How does network centric organization of sensors affect time critical fusion of dynamic spatial-temporal events in urban environments? This question has become more pertinent today as sensor networks enable automation that holds the promise of monitoring and localizing terrorist threats in complex urban environments by harnessing the power of many inexpensive electromechanical devices like cameras, chemical and IR sensors. This paper presents innovative methods for exploiting the existing urban infrastructure to design and operate flexible sensor network topologies that utilize constrained resources effectively to support dynamic fusion needs. As a network of physically embodied intelligent agents that interact physically with the sensed phenomena, the cognitive abilities of the sensors are constrained by the laws of physics. As goal directed agents they evolve in real world operations to adapt to change in their physical environment and to collaboratively improve task performance. We present formal models and methods to adaptively reconfigure the urban network design space, in-situ, to capture the statistical characteristics of emerging change in the information dynamics. Building on this framework, we present several challenging urban area applications that require adaptive sensor networks to dynamically cluster sensing, processing and communications resources in space-time neighborhoods of emergent hotspots for progressively fine grained sampling and prediction, and collaborate with other dynamic clusters for event tracking. The urban sensor network can, thus, support high predictability of microscopic and macroscopic dynamic events.