Abstract

Automated avalanche terrain mapping has become an established approach for communi-cating avalanche terrain exposure to public and professional backcountry users. Several groups have developed automated workflows that implement the Avalanche Terrain Exposure Scale (ATES) framework at scale by identifying start zones, simulating runouts, and assigning ATES classes. These approaches share the same class definitions but differ in how they parameterize release areas, represent forest, and translate terrain and avalanche characteristics into class ratings. This study compares four automated implementations at Connaught Creek in Rogers Pass, British Columbia, Canada. All products were generated on shared 5 m elevation and forest-cover data and compared against an ATES map produced manually and independently by three experts. Six avalanches from a high-magnitude rain-on-snow cycle, mapped by the local forecast team, provide an independent reference for the runout layers. All products show moderate to strong ordinal agreement with the expert map (quadratic-weighted kappa 0.59–0.73, which treats adjacent-class disagreements as less severe than distant ones); most differences are a single ATES class. The products differ more in the direction of disagreement than in its magnitude, reflecting the slope angle at which each escalates ratings and whether forest is used to reduce ratings. Release areas diverge most below 30°; runout layers are more similar, with typical-magnitude layers reaching 70–79% of the mapped valley-bottom avalanches and larger-magnitude layers reaching 90–98%. The closest automated product matches the level of agreement among the independent mappers themselves (exact agreement 65% versus 63%, weighted kappa 0.73 versus 0.71). The differences between each group’s map outputs are modeling choices shaped by regional snow climate and design philosophy, not errors in applying a shared ATES framework. The comparison does not rank the products; it clarifies the trade-offs so that shared, flexible workflows can support regionally appropriate maps while drawing on collective institutional knowledge.

You can download John’s paper by clicking here.