Item

Abstract

Traditionally flood classification is used to overcome the violation of homogeneity assumption in flood frequency analysis. Further, it also paves the way for implementing process-based flood frequency analysis (FFA). However, flood classification leads to insufficient flood samples in the flood types, which causes poor distribution fitting. Thus, this study amalgamates the ideas of Flood classification, Process-based FFA, and data pooling to estimate the process-based flood return value with better distribution fitting. We propose the peak-detection flood separation algorithm to separate the flood events from the continuous discharge time series, and we classify the flood events into groups based on their hydrograph characteristics. It is followed by data pooling based on the UNprecedented Simulated Extreme ENsemble (UNSEEN) approach. Finally, we use statistical-based mixed distribution to derive the single return period value from the flood clusters. We tested the proposed methodology to one of the high-elevated sites in Germany. Results show a relative difference of 25% for the 100-year return period discharge between the classical approach and the proposed methodology. Thus our study addresses the flood inadequacy problem in the process-based FFA by exploiting the existing ensemble reforecast datasets.