New statistical methodologies (to specify upper tails in flood probability distribution functions)

The occurrence of heavy tails in flood statistics is most relevant for characterising the probabilities of extreme floods. In the first phase, patterns of heaviness of tails were detected in precipitation and flood series for selected river basins e.g. Mulde and Inn (SP1) as well as for many other catchments in Germany (SP3). In the second phase, these findings will be generalized in a cooperation between SP 1, 3 and 7. The knowledge obtained in the first phase by SP1 and SP3 is condensed to the following hypotheses:

- Heavy flood tails are inherited from heavy tails of rainfall,
- The non-linear runoff generation response to precipitation creates heavy tails,
- Distinct flood types dominate the right tail,
- The mixture of flood event types generates heavy flood tails,
- Dryer catchments exhibit heavier flood tails, whereas lighter tails pertain to the wetter and snow dominated catchments,
- Larger catchments have lighter tails due to dampening of non-linear processes.

A special focus will be on the impact of flood types on the tail behaviour. The mixture model developed by SP1 will be used to quantify this impact in a spatially distributed fashion. The results will be used by SP3 to provide a linkage with the catchment state and runoff generation. For this purpose, the detected patterns of the first phase can be linked directly with the statistical quantiles. The results will be compared with the findings of homogeneous groups of gauges within the regionalisation approach of SP7. Moreover, the temporal patterns can be included in the non-stationary methods used by SP6 and 7. In SP1, the flood tail behaviour will be related to the catchment state and the runoff generation by using machine learning tools. Regional differences can be compared with the regionalisation results of SP7, e.g. the dependence on antecedent states. These considerations lead to a multivariate treatment of catchment state, precipitation and flood event characteristics. SP1 will develop critical values for multivariate models to define multivariate tail ranges.

The head of the task force will be S. Fischer (SP1), members are B. Merz (SP3), G. Blöschl (SP6) and U. Haberlandt (SP7)