This paper presents the development of an AIoT-based early flash-flood warning system to enhance disaster resilience in smart rural communities. The framework integrates multi-source hydrological sensors, AI-enabled edge–cloud computing, and a mobile alert application to provide real-time monitoring and short-term flood forecasting, and includes an intelligent hybrid model combines YOLOv10 for visual water-level detection from CCTV imagery with a long short-term memory (LSTM) network for hydrological time-series prediction. The system was deployed and evaluated at two sites in Thailand: the Ban Luang station in Chiang Mai and the Chumkho station in Chumphon. The experimental results show near-perfect detection performance by YOLOv10, with precision and mAP@0.5 exceeding 0.99 across varying water-level conditions. The LSTM model achieved high forecasting accuracy, with an R2 of 0.987 at Ban Luang and 0.781 at Chumkho, reflecting site-specific hydrodynamic complexity. The results confirm that integrating AIoT-based visual sensing with data-driven forecasting significantly improves the reliability, responsiveness, and robustness of early flash-flood warning systems in rural environments.
Wiangnak et al. (Tue,) studied this question.