Avalanche Debris Detection Using Satellite- and Drone Based Radar and Optical Remote Sensing

Wednesday, 17 December 2014: 8:15 AM
Markus Eckerstorfer, Eirik Malnes, Hannah Vickers, Stian A Solbø and Andreas Tøllefsen, NORUT Northern Research Institute, Tromsø, Norway
The mountainous fjord landscape in the county of Troms, around its capital Tromsø in Northern Norway is prone to high avalanche activity during the snow season. Large avalanches pose a hazard to infrastructure, such as buildings and roads, located between the steep mountainsides and the fjords.

A prolonged cold spell during January and February 2014 was followed by rapid new-snow loading during March 2014, inducing a significant avalanche cycle with many spontaneous, size D4 avalanches that affected major transport veins.

During and shortly after the avalanche cycle of March 2014, we obtained 11 Radarsat-2 Ultrafine mode scenes, chosen according to reported avalanche activity. We further collected four Radarsat-2 ScanSAR mode scenes and two Landsat-8 scenes covering the entire county of Troms. For one particular avalanche, we obtained a drone-based orthophoto, from which a DEM of the avalanche debris surface was derived, using structure-from-motion photogrammetry. This enabled us to calculate the debris volume accurately.

We detected avalanche debris in the radar images visually, by applying two detection algorithms that make use of the increased backscatter in avalanche debris. This backscatter increase is a product of increased snow water equivalent and surface roughness, roughly of the order of 3 dB. In addition, we applied a multi-temporal approach by repeatedly detecting avalanche debris at different acquisition times, as well as a multi-sensor approach, covering similar areas with different sensors.

This multi-temporal and multi-sensor approach enabled us to map the spatial extent and magnitude of the March 2014 avalanche cycle in the county Troms. With ESA’s Sentinel-1 satellite, providing high-resolution, large swath radar images with a short repeat cycle, a complete avalanche record for a forecasting region could become feasible. In this first test season, we detected more than 550 avalanches that were released during a one-month period over an area of roughly 3000 km2.