Automatic calibration and uncertainty quantification in waves dynamical downscaling

Abstract

Automatic calibration of numerical models and systematic quantification of the uncertainties associated with their use is not yet a widespread practice in coastal engineering. This paper proposes the use of a Bayesian algorithm for the automatic calibration and uncertainty quantification of a wave model, in the framework of the dynamical downscaling of off-shore waves to a nearshore project site, where a set of wave measurements is available. A spectral error was defined and used for the definition of the likelihood function used by the algorithm; this allows for improvements in terms of the wave spectra and not only in terms of a restricted set of wave parameter. In addition to the calibration of the model parameters, the methodology also addresses errors coming from boundary conditions; to this end it distinguishes between different wave systems and a set of parameters are defined for the correction of each wave system, reducing the errors introduced into the model by the boundary conditions. A case study in the South Atlantic coast showed the ability of the methodology to calibrate the model, resulting in simulations that properly fit the available measurement, while also providing an estimation of the uncertainties associated with the model results that can be straightforwardly used for probabilistic analysis in the coastal environment.