Radars have gained increased popularity as sensing devices due to their unique capability to sense objects of interest at very long distances and without being severely limited by weather conditions. Advances in technology have led to the possibility of choosing the sensing parameters of a radar in order to further improve its performance. Especially in the class of active phased array radars, the control of the agile beam is of paramount importance. By controlling the radar beam improved estimation results can be achieved leading to better situation awareness. In the literature, several approaches to sensor (including radar) management can be found. These can be roughly grouped into: a) rule-based or heuristics; b) task-based; c) information-driven; and d) risk/threat-based. These approaches are compared in this thesis and it is found that there is not a single approach that is both Bayes-optimal and takes into account explicitly the user requirements in different operational contexts. In order to overcome the challenges with the existing approaches, this thesis proposes managing the uncertainty in higher-level quantities (as per the JDL model) that are directly of interest to the operator and directly related to the operational goal of the radar system. The proposed approach is motivated by the threat assessment process, which is an integral part of defence missions. Accordingly, a prominent example of a commonly used higher-level quantity is the threat-level of a target. The key advantage of the proposed approach is that it results in Bayes-optimal sensor control that also takes into account the operational context in a model-based manner. In other words: a) a radar operator can select the aspects of threat that are relevant to the operational context at hand; and b) external information about the arrival of targets and other scenario parameters can be included when defining the models used in the signal processing algorithms, leading to context-adaptive sensor management. The proposed approach is initially used in simple tracking examples in order to demonstrate its potential and flexibility. Subsequently, it is used for controlling an agile radar beam such that multiple targets can be tracked while taking into account detection uncertainty and presence of spurious measurements. In these examples, a state-of-the art signal processing algorithm is used, i.e. a CB-MeMBer filter. Finally, the proposed approach is used for area surveillance, i.e. for detection and tracking of multiple targets while taking into account detection uncertainty and presence of spurious measurements. In this context, a density that estimates where any undetected targets might be (denoted as unDTD) plays a key role in balancing the search-to-track time ratio. The presented examples have been drawn both from the civilian and the military domain. From the civilian domain, air-traffic-control examples are shown where threat is modeled based on how fast and how close to each other two aircrafts might come. From the defence domain, asset protection examples are shown where threat is modeled based on how fast and how close to an asset of interest a target might come. Furthermore, the deviation from expected trajectories has been modeled because it can be of interest for anomaly detection purposes. The proposed approach has outperformed all the other approaches in the simulated examples presented in this thesis in achieving lower uncertainty in the threat-level of all targets. In all examples, the proposed approach has outperformed naïve approaches, such as periodic or random selection of sensing actions, in a) estimating the correct number of targets present in the considered scenarios; b) localizing the detected targets; and c) maintaining less tracks, thus lowering the computation time at the update step. When only tracking of targets is considered, the proposed approach was only outperformed in tracking accuracy by a scheme that minimizes the expected variance of the estimated number of targets present in the considered scenario and by a derived rule-based scheme. The main challenge when implementing the proposed approach is the mathematical description of threat. Several interesting aspects of threat have been modeled in this thesis but there are even more to be modeled. Taking into account non-measurable aspects of threat poses an added challenge. Other challenges that might be encountered are a) lower tracking accuracy; and b) higher computational complexity, when compared to other sensor management schemes. The presented research can be extended both within the radar domain and by exploring its application to other domains. Two prominent extensions of interest within the radar domain are: a) taking more aspects of threat into account; and b) addressing the target classification problem. Robotics applications, such as autonomous robot path-planning, offer a promising alternative domain for applying the proposed method.