Data Mining as Tool for Protection against Avalanches and Landslides

Abstract

Avalanches of snow, stone or mud (landslides) have always posed serious dangers both to individuals and to villages in ountainous regions all over the world. Many decisions to build a house or village in a certain location, or to embark on a hike or skiing tour, etc., have been traditionally influenced by the knowledge of local experts. Such local experts have learnt over generations to judge parameters that indicate potential dangers. Dramatic changes in weather patterns, often attributed to global warming, are endangering age-old knowledge parameters. It is hence essential to find new ways of determining whether a location or route is safe or has to be considered in danger, for a limited period of time or for good. We claim in this paper that the techniques to handle this new situation that have been deployed so far were often based on physics, on models of terrain, temperature, precipitation, etc. We believe that newer approaches that can be summarized under the buzz-word of data mining will gain more and more importance. We will explain this in some detail in the paper and hope to present some convincing arguments. One of the basic approaches we recommend we call “after event data mining”. To be more concrete let us consider one specific case: suppose a mud slide or snow avalanche has damaged a village in a location considered safe in the past. If analyzing the environmental data of past years (meteorological parameters as well as physical ones, like increase of deforestation) shows noticeable changes, observing analogous changes in some village still intact would be definite warning signals. They might lead to actions that preserve human lives and economic values. Note in passing that predictions based on this instrument of data mining might also have the reverse effect, e.g. demonstrating that a location unsafe before (because of the likelihood of snow avalanches) can now be considered safe (since the snowfall is dwindling in this location).