Avalanche risk evaluation: Computational intelligence and human expertise

After having threatened houses and communication ways, avalanches became a worrying risk for recreational activities in ski resorts or in the backcountry. In France, research started organizing after disasters occurred in 1970 (111 fatalities in 2 avalanches). In addition to observations, modeling experienced a strong development in order to improve risk management between 1990 and 2000: statistical and deterministic models and expert systems. Observation advances allowed producing useful risk maps, but the release time forecasting remains unsatisfactory. No model is able to systematically and accurately predict triggering times and places and avalanche sizes, probably due to the high complexity of slab triggering mechanisms, that belong to the family of so-called critical phenomena. This is why the best results were obtained so far through human experience, that combine field observations and use of simple tools for data recording and exploitation. Artificial intelligence should help experts in complex analysis, and in prevention of inevitable traps arising from subconscious.