There was a major rockslide in Tyrol [1,2]. Now, it’s 2018 (okay it was in 2017) and we are still not dealing with geohazards properly and hence it they are affecting our life. Something in risk estimation and mitigation goes wrong. From personal experience I would say that Austrian regulations and building codes regarding rockfall mitigation are quite reasonable. However, some things are still not the way they should be.

Contents

Comments on the Rockslide in Tyrol

The slope was known to be unstable according to this comment by ORT.AT [3]. In addition, they faced avalanche risk.

Hence, we have to ask ourselves:

  1. could the incident be prevented with reasonable geotechnical engineering on a reasonable budget?
  2. could the consequences be minimized?

This is a rockslide not a rockfall, therefore it is much more difficult and more expensive to build measures to prevent the rock slide. Moreover, it is an avalanche zone. Hence, we would have to establish additional measures to prevent (uncontrolled) avalanche release. Avalanche mitigation by releasing them artificially.

What else could be done? The problem with geohazards is usually not assessing if something happens but when. At this point advanced and realistic modeling can help to assess it. In the case of the rockslide in Tyrol we could say that it was likely that it happens. Hence, it might be more useful on focusing on disaster handling/reduction by having some kind of measures prepared to ensure safe access after the rock slide. Hence, disaster management should focus a bit more on these things.

Terminology

Dealing with geohazards requires some basic terminology. Many people working in the field are mixing them up or use them interchangeably. Especially hazard and risk. Therefore, I define the terms in which I use them.

Hazard

The clearest definition of “hazard” is the total probability that an event of a certain size will occur. The problem is that we are dealing with poor databases and very subjective data from field work. Therefore, “hazard” can be defined as a (qualitative; non-Bayesian) likelihood.

Vulnerability

Vulnerability describes the potential damage/consequences in some monetary unit. This includes not only costs related to direct damage but loss of productivity (e.g. closed roads lead to detours) as well.

Risk

Once we quantify hazard and vulnerability it is relatively simple to estimate the risk by multiplying both:

risk = hazard x vulnerability

If we calculate the risk for many sides in our area of responsibility, we can compare them and fix the one with the highest risk rating first.

Another approach is to use risk matrices and classify vulnerability and hazard categories. This takes the fuzziness of geological data into account.

States of Mind

Whenever we are dealing with geohazards, then we are enjoying the interaction with people haven very different states of mind. Here is a list of (sarcastic) character trades I experienced in the past in the fields of geosciences, engineering and geohazard mitigation ;).

The Geologist

“It’s geology - it can be this way or another.”

“You can’t predict anything in geology.”

“What the fuck are n-dimensional probability spaces?”

“It looks dangerous.”

The Engineer

“I am happier with numbers”

“What the fuck is a probability space?”

“Let’s remove the rock slope and replace it with some concrete. I know how to maintain it (because some building code tells me what to do).”

The Landlord/the Public

“I’m living here for 20+ years. Nothing has ever happened and therefore never will happen!”

The Politician

(in addition to landlord/the public):

“Elections are coming up.”

“There are more important things to finance (infrastructure maintenance is not sexy) - elections are coming up.”

Recent Advances in Predictive Modelling

Whenever we model something we want to predict the outcome given some variations of the input. To make it short: we want to predict the future. This is a bit more challenging for geologic material but not impossible.

Rockfall modelling

Some advertising for my own work: I have a paper coming up on this. Once, it is published you’ll find it in the publications section - stay tuned.

How can we assess geohazards due to rockfalls? We have basically three choices:

First, we can do basic spatial statistics over a given area and predict the spatial probability of an occurring rockfall. This requires past events and some data on parameters influencing rockfalls. Depending on our input and our underlying model we may predict on a map (2D) or we use a DEM (Digital Elevation Model) and predict it in 3D. However, it will lead to either probabilities or likelihoods (quantitative, non-Bayesian) that a rockfall could occur at a certain place. What is missing? The extent of the event. What extent has our event? What are our transport and deposit zones? Will anything be hit?

We can assess the latter with this method if and only if we have enough data on past events. We would have to use the third approach otherwise. To sum it up: this would be more a look into a crystal ball than useful hazard assessment.

The second approach is to utilize empirical models to predict affected areas of rockfall at one slope. These models come from times when computing power was very low and rough estimates were required to make at least some form of a hazard assessment. Well, we can do this in 2D (pseudo 1D) and 3D.

The third approach is to perform numerical simulations of rockfall events. Such numerical simulations moved from 2D (pseudo 1D; profile of a slope) to 3D. Popular examples of the first are CRSP 4.0 (Colorado Rockfall Simulation Program). However, there are some important limitations. The biggest issue with rockfall simulations on profiles is that the profile has to be hand picked first. Depending on impact conditions (energy, impact angle of rock and surface, rock geometrie, surface material) rocks may deviate from the path investigated and could hit something else. Furthermore, the 2D simulations are not designed for real rock geometries but work with much simpler geometries. Therefore, rockfall simulations with 3D terrain models are much more useful. Two popular examples are CRSP-3D (CRSP 5.0) and RAMMS::ROCKFALL. With 3D rockfall simulations it is possible to “launch” rockfalls with many variations of initial conditions (e.g. initial orientation of a rock with realistic geometry). This allows for statistical analysis of the numerical simulations and provides better understanding the hazard at a slope and therefore improve vulnerability assessment. Nevertheless, there is still one thing that is missing. What happens if a falling rock fractures and the fractured rocks are breaking apart and we may end up with completely different trajectories. Such an implementation should be relatively easy considering that 3D software uses DEM. The tricky thing is to assess structural properties of a flying rock.

However, there is one BIG problem with all theses models: we predict what would happen when a rockfall occurs, but can’t predict the time of occurrence (so far). A combination of numerical models with a great amount of parameter variation, very detailed models of rock slopes where rockfalls occurred recently (before and after the event) covering all impacting parameters and a wide sensor network for real-time monitoring it should be possible to utilize deep neural networks to predict the time of rockfall events with reasonable precision.

Compared to standard models used in geostatistics AI and especially deep learning shows promising signs of changing predictive modelling completely. Perhaps this helps to solve many of the remaining problems.

References

[1] Spiegel Online (2017): Felssturz in Österreich. Dutzende Menschen von Außenwelt abgeschnitten. https://www.spiegel.de/panorama/gesellschaft/tirol-menschen-nach-felssturz-von-aussenwelt-abgeschnitten-a-1184979.html

[2] Kurier.at (2017): Felssturz in Tirol: Notwege werden eingerichtet. https://kurier.at/chronik/oesterreich/felssturz-in-tirol-notwege-werden-eingerichtet/303.760.598

[3] ORF.at (2017): Felssturz Verkettung mehrerer Umstände. https://tirol.orf.at/news/stories/2885967/