Revisiting statistical-topographical methods for avalanche predetermination: Bayesian modelling for runout distance predictive distribution
Résumé
Return period is a classical tool for avalanche hazard mapping but is often poorly defined. To reduce ambiguity, high quantiles of a given quantity should be preferred. Inspired by the statistical-topographical "Norwegian" approaches and concepts developed by Ancey and Meunier, this paper presents a new method for computing the predictive distribution of snow avalanche runout distances. We evaluate the uncertainties associated with design values using a very simple propagation operator and minimal statistical hypotheses. Only release and runout altitudes are necessary, allowing the model to work with the French historical avalanche database. We propose a stochastic model flexible enough to reasonably capture avalanche data variability and to express inter-variable correlations. The Bayesian framework facilitates parameter inference and allows taking estimation error into account for predictive simulations.