arXiv:2509.17676v3 Announce Type: replace Abstract: The rapid advancement of Low-Power Wide Area Networks (LPWANs), particularly Long Range (LoRa) technology, has positioned them as a cornerstone of Next-Generation Internet of Things (NG-IoT) applications. Despite their long-range and low-power advantages, achieving high energy efficiency (EE) in LoRa networks remains a significant challenge in highly dynamic environments. Traditional terrestrial gateway deployments often suffer from coverage gaps and non-line-of-sight (NLoS) propagation, while satellite-based alternatives incur excessive energy consumption and prohibitive latency. To address these limitations, we propose a multi-UAV architecture in which unmanned aerial vehicles (UAVs) serve as mobile LoRa gateways to dynamically collect data from LoRa end devices (EDs). We formulate a joint optimization problem to maximize the weighted system EE by jointly optimizing spreading factors, transmission powers, UAV trajectories, and ED-UAV associations. This problem is transformed into a partially observable stochastic game (POSG), which we solve using our proposed Green LoRa Multi-Agent Proximal Policy Optimization (GLo-MAPPO). We adopt PPO because its clipped surrogate objective discourages excessively large policy updates, thereby promoting stable learning as multiple UAV agents simultaneously adapt their behaviors. Combined with centralized training and decentralized execution (CTDE), this enables coordinated policy learning while allowing each UAV to make decisions independently during execution. Simulation results show that GLo-MAPPO consistently outperforms state-of-the-art multi-agent reinforcement learning (MARL) benchmarks in terms of EE and power consumption across varying network densities. We further validate the trained policy through closed-loop, packet-level co-simulation using the OMNeT++/FLoRa LoRa simulator. Finally, ....