G Plaut

Due to orographic effects, the European Alps are particularly sensitive to floods, in par- ticular the ones following heavy precipitation events (HPEs). During recent years, there have been numerous damaging floods in the Alps, especially on the southern flank, but also intense snowfall, with attendant avalanches and loss of life on the northern flanks. Here, we objectively classify (in the space of their leading EOFs) the large-scale circulation (LSC) selected patterns giving rise to HPEs over alpine sub-regions. The introduction of a 'classifiability index' allows one to objectively choose the best number of clusters. Strikingly robust cluster centroids are found in most cases. Some of the clusters have a high 'discriminating power': whenever a daily LSC looks similar enough to the cen- troid of one of these clusters, the probability of an HPE is highly enhanced. This suggests building downscaling algorithms and semi-empirical HPE forecast schemes. The latter would be based on a comparison between the dynamically forecasted LSCs and the cluster centroids and would benefit both from the present skill of LSC forecasts (LSC forecasts are often better than direct precipitation forecasts, especially in mountainous terrains) and from the existence of cluster centroids with high discriminating power. The downscaling algorithms could also provide accurate tools for the exami- nation of the possible small-scale consequences of global change, as simulated by GCMs at a larger scale. The question of the robustness of the cluster centroids is hence an important one and is further investigated through sensitivity experiments. Using half-period precipitation data, the same patterns are found. The precise number of EOFs kept to perform the classification is also irrelevant. Classifi- cation of HPE LSCs into clusters is also compared to alternative approaches.