Abstract

Avalanche forecasting at the regional scale requires translating spatially heterogeneous snowpack conditions into coherent hazard assessments. Distributed snowpack simulations driven by numerical weather prediction models offer spatially continuous snow and hazard information, but how this information is sampled and aggregated remains a key methodological challenge. Avalanche warning services have adopted different strategies, ranging from semi-distributed approaches to fully distributed, high-resolution simulations. However, how these choices affect the information retained for regional hazard assessment remains unclear. This study explores how different levels of spatial aggregation affect the information represented by distributed SNOWPACK simulations. Simulations are conducted across four operational avalanche forecasting regions in western Canada representing maritime, transitional, and continental snow climates: Whistler Blackcomb and Whistler Heliskiing, Glacier National Park, Mike Wiegele Heliskiing, and Banff National Park. Three configurations of HRDPS forcing (2.5 km resolution) are compared: (i) a fully distributed configuration retaining all HRDPS grid points, (ii) a semi-distributed configuration aggregating forcing within 15-km hexagons and elevation bands, and (iii) a single-point configuration representing each operation by one aggregated forcing series per elevation band. We compare 24-h new snow amounts (HN24) over the 2025/26 winter season and snowpack stability characteristics derived using QMAH and AvaPro during a selected late-December storm event. The aggregated configurations generally captured typical HN24 conditions and spatially coherent loading events but represented a narrower range of spatial variability. Information retained varied between operations and elevation bands, with local precipitation extremes increasingly lost through aggregation. During the stability case study, both aggregated configurations captured the dominant instability signal during widespread loading and followed the general stabilization trend. Differences became most relevant during stabilization, when aggregated configurations progressively lost the unstable tail of the full-grid distribution representing spatially localized instability.

The results highlight the trade-off between computational demand and retained spatial information. The 15-km semi-distributed configuration provides a promising compromise, retaining spatial information beyond a single representative profile while substantially reducing computational demand compared with the fully distributed configuration. However, the appropriate sampling density depends on the spatial variability of the operational domain and the information required for operational hazard assessment.

You can download Martin’s paper by clicking here.