Item: AN APPLIED VALIDATION AND VERIFICATION FRAMEWORK FOR COMPUTER-ASSISTED PUBLIC AVALANCHE FORECASTING
-
-
Title: AN APPLIED VALIDATION AND VERIFICATION FRAMEWORK FOR COMPUTER-ASSISTED PUBLIC AVALANCHE FORECASTING
Proceedings: Proceedings, International Snow Science Workshop, Whistler, BC, Canada, 2026
Authors:
- McKinley Talty [ Utah Avalanche Center, Utah USA ]
- Travis J. Morrison [ Utah Avalanche Center, Utah USA ]
- Jack Eskeland [ Utah Avalanche Center, Utah USA ]
Date: 2026-09-28
Abstract: Computer-assisted avalanche forecasting models are increasingly used to support operational avalanche forecasting, yet robust and repeatable validation frameworks remain limited. This work presents a unified validation and verification system designed for operational public avalanche forecast teams to continuously validate snowpack structure, an instability model, and an avalanche danger rating model. We propose multiple verification use cases that serve as benchmarks for the validation system. Snowpack simulations are evaluated against manual snowpit observations using methods built upon dynamic time warping. Additionally, modeled temperature profiles are evaluated against measured temperatures observed throughout the snowpack. Modeled instability metrics and danger level forecasts were evaluated against backcountry observations, which were used to derive the Avalanche Activity Index (AAI) and human forecasted danger levels. AAI was calculated using the weighted sum of the frontal areas of reported avalanches. Results demonstrate that validating SNOWPACK against manual observations and measured snowpack temperatures leads to measurable comparisons in snowpack structure representation, while instability and danger level validation contextualize modeled outputs and foster a benchmark for performance. The proposed framework emphasizes relative performance, operational relevance, and repeatability rather than exact layer-by-layer model agreement, aligning model evaluation with the needs of practicing forecasters. This work provides a foundation for developing automated, scalable validation systems that support more reliable operational avalanche forecasting.
Object ID: ISSW2026_P2.3.pdf
DOI: https://doi.org/10.15788/1790098962
Language of Article: English
Presenter(s): Mckinley Talty
Keywords: Artificial intelligence, avalanche forecasting, modeling and quantitative forecasting, validation and verification
Page Number(s): 71 - 78
-