PNNL

Abstract

Numerical simulations have been used for estimating CO2 injectivity, CO2 plume extent, pressure distribution, and area of review (AOR), and for the design of CO2 injection operations and monitoring network for the FutureGen project. The simulation results are affected by uncertainties associated with numerous input parameters, the conceptual model, initial and boundary conditions, and factors related to injection operations. Furthermore, the uncertainties in the simulation results also vary in space and time. The key need is to identify those uncertainties that critically impact the simulation results and quantify their impacts. We introduced the local sensitivity coefficient (LSC) defined as the percent response of the output. The LSC is scalable to the error of an input and the composite sensitivity of an output to a subset of inputs can be calculated by summing the individual LSC values. We apply the local sensitivity approach to the FutureGen 2.0 Site in Morgan County, Illinois, USA, to investigate the sensitivity of input parameters and initial conditions. The conceptual model consists of 31 layers, each of which has a unique set of input parameters. The sensitivity of 11 parameters for each layer and 7 inputs as initial conditions is then investigated. For CO2 injectivity and plume size, about half of the uncertainty is due to only 4 or 5 of the 348 inputs and 3/4 of the uncertainty due to about 15 of the inputs. The initial conditions and the properties of the injection layer and its neighbor layers contribute most the sensitivity. Overall, the simulation outputs are very sensitive to only a small fraction of the input. However, the importance of inputs for controlling CO2 injectivity and plume size is not all the same. The results will be useful for AOR uncertainty estimation, cost-effective site characterization, injection well design, and monitoring network design.