Preprints
https://doi.org/10.5194/egusphere-2024-2374
https://doi.org/10.5194/egusphere-2024-2374
06 Aug 2024
 | 06 Aug 2024

Assessing the performance and explainability of an avalanche danger forecast model

Cristina Pérez-Guillén, Frank Techel, Michele Volpi, and Alec van Herwijnen

Abstract. During winter, public avalanche forecasts provide crucial information for professional decision-makers as well as recreational backcountry users in the Swiss Alps. While avalanche forecasting has traditionally relied exclusively on human expertise, the Swiss avalanche warning service has recently integrated machine-learning models to support the forecasting process. This study assesses a random forest classifier trained with weather data and physical snow-cover simulations as input for predicting dry-snow avalanche danger levels during the initial live-testing in the winter season of 2020–2021. The model achieved ∼70 % agreement with published danger levels, performing equally well in nowcast- and forecast-mode. By using model-predicted probabilities, continuous expected danger values were computed, showing a high correlation with the sub-levels as published in the Swiss forecast. The model effectively captured temporal dynamics and variations across different slope aspects and elevations, though it decreased the performance during periods with persistent weak layers in the snowpack. SHapley Additive exPlanations (SHAP) were employed to make the model's decision process more transparent, reducing its 'black-box' nature. Beyond increasing the explainability of model predictions, the model encapsulates twenty years of forecasters' experience in aligning weather and snowpack conditions with danger levels. Therefore, the presented approach and visualization could also be employed as a training tool for new forecasters, highlighting relevant parameters and thresholds. In summary, machine-learning models as the danger-level model, often considered 'black-box' models, can provide high-resolution, comparably transparent "second opinions" that complement human forecasters' danger assessments.

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this preprint. The responsibility to include appropriate place names lies with the authors.
Cristina Pérez-Guillén, Frank Techel, Michele Volpi, and Alec van Herwijnen

Status: final response (author comments only)

Comment types: AC – author | RC – referee | CC – community | EC – editor | CEC – chief editor | : Report abuse
  • RC1: 'Comment on egusphere-2024-2374', Simon Horton, 14 Aug 2024
  • RC2: 'Comment on egusphere-2024-2374', Spencer Logan, 22 Oct 2024
  • RC3: 'Comment on egusphere-2024-2374', Karsten Müller, 08 Nov 2024
Cristina Pérez-Guillén, Frank Techel, Michele Volpi, and Alec van Herwijnen

Model code and software

deapsnow_live_v1 Cristina Pérez-Guillén, Martin Hendrick, Frank Techel, Tasko Olevski, and Michele Volpi https://gitlabext.wsl.ch/perezg/deapsnow_live_v1

Cristina Pérez-Guillén, Frank Techel, Michele Volpi, and Alec van Herwijnen

Viewed

Total article views: 329 (including HTML, PDF, and XML)
HTML PDF XML Total BibTeX EndNote
218 87 24 329 5 9
  • HTML: 218
  • PDF: 87
  • XML: 24
  • Total: 329
  • BibTeX: 5
  • EndNote: 9
Views and downloads (calculated since 06 Aug 2024)
Cumulative views and downloads (calculated since 06 Aug 2024)

Viewed (geographical distribution)

Total article views: 338 (including HTML, PDF, and XML) Thereof 338 with geography defined and 0 with unknown origin.
Country # Views %
  • 1
1
 
 
 
 
Latest update: 08 Dec 2024
Download
Short summary
This study assesses the performance and explainability of a random forest classifier for predicting dry-snow avalanche danger levels during initial live-testing. The model achieved ∼70 % agreement with human forecasts, performing equally well in nowcast and forecast modes, while capturing the temporal dynamics of avalanche forecasting. The explainability approach enhances the transparency of the model's decision-making process, providing a valuable tool for operational avalanche forecasting.