Comparison of Local and Global Optimization Methods for Calibration of a 3D Morphodynamic Model of a Curved Channel

Abstract

In curved channels, the flow characteristics, sediment transport mechanisms, and bed evolution are more complex than in straight channels, owing to the interaction between the centrifugal force and the pressure gradient, which results in the formation of secondary currents. Therefore, using an appropriate numerical model that considers this fully three-dimensional effect, and subsequently, the model calibration are substantial tasks for achieving reliable simulation results. The calibration of numerical models as a subjective approach can become challenging and highly time-consuming, especially for inexperienced modelers, due to dealing with a large number of input parameters with respect to hydraulics and sediment transport. Using optimization methods can notably facilitate and expedite the calibration procedure by reducing the user intervention, which results in a more objective selection of parameters. This study focuses on the application of four different optimization algorithms for calibration of a 3D morphodynamic numerical model of a curved channel. The performance of a local gradient-based method is compared with three global optimization algorithms in terms of accuracy and computational time (model runs). The outputs of the optimization methods demonstrate similar sets of calibrated parameters and almost the same degree of accuracy according to the achieved minimum of the objective function. Accordingly, the most efficient method concerning the number of model runs (i.e., local optimization method) is selected for further investigation by setting up additional numerical models using different sediment transport formulae and various discharge rates. The comparisons of bed topography changes in several longitudinal and cross-sections between the measured data and the results of the calibrated numerical models are presented. The outcomes show an acceptable degree of accuracy for the automatically calibrated models.