A method is provided including receiving data corresponding to a three-dimensional physical representation of a porous rock sample; calculating a low-dimensional representation of a pore network in the porous rock sample based on the three-dimensional physical representation; extracting one or more geometrical parameter from the low-dimension representation; generating a capillary network model of the porous rock sample based at least on the at least one geometrical parameter for simulating fluid flow inside the porous rock sample; and performing at least one simulation of a flow of the fluid through the capillary network model of the porous rock sample with a fluid additive to provide a predicted enhanced fluid recovery efficiency.