Flood frequency analysis integrated with unprecedented flood samples and mixed probability distribution

Different flood-generating mechanisms are responsible for high flows in different catchments. This mixture of generating mechanisms could violate the homogeneity assumption of the extreme value distribution used often in flood frequency analysis. Thus, this study aims to classify flood samples into homogenous process-based groups and estimate the flood quantiles for different return periods. Furthermore, this study also deals with the sample inadequacy in the flood classification by pooling ensemble reforecast datasets based on the UNprecedented Simulated Extreme ENsemble (UNSEEN) approach. The Dresden gauge in the Elbe River is selected as the study site. Daily discharge data are extracted from the GRDC, and flood events are separated based on our proposed ‘Peak-identification flood separation algorithm’, which follows four steps: 1. Identification of peaks, i.e., points with a higher streamflow value than its prior and next values, 2. Pruning based on 90^th percentile threshold value, 3. Application of independence criterion, 4. Identification of flood starting and ending position. After flood separation, hydrograph features-based flood grouping and ensemble data pooling are performed. We observe the difference in the distribution characteristics of the observed in comparison to the pooled datasets. A relative difference of 0.25 (cumecs/cumecs) is noticed for the 100-year return level between observed and pooled data. As our key contribution, we address the sample mixing problem using the flood classification technique and establish the importance of data pooling.