Assessing snowpack stratigraphy accuracy based on different input data: Insights for operations avalanche forecasting

Avalanche forecasters and snow scientists use physically based snow stratigraphy models to fill spatial and temporal gaps in field-based snow profile observations. These models generate stratigraphy predictions using meteorological input from automated weather stations (AWS) or numerical weather prediction (NWP) models. The choice of input data is often determined by data availability or convenience instead of giving full consideration to the most appropriate source for a particular application. For example, while AWS may provide weather observations that better represent a particular site, they have large up-front costs and require specialized personnel to service and maintain. The goal of this study is to quantify the accuracy of snow stratigraphy produced by the SNOWPACK model driven by different input data, with a particular focus on cost-benefit analysis for operational avalanche forecasting. We generate modeled snow profiles at a field site in the Bridger Range of southwestern Montana, USA, using a) observations from an AWS at the field site and b) NWP output from the NOAA High-Resolution Rapid Refresh (HRRR) model. Validation data consist of a season-long time series of 10 manual snow profiles. We use dynamic time-warping (DTW) to quantify the overall and grain-type categorized similarities between modeled and in-situ observed profiles that are collocated in time and in space. Based on the similarity results, we present a cost-benefit analysis that considers the cost of installing and maintaining an AWS alongside the improved representation of snow depth, grain size, and weak layer types.