Quantifying hazard and risk at Brienz/Brinzauls, as well as similar sites, requires an understanding of the full range of potential emplacement scenarios. Constraints on this can be obtained by looking at past landslide which have occurred at the site. The Igl Rutsch landslide (Fig. ) was a moderate to rapid flowlike landslide that occurred at Brienz/Brinzauls between 1879 and 1881 (Heim, 1881). Surface mapping and boreholes through this landslide (Fig. ) indicate that it initiated in the geological units of the Raibler and Allgäu formation, and transitioned from disturbed Allgäu schists in the proximal zone into clayey soil deposits at the distal toe, with thicknesses of 20.5 and 30 m. We used this surface and subsurface data, as well as a geomorphic interpretation, to reconstruct the rupture surface and pre-failure source zone geometry of the Igl Rutsch landslide (Fig. ). In particular, a flat area, on which the village is founded, can be seen to run under the deposit (Fig. ). We adjusted the present-day contours to extrapolate this flat area under the deposit, and to maintain deposit thickness estimates consistent with the boreholes. We further interpreted the pre-failure geometry in the source zone, based on the present day morphology of the head scarp. This led to a reconstructed deposit volume of 5.2 Mm3, and source volume of 4.2 Mm3, which results in a bulking factor of 1.2.

The emplacement of the Igl Rutsch landslide was documented by the Swiss geologist Albert Heim, in a report which describes the nearly 3 year emplacement process (Heim, 1881). In his report, he compares the emplacement of the Igl Rutsch landslide to the movement of a glacier, both because of the internal deformation structures that developed during emplacement, as well as the slow, flowlike movement. Albert Heim's account includes some quantitative and qualitative indications of the landslide velocity, which we interpret and summarize on Fig. . Heim (1881) describes the landslide as moving by 1 to 2 md-1 in the winter of 1878. Following this, velocities are described more qualitatively, but clearly indicate that the landslide has slowed down on average, while continuing a long-term displacement trend with seasonal cycles (interpreted here as fluctuations between ∼ 0.125 and 0.75 md-1, albeit with significant uncertainty). Finally, Heim (1881) states that by the summer of 1881 the movement rate had become imperceptible. Heim (1881) suggested that the landslide would continue to periodically move, potentially reaching the Albula river in a few hundred years. However, the landslide has remained relatively stable for the nearly 150 years since movement ceased, with modern velocities on the order of a few centimetre per year.

The Insel failure, which occurred on 15 June 2023, strongly contrasts the slow, viscous behaviour exhibited by the Igl Rutsch. Starting in 2022, the Insel compartment became distinct from its surroundings, and started to accelerate towards the village of Brienz/Brinzauls. The installed early warning system indicated that the compartment was heading towards a catastrophic failure, so in mid May, 2023 the village was evacuated (Loew et al., 2024). The compartment then failed during a short period on the evening of 15 June 2023, and reached extremely-rapid velocities which were 5 orders of magnitude greater than Igl Rutsch. Images of the deposit are shown on Fig. A and B, and test pits through the toe of the deposit are primarily composed of colluvium derived from the Allgäu formation, with a thin mantling of blocks of Vallatscha (Fig. C). The debris further upslope and in the source zone is primarily composed of colluvium derived from the Vallatscha formation (Loew et al., 2024).

The Insel failure was documented by an extensive monitoring system, which included a georadar, seismic stations, and imagery from immediately before and after the main failure. This event occurred at night time, so no optical data of the failure exists. The seismic and radar data is plotted together on Fig. . Our interpretation, supported later by numerical modelling, is that the failure mainly occurred during a 120 s period, albeit with many precursory events (i.e. small volume slides and rock fall at the toe of the failing Insel compartment). This can be seen by the strong increase in seismic normalized amplitudes, as well as the decorrelation pattern visible in the georadar data, in particular the downslope extension of the area that is decorrelated (Fig. ). The available monitoring data cannot distinguish the failure sequence within this 120 s period, and one main failure as well as two failures separated by a few tens of seconds are both viable hypotheses (tested with numerical modelling below).

High resolution topography of the site is available prior to and after the 2023 failure of the Insel compartment, however a significant volume of material is deposited on the rupture surface. For the dynamic models that follow, we used a 3D extrapolation of the interpreted rupture surface given by Loew et al. (2024). This results in a source volume of 2.1 Mm3, and a deposit volume of 2.5 Mm3.

We back-analysed the Igl Rutsch and Insel case histories in order to derive model parameters that can be used for forecasting events that have similar types. For Igl Rutsch, we used the new semi-implicit runout model (Eqs.  and ), as well as the failure geometry shown on Fig. . We used timesteps of 720 s, 4000 particles, a topographic resolution of 10 m, and performed a probabilistic calibration (described below) in order to calibrate the friction angle and viscous parameter (Eq. ). We used a numerical resolution of 4000 particles as the results are relatively insensitive to this choice (Aaron, 2023), and it keeps model runtimes low. As part of this calibration, we compared measured and simulated velocities for the event (Fig. ). The latter velocities were obtained by calculating the velocity of the simulated front as it moves over the slope, as this provides the closest simulated quantity to the velocities estimated from eye witness accounts.

For the Insel event, we again calibrated using 4000 SPH particles and a topographic resolution of 5 m, but this time used the default explicit timestepping scheme in Orin3D. For this event there is some uncertainty regarding the failure sequence, and it is possible that the main failure occurred as two events separated in time, or as one event. In the present work, we simulated two end-member scenarios: 1.

The event occurred in two separate stages, and there was no dynamic interaction between the two stages (i.e. the first stage deposits and is then overrun by the second stage). Both stages had similar volumes.

We then calibrated the model for both these scenarios using the Coulomb frictional and Voellmy rheologies (Eqs.  and ). Both rheologies were used in order to provide calibrated parameters that are comparable to past case histories. The calibration results were assessed by comparing the simulated and observed deposit distributions and source zone lithology deposit locations. This latter observation was simulated by tracking the trajectories of individual particles in the SPH simulation. We then analysed where particles, which initiate in the Valatscha and Allgäu deposit zones, are simulated to deposit, and compared this to field observations of the deposit. We note that our numerical model is depth averaged, and each particle represents a substantial volume of material. The simulated deposit distribution is therefore representative of average behaviour, and cannot resolve smaller scale details such as thin surface deposits (Fig. ).

The main goal of the back-analyses performed in the present work is to derive basal resistance parameters that can be interpreted to infer flow mechanics. It is therefore important that uncertainties in the calibrated parameters are quantified, such that they can be considered in forecasts. To do this, we used the calibration methodology that is detailed in Aaron et al. (2019), which can be used to estimate a posterior probability density function of the unknown bulk material parameters that govern the user-specified rheologies. Briefly, this methodology uses estimates of impact area, deposit distribution and velocity to calculate a model fitness number for a given input parameter set. We then run the model a large number of times on a regular grid of parameters, in order to calculate the fitness numbers for a discretization of the entire parameter space, and use these numbers and assumed standard deviations to calculate a posterior probability density function. The parameter ranges explored for the various rheologies are given in Table .

The best fit parameters for the Igl Rutsch event are summarized on Table . Overall, it was found that the available velocity data provides a strong constraint on the best fit viscous parameter, and the deposit distribution well constrains the friction angle. The best fit simulation results are shown on Fig. , where it can be seen that the overall impact area (solid black) is well reproduced by the simulation. Too much material is deposited on the proximal slope in source zone, as compared to the field estimate, and the distal deposits are generally thinner than that estimated from the pre-event topo reconstruction.

Timelapse deposit depths taken every 6 months (Fig. ) shows that the mass spreads quickly during the first year, before reaching a steady state velocity, and finally coming to rest after about 3 years. This is also shown on the simulated velocities (Fig. ), which compare well with those inferred from eye witness accounts in terms of the trend and the absolute values. Our simulations have initially high velocity values, on the order of 1 md-1, which well matches those reported by Heim (1881). They then slow down, again in accordance with the previously cited account, before movement reaches levels of a few centimetre per day, and eventually ceasing altogether.

The best fit parameters for the different failure Insel scenarios are shown in Table , and the corresponding simulation results using the frictional and Voellmy rheologies are given on Fig. . The simulated material distribution for the one and two stage results is shown on Fig. . All simulations can reproduce the observed deposit distribution reasonably well, albeit with differences in the volume of material deposited in the source zone. Furthermore, both drag laws lead to very similar results, because the best fit turbulence coefficient is high. Figure  shows that the simulation scenarios result in different distributions of the source zone lithologies, with the two-stage results simulating deposition of Allgäu schist in the source zone, and a mixing of Allgäu schist and Vallatscha dolomite at the toe, whereas the single-stage simulations predict mainly Vallatscha dolomite in the source zone, and Allgäu schist at the toe. This difference occurs because, in the one stage simulations, the upslope Vallatscha compresses the downslope Allgäu units, leading to most of the Allgäu vacating the source zone. Geological interpretations of the post-failure situation indicate that the source zone deposit is mainly Vallatscha dolomite (Loew et al., 2024), and the test pits in the debris indicate that the toe is primarily composed of highly disturbed Allgäu schist (Fig. ). This suggests that the single-stage failure scenario better matches the observed distribution of lithologies in the deposit.

Timelapse simulation results are shown on Fig.  for the single-stage frictional simulations. As can be seen, the simulation takes about 50 s to come to rest, and leaves a significant deposit in the source zone, as observed based on the field evidence. It should be noted that the thick deposit along the margins of the source zone are an artefact of the interpolation methodology used in SPH.

Numerical models of landslide motion have typically been implemented to simulate the motion of extremely-rapid, flowlike landslides (e.g. Aaron and Hungr, 2016; Bouchut et al., 2003; Christen et al., 2010; Hungr, 1995; Hungr and McDougall, 2009; Pirulli, 2005; Pudasaini and Mergili, 2019). Few researchers have attempted to simulate the motion of slower events, and those that have often use simplified 1D models that cannot account for spreading (e.g. Ranalli et al., 2010). This is likely because of the extremely small timesteps required by explicit solvers that are needed to avoid unphysical oscillations in the simulated velocities that lead to numerical instabilities. These timesteps would result in decadal to century simulation times using a CPU implementation of our numerical model.

We have overcome these numerical constraints by using a highly optimized GPU implementation of SPH, and a semi-implicit timestepping scheme. Our validation indicates that the assumptions we make in the derivation are appropriate for simulating the rheology we use to attain the moderate to rapid velocities. It is interesting to note that our simulations require the use of both implicit timestepping and GPU computing. This is because our implicit timesteps cannot get so big, due to the assumptions regarding the internal pressure distribution used to derive the governing equations. However, the use of a GPU allows for relatively small timesteps (∼ 720 s) to be used to solve 3 years emplacement duration in a few tens of minutes using a desktop GPU (Nvidia Geforce RTX 3070 Ti).

The ability of our new numerical scheme to overcome the computational challenges described above is likely due to a few unique characteristics that govern the emplacement of flowlike landslides. In particular, most source geometries feature a much larger spatial area than height, leading to relatively small internal pressure gradients (∂h∂x and ∂h∂y terms in Eqs.  and ), making it acceptable to use relatively large timesteps in the newly implemented implicit timestep scheme. With the model validated and verified (Figs.  and ), there is potential to simulate other flowlike landslides of this type such as earthflows (e.g. Aaron et al., 2021; Keefer and Johnson, 1983; Mackey and Roering, 2011; Pudasaini and Mergili, 2025). Pore pressure data is occasionally available for these other landslide types, which could justify the use of more complex, time dependent rheological parameters. This will be a subject of future work, and the present implementation fills an important existing gap in the analysis of landslide motion.

Large landslides in rock have been typically documented to experience extremely-rapid velocities (e.g. Aaron and McDougall, 2019; Coe et al., 2016; Hungr and Evans, 2004) or very slow to slow velocities (e.g. Ranalli et al., 2010; Wolter et al., 2020). However, our qualitative and quantitative analyses of the Igl Rutsch landslide clearly show that large displacement, flowlike landslides can occur in rock with velocities intermediate between these two end members. The mechanisms by which this can occur are this critical for understanding the behaviour of other, similar landslides.

Our Igl Rutsch simulation results, which are verified by the eye witness accounts, suggest that this landslide experienced significant viscous resistance during its emplacement, which lead to its low velocities and 3 year emplacement duration (Fig. ). We infer that the source of the viscous resistance experienced by this landslide is the clay-rich Allgäu Schists which are present in the source zone (Fig. ). Our back-analysis results further imply that these units can experience relatively low dynamic friction angles (∼ 23°), although these are consistent with some other reported values (Ranalli et al., 2010; Skempton et al., 1997).

The simulated mass comes to rest due to a combination of spreading (which reduces internal pressure gradients), and changes in the topographic slope angle. Remarkably, it was not necessary to account for pore pressure generation and dissipation to explain the emplacement of the Igl Rutsch failure. Most landslides with moderate to rapid velocities, such as earthflows (e.g. Mackey and Roering, 2011), are acutely sensitive to pore pressure dynamics. The fact that we could reproduce many features of Igl Rutsch without explicit consideration of pore pressure suggests its dynamics are different from these events, despite superficial similarities. However, Fig.  shows that, between Jan 1879 and July 1881, the landslide likely went through seasonal acceleration phases, which are still present in the inclinometer data measured in some of the boreholes (albeit at much smaller magnitudes). This suggests that seasonal accelerations due to pore pressure dynamics may overprint the bulk behaviour, and that this is not captured in our numerical simulations. This could potentially be addressed by employing time-varying parameters in our proposed basal drag law (Eq. ), though this would complicate the calibration procedure.

Our simulations result in too much deposition in the source zone, as compared with that inferred from the topographic reconstruction (Fig. ). This could be caused by uncertainties in the topographic reconstruction, as well as the dam break initial condition that is used in our model, whereby it is assumed that the mass is instantaneously in a state of distributed shear failure. In reality, the mass likely moved some distance as a rigid body, with internal failure and flowlike movement only occurring sometime later, along with loosening and softening of the failed material which transformed it from more intact schist into flowing soil that is rich in clay (e.g. Coe et al., 2016). Additionally, the eastern part of the moving mass is transported over stable ground, whereas the western part is transported over the moving mass of the Rutschung Dorf, which may have reworked the deposit after the event.

Our analysis demonstrates that, compared to many other large volume catastrophic rock slope failures which transitioned into rock avalanches (e.g. Aaron, 2017; Aaron and McDougall, 2019), the Insel event exhibited low mobility, consistent with that expected for a dry granular flow of rock fragments (e.g. Heim, 1932). This is demonstrated by its high H/L value (0.6), the observed limited runout beyond the toe of the slope (Fig. ), and best-fit Voellmy resistance parameters that are much more resistive than typical for rock avalanches (Aaron and McDougall, 2019). Previously, the relative low mobility of some events had been explained with reference to disintegration into a series of small volume failures over a long time (e.g. Eberhardt et al., 2004). However, our analysis of available monitoring data, as well as our simulation results, support an interpretation of the Insel event whereby the majority of the mass failed during an ∼ 2 min period starting at approximately 23:37 LT (local time). Our results further suggest that, if multiple failures happened, they were not separated substantially in time and featured significant dynamic interaction, which is necessary to reproduce the observed deposit distribution. This interpretation is consistent with the seismic data, as it shows a pronounced spike at this time, the georadar, which shows downslope decorrelation at this time, as well as the simulation results, which show a better fit to observed deposit lithologies if the event is simulated as a single slope failure. We therefore interpret the thin mantling of blocks of Vallatscha dolomite on the deposit (Fig. ) as the result of dynamic interaction and overriding of the upslope Vallatscha unit onto the downslope Allgäu unit.

Our interpretation of these results is that they provide a clear demonstration of the importance of extrinsic factors, specifically the path conditions, in governing flowlike landslide mobility. Aaron et al. (2022) recommends dividing the potential runout path into a “source” and “path”, and assessing the shear strength differently in these two zones. In the case of Insel, it appears as though no sudden weakening occurred in the source zone, which is consistent with the large pre-failure displacements. Further, the path material encountered was not composed of weak substrate, such as saturated, liquefiable material, snow or ice, which would have significantly enhanced the mobility of the event. This suggests that, if the failure had happened when the site was snow covered, or when the colluvium on site was saturated, the dynamics could have been different. The Insel failure thus serves as an important end member case history when assessing rock slope mobility in the context of a hazard analysis, as it demonstrates that large volume catastrophic rockslope failures do not necessarily exhibit excessive mobility.

Our detailed analysis of the Igl Rutsch and Insel failures has demonstrated remarkable differences between the two events, despite their source zones being located a mere 500 m apart, within the same geologic sequences and on the same slope. Whereas Igl Rutsch took 3 years to emplace, Insel took only 2 min, a difference of 5 orders of magnitude. The likely difference between these two failures is the proportion of clay-rich Schist present within the failed mass. At Igl Rutsch, the majority of the mass is in these clay-rich units (Fig. ), which likely resulted in its viscous shear behaviour. Conversely, Insel contained a large portion of brittle dolomites from the Vallatscha formation (Loew et al., 2024, p. 20), and consequently underwent a more catastrophic failure process as compared to the Igl Rutsch. Interestingly, the Insel rupture surface daylighted in clay-rich Schist, which allowed it to accommodate the observed large pre-failure displacements (> 50 m) that occurred prior to failure (Kenner et al., 2025). These results therefore highlight the critical importance of accounting for source zone lithology when understanding and forecasting the future behaviour of deforming rock slopes and their emplacement after failure. They further highlight the sensitivity to relatively minor changes in the proportion of ductile and brittle units in the source zone.