False alarms from individual sensors and duplicate detections from sensors in close proximity provide a user with inaccurate and superfluous information. Responding to all of these detections distracts the user, and also wastes system resources including bandwidth and storage capacity. Fusion can reduce these implications of an unattended sensor system and can occur on three different levels: multi-modal fusion of multiple transducer-based algorithms (including PIR, magnetic, seismic, acoustic, video, and biological/chemical) within a single remote ground sensor to provide a single detection; multi-sensor fusion which combines the detections of several remote ground sensors in a network; and multi-resource fusion which combines multi-modal and multi-sensor fusion with other resources such as historical, media, statistical, and user-defined information. Each level of fusion may be used together or separately to allow for increased sensitivity and a low false alarm rate while conserving power, bandwidth, storage capacity, and user interaction. This paper will describe the design and usage of each level of fusion in a networked unattended ground sensor system.