A GB-SAR Processor for Snow Avalanche Identification

An algorithm for the automatic detection and classification of snow avalanches has been developed and assessed through the archive of synthetic aperture radar (SAR) images acquired during six winter campaigns with a ground-based linear SAR. The scheme, based on the following classic steps: 1) detection; 2) features extraction; and 3) object classification, is fully described in this paper, representing the first attempt to implement a semiautomatic near-real-time operational snow avalanche monitoring tool by SAR. Detection of possible avalanche events is performed by the combined application of thresholding and morphological filters to the differential coherence of consecutive images. Classification of events likely to be snow avalanches is based on statistics extracted from the whole image and features associated to the single regions. Tests have been conducted over more than 60 000 images. Results show a drastic reduction on the images to manually supervise (2.2%). With a 9% false-negative rate and 60% accuracy over the 2.2% of images to examine, the processor represents an interesting support tool for the daily operations of avalanche risk assessment in a commercial ski resort.