We consider the problem of optimizing a novel acoustic leakage detection system for an urban water distribution network composed of a number of detectors and transponders to be placed in a choice of hydrants such as to provide a desired covering under given budget restrictions. The problem is modeled as a particular Prize Collecting Steiner Arborescence Problem. We present a branch and cut and bound approach taking advantage of the special structure at hand which performs well when compared to other approaches. Furthermore, using a suitable stopping criterion, we obtain approximations of provably excellent quality (in most cases actually optimal solutions). The test bed includes the real water distribution network from the Lausanne region as well as carefully randomly generated realistic instances.