Optimal calibration of hydrological models for derived flood frequency analyses using stochastic rainfall - revisited

Derived flood frequency analysis (DFFA) allows the estimation of design floods with hydrological modelling for both poorly observed basins and for catchments under nonstationary conditions. For mesoscale catchments long records of sub-daily precipitation are required. As these are usually not easily available, stochastic weather data can be used as an alternative. Objective of this research is to find the optimal calibration strategies of a hydrological model for DFFA using stochastic weather data as input by comparing various calibration alternatives. The optimal calibration of the hydrological model should a) consider long records regarding robust estimation of the extremes b) select the most informative parts from these records and c) utilise the stochastic input data.

Hourly climate variables are disaggregated from long daily records using a k-nearest neighbour approach. For hydrological modelling the semi-distributed conceptual HBV model is used. The model is calibrated alternatively on observed flow data and on various flow statistics considering different temporal discretisations and time periods. The main validation of the hydrological model is based on long term flood statistics. The calibration approaches are tested for several mesoscale catchments of the Mulde River basin in Germany. The results will reveal the advantages and disadvantages of the different calibration strategies and if there is an optimal approach.