Avalanche warning requires knowledge concerning the snow characteristics and experience regarding topographic conditions. The conclusions of the different assessments are summarized amongst others in the avalanche bulletin, which includes a description of the current situation and its temporal changes as well as the prevailing danger levels and avalanche problems. The process follows a set of rules, and it is expected that machine learning can learn the process at least partially. Such an approach is implemented in the model s-now* (see contribution Fromm et al.).
A v straight forward validation using statistical methods cannot consider all features of the spatiotemporal output of the model. Therefore, a methodology is shown that allows identifying critical cases against which the quality of the model is assessed. These cases cover all avalanche problems and all danger levels. A selection of cases will be used for the discussion. Modelled avalanche danger levels and modelled avalanche problems will be shown on maps and compared with data from the bulletins. The location and the date of avalanche accidents will be the basis for detailed snow profile analysis using the output of s-now*.
