Enhancing Hydrological Model Performance Through Clustering of Dynamic Catchment Characteristics and Parameter Discretization

Abstract A common structural deficiency in hydrological modeling is the inability to represent the temporal variations within the catchment due to an inadequate understanding of the physical mechanisms behind these processes, leading to generalized runoff simulations. To address model structural deficiencies, we propose a novel approach called Clustering of Dynamic Catchment Characteristics and Parameter Discretization (CDCC). This approach establishes a comprehensive index system describing the meteorological and land‐surface conditions of the catchment, using data mining techniques to extract dynamic characteristics from the index system and cluster the time series into sub‐periods with similar hydrological processes. Then the sub‐period calibration technique is used to couple the extracted catchment dynamics and hydrological models, with parameters discretized and calibrated in each sub‐period accordingly. The results show that the CDCC approach effectively identifies temporal variations in dominant hydrological processes using the Model Parameter Estimation Experiment data set. When applied in conjunction with the HYMOD model, the CDCC approach improves the median Nash‐Sutcliffe Efficiency from 0.55 to 0.67, while significantly enhancing model performance across different flow phases with reduced Root Mean Square Error. Although poor parameter responses to catchment dynamics may arise due to optimization algorithm convergence, parameter boundary constraints, equifinality, or model structure limitations, the dynamic parameter sets effectively capture temporal variations and improve the simulation of various flow phases. The CDCC method would enhance the structural flexibility of current hydrological models, extending their applicability across diverse hydrometeorological conditions and supporting more efficient water resource management under changing environment.