If you want your boundary to be far enough away to encompass a CO2 plume you are injecting, I wrote a little article about that a few years ago ( Doughty, C., (2008). Estimating plume volume for geologic storage of CO2 in saline aquifers, Groundwater, 46:6, 810-813. ) But you probably also want your grid to be big enough to encompass the pressure signal, which goes out much farther than the CO2 itself. A simple rule of thumb that I follow is to make my sure my boundary is at least as far away from the injection well as the hydraulic diffusion distance, which I calculate as L=sqrt(4(T/S)t), where T is transmissivity, S is storativity, and t is time. T/S is hydraulic diffusivity and should be in units of m^2/s. To relate to TOUGH input variables: T=k h/mu, where k is permeability in m^2, h is thickness in m, and mu is viscosity in Pa-sec; S=phi C h, where phi is porosity, C is compressibility in 1/Pa, and h is thickness in m. Note that the h terms cancel out, so you don't need to worry about them. C is the sum of fluid compressibility and rock compressibility. If you are injecting CO2 into water, then you should use water properties for viscosity and fluid compressibility, since most of the pressure pulse will be traveling through water beyond your CO2 plume. For long times, L can get pretty big, but I usually let my grid get gradually coarser beyond the region where I expect the CO2 to go, which I keep fine.
Good luck with your simulations!