Overcoming equifinality: time-varying analysis of sensitivity and identifiability of SWAT runoff and sediment parameters in an arid and semiarid watershed

The phenomenon of “equifinality for different parameters” limits the link between parameters and catchment characteristics; however, solving the equifinality problem is a major challenge in the development, generalization, and application of a model. This study focused on the Yanhe River Watershed to investigate the time-varying characteristics of sensitivity and identifiability of SWAT (Soil and Water Assessment Tool) runoff and sediment parameters based on the Sobol’ and generalized likelihood uncertainty estimation methods. The results indicate that (i) the nondominated sorting genetic algorithm-II has good adaptability and reliability in parameter calibration of the SWAT model in the Yanhe River Watershed. The evaluation indicators (Nash–Sutcliffe efficiency, R², and percent bias) of monthly runoff and sediment in the Ganguyi hydrological station were all satisfactory per the SWAT model during the calibration and validation periods. (ii) The interaction between runoff and sediment parameters is a crucial reason for parameter sensitivity, which has obvious time-varying characteristics and is largely dependent on precipitation in the Yanhe River Watershed. Temporal and spatial variability of precipitation should be considered in the detailed analysis of parameter identifiability, and watershed managers should not ignore changes in the runoff process when regulating sediment. (iii) Only a relatively small number of parameters can be identified in the runoff and sediment simulation process of the Yanhe River Watershed, such as CN2 (initial soil conservation service runoff curve number for moisture condition II), CH_K2 (effective hydraulic conductivity in main channel alluvium), ALPHA_BF (baseflow alpha factor), USLE_C (cover and management factor), USLE_P (support practice factor), and USLE_K (soil erodibility factor), due to high surface runoff, reduced lag time, reduced low flows, increased peak flows, and channel erosion, respectively. More importantly, there is a strong positive correlation between parameter identifiability and parameter sensitivity. Both are effective methods of parameter diagnosis, but the identifiability of parameters is not equivalent to its sensitivity. Our results strongly suggest that a detailed parameter sensitivity and identifiability analysis is a critical step in improving hydrological model performance to reduce the risk of “equifinality for different parameters” while articulating all relevant hydrological processes.