Global sensitivity analysis of reactive transport modelling for the geochemical evolution of a high-level radioactive waste repository

Abstract

Reactive transport models are used for the performance assessment of high-level radioactive waste repositories (HLW). Quantifying model uncertainty requires the identification of the parameters having the most significant effect on model outputs and the largest impact on prediction uncertainties. Sophisticated sensitivity analysis methods have been developed for numerical models based on the analysis of model results such as the Morris elementary effects and the High Dimensional Model Reduction (HDMR) method proposed by Sobol. VARS (Variogram Analysis of Response Surfaces) is a variance-based method to analyze the structure of a model output. Here we report the application of VARS and HDMR for evaluating the global sensitivities of the predictions of a reactive transport model of the metallic canister and bentonite buffer of a HLW repository in granitic rock. Global sensitivities are analyzed for pH, redox potential and volume of corrosion products at the canister/bentonite interface at t = 1·10⁴, 2.5·10⁴ and 5·10⁴ years. The input parameters include: the corrosion rate of the metallic canister (CR), the bentonite porewater diffusion coefficient (D_e), the groundwater flow through the granitic host rock (Q_gra), the cation exchange selectivity of Fe (K_Fe) and the solubility of magnetite (Log K). VARS and HDMR results show that the most influential input parameters for pH are CR, Q_gra and K_Fe. The sensitivity indexes of pH for parameters CR and K_Fe increase with time while that of log K of magnetite decreases. The sensitivity index of D_e, however, remains constant in time. Similar to pH, the most influential parameters for Eh are CR and Q_gra. The sensitivity indexes of Eh for CR and log K of magnetite decrease with time while that of Q_gra increases with time. The main interactions between parameters computed with HDMR indexes for pH and Eh at t = 5·10⁴ years occur between CR and Q_gra. Similar to pH and Eh, CR is the most influential parameter for the volume of corrosion products while D_e is the second most relevant. All the sensitivity indexes for the volume of corrosion products increase with time except for CR. Parameter rankings of Sobol and VARS are nearly identical.