| dc.description.abstract | Using eCognition we developed an algorithm to automatically detect and map avalanche deposits in Very High Resolution (VHR) optical remote sensing imagery acquired from satellites and airplanes. The algorithm relies on a cluster-based object-oriented image interpretation approach which employs segmentation and classification methodologies to identify avalanche deposits. The algorithm is capable of detecting avalanche deposits of varying size, composition, and texture. A discrete analysis of one data set (airborne imagery collected near Davos, Switzerland) demonstrates the capability of the algorithm. By comparing the automated detection results to the manually mapped results for the same image, 33 of the 35 manually digitized slides were correctly identified by the automated method. The automated mapping approach characterized 201 667 m2, of the image as being representative of a fresh snow avalanche, roughly 8.5% of the image. Through a spatial intersection between the manually mapped avalanches and the automatically mapped avalanches, 184 432 m2, or 89%, of the automatically mapped regions are spatially linked to the manually mapped regions. The rate of false positive was less than 1% of the pixels in the image. The initial results of the algorithm are promising, future development and implementation is currently being evaluated. The ability to automatically identify the location and extent of avalanche deposits using VHR optical imagery can assist in the development of detailed regional maps of zones historically prone to avalanches. This in turn can help to validate issued avalanche warnings. | en_US |
