Click here to download Simon’s paper.

Abstract

Spatially distributed snowpack models are increasingly being adopted by avalanche forecasting agencies, but forecasters could benefit from simpler methods to interpret the complex model output. We introduce a statistical clustering technique that summarizes model output by predicting forecast regions with similar hazard characteristics. The method derives five metrics from simulated profiles to summarize different components of hazard: snow depth, new snow, wind-drifted snow, persistent weak layers, and wet snow. These metrics, along with spatial arrangement information, are combined into a distance metric that is fed into a fuzzy clustering algorithm to group small regions into larger spatially contiguous regions with distinct hazard characteristics. Our application of the method to western Canada during the 2023-24 winter produced regions that closely aligned with the regions in Avalanche Canada’s daily public forecasts. The method’s flexibility in considering various snowpack properties and spatial patterns makes it an attractive option for decision support.