Item: A PRELIMINARY STATISTICAL ASSESSMENT OF INSTABILITY PREDICTABILITY IN OBSERVED SNOW PROFILES
-
-
Title: A PRELIMINARY STATISTICAL ASSESSMENT OF INSTABILITY PREDICTABILITY IN OBSERVED SNOW PROFILES
Proceedings: Proceedings, International Snow Science Workshop, Whistler, BC, Canada, 2026
Authors:
- Sondre Wold [ Norwegian Water Resources and Energy Directorate (NVE), Oslo, Norway ]
Date: 2026-09-28
Abstract: Observational recordings of snow conditions help avalanche forecasters assess snowpack instability over large areas. In Norway, trained observers submit snow profiles and Extended Column Test (ECT) results through the RegObs platform as part of the national avalanche forecasting service. However, observations collected under field conditions contain unavoidable noise, raising the question of how much information about snow instability is preserved in recorded snow profiles. Using six seasons of observational data from Norway, we investigate this question using ECT outcome as a proxy for snow instability. We train a suite of machine learning classifiers to predict ECT outcome from recorded profiles, using classifier performance as an empirical estimate of the predictive signal in the data. We find that models that process the snowpack layer-by-layer outperform models based on profile-level aggregates, showing that the layered structure of the snowpack carries information that can be lost through aggregation. We also provide an estimation of the association between different grain types and ECT outcome using the pointwise mutual information.
Object ID: ISSW2026_P1.38.pdf
DOI: https://doi.org/10.15788/1790098791
Language of Article: English
Presenter(s): Sondre Wold
Keywords: instability tests, sequence modeling, data validation, machine learning
Page Number(s): 1259 - 1265
-