Gaussian Process Emulators for Quantifying Uncertainty in CO2 Spreading Predictions in Heterogeneous Media

摘要: Abstract We explore the use of Gaussian process emulators (GPE) in numerical simulation CO 2 injection into a deep heterogeneous aquifer. The model domain is two-dimensional, log-normally distributed stochastic permeability field. first estimate cumulative distribution functions (CDFs) breakthrough time and total mass using computationally expensive Monte Carlo (MC) simulation. then show that we can accurately reproduce these CDF estimates with GPE, only small fraction computational cost required by traditional MC In order to build GPE predict simulator output from field consisting 1000s values, truncated Karhunen-Loeve (K-L) expansion field, which enables application Bayesian functional regression approach. perform cross-validation exercise give an insight optimization experiment design for selected scenarios: find it sufficient 100s values size training set adequate as few 15 K-L components. Our work demonstrates be effectively applied uncertainty analysis associated modelling multiphase flow transport processes media.