Bayesian Risk Assessment of a Tsunamigenic Rockslide at knes
Chapter
Overview
Research
Identity
Additional Document Info
Other
View All
Overview
abstract
Springer International Publishing Switzerland 2016. This chapter introduces a comparison between two methods for estimating the risk of a tsunamigenic rockslide at knes, Norway. The first method follows a classical approach based on best estimates of the risk factors (i.e., hazard, vulnerability, and elements at risk). The second method follows a more recent approach based on Bayesian networks, which introduces the notion of causal effects and defines full probability distributions for each risk factor. The Bayesian approach is thought to be more powerful in terms of number and quality of inferences. It allows for conducting diagnosis and prognosis risk assessment analyses and it traces the influence of new evidence as it becomes available, either from experimental observations, model predictions, informed expert beliefs, a combination of them, or even interventions in the model to reproduce optimal decision-making processes (e.g., by introducing the stochastic model of an early warning system). Both methods illustrate the interaction of multiple natural threats when implemented in the Storfjord area where the knes rockslide is located. Results generated from the proposed methods are based on available evidence; however a key component on both approaches is the evidence assimilation from the experts who provided technical information, but also their beliefs in terms of probability measures (i.e., informed experts beliefs). The use of informed experts beliefs introduced a unique approach for incorporating fine engineering judgment into risk measures in a systematic manner. Results obtained from each method showed significant qualitative differences in terms of inference capabilities, but in terms of the expected risk estimates, their orders of magnitude were relatively similar, which validated the state of risk at the knes rockslide.
