| Geological carbon sequestration, like any subterranean activity, is associated with large uncertainties. In particular, numerical models require many input parameters, many of which need to be estimated because they cannot be measured or observed directly. Examples of these parameters include the physical and flow properties of the porous media in which the CO2 is injected. In order to estimate these parameters, we try to select the combination of parameter values that results in the best fit by the model, via an optimization algorithm. Measurement data carries uncertainty often due to imperfect measurement techniques and these errors need to be taken into account. In addition, the computational expense associated with geological carbon sequestration process-based models restricts the parameter estimation methodology to optimization algorithms that can provide a good set of parameters without requiring too many forward simulations of the carbon sequestration transport model. In this work, we compare the efficiency of optimization algorithms on three geological carbon sequestration process-based models with varying numbers of parameters. We show that response surface based algorithms are best tailored for complex multiphase heterogeneous models because they yield multi-modal objective functions. Derivative-based and local methods however are best suited for simpler homogeneous models. Finally, we quantify the uncertainty of the input parameters stemming from measurement uncertainty and translate it into CO2 plume position uncertainty. We show that the SOARS algorithm is more computationally efficient that the standard MCMC method and that it can be used in geological carbon sequestration process-based models with single and multiple outputs. This uncertainty analysis is reproduced for different monitoring schemes and tied to the uncertainty in the CO2 plume position. |
