YOLOv8-Based River Debris Monitoring System With Meteorological Data for Flood Mitigation

This study proposes an integrated flood potential monitoring system that combines river debris detection with meteorological data, driven by Indonesia's persistent flooding disasters and the limitation of existing monitoring systems that rely on only a single type of hydrological data. Utilizing the YOLOv8 deep learning model trained on a custom dataset of 2,249 images, the study detects and quantifies floating debris density on river surfaces. Unlike prior studies that focus solely on debris detection or hydrological monitoring in isolation, this study integrates debris density with precipitation data from BMKG using the Simple Additive Weighting (SAW) method to generate a multi-parameter flood potential indicator, further considering the river's cross-sectional area to improve estimation accuracy. Results show that the YOLOv8 model achieved a precision of 0.788, a recall of 0.773, and an mAP50 of 0.814, with an inference speed of approximately 50 FPS, indicating strong detection capability alongside a favorable accuracy-speed trade-off compared to prior models such as A2ANet (mAP50 0.887 at ~18 FPS) and YOLOv7 (mAP50 0.760 at ~45 FPS). The wood and paper classes recorded the strongest detection performance, while the "other" class showed relatively lower accuracy due to overlapping visual characteristics. The integration of debris density and rainfall data via SAW produced flood risk classifications (SAFE, FLOOD WARNING, FLOOD DANGER) that were consistent with observed river conditions during testing. The findings were consolidated into a dashboard-based monitoring platform to support early flood warning. This study extends existing single-parameter flood monitoring approaches by synthesizing computer vision-based debris detection with a multi-criteria decision-making method, offering local governments and disaster management teams a practical, remotely accessible tool to identify blockage risks and schedule preventive river-cleaning actions before heavy rainfall occurs.