The technology of base stations on board unmanned aerial vehicles, also known as aerial base stations (ABSs), promises to deliver cellular connectivity in areas where the terrestrial infrastructure is overloaded, damaged, or inexistent. A central problem in this context is to determine the locations where these ABSs must be deployed to serve a set of users on the ground given the positions of the latter. However, existing schemes assume that the channel gain depends only on the length and (possibly) the elevation of the link. To alleviate this limitation, this paper proposes a scheme that accommodates arbitrary channel gains by means of a propagation radio map of the air-to-ground channel. The algorithm finds the locations of an approximately minimal number of ABSs to serve all ground terminals with a target rate while meeting the given constraints on the capacity of the backhaul links and respecting no-fly regions. A convex-relaxation formulation ensures convergence and the alternating-direction method of multipliers is utilized to derive an implementation whose complexity is linear in the number of ground terminals. Numerical results with tomographic as well as ray-tracing channel models corroborate the strengths of the proposed scheme.