Event detection and recognition are important for environmental monitoring in the Internet of things and cyber-physical systems. Low power wide area (LPWA) networks are one of the most powerful wireless sensor networks to support data gathering; however, they do not afford peak wireless access from sensors that detect significant changes in sensing data. Various data gathering schemes for event detection and recognition have been proposed. However, these do not satisfy the requirement for the three functions for the detection of the occurrence of an event, the recognition of the position of an event, and the recognition of spillover of impact from an event. This study proposes a three-stage data gathering scheme for LPWA. In the first stage, the access limitation based on the comparison between the detected sensing data and the high-level threshold is effective in reducing the simultaneous accessing sensors; thus, high-speed recognition of the starting event is achieved. In the second stage, the data centre station designates the sensor to inform the sensing data to achieve high accuracy of the position estimation of the event. In the third stage, all the sensors, except for the accessing sensors in the early stage, access the data centre. Owing to the exhaustive gathering of sensing data, the spillover of impact from the event can be recognised with high accuracy. We implement the proposed data gathering scheme for the actual wireless sensor system of the LPWA. From the computer simulation and experimental evaluation, we show the advantage of the proposed scheme compared to the conventional scheme.