Sustainable Hierarchical Networking for Extreme-Edge AI-Based Wildlife Monitoring in Resource-Constrained...

Deploying artificial intelligence (AI) for wildlife monitoring in remote environments remains challenging due to limited energy, computational resources, and unreliable communication. Conventional edge AI systems rely on continuous sensing, high-power communication, and full-precision models, resulting in excessive energy consumption. This paper presents a green-energy-based hierarchical network architecture for sustainable wildlife monitoring at the extreme edge. The framework distributes sensing, communication, and AI inference across edge, far-edge, and extreme-edge layers. A low-power IEEE 802.11ah network with on-demand channel allocation reduces transmission overhead, while passive infrared (PIR)-based event-driven activation replaces continuous camera operation. Model quantization (FP32, FP16, INT8) and distributed split inference improve computational efficiency, and tracking-assisted communication eliminates redundant transmissions. Experimental results demonstrate significant sustainability improvements. PIR-based operation reduces daily energy consumption by 95.11% compared with continuous operation. Tracking-assisted communication reduces both bandwidth requirements and energy consumption by more than 99% (p<0.001). Token-based channel allocation provides 39.4% energy savings compared with Target Wake Time (TWT)-based scheduling (sensitivity range: 29.2-45.1%). INT8 quantization delivers 4.7× higher throughput, 61% lower memory usage, and 39% lower power consumption, with a 2.6% mean average precision (mAP) reduction. Field validation confirms the INT8 model's effectiveness with an overall mAP of 82.89% (86.40% daytime, 79.37% nighttime) and precision above 96%. Combined with PIR activation, the INT8 system achieves 63.76 Wh/day-a 95.17% reduction versus FP32 continuous operation. The integration of renewable energy harvesting, efficient sensing, lightweight AI, and communication optimization enables scalable, autonomous, and sustainable wildlife monitoring in resource-constrained environments.