data Fig. 7), we use a brute-force procedure by exploring at regular steps different parameters. First, we iterate on the pixels of the topography. For each pixel, we iterate between the extreme geometric limits of the ground trace, calculating the average aspect and steepest slope over the given scale. For each explored ground window, we iterate on the parameters (a, b, c and β) of the ellipsoids that fit the ground trace and calculate the SF (failure surfaces showing an overhanging portion are rejected). For each pixel, the parameters giving the minimum SF value are retained and the associated potential rockslide is located at the coordinates of its centre of mass. Finally, in the result grid, if a pixel contains several centres of mass of potential rockslides, we keep only the one with the lowest SF.

Before applying our model to the glaciated Annapurna area or to the fluvially dominated region farther south (Supplementary Information SI-6), we tested it on a benchmark case and ensured that the searching procedure was finding: (1) similar solution (SF value) as previous studies65,66 and (2) optimal ellipsoidal failure geometries not too far from the semianalytic circular68 or log-spiral69 solutions. About this later point, our optimal surfaces are slightly more concave than the semianalytic solutions, but the SF values match to within 10%.

To explore the mechanic stability of the Annapurna range, we consider the steep areas within similar elevation range and lithology as the Annapurna IV collapse, that is, for elevations higher than about 4,000 m and for TSS units, excluding the flat cirque bottom (Extended Data Fig. 8). For the SF calculation, a fixed value of the internal friction angle φ = 35° (ref. 70) is considered. Only cohesion values were varied. For each cohesion value, we plot the distribution of SF values. If most prominent cliffs are near the verge of failure as expected in mountains with both high uplift and high erosion16, then the distribution mode has to be slightly above the critical value of 1, that is, for an apparent cohesion ≥1 MPa for the high relief of the Annapurna range.

Locally, the geometry of gravity instabilities is most often related to the orientation of fracture families and/or to schistosity: at the scale of our approach, such a consideration of local factors is not possible, and the explored cohesion value should be viewed as an average estimate. Furthermore, as the effects of pore pressure, seepage forces or seismic acceleration are not taken into account, this value can be seen as a minimum46 for the glaciated relief of the Annapurnas. Notably, because the internal friction angle has not been explored, the cohesion values tested in our procedure must be viewed as an apparent value. However, the precise value is of little importance insofar as we are looking for a relative criterion to test realistic topographies of the palaeo-Annapurna IV in a second step (Supplementary Information SI-7) or to compare apparent cohesion values between glaciated and unglaciated areas (Supplementary Information SI-6).

Palaeotopographic reconstruction

To explore a family of possible geometries for the Annapurna IV palaeo-summit, we generated 2,000 topographies from fixed ground support points along the northern ridge and from random paths for the crest and the southern basal footprints (see also computing chart in Supplementary Information SI-7). Whereas the elevation of the southern basal footprints is interpolated from the initial cirque surface (Extended Data Fig. 4a), the crest line elevation is randomly varied within a fixed range (the northern slopes between the present head scarp and the crest are allowed to vary between 35° and 50°, whereas the horizontal path is oriented in continuity with the present ridgelines at their eastern and western terminations). Intermediate points with randomly varying elevations are also generated before performing surface interpolation and volume calculation.

To consider a palaeotopography as acceptable, besides producing a rockslide volume between 20 and 26.5 km3, its hillslopes must be stable according to the SF criterion defined for the Annapurna region. For each randomly generated palaeotopography, we thus apply the same procedure described in the previous section (Extended Data Fig. 7) and derive the distribution of minimal SF values. For the high relief of the Annapurnas and assuming φ = 35°, all SF values were found to be ≥1 for a cohesion value C = 2 MPa. On the basis of this absolute cutoff, any palaeotopography presenting SF values lower than unity for C = 2 MPa were rejected. From the final set of acceptable geometries, we calculate a mean topography, a distribution of volumes, of crest profiles and of palaeo-summit elevations (Fig. 3d and Extended Data Fig. 9).

It should be noted that, for the modelled palaeotopographies, the optimal ellipsoidal failures predict a steeper and shallower rupture surface than that actually observed and preserved in the present topography beneath Annapurna IV. Applying Hovland’s formula67 (Eq. 1) on the actual failure surface, and exploring various sliding directions to reach a minimal stability, leads to SFfs values of 1.42 ± 0.1 for C = 2 MPa (vertical red line in Extended Data Fig. 9g), well above the collapsing threshold of 1. Although this high value results in part from the surface roughness compared with that of a smooth ellipsoid, it also reflects the likelihood that the rupture was strongly guided by existing schistosity planes, especially in the western part of the collapse. It is suggested that these schistosity planes exhibit friction angles and cohesion substantially lower than 35° and 1–2 MPa, respectively. Nevertheless, if we attempt to reduce the set of possible palaeotopographies, we could potentially add a third criterion that the SFfs along the failure surface has to be minimal or not too high: we might, therefore, preferentially select a subset of palaeotopographies with SFfs ≤ 1.42 (for C = 2 MPa), which would be more prone to collapse (dark-red histogram in Fig. 3d).