Air-to-Ground Situation Assessment (SA) requires gathering information on the entities evolving on the ground (e.g., people, vehicles), and inferring the relations among them and their final intent. Several airborne sensor data might concur in the compilation of such high-level picture, which is aimed at identifying threats and promptly raising alarms. However, this process is intrinsically prone to errors: as the evidence - provided to the SA algorithm - originates from noisy sensor observations, the final outcome is also affected by uncertainty. High-level inferred variables, such as "Situation" and "Threat Level", can be seen as error-affected, incomplete estimates of the ground truth, hence they are subject to improvement by properly steering the data acquisition process. In this paper we address the problem of optimizing the air route of the sensing platform, in order to reduce the number of false declarations or the delay in threat declaration in the SA stage. Specifically, we consider the problem of detecting a hostile behavior between pairs of ground targets by exploiting track data generated from airborne bearings-only measurements. We model the optimization problem with a sequential Markov Decision Process (MDP): the platform sequentially selects the best maneuver (i.e., its acceleration vector) in order to maximize the total reward over an infinite horizon. We define the potential contribution of an action as a function of the expected environmental conditions (e.g., obscurations of the line-of-sight) and the improvement of the localization accuracy achievable for the tracked objects. We demonstrate that following the optimized trajectory the delay in the declaration of a hostile behavior between two targets is reduced at the same erroneous declaration rate.