Recent Developments of Avalanche Forecasting by Discriminant Analysis Techniques: A Methodological Review and Some Applications to the Parsenn Area (Davos, Switzerland)

Different statistical methods have been tested to answer the challenging problem of forecasting avalanche activity. For each approach, the theoretical background is briefly described, and the main advantages and drawbacks are discussed. The first method consists of a simple discriminant analysis applied to a sample of avalanche days against a sample of non-avalanche days. The second approach tries to take into account different types of avalanche phenomena associated with different types of snow and weather situations. It requires the development of an avalanche typology compatible with the available variables, and leads to a two-stage decision model. A given day is first allocated to a weather type, within which the proper model avalanche-non-avalanche is then processed. A third method, a local non-parametric one, consists of drawing, for the day under study and in an appropriate predictor space, its nearest neighbours from the sample file in order to get an estimate of the probability of avalanche occurrence. For each approach, the explanatory variables may be processed directly as quantitative continuous data or as qualitative categorized data. This removes the problems associated with the very asymmetric distribution of half of them, at the cost of a moderate loss of information. As a rule, the methods were calibrated and then applied to the winters 1972–73 and 1973–74 used as a test sample, thus allowing comparison of their respective potentials in operational forecast.