Author(s): Andres Heredia; Silke Wieprecht; Sebastian Schwindt

Linked Author(s): Silke Wieprecht, Sebastian Schwindt

Keywords: Bayesian calibration; Friction zones; Gaussian process; Telemac2d

Abstract: This study presents a metamodeling approach using multioutput Gaussian process emulators (MO-GPEs) within a Bayesian active learning (BAL) framework. We use this approach to optimize spatially varying Nikuradse roughness coefficients in a 2d hydrodynamic model of a near-natural fishway with a gravel bed. MO-GPEs learn relationships among deterministic numerical model outputs (i.e. calibration targets), to improve predictions of response surfaces across the (roughness) parameter space. This supervised learning procedure represents a simultaneous stochastic calibration of four Nikuradse roughness coefficients and provides uncertainty quantifications of the calibration parameters. The results show a reduction of 4.5% in normalized root mean square error (NRMSE) compared to a human-calibrated benchmark model. Uncertainty, measured by the width of posterior distributions, also reduces for all four calibrated roughness coefficients.

DOI:

Year: 2026