Use of Expert Knowledge in Avalanche Forecasting .

Abstract

Two computerised modules for avalanche forecasting, including the results of field testing and experience with the products have been discussed. The modules refer to numerical prediction system uses discriminant analysis (both parametric and nonparametric) and Bayesian statistics to yield estimates of the probability of avalanching based on calculations in 6-7 dimensional discriminant space. Two years of field testing (1992-94) has shown that the predictive capability of the system is very consistent, an accuracy of about 80 per cent has been achieved. Rule-based expert system interprets snowpack structure from snow pit profiles. This part of the system consists of non-numerical algorithms. The system has been field-tested in four field areas during the winter of 1993-94. This has resulted in the refinement of the rules and the product is considered to be suitable for operational use.