Hypothesis-testing and assisted-history-matching applied to evaluate uncertainty of model selection and parameter values

Pore pressure evolution and hydraulic flow are two important physical features to consider in barrier integrity and radionuclide transport applications. Existing literature suggests a potentially non-negligible impact of thermo-osmosis on pore-pressure evolution due to thermal gradients in clay rocks without arriving at a consensus. Numerical experiments based on models are a widely used method in the geosciences to test the relevance of physical features under specified conditions. Mismatch between the observations and the simulation can be partially attributed to the assumptions and simplifications made when developing those models. Their impact on the results and conclusions drawn from the numerical experiments can be significant and thus needs to be explored. Such exploration can proceed by jointly investigating the uncertainty associated with modelling choices, process selection and calibration. This study applies a hypothesis-testing method based on an assisted-history-matching workflow to integrate uncertainty evaluation of model selection, process representation and parameter identification. Three models were compared, representing, respectively, the correct, approximate and wrong hypotheses with respect to a synthetic data set resembling the ATLAS experiment, an in-situ, full-scale heating experiment performed at the HADES underground laboratory in Mol, Belgium. We show that the approach can recover the correct modelling hypothesis in the presence of parameter uncertainty despite competing hypothesis with the same amount of parametric degrees of freedom

Econometrics · Matching (statistics · Model selection · Selection (genetic algorithm · Statistics · Computer Science · Mathematics · Reservoir Engineering and Simulation Methods · Artificial Intelligence