Automated snow avalanche release area delineation in data sparse, remote, and forested regions

Abstract

Abstract. Potential avalanche release area (PRA) modelling is critical for generating automated avalanche terrain maps which provide low-cost large scale spatial representations of snow avalanche hazard for both infrastructure planning and recreational applications. Current methods are not applicable in mountainous terrain where high-resolution elevation models are unavailable and do not include an efficient method to account for avalanche release in forested terrain. This research focuses on expanding an existing PRA model to better incorporate forested terrain using satellite imagery and presents a novel approach for validating the model using local expertise, thereby broadening its application to numerous mountain ranges worldwide. The study area of this research is a remote portion of the Columbia Mountains in southeastern British Columbia, Canada which has no pre-existing high-resolution spatial data sets. Our research documents an open source workflow to generate high-resolution DEM and forest land cover data sets using optical satellite data processing. We validate the PRA model by collecting a polygon dataset of observed potential release areas from local guides, using a method which accounts for the uncertainty of human recollection and variability of avalanche release. The validation dataset allows us to perform a quantitative analysis of the PRA model accuracy and optimize the PRA model input parameters to the snowpack and terrain characteristics of our study area. Compared to the original PRA model our implementation of forested terrain and local optimization improved the percentage of validation polygons accurately modelled by 11.7 percentage points and reduced the number of validation polygons that were underestimated by 14.8 percentage points. Our methods demonstrate substantial improvement in the performance of the PRA model in forested terrain and provide means to generate the requisite input datasets and validation data to apply and evaluate the PRA model in vastly more mountainous regions worldwide than was previously possible.

Automated avalanche hazard indication mapping on state wide scale

“…Such maps can be calculated for any area where high spatial resolution digital elevation models (DEM), information on the snow climatology and the protective effects of the forest are 35 available. Today, such information can be deducted from satellite data (Sykes et al, 2021). Any available avalanche observations in the region are very helpful to check the validity of the results.…”

“…Furthermore, the presented approach can be applied to regions worldwide, where a high spatial resolution DTM, information on the protective effects of present forest and information on extreme snow depth increase is available. We already successfully applied this approach to the regions Trentino, Livigno, Aosta and Langtaufers in Italy (Monti et al, 2018;15 Maggioni et al, 2018;Bühler et al, 2018a) as well as regions in Canada (Sykes et al, 2021), Alaska, India, Afghanistan, and New Zealand. An adapted version of this approach was applied to simulate typical skier avalanches over entire Switzerland (Harvey et al, 2018).…”

Development and validation using ground truth of a method to identify potential release areas of snow avalanches based on watershed delineation

“…Snow avalanche PRA detection methods from terrain analysis belong to the class of susceptibility mapping 45 methods, which are widely used for many mountain hazards (Bertrand et al, 2013;Eckert et al, 2018). Several automatic snow avalanche PRA detection methods are now available in the literature, and, since first proposals (Maggioni et al, 2002;Maggioni and Gruber, 2003), different extensions have been implemented (e.g., Sykes et al, 2021). For example, while PRAs were historically assessed independently from snow and weather conditions, Chueca Cía et al ( 2014) developed a multi-criteria analysis for snow avalanche susceptibility mapping that uses 50 wind directions and snowdrift to identify PRAs in a dynamic way.…”