A model selection process based on multi-objective optimization using a fast proxy is presented that chooses geologic models conditioned to observed flow and geomechanical responses. The responses of geologic models to injection of large volumes of CO2 are evaluated using a proxy that approximates pressure distribution using a random walk particle tracking algorithm and computes surface deformation using a stress-field solver. The geologic models showing similar proxy responses are grouped into clusters by invoking multi-dimensional scaling and k-means clustering. A representative model of each cluster is chosen and its simulation results are compared to the observation. The resultant posterior ensemble consists of models that belong to one or more clusters whose representative models are not only in good agreement with given observation data but also non-dominated to other representative models. The usefulness of the model selection approach is demonstrated on CO2 sequestration of a fractured gas reservoir at In Salah, Central Algeria. The explicit coupling of the connectivity estimator and the stress solver reproduces similar migration patterns of CO2 plume to those obtained using full-physics numerical simulations. Incorporating time-lapse observations measured using satellite-based interferometric synthetic aperture radar imposes additional constraints to refine geologic models by improving the matching quality of bottomhole pressure, and thereby contributes to a further reduction in geologic uncertainty. The multi-objective optimization process implemented in the model selection process yields a diversified model set than those obtained using a global optimization procedure. The posterior ensemble exhibits consistent geologic characteristics in terms of spatial distribution and orientation of migration pathways. The posterior ensemble is then used as an initial population for a multi-objective history matching procedure. The results of model selection and expansion process indicate that the implementation of Pareto-optimality is advantageous to realize diversified geologic models that yield unbiased assessment of uncertainty associated with prediction of future CO2 plume migration.