A digital similar (DS) of a population of a region is a common starting point for agent-based modeling and simulation. Here, an integer linear programming-based algorithm is presented that refines an existing, high-resolution methodology for constructing DSs. The extension consists of constructing a household-to-residence mapping that maximizes the correlation between household income of individual households and residence property values of individual residences. The algorithm is applied to a coastal region of Virginia (USA) where we demonstrate that new household-to-residence assignment generates significantly different outcomes than the existing approach which is random assignment at blockgroup level. Using the context of road inundation and measures such as time to evacuate and time to reach critical care, significant differences across household income segments are demonstrated with the new method, while no such difference is established with the prior method.