Item: MACHINE LEARNING FOR AVALANCHE ACTIVITY FORECASTING: COMBINING THE KNOWLEDGE OF PAST AVALANCHE ACTIVITY AND SNOW MODELLING
-
-
Title: MACHINE LEARNING FOR AVALANCHE ACTIVITY FORECASTING: COMBINING THE KNOWLEDGE OF PAST AVALANCHE ACTIVITY AND SNOW MODELLING
Proceedings: International Snow Science Workshop 2024, Tromsø, Norway
Authors:
- Léo Viallon-Galinier [ Univ. Grenoble Alpes, Université De Toulouse, Météo-France, CNRS, CNRM, Centre D’études De La Neige ]
- Marianne Poussereau [ Univ. Grenoble Alpes, Université De Toulouse, Météo-France, CNRS, CNRM, Centre D’études De La Neige ]
- Pascal Hagenmuller [ Univ. Grenoble Alpes, Université De Toulouse, Météo-France, CNRS, CNRM, Centre D’études De La Neige ]
- Matthieu Lafaysse [ Univ. Grenoble Alpes, Université De Toulouse, Météo-France, CNRS, CNRM, Centre D’études De La Neige ]
Date: 2024-09-23
Abstract: Predicting avalanche activity is critical in mountainous areas to support early warning services. For this purpose, and to complement field observations, snow cover modelling provides information otherwise unavailable on the present and future state of the snowpack. We gather observations of avalanche activity, mainly from ski resorts, and use machine learning to relate the observations and modelled snow conditions. The model brings information on snowpack evolution along time, with some specific attention paid to indicators dedicated for avalanche activity while observations provide the feedback from the terrain. The relation between modelled snow conditions and avalanche activity is optimized by the machine learning model to provide relevant information on avalanche activity. We provide estimations of the avalanche activity based on locally observed avalanches but also use the relation between modelled snow conditions and avalanche activity to transfer information across space. The model could be trained in data-rich regions and then applied on areas with less information available on past avalanche activity as we try to be as close as possible to the relation between snow conditions and avalanche activity. The results were evaluated against different dataset available in France: both data from ski resorts providing daily observations of avalanche activity and avalanches recorded in pre-defined avalanche paths by forest rangers (EPA from ONF - INRAE). Such method allow to transfer the information from data-rich regions to data-poor regions and thus improve globally the knowledge of avalanche conditions. More generally, it opens perspectives to provide new tools to forecasters in charge of the redaction of avalanche bulletins.
Object ID: ISSW2024_O3.9.pdf
Language of Article: English
Presenter(s): Léo Viallon-Galinier
Keywords: Avalanche, Machine learning, Snow modelling
Page Number(s): 449 - 456
-