Bayesian inference of numerical modeling-based morphodynamics: Application to a dam-break over a mobile bed experiment

Abstract

Numerical modeling of morphodynamics presents significant challenges in engineering due to uncertainties arising from inaccurate inputs, model errors, and limited computing resources. Accurate results are essential for optimizing strategies and reducing costs. This paper presents a step-by-step Bayesian methodology to conduct an uncertainty analysis of 2D numerical modeling-based morphodynamics, exemplified by a dam-break over a sand bed experiment. Initially, uncertainties from prior knowledge are propagated through the dynamical model using the Monte Carlo technique. This approach estimates the relative influence of each input parameter on results, identifying the most relevant parameters and observations for Bayesian inference and creating a numerical database for emulator construction. Given the computationally intensive simulations of Markov chain Monte Carlo (MCMC) sampling, a neural network emulator is used to approximate the complex 2D numerical model efficiently. Subsequently, a Bayesian framework is employed to characterize input parameter uncertainty variability and produce probability-based predictions.

References in corpus (4)

• The Expressive Power of Neural Networks: A View from the Width
• Efficient Learning of Accurate Surrogates for Simulations of Complex Systems
• Auto-weighted Bayesian Physics-Informed Neural Networks and robust estimations for multitask inverse problems in pore-scale imaging of dissolution
• Application of a dispersive wave hydro-sediment-morphodynamic model in the Ria Formosa lagoon