Snow Profile Alignment and Similarity Assessment for Aggregating, Clustering, and Evaluating of Snowpack Model Output for Avalanche Forecasting

Abstract:

Detailed snowpack models that simulate the evolution of the snow stratigraphy in support of operational avalanche forecasting have been around for close to 20 years. Despite their ability to complement the traditional data streams used in avalanche forecasting, the adoption of snowpack models has been limited. Informal conversations with forecasters highlight two main issues: 1) the overwhelming volume of data produced by the models, and 2) validity concerns due to cumulative impact of potentially inaccurate weather inputs. Hence, enhancing the operational value of snowpack models critically requires the development of processing tools that reduce data volume, and allow forecasters to explore the simulations in a way that builds trust.

In this presentation, we introduce a numerical method for processing and summarizing large numbers of snow profiles that emulates the data assimilation process of avalanche forecasters. Our approach exploits Dynamic Time Warping, a well-established data science method, to align profiles by matching layers between them based on hardness, grain type and optionally deposition date. The similarity of the aligned profiles is then evaluated with an independent similarity measure that focuses on snowpack features relevant for avalanche forecasting, which creates new opportunities for automated summarizing and validating of snowpack model output. First, the similarity measure provides the necessary quantitative link to data clustering and aggregating methods, which can be used to meaningfully group and summarize snowpack information. Second, the similarity measure can be used to computationally compare model output with human snow profile observations and objectively quantify the degree of agreement.

Our algorithm aims to promote the operational application of snowpack models in avalanche warning services by providing analysis and validation tools that enable forecasters to continuously interact with snowpack simulations in familiar and accessible ways that offer direct operational value. This allows forecasters to develop an in-depth understanding of how to interpret the simulations and when to trust them, which is critical for a meaningful integration of snowpack models into operational avalanche forecasting.