Detection of drones carries critical importance for safely and effectively managing unmanned aerial system traffic in the future. Given the ubiquitous presence of the drones across all kinds of environments in the near future, wide area drone detection and surveillance capability are highly desirable, which require careful planning and design of drone sensing networks. In this paper, we seek to meet this need by using the existing terrestrial radio frequency (RF) networks for passive sensing of drones. To this end we develop an analytical framework that provides the fundamental limits on the network-wide drone detection probability. In particular, we characterize the joint impact of the salient features of the terrestrial RF networks, such as the spatial randomness of the node locations, the directional 3D antenna patterns, and the mixed line of sight/non line of sight (LoS/NLoS) propagation characteristics of the air-to-ground (A2G) channels. Since the strength of the drone signal and the aggregate interference in a sensing network are fundamentally limited by the 3D network geometry and the inherent spatial randomness, we use tools from stochastic geometry to derive the closed-form expressions for the probabilities of detection, false alarm and coverage. This, in turn, demonstrates the impact of the sensor density, beam tilt angle, half power beam width (HPBW) and different degrees of LoS dominance, on the projected detec?tion performance. Our analysis reveals optimal beam tilt angles, and sensor density that maximize the network-wide detection of the drones.