Impact of robustness of hydrological model parameters on flood prediction uncertainty

Abstract

The robustness of hydrological model parameter values in flood predictions is a known area of concern, but there is a lack of a comprehensive approach to the handling of model parameter robustness, model simulation uncertainty and multiobjective model calibration when calibrating multiple flood data sets. For investigation of the impact of robustness of hydrological model parameters on flood simulation uncertainty, this paper develops a Minimax‐Regret robust multiobjective optimisation framework for robust hydrological model parameter calibration and uncertainty analysis. The robustness is considered as an objective function in this study instead of a constraint as in previous research. A physically based semi‐distributed hydrological model is employed to illustrate the proposed framework in a midscale catchment. Results show that the proposed framework can effectively explore robust hydrological model parameter values and quantify flood simulation uncertainty. A trade‐off between robustness and Nash–Sutcliffe efficiency is found, implying that the better the Nash–Sutcliffe efficiency, the less robust the non‐dominated parameter values and that improving robustness alone cannot guarantee narrower uncertainty intervals and greater containing ratios. These results reveal that robustness should not be used alone to select behavioural parameter sets, and a balance has to be made between robustness and Nash–Sutcliffe efficiency.