Event clustering approach for flood frequency analysis

Design flood obtained under stationary conditions is obsolete in capturing the modulations induced by anthropogenic climate change. This leads researchers to employ the non-stationary approach to analyze the flood frequency. In that regard, hydroclimatic connections between extreme events and large scale climate drivers can be incorporated in flood frequency analysis by conditioning the distribution parameters with the climate covariates. However, more than one mechanism can drive floods, thus consideration of flood events as one category may neglect potential links or give rise to spurious connections. This necessitates the clustering of flood events based on its flood generating mechanisms to incorporate the climate drivers better. Therefore, this study focuses on grouping flood events based on the shape of hydrographs by using the time series clustering technique, based on the concept that the statistical shape of the flood hydrograph represents the hydrologic processes over the region. Germany, divided into three different streamflow regimes & driven by multiple flood mechanisms, is considered as the study area. Climate network is employed to identify the climate drivers of each cluster. The non-stationary flood frequency analysis is then carried out on the individual clusters using the identified climate covariates. Return flood values obtained from this study’s results are compared against the traditional stationary and climate conditioned non-stationary values. Thus, this study better represents the local/climate drivers' influence on the flood frequency at the changing climate conditions.