Deriving probabilistic regional envelope curves with two pooling methods

摘要

A probabilistic regional envelope curve (PREC) assigns a recurrence interval to a regional envelope curve. A central point of this method is the determination of homogeneous regions according to the index flood hypothesis. A flood discharge associated with the recurrence interval (PREC flood quantile) is estimated for each gauge of a homogeneous region. In this study, the influence of two pooling methods on PREC for a large group of catchments located in the south-east of Germany is investigated. Firstly, using cluster analysis, fixed homogeneous regions are derived. Secondly, the Region of Influence (RoI) approach is combined with PREC. The sensitivity of PREC flood quantiles with respect to pooling groups is evaluated. Different candidate sets of catchment descriptors are used to derive pooling groups for both pooling methods. Each pooling group is checked by a homogeneity test. PRECs are then constructed for all homogeneous regions. The ensemble of PREC realisations reveals the sensitivity of the PREC flood quantiles. A comparison with the traditional index flood method ascertains the suitability of the pooling methods. A leave-one-out jackknifing procedure points out a similar performance of cluster analysis and RoI. Furthermore, a comparison of different degrees of heterogeneity for deriving pooling groups reveals that the performance of PREC for ungauged catchments decreases in more heterogeneous pooling groups.