New control paradigms are needed for large networks of wireless sensors and actuators in order to efficiently utilize system resources. In this paper we consider when feedback control loops are formed locally to detect, monitor, and counteract disturbances that hit a plant at random instances in time and space. A sensor node that detects a disturbance dynamically forms a local multi-hop tree of sensors and fuse the data into a state estimate. It is shown that the optimal estimator over a sensor tree is given by a Kalman filter of certain structure. The tree is optimized such that the overall transmission energy is minimized but guarantees a specified level of estimation accuracy. A sensor network reconfiguration algorithm is presented that leads to a suboptimal solution and has low computational complexity. A linear control law based on the state estimate is applied and it is argued that it leads to a closed-loop control system that minimizes a quadratic cost function. The sensor network reconfiguration and the feedback control law are illustrated on an example.