Fast assessment of uncertainty in buoyant fluid displacement using a connectivity-based proxy

It is crucial to estimate the uncertainty in flow characteristics of injected fluid. However, because a large suite of geological models is probable given sparse static data, it is impractical to conduct full physics flow simulations on the entire suite of models in order to quantify the uncertainty in fluid displacements. Thus a fast alternative to a full physics simulator is necessary to quickly predict the fluid displacements. Most of the proxies proposed thus far are inappropriate to approximate the buoyant flow of injected fluid for 3D heterogeneous rock during the injection period. In this dissertation, a new proxy will be proposed to quickly predict the buoyant flow of injected fluid during CO2 sequestration. The geological models are ranked based on the extent of the approximated CO2 plumes. By selecting a representative group of models among the ranked models, the uncertainty in the spatial and temporal characteristics of the CO2 plume migrations can be quickly quantified. About 90% of the computational cost of quantifying the uncertainty in the extent of CO2 plumes was saved using the proposed connectivity based proxy. In a geological carbon storage project, the spatial and temporal characteristics of CO2 plume migrations can be monitored by 4D seismic surveys. The images of CO2 plumes obtained from 4D seismic surveys are used as observed data to find subsurface models honoring the spatial and temporal characteristics of the observed CO2 plumes. However, because manually comparing an observed CO2 plume and prior CO2 plumes in a large suite of subsurface models is inefficient, an automatic measure to calculate the dissimilarity between the CO2 plumes is necessary. The most intuitive way to calculate the dissimilarity is the Euclidean distance between vectors representing CO2 plumes. However, this is inappropriate to measure the dissimilarity between CO2 plumes because it does not consider spatial relation between the elements of the vectors. The shape dissimilarity between the CO2 plumes that reflects the spatial relation can be calculated using the Hausdorff distance. The computational cost of calculating the shape dissimilarity between CO2 plumes is significantly reduced by calculating the Hausdorff distance between the representations of the CO2 plumes such as perimeter, surface, and skeleton instead of the original CO2 plumes. An appropriate representation should be chosen according to the spatial characteristics of CO2 plumes.