Edge AI–LoRaWAN Framework for Sustainable Aquaponics: Real-Time Monitoring and Prediction in Tropical Environments
Abstract
Sustainable aquaponics in tropical environments faces significant challenges from traditional cloud-dependent IoT systems, including communication latency, bandwidth constraints, and connectivity disruptions. This study introduces an Edge AI-LoRaWAN framework for real-time monitoring and autonomous control in tropical aquaponics systems. We conducted a comprehensive 60-day experimental validation using a 500-liter vannamei shrimp-lettuce aquaponics system in tropical lowland Indonesia, equipped with industrial-grade sensors that monitored water temperature, pH, dissolved oxygen, salinity, and environmental conditions at one-minute intervals. Random Forest regression models ran on Raspberry Pi 4 edge devices and were compared with cloud-based processing on AWS EC2, generating 86,400 total records across seven parameters. Data partitioning followed temporal ordering with 70% training, 10% validation, and 20% testing datasets to prevent information leakage. Five-fold time-series cross-validation ensured model stability and generalization performance across different temporal segments. Comparative evaluation with an LSTM model showed slightly higher predictive accuracy but at the cost of substantially increased latency and computational requirements. The Edge AI framework achieved statistically equivalent prediction accuracy (R² > 0.90) for critical water quality parameters while demonstrating superior operational performance: 93.3% latency reduction (27.3 ms versus 402.8 ms), 91.8% bandwidth savings (18.2 MB versus 219.6 MB daily), and 38.5% energy efficiency improvement (240 versus 390 mAh daily). Controlled stress testing under three scenarios showed improved system resilience, with recovery success rates rising from 81.3% (cloud) to 96.7% (edge) and preventing all critical parameter threshold breaches, compared with 12/30 failures in cloud systems. Results demonstrate that Edge AI-LoRaWAN integration provides autonomous, resource-efficient monitoring capabilities essential for sustainable aquaponics deployment in resource-constrained tropical environments, establishing distributed intelligence as a viable alternative to centralized processing paradigms for agricultural IoT applications.