Uncertainty reduction by DoE-based history matching of the FE-experiment at Mt. Terri

For the evaluation of the performance of nuclear waste repositories, a detailed investigation of the sensitivity and uncertainty of the underlying processes is essential. The verification and validation of the numerical tools under realistic conditions using experimental data from underground research laboratories is all the more crucial because a comprehensive experimental analysis or investigation of the final repository site is impractical for a variety of reasons. In our contribution, we apply design-of-experiment-based history matching to reduce and examine parameter uncertainties with pressure and temperature data from the FE experiment. The FE experiment is a full-scale multiple heater test in the Opalinus clay at the URL site in Mount Terri in Switzerland. We use the open-source program OpenGeoSys for the deterministic thermo-hydro-mechanical modeling. An initial parameter screening to identify heavy hitters was conducted prior to proxy building using a Gaussian Proxy model. The proxy model is then used for experiment-matching, employing Monte-Carlo sampling over the entire remaining parameter space. The obtained parameter boundaries were applied to a global sensitivity analysis based on the proxy model to quantify uncertainty and show how parameter sensitivities affect results.