Effective applications of M2M (Machine-to-Machine) systems for disaster prevention, such as an early warning system for flood and landslide, are highly expected. In particular, in applications for flood disaster prevention, an early warning of the flash flood caused by local torrential rain is required as well as a flood impact analysis based on water level and rainfall monitoring in the whole basin. However, in a conventional server-centric M2M system, an increase of network load and a delay of feedback control has become one of the most critical issues. Therefore, it is difficult to apply the conventional server-centric M2M system to applications of flood disaster prevention, in which both a real-time flood alert and a flood impact analysis are required. In this paper, we propose a distributed cooperative M2M system to solve such problems, and make an attempt to apply the proposed system to applications for flood disaster prevention. This proposed system provides an event-driven data collection mechanism for efficient data collection in low-bandwidth networks on the Never Die Network to achieve disaster tolerance of the access network. In addition, a rule-based autonomous control mechanism for a real-time flood alert in the M2M gateway is offered.