HydroBayesCal: Surrogate-Assisted Bayesian Calibration
HydroBayesCal is a Python package for the surrogate-assisted Bayesian calibration of computationally expensive hydro- and morphodynamic models. Calibrating a full-complexity numerical model directly is often infeasible, because a single simulation can take hours to days and stochastic calibration needs many runs. HydroBayesCal sidesteps this cost by training a Gaussian Process Emulator (GPE) as a fast surrogate of the numerical model from a small set of strategically sampled simulations, and then refining it with Bayesian Active Learning (BAL) — iteratively adding the training points that maximise the information gain (relative entropy) and Bayesian model evidence for the calibration. The package supports both single-output and multi-output GPEs, and couples to open-source modelling software through a common binding layer: TELEMAC (2D/3D) is fully supported, an OpenFOAM binding is under active development, and a Delft3D-FLOW binding is planned. Adding bindings for further solvers — or swapping the experimental-design backend (BayesValidRox) — is designed to be straightforward.
Scientific background and references
The methods implemented here build on the Bayesian active-learning framework of Oladyshkin et al. (2020) and on Gaussian-process regression as described by Rasmussen & Williams (2006). The calibration strategy and its application to reservoir sedimentation and three-dimensional reservoir hydrodynamics are documented in Mouris et al. (2023) and Schwindt et al. (2023).
The full bibliography, including DOIs, is collected on the References
page. (The development repository keeps local copies of these works in a
git-ignored ExportedItems/ folder for convenience; they are not
redistributed.)
Note
HydroBayesCal is research software developed at the University of Stuttgart and collaborators. It is provided under a BSD 3-Clause license (see Disclaimer and License).
Getting started
User guide