Understanding the impact of terrain and avalanche characteristics on the detection of avalanche debris in Sentinel-1 imagery

Abstract:

Snow avalanches are a winter hazard that impacts infrastructure, transportation corridors, and endangers lives. Having more spatially and temporally complete datasets of avalanche occurrences is vital for avalanche research and for validation of regional avalanche forecasts. However, current datasets are typically limited to point-based field observations that are inherently limited by the need for a human observer. Satellite imagery-based identification of avalanche debris has been proposed and studied as a method to expand existing avalanche databases. Recent papers have shown that avalanches can be detected using the European Space Agency Sentinel 1 imagery but have not explored the terrain or avalanche parameters that control when and if avalanche debris can be detected with this sensor.

This study examined the effects of avalanche factors (d-size, moisture content, avalanche type, nature of triggering), meteorological history (accumulated precipitation, maximum temperature between avalanche and activity image) , and terrain characteristics (tree-cover, curvature, radar shadow and layover) on the detectability of avalanche debris in Sentinel 1 imagery. Using a set of field-observed avalanche databases from the Bridger-Teton Avalanche Center, Utah Department of Transportation, and Sierra Avalanche Center this study investigated if avalanche paths that had field recorded events had manually detectable avalanche debris in Sentinel 1 imagery. In total, ­­­­­­­250 paths, with 119 manually observed avalanches were examined in Sentinel 1 imagery. The detected vs undetected events were then separated to look for significant differences in avalanche, meteorological, and terrain variables. Significant differences between the detected vs undetected avalanches were d-size, maximum temperatures in the path, and radar layover. This study expands our understanding of when we may be able to detect avalanches using Sentinel 1 imagery and some of the limitations of this method for creating complete avalanche occurrence datasets.