6G-Enabled FANET–IoT Framework for Intelligent Watershed Monitoring Using Multi-Agent Deep Reinforcement Learning

Abstract

Water ecosystems face increasing threats from pollution, climate change, and extreme hydrological events, while conventional monitoring systems often provide limited adaptability and spatial coverage. This paper proposes a 6G-enabled smart watershed monitoring framework integrating Flying Ad Hoc Networks (FANETs), Internet of Things (IoT) sensors, deep learning, and Multi-Agent Deep Reinforcement Learning (MADRL). Cooperative unmanned Aerial Vehicles (UAVs) interact with terrestrial IoT nodes through a simulated 5G/6G communication environment, while deep learning models support water-quality prediction and ecological-risk assessment. The MADRL framework enables UAV agents to collaboratively optimize sensing coverage, data collection, energy consumption, and pollution-event response under dynamic environmental conditions. The framework is evaluated in a Python-based simulation environment using five UAVs and distributed IoT sensing nodes under normal and pollution-affected watershed scenarios with communication impairments and environmental disturbances. Performance is compared with centralized static monitoring, rule-based UAV patrol, and single-agent reinforcement learning using spatial coverage, pollution-detection latency, prediction accuracy, false-alarm rate, energy consumption, and communication metrics. Results show approximately 40% higher spatial coverage, 60% faster pollution-event detection, 15% higher water-quality prediction accuracy, and 35% lower false-alarm rates than the considered baselines. The results also indicate improved communication resilience and energy-aware UAV coordination. As the evaluation is simulation-based, these findings demonstrate computational feasibility and comparative effectiveness under the specified assumptions rather than physical deployment. Real-world UAV testbed and watershed validation remain important future directions.