Previous studies have shown that the seasonal dynamics of model parameters can compensate for structural defects of hydrological models and improve the accuracy and robustness of the streamflow forecast to some extent. However, some fundamental issues for improving model performance with seasonal dynamic parameters still need to be addressed. In this regard, this study is dedicated to (1) proposing a novel framework for seasonal variations of hydrological model parameters to improve model performance and (2) expanding the discussion on model results and the response of seasonal dynamic parameters to dynamic characteristics of catchments. The procedure of the framework is developed with (1) extraction of the dynamic catchment characteristics using current data-mining techniques, (2) subperiod calibration operations for seasonal dynamic parameters, considering the effects of the significant correlation between the parameters, the number of multiplying parameters, and the temporal memory in the model states in two adjacent subperiods on calibration operations, and (3) multi-metric assessment of model performance designed for various flow phases. The main finding is that (1) the proposed framework significantly improved the accuracy and robustness of the model; (2) however, there was a generally poor response of the seasonal dynamic parameter set to catchment dynamics. Namely, the dynamic changes in parameters did not follow the dynamics of catchment characteristics. Hence, we deepen the discussion on the poor response in terms of (1) the evolutionary processes of seasonal dynamic parameters optimized by global optimization, considering that the possible failure in finding the global optimum might lead to unreasonable seasonal dynamic parameter values. Moreover, a practical tool for visualizing the evolutionary processes of seasonal dynamic parameters was designed using geometry visualization techniques. (2) We also discuss the strong correlation between parameters considering that dynamic changes in one parameter might be interfered with by other parameters due to their interdependence. Consequently, the poor response of the seasonal dynamic parameter set to dynamic catchment characteristics may be attributed in part to the possible failure in finding the global optimum and strong correlation between parameters. Further analysis also revealed that even though individual parameters cannot respond well to dynamic catchment characteristics, a dynamic parameter set could carry the information extracted from dynamic catchment characteristics and improve the model performance.
This item's license is: Attribution 4.0 International