Automatic detection of avalanches in high-resolution optical satellite data

We demonstrate that avalanches could be successfully detected and mapped from high resolution optical satellite imagery. The key part of the detection algorithm is a texture segmentation step, which distinguishes the avalanches from other objects such as smooth and rugged snow, trees and rock. Two different approaches are investigated: a method based on gray‐level co‐occurrence matrices (GLCM), and a method based on directional filters. To further enhance the performance we propose to process the mapped avalanche objects in feature extraction and classification stages. The algorithms are developed and trained on a Quickbird image of a Norwegian mountain area which contains several avalanches. The segmentation algorithms detect parts of all avalanches. The directional filter method was also tested and validated on another Quickbird image, covering a different scene in Norway. The GLCM approach has a higher rate of false detections than the directional filters approach, but maps the outline of the avalanches better. A brief demonstration of feature extraction shows that context and shape of detection objects may provide important information to further enhance the performance by reducing the number of false detections and refining the outline. From this case study, we believe that avalanche mapping in high resolution optical images is possible.