Synthetic Aperture Radar (SAR) Monitoring of Avalanche Activity: An Automated Detection Scheme

Snow avalanches are a recurring hazard during the winter months in mountainous regions such as Troms county. Monitoring of their occurrence has however, first become feasible through the launch of the Sentinel-1A and 1B Synthetic Aperture Radar (SAR) satellites which provide near-daily coverage of the area surrounding Tromso. With the large areas covered by a single SAR image and the short times between repeat acquisitions, an enormous amount of data is now available, providing an ideal opportunity for operational monitoring of avalanche activity on a global scale. Such a system requires automated detection of avalanches since it is unrealistic to perform this task manually. A test version for an automatic avalanche detection algorithm based on change detection and K-means classification methods was developed and tested on the Tamokdalen area in Troms using Sentinel-1A images at 20 m resolution. However, the algorithm was not robust under variations in snow and weather conditions between acquisition dates and as such, we have revised the algorithm to address this. In the updated version we have retained several procedures of the first version, but the main difference being the replacement of the pre-classification change detection with a post-classification change detection scheme together with improved filtering techniques. Results are shown for examples acquired from the 2017 winter season and we summarize the main improvements and future requirements of the revised detection algorithm.