Comparative research of optimization algorithms for parameters calibration of watershed hydrological model

Abstract

Parameter optimization and calibration of the hydrological model has been one of the important research fields in hydrological forecasting. This paper is written to address the inherent defects that traditional parameter optimization of Xinanjiang hydrological model with a single objective entails. These methods cannot fully exploit hydrological characteristics information from hydrological observation. We selected the Nash Sutcliffe coefficient, which is known to be biased for high flows and the logarithmic form of the Nash Sutcliffe coefficient that emphasize low-flow values as the objective functions. Then, we adopted the multi-objective optimization algorithms, such as the Nondominated Sorted Genetic Algorithm-II (NSGAII) and the Third Evolution Step of Generalized Differential Evolution (GDE3), and the single-objective optimization algorithm, Simulated Annealing (SA). These algorithms were applied in Heihe River Basin to calibrate parameters of the Xinanjiang hydrological model for long-term prediction of river discharges. Through the evaluation of the Pareto optimal parameter set derived from multi-objective optimization algorithms and the optimal solution obtained from the single objective algorithm, the results showed that the multi-objective optimization algorithms, in particular the NSGA-II algorithm, perform best to locate the Pareto optimal solutions in the parameter search space. They can also obtain better results with respect to the model parameters calibrated by the single objective algorithm. The major contribution of this work is the comparative application research of single-objective optimization with the multi-objective optimization algorithms for the parameters optimization of the Xinanjiang model in the Heihe River basin.