Hydrological model calibration by sequential replacement of weak parameter sets using depth function

Abstract

It is always a dream of hydrologists to model the mystery of complex hydrological processes in a precise way. If parameterized correctly, a simple hydrological model can represent nature very accurately. In this study, a simple and effective optimization algorithm, sequential replacement of weak parameters (SRWP), is introduced for automatic calibration of hydrological models. In the SRWP algorithm, a weak parameter set is sequentially replaced with another deeper and good parameter set. The SRWP algorithm is tested on several theoretical test functions, as well as with a hydrological model. The SRWP algorithm result is compared with the shuffled complex evolution-University of Arizona (SCE-UA) algorithm and the robust parameter estimation (ROPE) algorithm. The result shows that the SRWP algorithm easily overcomes the local minima and converges to the optimal parameter space. The SRWP algorithm does not converge to a single optima; instead, it gives a convex hull of an optimal space. An ensemble of results can be generated from the optimal space for prediction purpose. The ensemble spread will account for the parameter estimation uncertainty. The methodology was demonstrated using the hydrological model (HYMOD) conceptual model on upper Neckar catchments of southwest Germany. The results show that the parameters estimated by this stepwise calibration are robust and comparable to commonly-used optimization algorithms. SRWP can be an alternative to other optimization algorithms for model calibration.