BOAR: Bayesian optimization for automated roughness calibration in two-dimensional hydrodynamic models

BOAR is an open-source Python framework for automated hydraulic roughness calibration in two-dimensional shallow-water numerical solvers. BOAR combines Bayesian optimization with conditional sampling to integrate expert knowledge, including feasible parameter ranges and inter-parameter constraints, thereby reducing the number of model evaluations. It standardizes calibration workflows, improves reproducibility, and supports flexible, user-defined loss functions. Applied to 25 laboratory vegetated compound-channel flow cases, it converged within a median of 9 model runs for a depth-error of 0.4%–1.5% tolerance. The current implementation interfaces with BASEMENT and can be extended to other river-modeling tools, providing a flexible and extensible solution for hydraulic modeling.