About probabilistic graphical models in probabilistic avalanche science: The case of stop or go

• Mario Pichler
• Sebastian Eder
• Michael Larcher

In order to provide practically applicable tools for winter mountaineers to manage uncer tainties and to reduce associated risks underneath an acceptable residual risk level, several risk strategies and decision frameworks have been developed. While all of these frameworks are comprehensively described in books, booklets and small paper cards, no formal representations of this expert knowledge are available for the broad public to be exploited for a) the development of novel (interactive) training material, and b) to exemplify and allow analysis of causal relationships of decision-relevant influence factors. The current avalanche science paradigm, initiated with the Reduction Method and pursued by others (e.g. Stop or Go, SnowCard), clearly propagates a probabilistic risk management approach, however, no approaches are available that try to formally capture this expert knowledge by means of – potentially promising – probabilistic graphical models. This paper introduces a formal, computer-based representation of the “Stop or Go” avalanche risk management strategy that allows for interactive experimentation to gain a deeper understanding about mutual influences of different factors in avalanche decision making. We de-scribe the development and experiments with a Bayesian network representation of “Stop or Go”. The case of “Stop or Go” represents a first step approaching a broader formalization of avalanche expert knowledge by means of probabilistic graphical models. Such formal representations facilitate the development of interactive training material – e.g., mobile decision support apps, as done in this work – for raising the level of knowledge concerning dealing with probabilities during winter mountaineering, and allow for computer-based reasoning support.