Event records and characteristics

A destabilization sequence lasting more than two weeks preceded the main collapse. Rockfalls from the Kleines Nesthorn onto Birch Glacier intensified on 14 May as reported by the local authority (Fig. 1a). Because these early events were continuously recorded with regional seismic stations, they form a unique, high-resolution dataset for precursory sequence analysis. The increased rockfall activity prompted authorities to evacuate the 300 residents of Blatten as a precaution. The deposition of debris subsequently destabilized the Birch Glacier, leading to accelerated ice flow with velocities of tens of meters per day. On 27 May, an ice avalanche involving several thousands of cubic meters, according to the Lötschental Regional Command Staff33, detached from the glacier terminus, coming to rest roughly 400 m above the Lonza River flowing along the Lötschental valley bottom (Fig. 1a).

The main collapse on 28 May mobilized ≈ 9.3 × 106 m3 of material, including ≈ 3.0 × 106 m3 of ice32, leaving a head scarp on the Kleines Nesthorn and an adjacent glacier-collapse scarp (Fig. 1b and f). The avalanche mobilized a maximum detachment thickness of 130 m at Kleines Nesthorn and 51 m at the glacier (Fig. 1c–e). Upon detachment, the rock-ice mass descended a narrow gorge (Fig. 1h), traversed the valley bottom and ran more than 200 m up the Weissenried counterslope (Fig. 1g). The final deposits on the valley floor were up to 34 m thick (Fig. 1d and g). The deposit dammed the Lonza River, forming a lake that flooded previously unaffected parts of the village.

The avalanche-generated ground motion was equivalent to an ML 3.1 earthquake, where ML is the local magnitude34. It generated strong long-period (15–180 s) surface waves that were recorded at distances exceeding hundreds of kilometers (Fig. 1j). Station LAUCH (Fig. 1a) (≈ 5 km away) recorded a signal of around 100 s duration, with an emergent onset and peak ground velocities exceeding 10−4 m/s (Fig. 1i).

Precursory destabilization sequence

We used a self-supervised clustering algorithm35 to quantify the precursory seismic sequence and constrain the nature of the events using continuous vertical-component data from station LAUCH (Fig. 1a) between 1 January and 1 June 2025, segmented into 40,574 180  s time windows. The method identified 236 clusters, 17 of which showed a clear increase in event counts (≈ 700 events, refer to “Methods”). These form two families based on their temporal evolution. Family 1 follows an S-shaped temporal pattern, with activity increasing after 19 May, peaking on 22 May, and then plateauing (e.g., Fig. 2e and g and Supplementary Fig. S2). Family 2 exhibits a gradual increase of event rate from 19 to 28 May (e.g., Fig. 2f and f and Supplementary Fig. S3). Whereas event frequency tracks the progression of instability, the total mobilized mass can evolve independently because the volume of individual events varies36. Furthermore, distinguishing the source locations is essential in this complex setting, as failures could originate from either the Kleines Nesthorn or the Birch Glacier.

Fig. 2: Seismic analysis of the precursory destabilization sequence.Fig. 2: Seismic analysis of the precursory destabilization sequence.

The left orthophotos (SWISSIMAGE 10 cm, 2024), a, b show the locations of rockfall (teal) and glacier (pink) sources. c, d Energy-duration scaling relations used to infer relative mobility for c Family 1 (rockfalls depositing on the glacier) and d Family 2 (small rock-ice avalanches). Observed seismic events (black triangles) are directly compared to granular flow simulations (colored dashed lines and scatter points). Each color corresponds to a distinct friction angle and each point represents one simulation at a specific volume. e, f Representative seismic waveforms and spectrograms for Family 1 and Family 2, respectively. g, h Temporal evolution of the accumulated volume and event number of Family 1 and Family 2. The gray shaded envelopes represent the propagated uncertainty bounds for the volume estimate. Inset of (h) highlights cluster 193 with the y-axis being the individual volume, showing a clear trend of increasing individual event volume preceding the main collapse.

We used a method30,31,37 to quantify both the volume and characteristic friction of an event series based on the high-frequency seismic signals coupled with numerical simulations. This is because these small events do not produce low-frequency waves that can be reliably resolved at distant stations. This approach assumes that the scaling relation between radiated seismic energy (Es) and signal duration (ts) follows a power law, \({E}_{s}=\alpha {t}_{s}^{\beta }\). The exponent β serves as a proxy for flow mobility, where higher values indicate stronger energy dissipation and lower mobility (refer to “Methods”). Calculating this relation reveals different behaviors for the observed events: Family 1 shows β = 2.38 ± 0.18, whereas Family 2 gives a lower β = 1.80 ± 0.15 (Fig. 2c, d).

An equivalent power-law relation, \({E}_{p}=\alpha {\prime} {t}_{p}^{\beta {\prime} }\), links the loss of potential energy (Ep) to flow duration as shown for lab- and field-scale granular flows30,37. To recover these power laws, we performed a series of simulations using the shallow water depth-averaged numerical model (SHALTOP)38,39, which solves the shallow-flow conservation equations for dense granular flows over complex terrain17,22,38,39,40,41,42,43. To distinguish source locations, we defined two possible source areas: Kleines Nesthorn (Fig. 2a) and the Birch Glacier terminus (Fig. 2b). We simulated 55 events (500–300,000 m3) at Kleines Nesthorn and 35 events (500–30,000 m3) at the Birch Glacier terminus, as smaller volumes were reported for the glacier failures.

Assuming that the seismic duration equals the flow duration for each event, we calibrate an energy-conversion ratio \({R}_{ps}=\frac{{E}_{p}}{{E}_{s}}\) by matching the duration-energy scaling exponents (β) between observations (\({E}_{s}\propto {t}_{s}^{\beta }\)) and SHALTOP simulations (\({E}_{p}\propto {t}_{p}^{\beta }\)). The β values are then derived from simulations using a constant Coulomb friction rheology with friction angles δ ∈ [10°, 20°, 30°] over a range of event volumes and for two source locations. For Kleines Nesthorn, the β values are 2.07 ± 0.24, 2.33 ± 0.20, and 3.01 ± 0.27 (Fig. 2c). For Birch Glacier terminus, they are 1.78 ± 0.15, 2.10 ± 0.19, and 2.46 ± 0.21 (Fig. 2d). Matching the seismic and simulated power laws requires the Kleines Nesthorn source and a friction angle δ = 20° for Family 1. These events are thus most probably rockfalls. In these simulations, the flow typically comes to rest and deposits on the glacier. Family 2 requires a friction angle of δ = 10° and events originating from the glacier terminus, thus corresponding to small rock-ice avalanches. Because SHALTOP computes flow kinematics independent of material mass, its raw output represents kinematic potential energy per unit density. To plot this simulated energy (Ep) in absolute Joules (Fig. 2c, d), we scaled the raw outputs by the mean bulk density of each respective family (refer to “Methods”). Because density acts as a scalar multiplier, this unit conversion has no effect on the calculated scaling exponents (β).

We compared the constant friction results against the empirical volume-dependent law \(\mu (V)=\tan \delta ={V}^{-0.0774}\), which is calibrated for standard dry granular mass movements42. For Family 1, our best-fit constant friction δ = 20° matches the μ(V) equivalent friction angle only at the maximum simulated volume of V = 300,000 m3 (\(\arctan (300,00{0}^{-0.0774})\approx 20.{6}^{\circ }\)). This implies that the smaller precursors were more mobile than expected for typical dry granular flows, likely due to pre-existing moisture in the release mass and/or the presence of rainwater as over ≈ 70 mm of rainfall accumulated in May (before 28 May)44. Furthermore, for this family, the simulated β exponent predicted by using the μ(V) law in the simulations across the 500–300,000 m3 volume range (2.73 ± 0.25) overestimates the observed β (2.38 ± 0.18), which confirms that these events are not dry31. For Family 2, the observed β (1.80 ± 0.15) is also substantially smaller than the μ(V) prediction for its 500–30,000 m3 volume range (2.71 ± 0.20), requiring a much lower friction angle of 10° as found with the constant Coulomb friction simulations. Recent numerical simulations by ref. 45 demonstrate that adding 70% ice to a granular mixture can reduce its friction angle by half compared to pure rock. Therefore, these Family 2 events are likely small rock-ice avalanches comprised of glacier ice and deposited debris from the slope.

We estimated the individual release volume by linking its radiated seismic energy to the physical loss of potential energy (refer to “Methods”). By accounting for variations in source mass density and energy conversion ratios, we established confidence envelopes for these volume estimates (refer to “Methods”). The cumulative seismic release for Family 1 leads to a cumulative volume reaching 3.91 × 106 m3 (uncertainty range: 3.54 × 106 to 4.77 × 106 m3) by 28 May. The S-shaped growth curve indicates rapid mass wasting from the Kleines Nesthorn of 3.75 × 106 m3 (uncertainty range: 3.40 × 106 to 4.57 × 106 m3) released by 23 May (Fig. 2g), aligning closely with the 3.50 × 106 m3 mass loss estimated from digital elevation models (DEM) differencing. The cumulative volume of family 2 grew steadily , reaching 1.09 × 106 m3  (undertainty range: 0.95 × 106 m3 to 1.32 × 106 m3) by 28 May. This estimate represents the total mobilized volume of small rock-ice avalanches, including both glacier ice and previously deposited rock debris. Therefore, the corresponding ice volume was likely substantially smaller, probably on the order of several 105 m3, depending on the rock-ice fraction. Within Family 2, Cluster 193 highlights this trend with increasing individual event volumes preceding the final collapse (Fig. 2h, inset). However, other clusters lack this clear escalation in individual event size, indicating that their cumulative mass wasting volume was primarily driven by event frequency.

Seismic inversion

To resolve the dynamics of the main rock-ice avalanche, we inverted low-frequency (15–180 s) seismic waveforms (Fig. 3a) recorded at distances between 33 and 110 km. We used the lsforce code46,47 to estimate the three-dimensional (3D), time-varying equivalent single force exerted on the Earth’s surface by the avalanche (Fig. 3c). At long periods, this force is equal to but opposite in direction to the product of the acceleration of the center of mass of the landslide and its total mass23,24,48.

Fig. 3: Seismic force history of the 28 May 2025 main collapse.Fig. 3: Seismic force history of the 28 May 2025 main collapse.

a Seismic waveforms from two close broadband stations, LAUCH (≈ 5 km) and WIMIS (≈ 33 km), shown in three frequency bands. Roman numerals denote four key kinematic instances discussed in the main text. b Shaded relief map (swissALTI3D, 2024) of the avalanche path. The approximately inferred location of the avalanche is shown for each instant with blue arrows indicating the direction and magnitude of the force. c Three-component inverted force history. d Magnitude of the total inverted force vector. The error bands in (c) and (d) represent the uncertainty range estimated using a jackknife resampling scheme. The central solid lines in (c) and (d) denote the “best” inversion, computed using the optimal Tikhonov regularization parameter selected via the L-curve method46. The dashed line in (d) represents the mean magnitude of the jackknife ensemble. e Vertical angle and azimuth of the inverted force vector. The scattered points are color-coded with force magnitude. As the avalanche decelerates and force magnitudes approach zero towards the end of the timeframe, the markers become nearly transparent.

We interpret four phases in the force history (Fig. 3b–e). During Phase 1 (0–51 s), the mass accelerates as the force magnitude grows steadily, reaching 6 × 1010 N. The force is directed southeast as shown at t = 24 s (Fig. 3b and e), opposite, as expected, to the initial acceleration of the mass. Minimal high-frequency (> 1 Hz) energy detected at LAUCH at 5 km distance during this phase indicates limited mass fragmentation19,27,49,50. During Phase 2 (51–70 s), the mass transits the gorge and collides with the valley floor. At ≈ 51 s, the force azimuth abruptly rotates by ≈ 180° as the flow enters the narrow gorge. This reversal marks the transition from acceleration (0–II) to deceleration (IIb, 54 s) and captures the strong centripetal forces during flow confinement. The first burst of high-frequency energy (> 0.3 Hz) appears at this instant (Fig. 3a), indicating internal collisions of the rock-ice mass. As the avalanche exits the gorge and collides with the Weissenried counterslope (III, 70 s), the force reaches 5 × 1010 N and the flow radiates elevated high-frequency seismic energy. Phase 3 (70–92 s) consists of flow redirection and secondary runup. Following the impact of the avalanche with the Weissenried counterslope, the force vector swings southwest during the backflow of material toward Blatten and the lateral spreading of the mass. A secondary force peak near 92 s corresponds to a smaller runup towards the valley side of initial descent and southwestward azimuth change, consistent with observed deposit geometry and high-frequency bursts25. Phase 4 (> 92 s) describes the spreading and final mass deposition. The force magnitude decays as the bulk of the rock-ice avalanche comes to rest. High-frequency radiation ceases, indicating that energetic collisions have ended and motion is dominated by the slow spreading of the depositing flow.

Rock-ice avalanche modeling

Whereas most landslide back-analyses rely on one runout model, we used two complementary numerical runout models to test different rheologies to get model-independent insight into the flow properties. SHALTOP provided an efficient framework to explore the rheological parameter space, whereas a Material Point Method (MPM) model51,52 was used to assess the consistency of the resulting flow kinematics in 3D. The MPM model, which solves the mass and momentum balance equations in an Eulerian-Lagrangian framework and resolves the full 3D stress field, can capture detachment, internal shearing, and curvature effects beyond the reach of depth-averaged models. We modeled the long-period force histories predicted by each model simulation based on the center of mass trajectories.

The granular flow models were constrained with independent observations: (i) mapped deposit distribution and flow-related features from DEM differencing between 23 May and 29 May 2025 and orthophotos53 (Fig. 4a, b), and (ii) the inverted seismic force history (Fig. 4k). Model performance was evaluated using four control metrics (refer to Supplementary Information S1): (1) spatial overlap of observed deposit outlines and maximum spatial extent of the simulated deposits (Trimline ratio); (2) Root Mean Squared Error of deposit thickness (Thickness RMSE); (3) normalized cross correlation between the simulated and inverted force histories (Cross-corr); and (4) RMSE of simulated versus inverted force amplitudes after aligning for the maximum cross correlation (Amp RMSE). Because these four metrics possess different units and scales, direct arithmetic combination is invalid. To resolve this, we transformed each raw metric into a z-score, which quantifies standard deviations from the mean performance across all simulations54. We assigned a combined weight of 0.7 to the deposit metrics (Trimline ratio and Thickness RMSE) and 0.3 to the seismic metrics (Cross-corr and Amp RMSE). This uneven weighting is justified for two reasons. First, high-resolution topographic differencing provides direct, low-uncertainty measurements, whereas seismic inversions are indirect calculations with accumulated wave-propagation and point-source uncertainties. Second, seismic force history reflects bulk center-of-mass accelerations that are most sensitive to the highly mobile phases of the event. As the mass loses momentum, the signal-to-noise ratio drops, leaving the late-stage depositional dynamics partially unresolved.

Fig. 4: Comparison of simulation results with deposit and seismic constraints.Fig. 4: Comparison of simulation results with deposit and seismic constraints.

a Observed final deposit thickness. b Mapped flow lines based on geomorphological features from post-event (30 May, 2025) aerial view (data sources as in Fig. 1). Colors indicate relative flow sequence, from early (warm colors) to later (cool colors) stages. c–i Simulated final deposits from SHALTOP and MPM across different rheologies, each representing the specific simulation that achieved the best overall metric z-scores. Black contours delineate observed deposit outlines. j Quantitative comparison of model performance against the final deposit (Trimline ratio (Trim), Thickness RMSE (TR)) and seismic force history (Cross-correlation (CC), Amplitude RMSE (AR)) as well as the weighted z-score. The best scores among the simulations are bolded. k Comparison between the three components of the inverted seismic force and the forces predicted by each model simulation. The error bands represent the uncertainty range for seismic inversion estimated using a jackknife resampling scheme. The central solid lines in denote the “best” seismic inversion, computed using the optimal Tikhonov regularization parameter selected via the L-curve method46.

We tested different rheologies that represent distinct assumptions and complexities about frictional behavior: a constant friction Coulomb model, a two-parameter Voellmy model, volume-dependent μ(V) law and laws with variable friction coefficient: velocity (u)-dependent μ(u) law, and the μ(I) rheology, which depends on the inertial number I55. The Coulomb model assumes time-invariant basal friction. The Voellmy model adds a velocity-squared turbulent drag term to account for increased resistance at high velocity. μ(V) assumes friction is constant in time and space but varies with event volume, whereas μ(u) and μ(I) permit friction to evolve dynamically with flow velocity or inertial number (refer to Supplementary Information S2). Given its well-established integration with seismic force inversion [e.g., refs. 21,22,26,29], we used SHALTOP to perform the primary sensitivity analysis across the listed basal friction laws. Although fully coupled two-phase thermo-hydro-mechanical models could, in principle, describe pore pressure effects, low ice friction and melt-induced frictional weakening56,57,58,59,60,61, these processes, the associated material parameters, and the initial conditions remain difficult to constrain in natural terrain. Therefore, we used the aforementioned one-phase frictional models and treat μ as a parameter that is calibrated to reproduce deposit geometry and seismic data. Because the exceptionally well-documented Blatten event was captured by multiple independent observations, it provides a robust basis for determining the frictional model that best reproduces the data, and constraining its parameters, resolving the flow dynamics. The resulting value of μ reflects effective energy dissipation, with lower μ indicating frictional weakening mechanisms that enhance flow mobility.

When calibrating different models to maximize their fit to the observations, the metrics reveal distinct strengths and limitations across different rheologies (Fig. 4j, refer to Supplementary Data 1 for the full sensitivity analysis). For a release volume of ≈ 9.3 × 106 m3, the μ(V) law, empirically fitted on essentially dry landslides, predicts δ ≈ 16° with confidence intervals from 8° to 20°42. However, simulations using these values underestimate the observed runout, indicating that the Blatten event was more mobile than expected for dry landslides of comparable size (Fig. 4c). Although Coulomb rheology with δ = 8° approximates the distal extent of the deposits (Fig. 4d), it yields the poorest match for the deposit distribution. The μ(u) rheology achieves the highest performance against the seismic constraints (Fig. 4j, k). However, it fails to accurately distribute the mass and gives a high Thickness RMSE and the lowest Trimline ratio among the available models (Fig. 4f and j). The Voellmy rheology provides an intermediate fit but still underestimates lateral spreading (Fig. 4g and j). Only the μ(I) parameterization (Fig. 4e) captures the late-stage flow surge that produces the thicker deposit distribution at the southwestern margin, where friction decreases toward its lower bound μ1 during the depositional stage of the simulation (Supplementary Fig. S4). We then ran the 3D MPM model for Coulomb and μ(I) to also simulate the depth-resolved flow kinematics (Fig. 4h, i). For the Coulomb model, the best-fit basal friction angle is δ = 11.8° for the MPM simulation, which is also near the lower bound of empirical estimates for rock-ice avalanches42,62. However, similar to the SHALTOP results, this constant friction fails to reproduce the observed deposit geometry. For the μ(I) rheology, the bounds δ1 = 2° and δ2 = 12° were adopted in SHALTOP and δ1 = 1° and δ2 = 16° for MPM, where δ1 and δ2 denote the lower and upper limits of the μ(I) law. Also in the MPM simulations, the μ(I) parameterization provides a better reproduction of the observations than a constant Coulomb friction model.

Crosscutting levees and lobes on the deposit allow the reconstruction of at least four flow pulses towards the southwest margin of the deposit. We refer to these pulses as flow order 1–4, defined by their relative timing inferred from crosscutting relationships: earlier deposits are overlain and locally incised by later, channelized surges (Fig. 4b). The first pulse determined the maximum southern extent, running ≈ 80 m (vertical) up in the direction of the source area before flowing farther downstream (flow order 1 in Fig. 4b). Superimposed on this initial deposit are elongated lobes with levees (flow order 2–4 in Fig. 4b) that extend almost 2 km downstream, beyond the village of Blatten. These channelized features incised the initial deposit from the first pulse, indicating a subsequent highly mobile surge. Because a constant-friction rheology fails to capture these complex morphological features, reproducing their extended runout requires the μ(I) rheology’s extreme drop in friction. The near-zero friction required during several critical stages (Supplementary Fig. S4) strongly indicates the material underwent frictional weakening mechanisms. We explicitly state that the friction coefficient represents a calibrated effective bulk value that suffers from equifinality. This equifinality is inherent to the use of one-phase models, in which different physical mechanisms are absorbed into an effective friction term. Beyond ≈ 92 s, the avalanche’s center-of-mass decelerates and the remaining motion transitions into slower, lateral spreading, which produces weaker and more complex force signatures. This complex force history evades the seismic constraints as the low-frequency seismic signals are too weak compared to the background noise.

Comparisons at four representative times (24, 51, 70, 92 s) (Fig. 5a, b) show consistent kinematics between SHALTOP and MPM, with MPM additionally resolving the dilated flow detaching from the topography in the narrow gorge (Fig. 5c, d). The flow front impacts the gorge wall, which causes the flow to climb over it. As the valley floor turns concave-downward, the flow partially detaches from the terrain, which is captured explicitly by MPM (Supplementary Video 2). Despite the different model dimensionalities, the velocity magnitudes of SHALTOP and MPM remain in close agreement throughout the phases. The consistency between the μ(I) simulations for SHALTOP and MPM across independent metrics and their best z-score for each model demonstrates that the observed runout and dynamics are only reproduced when δ1 is exceptionally low (2°/1°). This indicates that the effective friction varied significantly throughout the flow with persistent frictional weakening.

Fig. 5: Snapshots of simulated avalanche dynamics for the μ(I) rheology.Fig. 5: Snapshots of simulated avalanche dynamics for the μ(I) rheology.

The panels compare the velocity fields predicted by a SHALTOP and b MPM at key kinematic instances corresponding to the seismic analysis in Fig. 3: acceleration (I, 24 s), gorge transit (II, 51 s), valley-floor collision (III, 70 s), and flow redirection (IV, 92 s). c Close-up of the 3D MPM simulation during gorge transit, highlighting detached materials (dashed connectors indicate the view location). The eye symbol marks the approximate viewing direction of the video frame in (d). d Frame extracted from video (courtesy of the Canton of Valais) showing the observed detached particle cloud at a similar stage of the avalanche.

Mechanisms driving avalanche mobility

One or several frictional weakening mechanisms must have acted to produce the high mobility of the Blatten rock-ice avalanche42,63,64, extending beyond the physical assumptions of dry granular flows55. However, the calibrated friction values obtained from the one-phase models should be interpreted as effective bulk parameters. Here, we therefore evaluate the potential contributions of several weakening mechanisms qualitatively, using field observations, geotechnical measurements, energy-balance estimates, and model-inferred friction patterns as supporting evidence. Their relative contributions cannot be uniquely resolved with the available observations.

A primary factor is rock-ice friction reduction. The Blatten rock-ice avalanche involved a total volume of ≈ 9.3 × 106 m3, including ≈ 3.0 × 106 m3 of ice. The substantial ice content likely favored low-friction interactions within the mixture45,65,66,67. As demonstrated by ref. 45, an ice fraction of 32% can reduce the bulk friction angle by approximately 20%. This reduction may explain the best-fit constant Coulomb angles (8°/11.8°). However, the deposit is not reproduced with constant Coulomb friction; instead, it requires a variable friction as in the μ(I) law with extreme frictional drop to near-zero lower bounds (2°/1°). Therefore, ice content alone may not fully explain the large observed mobility.

The strong shearing of the rock-ice debris under high normal stress likely also led to dynamic fragmentation, further amplified by strong impact loading against the steep, irregular topography. The fragmentation process could rapidly produce fine particles within the rock-ice mixture and enhance mobility68. As shown in Supplementary Fig. S4, these expectations agree with our low friction modeled with μ(I) parameterization during failure, gorge transit, and collision with the counterslope (instances I, II, and III in Fig. 3, respectively).

Frictional and impact heating during rapid flow may also have generated meltwater, further reducing basal resistance62. Sosio et al.69 reported calibrated bulk basal friction angles as low as 2.75° for the Sherman ice-rock avalanche with volumes of order 107 m3 that involved 2 × 106 m3 of ice. Our energy-balance estimate suggests a potential meltwater volume of ≈ 0.88 × 105 to 2.21 × 105 m3 (see “Methods”). Together with water present in the release mass (which cannot be quantified from our observations), this meltwater may have increased pore water pressures in parts of the landslide. Elevated pore water pressures can reduce internal and basal shear stresses by lowering effective stresses in the solid phase58,70.

Field surveys conducted shortly after the event documented wet deposits (Fig. 6b), but the hazard posed by potential secondary landslides prevented direct measurements of water content and liquefaction. Consequently, the possible occurrence of elevated pore water pressures can only be inferred indirectly from geotechnical analyses of the deposited sediments that we collected later. Grain-size analysis of four samples from the counterslope and main deposit (Fig. 6a) reveal a high fine material content (≈ 15% silt and clay), classifying the material as well-graded (poorly sorted) gravel with fines, and consistent with the low permeability measured in the deposits (k0 ≈ 10−12 m2) (refer to “Methods”). These fine particles likely originate from both the rock/debris mass from the Kleines Nesthorn and subglacial till from the Birch Glacier. The sediment’s texture closely resembles that of another documented fine-rich rock-ice avalanche at Flat Creek (Alaska)71, where the high fraction of fines originated from subglacial till. Such fine-grained, low-permeability material can hinder dissipation of high excess pore pressures if these were generated during flow70. Impacts with the complex topography induce shearing and compression of the material. These processes may be at the origin of rapid pore pressure generation due to shearing-induced dilatancy effects or isotropic pressurization56. Because of the low permeability, these excess pore pressures would have dissipated slowly, allowing frictional weakening to persist56. These observations are therefore consistent with conditions that promote fluid-assisted weakening. However, they are not sufficient to determine whether pore pressure effects were dominant, nor to quantify their contribution relative to other processes.

Fig. 6: Grain-size distributions and field observations of the avalanche deposits.Fig. 6: Grain-size distributions and field observations of the avalanche deposits.

a Grain-size distributions of one sample from the Weissenried counterslope (blue) and three samples from the Blatten main deposit (orange). The collected samples included coarse gravel and small cobbles, up to ≈ 5 cm, but excluded larger boulders. The grain-size distribution is compared with data from the Flat Creek glacier detachment deposits reported by ref. 71, rescaled so that the maximum measured grain size corresponds to 5 cm for comparison. b Field photograph on 28 May showing the Blatten deposit with darker zones (black arrows) indicating locally wet conditions (ETH Zürich).

Alternative physics-based multiphase models could provide further insights into the processes responsible for friction reduction, including pore pressure weakening, dynamic fragmentation, rock-ice friction reduction, and meltwater generation56,57,58,59,60,61. However, the accurate description of these processes is still an open issue, and such models require knowledge of parameters that are both highly uncertain and, in some cases, effectively unmeasurable, yet potentially exert a strong influence on the modeled dynamics56,58,61. Clear examples include the initial water content, initial dilatancy angle, and permeability evolution during the flow, all of which can substantially affect the magnitude of excess pore pressures generated during motion61.