Detection of avalanche debris fields in high resolution SAR imagery

Abstract

During the winter of 2013, Northern Norway saw several large avalanche episodes that caused closed roads and loss of human lives. On Senja and Kval\o ya two large avalanches were reported and images from RADARSAT-2 were ordered in ultrafine mode with the goal of detecting these large avalanches in SAR images. Later that year another two images were ordered from RADARSAT-2 that were be used as reference images.

In this thesis an algorithm for automated detection of avalanches is created and applied to the SAR images from the two islands of Senja and Kval\o ya. This algorithm trains on several possible debris fields before the best average threshold, moving window size and filter is applied to the full SAR image and an avalanche mask is created. To further remove areas that have a low likelihood of avalanche debris fields, such as lakes and very steep slopes, a slope mask is created from a Digital Elevation Model and added to the avalanche mask.

In the final results several new avalanche debris fields are discovered and verified by applying the DEM to create 3D images, and identifying surrounding terrain as probable for the occurrence of avalanches. The sites that were first chosen as possible avalanche debris fields, are still visible in the final image when average variables have been used in the filtering and thresholding process of the full SAR images.