Gaussian Process Emulators for Quantifying Uncertainty in CO2 Spreading Predictions in Heterogeneous Media
(English)Manuscript (preprint) (Other academic)
We explore the use of Gaussian process emulators (GPE) in the numerical simulationof CO2 injection into a deep heterogeneous aquifer. The model domainis an uncertain two-dimensional log-normally distributed permeability eld. Weestimate the cumulative distribution functions (CDF) of the CO2 breakthroughtime and the total mass using a computationally expensive Monte Carlo (MC)simulation. We then show that we can accurately reproduce these CDF estimateswith a GPE, but using only a fraction of the computational cost comparedto Monte Carlo. In order to build a GPE that can predict the simulator outputfrom a permeability eld consisting of 1000s of values, we use a truncatedKarhunen-Loève expansion of the permeability eld, and then use a Bayesianfunctional regression approach. We explore issues encountered in using GPEsin uncertainty analyses of CO2 storage problems, including the optimization ofthe experiment design, and provide perspectives for future applications.
CO2, Bayesian, Permeability, KL expansion, Monte Carlo, Cumulative distribution function, Uncertainty analysis
IdentifiersURN: urn:nbn:se:uu:diva-298748OAI: oai:DiVA.org:uu-298748DiVA: diva2:947102
FunderEU, FP7, Seventh Framework Programme, 227286EU, FP7, Seventh Framework Programme, 282900EU, FP7, Seventh Framework Programme, 309067