Impacts of changes in the watershed partitioning level and optimization algorithm on runoff simulation: decomposition of uncertainties

Abstract

Hydrological modeling has provided key insights into the mechanisms of model state, such as the watershed partitioning level and optimization algorithm, and their impacts on the hydrological process, but the uncertainty of this impact is poorly understood. To this end, in this study, the effects of the watershed partitioning level and optimization algorithm for hydrological simulation uncertainty were assessed based on the semi-distributed TOPMODEL model, i.e., six watershed partitioning levels and three intelligent global optimization algorithms were used in the source region of the Yellow River. Meanwhile, the uncertainty contribution of the individual and interaction of the watershed partitioning levels and optimization algorithms on the hydrological process were dynamically evaluated using the variance decomposition method based on subsampling. Results showed that the impacts of the watershed partitioning level and optimization algorithm on the runoff simulation were particularly obvious for different characteristic periods. In the flood period, the optimization algorithm was the dominant factor affecting the runoff simulation uncertainty, with the proportion of up to 0.50, whereas the contribution of the watershed partitioning level was only 0.22. In the non-flood period, they contributed substantially to the uncertainty of the runoff simulation, accounting for about 0.30. Moreover, the interactions between the watershed partitioning level and optimization algorithm had a strong influence throughout the year, especially in the non-flood period, which may be because the hydrological model amplifies the output error and increases the interaction effect. Generally, the results shed important insight into reducing the uncertainty of the runoff simulation in future research.