Parameter sensitivity and uncertainty of a one-dimensional morphodynamic model in the Lower Yellow River

Abstract

One-dimensional morphodynamic models are effective tools to conduct flood risk assessment and sediment management of alluvial rivers. In order to gain better understanding of morphodynamic models and improve the simulation efficiency and accuracy, a comprehensive investigation into parameter sensitivity and uncertainty has been conducted for a one-dimensional morphodynamic model. This model was calibrated using the measured flow and sediment data in the Lower Yellow River (LYR). The variance-based Sobol' and GLUE (Generalized Likelihood Uncertainty Estimation) tools were applied to quantify the parameter sensitivity and uncertainty to the simulation of different hydrological indicators in the meandering reach, with various model response surfaces selected. Results show that the importance of input parameters varied with different model outputs. The simulated water level and discharge hydrographs were more sensitive to the roughness coefficient at middle and high flows, with high total- and first-order sensitivity indices to the NSEs and RMSEs, while the simulated sediment concentration was more sensitive to the input parameters related to sediment transport capacity (K and M) and recovery coefficient (bd). Notably, the pairwise interactions between K and different roughness coefficients were significant in the simulated sediment concentration hydrographs. Spatio-temporal variability in the effect of input parameters is also revealed. The most contributing factor was screened as K for the maximum sediment concentration in the uppermost and lowermost reaches of the meandering reach, which turned to be bd in the middle reach. The most sensitive parameter to the simulated sediment concentration was screened as K in the initial period and then turned to be the exponent of recovery coefficient in case of erosion (be), which was bd in the last period over the simulation period. An analysis of parameter uncertainty indicates that the morphodynamic model can obtain satisfactory accuracy with most observed data concentrating in the 95% confidence interval, and the uncertainty of simulation outputs increased downstream. Improved understanding of parameter sensitivity and uncertainty in morphodynamic modelling can provide the scientific guidance for factor prioritization in the future model calibration.