Surrogate-based Bayesian comparison of computationally expensive models: application to microbially induced calcite precipitation

Abstract Geochemical processes in subsurface reservoirs affected by microbial activity change the material properties of porous media. This is a complex biogeochemical process that currently contains strong conceptual uncertainty. means, several modeling approaches describing are plausible and modelers face uncertainty choosing most appropriate one. The considered models differ underlying hypotheses about structure. Once observation data become available, rigorous Bayesian model selection accompanied justifiability analysis could be employed to choose model, i.e. one describes physical best light available data. However, computationally very demanding because it conceptualizes different phases, biomass dynamics, geochemistry, precipitation dissolution Therefore, framework cannot based directly on full computational as this would require too many expensive evaluations. To circumvent problem, we suggest perform both after constructing surrogates for competing models. Here, will use arbitrary polynomial chaos expansion. Considering surrogate representations only approximations analyzed original models, account approximation error introducing novel correction factors resulting weights. Thereby, extend assess similarities We demonstrate method representative scenario microbially induced calcite medium. Our extension provides suitable approach comparison gives an insight necessary amount reliable performance.