Surrogate modeling-based calibration of hydrodynamic river model parameters

Abstract

As opposed to other disciplines, automated calibration procedures are not common practice for full hydrodynamic river models, mainly because of the long computation times impeding the accurate assessment of parameter values. Default or text-book values are therefore often used. This paper introduces a methodology to optimize hydrodynamic model parameter values, based on the use of a surrogate conceptual model. Thanks to the spatial lumping and the explicit calculation schemes of these conceptual models, very short calculation times and a large number of simulation runs can be achieved. The surrogate model is coupled with the Shuffled Complex Evolution Metropolis algorithm of the University of Arizona (SCEM-UA) to identify the optimal parameter sets and their uncertainty. Afterwards, the optimized parameter values are transferred to the full hydrodynamic model. The methodology is demonstrated on a case study of the river Molenbeek in Belgium, using streamflow, water level and gate level observations. Results show a decrease of the hydrodynamic model residuals by about 60 percent.