Abstract

Physically based snowpack models are increasingly integrated into operational avalanche hazard forecasting, yet fragmented evaluation practices reduce confidence in their outputs. Mismatched spatial and temporal scales, heterogeneous datasets, inconsistent terminology, and custom software pipelines limit reproducibility and prevent groups from transferring findings. To address these challenges, we introduce a structured evaluation framework for snowpack models in support of avalanche hazard forecasting. The framework organizes evaluation design across four core dimensions: Context (why and where an evaluation is done), Model (how the model chain is configured), Data (which observations are used), and Analysis (how model–data comparisons are performed). Applying this framework highlights the critical need to separate physical measurement agreement (Quality) from practical forecasting utility (Value). Effective evaluation requires assessing each stage of the model chain sequentially, aligning spatial and temporal scales, and reporting performance for specific scenarios rather than aggregate summaries. We translate these insights into actionable guidance for researchers, software developers, and forecasters to build trust and advance model-based decision support.

You can download Simon’s paper by clicking here.