Abstract ID: 3.13004 | Accepted as Talk | Talk | TBA | TBA

Winter sport activities taking place in unsecured mountainous terrain have gained in popularity in recent decades, but accident statistics show that these activities come with inherent dangers. Backcountry skiers in particular – who travel on unsecured slopes – voluntarily expose themselves to the risk of serious injury or death if they are caught by an avalanche. On average, 23 people die in avalanches each winter in Switzerland, and most victims trigger the avalanche themselves, highlighting the critical role of human factors in avalanche accidents. While literature shows the importance of heuristic-based decision-making in uncertain situations, it can lead to unconscious biases and systematic errors. These biases and errors are believed to be complicit in avalanche accidents, yet they are difficult to quantify or measure. We know remarkably little about the decision-making of those who are not involved in accidents. Spatially explicit user-generated content is a popular data source for studying humans in nature. Typical analysis of such data concentrates on the use of only two dimensions – space and time – exploring where and when backcountry skiers are in avalanche terrain. However, these data often also contain rich semantic data – for example in the form of textual descriptions of tours – offering valuable insights into perceptions of nature and the environment. Previous studies have shown that accident frequency varies significantly across time and space but does not always align with the baseline usage frequency. This suggests that decision making heuristics, which are shaped by perception of the current conditions and the environment, may be more or less robust according to the changing nature of avalanche problems. Understanding conditions under which decision-making heuristics are less robust therefore holds the potential to better understand behaviour in avalanche terrain and eventually mitigate avalanche accidents. Using a geo-referenced text corpus of more than 28’000 ski tour reports written by backcountry skiers since 2013 and published on a Swiss mountaineering website, we demonstrate how semantic user-generated content can be leveraged to study perception of risk and the environment in uncertain and risky conditions – and how these perceptions relate to avalanche hazard and accidents.