Bayesian inference in mass flow simulations - from back calculation to prediction

München

2017

Sonderdruck

202740

Mass flow simulations are an integral part of hazard assessment. Determining the hazard potential requires a
multidisciplinary approach, including different scientific fields such as geomorphology, meteorology, physics,
civil engineering and mathematics. An important task in snow avalanche simulation is to predict process
intensities (runout, flow velocity and depth, ...). The application of probabilistic methods allows one to develop
a comprehensive simulation concept, ranging from back to forward calculation and finally to prediction of mass
flow events. In this context optimized parameter sets for the used simulation model or intensities of the modeled
mass flow process (e.g. runout distances) are represented by probability distributions. Existing deterministic flow
models, in particular with respect to snow avalanche dynamics, contain several parameters (e.g. friction). Some of
these parameters are more conceptual than physical and their direct measurement in the field is hardly possible.
Hence, parameters have to be optimized by matching simulation results to field observations. This inverse problem
can be solved by a Bayesian approach (Markov chain Monte Carlo).