A new hierarchical Bayesian approach to analyse environmental and climatic influences on debris flow occurrence

• Published in: Geomorphology
• Main Authors: Jomelli, Vincent, Pavlova, Irina, Eckert, Nicolas, Grancher, Delphine, Brunstein, Daniel
• Other Authors: Laboratoire de géographie physique : Environnements Quaternaires et Actuels (LGP), Université Paris 1 Panthéon-Sorbonne (UP1)-Université Paris-Est Créteil Val-de-Marne - Paris 12 (UPEC UP12)-Centre National de la Recherche Scientifique (CNRS), Erosion torrentielle neige et avalanches (UR ETGR (ETNA)), Institut national de recherche en sciences et technologies pour l'environnement et l'agriculture (IRSTEA)
• Format: Article in Journal/Newspaper
• Language: English
• Published: HAL CCSD 2015
• Subjects: Debris flow activity; Regional scale; Climate change; Hierarchical Bayesian modelling; [SHS.GEO]Humanities and Social Sciences/Geography; permafrost
• Online Access: https://hal.science/hal-01483500; https://doi.org/10.1016/j.geomorph.2015.05.022

[Departement_IRSTEA]Eaux [TR1_IRSTEA]RIVAGE International audience How can debris flow occurrences be modelled at regional scale and take both environmental and climatic conditions into account? And, of the two, which has the most influence on debris flow activity? In this paper, we try to answer these questions with an innovative Bayesian hierarchical probabilistic model that simultaneously accounts for how debris flows respond to environmental and climatic variables. In it full decomposition of space and time effects in occurrence probabilities is assumed, revealing an environmental and a climatic trend shared by all years/catchments, respectively, clearly distinguished from residual "random" effects. The resulting regional and annual occurrence probabilities evaluated as functions of the covariates make it possible to weight the respective contribution of the different terms and, more generally, to check the model performances at different spatio-temporal scales. After suitable validation, the model can be used to make predictions at undocumented sites and could be used in further studies for predictions under future climate conditions. Also, the Bayesian paradigm easily copes with missing data, thus making it possible to account for events that may have been missed during surveys. As a case study, we extract 124 debris flow event triggered between 1970 and 2005 in 27 catchments located in the French Alps from the French national natural hazard survey and model their variability of occurrence considering environmental and climatic predictors at the same time. We document the environmental characteristics of each debris flow catchment (morphometry, lithology, land cover, and the presence of permafrost). We also compute 15 climate variables including mean temperature and precipitation between May and October and the number of rainy days with daily cumulative rainfall greater than 10/15/20/25/30/40 mm day(-1). Application of our model shows that the combination of environmental and climatic predictors ...