When is a rock slope going to fail?

Forecasting the time of an imminent slope failure is still a very challenging task, even with constantly improving monitoring techniques. How can we get reliable prospective forecasts despite the diverse geology and limited number of events?
When is a rock slope going to fail?

Why forecasting failure time?

During the last years, the focus of my work in the AlpSenseRely project was the real-time monitoring of a quite spectacular rock slope instability at a summit called Hochvogel (Leinauer et al. 2020). A wide-open crack is dividing this 2,592 m high mountain into a stable side and a 200,000-600,000 m³ big instability. Since we started working on that site, it drew much more media attention than we expected and actually everyone was sooner or later asking the same question:

When is the Hochvogel going to fail?

I personally think that I will outlive this slope instability as it looks now, but of course we want to know better. Very soon we concluded that we cannot forecast a failure time before a sustainable acceleration appears (e.g. Intrieri et al. 2019). However, if we start thinking about how to forecast the Hochvogel failure best once such an acceleration appears, it will be too late. We need to have something ready in the drawer that we can use with our high-resolution data.

Johannes on the way up to the Hochvogel summit (left in the background). Foto: Riccardo Scandroglio.
Johannes on the way up to the Hochvogel summit (left in the background). Foto: Riccardo Scandroglio.

The inverse velocity method

The established method to forecast slope failures is the inverse velocity method coming from the 1960s to 1980s (Saito 1969, Fukuzono 1985, Voight 1989). Later, it could be shown that very catastrophic slope collapses like the famous Vajont case could have been forecasted if the deformation data are analysed and interpreted accordingly (e.g. Kilburn & Petley 2003). The most challenging and tricky part, however, is to make trustworthy predictions prospectively before the failure happens. Clearly not an easy task (Kristensen et al. 2021), and this remains. Modern high-resolution data additionally add the complexity of necessary data smoothing and deciding, where to start your forecast. Finally, a forecast without a measure for its uncertainty is less valuable for any decision-maker.

To overcome these drawbacks, we had the opportunity to start working with a really nice dataset from the Swiss Alps, where a block failed with a GNSS-inclinometer station on top of it. The sensor on the Grabengufer block had delivered deformation data until the minute before failure (Cicoira et al. 2022). This helped much in developing our forecasting concept. Additionally, we collected all monitoring data we could find, where the failure time was known, and the data resolution was daily or better. These data are mostly not publicly available, not documented or even restricted. Looking on how much is monitored nowadays, there must be more datasets out there and they should somehow be made available. This would be of great benefit to the community!

Jan maintaining a sensor at the Grabengufer landslide. Foto: PermaSense project.

Jan maintaining the sensor at the Grabengufer landslide. Foto: PermaSense project.

Of course, landslides are diverse. They differ in volume, material, failure process, triggers and then they are monitored by different techniques and on different locations within the landslide. It is remarkable, how well our single concept can forecast the final failure despite the given variety of failure types and sensors. This is probably because the model looks at the phenomenological effect of slope displacement (= result of all influencing effects), regardless of the underlying failure process or geology and regardless of the measuring sensor. This might also be the reason why this method can deal with a great variety of slope failures. From our study we infer that the forecasting performance is mainly dependent on the monitoring frequency and the data quality (signal-to-noise ratio).

Our concept: the prospective failure time forecast model

Here, you can see a time-lapse real-time animation of a forecast of the last three weeks before the Preonzo 2012 failure (see the paper for details in the visualization). This is, how forecasts would appear to decision-maker. Colours represent different window lengths used for the calculation of the inverse velocity and the forecasts. Life expectancies (the expected time until failure) are only calculated after the detection of an onset of acceleration. Life expectancies are plotted against the time when the forecast is made in the centre plot. The grey area represents the uncertainty of the forecasts. It is high shortly after the OOA and finally converges towards the failure. In the right boxplots, all forecasts since the OOA are included to give some additional statistical information.

Our approach can overcome the major drawbacks of current retrospective failure time forecasting methods. We can achieve reliable results with a variety of slope failure processes, volumes, materials, and sensor types. Also important is the implementation of an automatic starting point definition, if we think of the future in wireless (near) real-time monitoring: multiple sensors delivering frequent data automatically. Finally, by providing uniform forecast information including an uncertainty estimation, our concept can become a key element or booster for future reliable and quantitative real-time natural hazard management. Its application might not be limited to rock slopes, but the underlying physical principle might also be valid at earth slopes, man-made slopes, artificial structures, and glaciers, thus supporting decision-makers in a multitude of critical situations.


Cicoira, A. et al. In situ observations of the swiss periglacial environment using gnss instruments. Earth Syst. Sci. Data 14, 5061–5091 (2022).

Fukuzono, T. A method to predict the time of slope failure caused by rainfall using the inverse number of velocity of surface displacement. J. Japan Landslide Soc. 22, 8–14 (1985).

Intrieri, E., Carlà, T. & Gigli, G. Forecasting the time of failure of landslides at slope-scale: A literature review. Earth-Sci. Rev. 193, 333–349 (2019).

Kilburn, C. R. & Petley, D. N. Forecasting giant, catastrophic slope collapse: lessons from Vajont, Northern Italy. Geomorphology 54, 21–32 (2003).

Kristensen, L. et al. Movements, failure and climatic control of the Veslemannen rockslide, Western Norway. Landslides (2021).

Leinauer, J., Jacobs, B., Krautblatter, M. Anticipating an imminent large rock slope failure at the Hochvogel (Allgäu Alps). Geomechanics and Tunnelling 13, No. 6, pp. 597–603 (2020).

Saito, M. Forecasting time of slope failure by tertiary creep. Proc. 7th Int. Conf. Soil Mech. Found. Eng. Mexico 2, 677–683 (1969).

Voight, B. Materials science law applies to time forecasts of slope failure. Landslide News 3, 8–11 (1989).

Please sign in or register for FREE

If you are a registered user on Earth and Environmental Sciences Community, please sign in