ABSTRACT
A methodology for the assessment of contaminant transport model predictive certainty is presented and its use illustrated in a case study of fluoride migration through clay barriers. Bayesian nonlinear regression using global probabilistic optimisation with the shuffled complex evolution algorithm is used to fit a finite element solution of the contaminant transport equation with coupled finite mass boundary conditions to experimental data. Model predictive certainty is assessed with the Metropolis algorithm. The methodology is applied to a laboratory experiment of fluoride diffusion in clays that is intended to provide parameters that will drive a model of landfill liner behaviour. It is shown that the commonly used linear sorption transport model is not able to describe the experiment. In contrast, the Langmuir sorption transport model is found to adequately represent the laboratory experiment behaviour. The modelling approach is able to quantify model predictive certainty and provide insight into experimental design and parameter uncertainty.
