Abstract
Wireless sensor networks (WSNs) for precision agriculture are constrained by limited node energy, weather-induced link variability, and latency requirements. This work introduces SWADS, a clustered WSN architecture in which an unmanned aerial vehicle (UAV) serves as a mobile sink and cooperates with two intelligence layers: (i) a long short-term memory (LSTM) forecaster for short-horizon weather prediction that triggers proactive UAV/fixed-sink handover during adverse conditions, and (ii) a reinforcement-learning (RL) policy for energy-aware cluster-head (CH) selection. In MATLAB simulations with 200 nodes over 8,000 rounds (first-order radio model, AWGN channel), the LSTM achieves ≈ 96% validation accuracy (> 97% training), enabling timely handovers that avoid predicted fades, while the RL policy selects near-optimal CHs with ≈ 95% accuracy, mitigating energy hotspots. SWADS sustains operation from first-node death (FND) at ~ 5,760 rounds to last-node death (LND) at ~ 7,032 rounds, demonstrating extended lifetime under clustered, mobility-aware routing. End-to-end delay remains low at ~ 1-1.2 ms on average, and packet loss is limited to ~ 6.04% despite channel noise, reflecting reduced contention via aggregation and shortened sink-CH distances. Throughput remains stable up to ~ 7,000 rounds with a peak of ~ 160 packets/round. Across baselines (static-sink LEACH-style, UAV mobile sink without weather awareness, and RL-based clustering without mobility), SWADS consistently delivers longer lifetime, lower delay, and more stable throughput. These results indicate that coupling weather-aware sink mobility with RL-driven clustering provides a robust and energy-efficient path to practical, long-lived agricultural WSN deployments.