Abstract

Snowpack models can provide detailed and continuous insight about the evolution of the snow stratigraphy in ways that are not possible with direct observations. However, the volume of data generated by the simulations can easily become overwhelming, and since simulated snow profiles are characterized by a rather complex, multidimensional data format, it is challenging to analyze the rich information manually. The available information is therefore commonly reduced to bulk properties and summary statistics of the entire snow column or individual grid points. This is only of limited value for operational avalanche forecasting where knowledge about thin, critical avalanche layers is important. In our opinion, the lack of efficient ways to access and mine large numbers of snow profiles is one of the key reasons for the limited operational use of spatially distributed snowpack simulations and ensemble systems so far.

We discuss recently developed tools for numerically processing snow profiles to make large volumes of snowpack model output more accessible for practitioners in relevant ways. This includes algorithms that compare and assess generic snow profiles by matching corresponding layers and aligning them before effectively synthesizing many profiles into a meaningful overall perspective. Our approach enables the computation of informative summary statistics and distributions of snowpack layers, as well as the dynamic clustering of profiles into groups with distinct conditions. Our algorithms are based on customized versions of Dynamic Time Warping (DTW) and DTW Barycenter Averaging (DBA), well established methods in the data sciences.