András Bárdossy, Faizan Anwar, Jochen Seidel
Traditionally, extremes are investigated using univariate extreme value statistics. This approach may be appropriate for discharges at selected sites, but not for precipitation where volumes correspond to space-time extent of the rainfall field. Further simultaneous occurrence of floods in different catchments may cause extreme floods downstream. Therefore, the investigation of the spatial extent of floods is also of great importance. The purpose of this project is to investigate spatial extent of extremes, using spatial copulas and entropy.
The dependence structure of high intensity precipitation is investigated on different temporal and spatial scales. As the spatial distribution of precipitation depends strongly on the weather situation, the investigation includes a classification of the atmospheric circulation patterns (CPs). These classifications are based on gridded sea level pressure data, and uses fuzzy rules for their definition. In order to obtain meaningful classes a semi-supervised learning procedure is used. For a given calibration period a simulated annealing based optimization of the fuzzy rules is performed so that the obtained CPs can explain both dry and wet conditions well. Validation periods are used to check the temporal transferability of these patterns. Spatial statistical investigations are conditioned on these CPs. The methodology is applied on a set of different catchments in Austria and in Germany.