New Models Boost Avalanche Prediction Accuracy

Recent major rock-ice avalanches in the Swiss Alps - especially the dramatic collapses near Brienz and Blatten - also indicate the need for even more advanced modelling approaches. More accurate models not only help improve the understanding and prediction of such natural hazards but also support more effective and safer management of these events in populated mountain regions.

"While classical depth-averaged models are very useful for first-order estimates, many have difficulties when dealing with rugged, irregular terrain like that seen in Blatten, where flow behaviour is highly three-dimensional." That's the conclusion drawn by Johan Gaume, Professor of Alpine Mass Movements at ETH Zürich and the WSL Institute for Snow and Avalanche Research SLF. Models technically termed 'depth-averaged' do not calculate every single movement within a landslide or avalanche. Instead, they estimate an average flow - essentially describing how fast and how thick the flow is, and in which direction the entire mass moves across the ground.

In 2022, Johan Gaume, and his colleagues published a landmark scientific paper entitled Towards a Predictive Model for Alpine Mass Movements and Process Cascades. Using a newly developed 3D simulation tool, they were able to realistically reproduce several catastrophic events, including the 2017 Piz Cengalo rock-ice avalanche, the 2019 Flüela Wisshorn rock-snow avalanche, the infamous 1963 Vajont landslide, followed by a flood wave in a reservoir, and the 2020 Whymper ice-snow avalanche.

While these simulations demonstrated the capabilities of their model, they were performed after the events, making it difficult to fully claim predictive power for real-world scenarios.

Brienz - the first true test of accuracy

The real test came in 2023, when the village of Brienz was evacuated due to the imminent collapse of a large section of the mountainside above. Relying on their model, the researchers conducted so-called 'blind simulations' - meaning they ran the model exactly as originally developed, without pre-adjusting any parameters to better fit the actual landslide that followed later. This demonstrated that the model was reliable and not just customised for a single situation.

The amount of material released was estimated based on how the surface of the mountain moved over time, while the resistance to rock sliding was estimated with caution using results from rock tests. "Our simulation predicted that the resulting avalanche would stop just tens of meters short of the first houses. These results were shared informally with cantonal authorities and ultimately matched the actual extent of the real landslide very closely," says Johan Gaume.

In 2025, another critical situation emerged in May, when the village of Blatten faced evacuation due to the risk of a massive rock-ice failure. Although not officially commissioned and without direct contact with the Valais authorities, Gaume and his colleagues started conducting simulations. Their goal was to further test their predictive model in a scenario even more complex and precarious than Brienz - because in Blatten, alongside rock and water, also ice was involved, and the terrain is extremely complex.

Accurate modeling of the Blatten rock-ice avalanche.

"Given the dramatic situation in Blatten and the novelty of our modelling approach, we proceeded with great caution and subjected the model to a rigorous verification process to ensure its accuracy and reliability," Gaume points out. The researchers modelled the release of a 10-million-cubic-meter rock-ice mixture, assuming that the falling rock would either entrain or trigger the collapse of the glacier.

This volume estimate was based on expert evaluations, which placed the rock volume between 3 and 5 million cubic meters, and on the known size of the glacier, estimated at approximately 5 million cubic meters. This estimate closely matched the 9.3-million-cubic-meter rock, debris and ice volume determined in the post-event analysis by ETH Zurich and SLF glaciologists led by Daniel Farinotti (see fact sheet in ETH News from 4 June 2025).

For the friction coefficient, which characterizes sliding resistance, the researchers selected a value of 0.2 - a cautious estimate well supported by experience from past rock-ice avalanches (see Figure 1). Although this is slightly lower than the value that best matched previous real-world events (0.25), the researchers justify their choice by pointing to historical variability, noting that some past avalanches exhibited even lower sliding resistance.

The model captures shockwave effects

In recent weeks, the team revisited the initial unpublished simulations and noted that using a slightly higher friction value of 0.23 would have improved the match even further. Moreover, it turned out once again that the model is capable of realistically handling cascading processes in complex, steep topography.

"Overall, we have achieved a level of predictive accuracy that enables our model to provide more precise estimates of complex alpine mass movements in the future - both in terms of how far they may extend downslope and how much of the valley floor they might cover," Gaume states.

"We now have a reliable, ready-to-use tool that enables us to support the authorities with simulations assessing the potential consequences of impending mass movements," he adds, clarifying that these scientific simulations have not been communicated to the Valais authorities and are not part of the ongoing official investigations and risk management efforts.

As in reality, the simulation results indicate that most of Blatten is destroyed, while the neighbouring area of Weissenried narrowly escapes the falling rock and ice. The model very precisely shows a runout of the collapsed mass of 1.2 kilometers on the southwest side of the valley and 700 meters on the northeast side - values that prove to be highly accurate when compared with the actual disaster.

One key factor in the case of the Birch Glacier above Blatten was the complexity of the terrain: the rock and ice flow began in a relatively open area, narrowed dramatically, and ended in a gorge that was not aligned with the initial direction of movement. This created a shockwave effect that caused part of the descending flow mass to become airborne (as captured in videos of the event) - a phenomenon that traditional models typically fail to capture. Notably, particles were reported to have reached heights exceeding 100 metres above the terrain surface.

Widely used tools in engineering practice for modelling snow avalanches, rock avalanches and debris flows are typically based on 2D depth-averaged methods. These assume that the rock and water flow is shallow and remains in constant contact with the terrain, resulting in continuous friction. "In contrast, our 3D model allows particles to detach from the surface, reducing ground friction and accurately capturing airborne phases - this is critical for simulating flow behaviour and runout in steep or complex terrain," Gaume explains.

Towards advanced modelling in hazard management

These models provide more realistic insights into flow dynamics, impact zones, and runout distances -ultimately enabling better-informed decisions and more effective risk mitigation. "Our aim is not to replace existing 2D tools, but to offer a complementary solution where classical models may reach their limits. We are actively working to make our model accessible and usable for practitioners and authorities," Gaume explains.

"We hold the authorities in the Lötschental and in Brienz in the highest regard for the exemplary way in which they have managed - and continue to manage - the situation, and we feel deep compassion for the residents who have lost their homes and belongings," Gaume emphasises. "Tragically, the glacier collapse also claimed one life, which reminds us of the very real human cost behind these natural disasters."

This further strengthens Gaume's resolve to do everything possible to ensure that forecasting and early warning of such events becomes even more effective in the future. Looking back at the early stages of modelling the rock-ice avalanche, Gaume recalls the uneasy awe he felt when the simulations first indicated a possible destruction of the village:

"On my side, the initial results I obtained seemed rather unrealistic, particularly due to the significant upslope flow toward Weissenried. Had I had the opportunity to visit the site before running the simulation, I would likely have found these results even less plausible, given the elevation of the village relative to the Lonza. I therefore felt it was essential to discuss them with my colleagues before taking any more formal steps."

With their newly developed model, the researchers from ETH and SLF have taken an important step toward making 3D simulation tools even more accurate for future hazard assessments - especially in complex alpine environments - and, hopefully, helping to reduce the extent of damage and loss in the future.