Assessing the temporally dynamic parameters in hydrological models: dynamic operations and evolutionary processes [post]

Tian Lan, Kairong Lin, Chong-Yu Xu, Zhiyong Liu, Huayang Cai
2020 unpublished
Abstract. The temporal dynamics of parameters can compensate for structural defects of hydrological models and improve the accuracy and robustness of the streamflow forecast. Given the parameters usually estimated by global optimization algorithms, a critical issue, however, which received little attention in the literature, is that the possible failure in finding the global optimum might lead to unreasonable parameter values. This may cause the poor response of the dynamic parameters to
more » ... arameters to time-varying catchment characteristics (such as seasonal variations of land cover). In this regard, we propose a framework for identifying the difficulty of finding the global optimum for dynamic hydrological model parameters by investigating their evolutionary processes. Specifically, the probability distributions of the violin plots and the divergence measure of the polylines in the parallel coordinates are applied and developed to configure the evolutionary processes in the individual parameter spaces and multi-parameter space, respectively. Also, a complete solution for the dynamic operation of parameters is proposed. Furthermore, clustering operations, calibration scheme and correlation between parameters are further discussed. The results showed that the performance of the hydrological model with dynamic parameters achieves a significant improvement. However, the response of individual parameters (even high-sensitive parameter) to dynamic catchment characteristics is generally poor. The main reasons can be primarily attributed to the complexly linear and nonlinear correlation between parameters and poor ability in finding the global optimum. In this regard, the dynamic parameter set instead of individual dynamic parameters is suggested to extract dynamic catchment characteristics. Importantly, we found that the properties of hydrological-model parameters, including identifiability, sensitivity, correlation and the ability to find global optimum, interact with the response of parameters to the dynamic catchment characteristics. The ability to find global optimum has a significant influence on the hydrological model performance with dynamic parameters. Hence, the ability to find global optimum is suggested as one of the essential properties of the hydrological model parameters. The study provides a valuable benchmark for temporally dynamic parameters in hydrological models.
doi:10.5194/hess-2020-195 fatcat:fm2ro5ealzdbhodvop5o7tybzy