Articles | Volume 12, issue 9
Nat. Hazards Earth Syst. Sci., 12, 2893–2906, 2012
Nat. Hazards Earth Syst. Sci., 12, 2893–2906, 2012

Research article 13 Sep 2012

Research article | 13 Sep 2012

Automated detection of snow avalanche deposits: segmentation and classification of optical remote sensing imagery

M. J. Lato1, R. Frauenfelder1, and Y. Bühler2 M. J. Lato et al.
  • 1Norwegian Geotechnical Institute, Pb. 3930 Ullevaal Stadion, 0806 Oslo, Norway
  • 2WSL Institute for Snow and Avalanche Research SLF, Flüelastrasse 11, 7260 Davos, Switzerland

Abstract. Snow avalanches in mountainous areas pose a significant threat to infrastructure (roads, railways, energy transmission corridors), personal property (homes) and recreational areas as well as for lives of people living and moving in alpine terrain. The impacts of snow avalanches range from delays and financial loss through road and railway closures, destruction of property and infrastructure, to loss of life. Avalanche warnings today are mainly based on meteorological information, snow pack information, field observations, historically recorded avalanche events as well as experience and expert knowledge. The ability to automatically identify snow avalanches using Very High Resolution (VHR) optical remote sensing imagery has the potential to assist in the development of accurate, spatially widespread, detailed maps of zones prone to avalanches as well as to build up data bases of past avalanche events in poorly accessible regions. This would provide decision makers with improved knowledge of the frequency and size distributions of avalanches in such areas. We used an object–oriented image interpretation approach, which employs segmentation and classification methodologies, to detect recent snow avalanche deposits within VHR panchromatic optical remote sensing imagery. This produces avalanche deposit maps, which can be integrated with other spatial mapping and terrain data. The object-oriented approach has been tested and validated against manually generated maps in which avalanches are visually recognized and digitized. The accuracy (both users and producers) are over 0.9 with errors of commission less than 0.05. Future research is directed to widespread testing of the algorithm on data generated by various sensors and improvement of the algorithm in high noise regions as well as the mapping of avalanche paths alongside their deposits.