Item: CHARACTERIZING CONSISTENCY OF SNOWPACK STABILITY ACROSS DIFFERENT OPERATIONAL MODEL CHAINS
-
-
Title: CHARACTERIZING CONSISTENCY OF SNOWPACK STABILITY ACROSS DIFFERENT OPERATIONAL MODEL CHAINS
Proceedings: Proceedings, International Snow Science Workshop, Whistler, BC, Canada, 2026
Authors:
- Martin Metzger [ ALPSolut, Livigno, Italy ]
- Michael Binder [ Avalanche Warning Service Tyrol, Innsbruck, Austria ] [ Institute of Atmospheric Physics, German Aerospace Center, Oberpfaffenhofen, Germany ]
- Martin Perfler [ Avalanche Research Program, Simon Fraser University, Burnaby, BC, Canada ]
- Christoph Mitterer [ Avalanche Warning Service Tyrol, Innsbruck, Austria ]
- Fabiano Monti [ ALPSolut, Livigno, Italy ]
Date: 2026-09-28
Abstract: Regional and site-specific avalanche forecasting increasingly relies on snowpack simulations to complement weather forecasts and sparse field observations. Their operational value, however, depends not only on the snow cover model but also on the meteorological forcing, the initialization strategy, and the postprocessing used to translate simulated stratigraphy into stability information. Retrospective evaluations quantify a model's past agreement with regional forecasts and observations, yet forecasters need an indicator of reliability during the forecasting process itself. Consistency across independent model chains may serve as such an indicator, helping to identify when simulated stability information can provide robust decision support. To investigate this, we contrast three model chains for the Euregio warning region (Tyrol, South Tyrol, Trentino): simulations forced by (i) automatic weather station measurements, (ii) numerical weather prediction (NWP) data, and (iii) NWP data but initialized with observed snow profiles. For the 2025–2026 winter we assess how forcing and initialization affect simulated stability and quantify the consistency among the chains over time. Stability is derived through two conceptually different postprocessing approaches: the physics-based SK38_CCL index, linking skier-induced failure initiation and crack propagation, and the machine-learning Punstable index, which estimates the probability of instability with a random forest trained on observed profiles. Building on the assumption that higher consistency indicates greater reliability, we identify situations in which the simulated information appears robust and periods in which chains or indices diverge. The three chains agree well and robustly capture the exceptional avalanche period of 18–25 February 2026; the profile-initialized chain, however, is reliable mainly where it is regularly refreshed with recent profiles. Consistency is largely governed by the agreement of the accumulation forcing (reflected in snow depth) and by profile freshness, and it drops transiently during rapid-loading events. These findings provide practical guidance on when modeled snowpack stability can support operational avalanche forecasts and when it should be interpreted with caution.
Object ID: ISSW2026_P2.60.pdf
DOI: https://doi.org/10.15788/1790099064
Language of Article: English
Presenter(s): Martin Metzger
Keywords: Avalanche forecasting, Snowpack stability, AWSOME, Consistency
Page Number(s): 2231 - 2238
-