Abstract
The agricultural sector is increasingly reliant on digital technologies to enhance efficiency and sustainability. Real-time variable rate irrigation (VRI) is crucial for optimizing water usage and crop yields, especially in regions facing water scarcity and fluctuating environmental conditions. However, the widespread adoption of advanced VRI systems is often hindered by unreliable network connectivity in remote agricultural areas. This paper addresses this challenge by proposing an optimized Edge AI framework designed for real-time VRI in low-connectivity environments. We develop a lightweight, energy-efficient AI model capable of on-device inference for processing sensor data (e.g., soil moisture, weather forecasts) and making immediate irrigation decisions. Our approach leverages model quantization and pruning techniques to reduce computational overhead and memory footprint, enabling deployment on resource-constrained edge devices. A simulated environment was used to evaluate the framework's performance against traditional cloud-based VRI systems and static irrigation schedules. Results demonstrate that the proposed Edge AI solution achieves comparable or superior irrigation accuracy while significantly reducing reliance on continuous network access. Specifically, the edge-based system exhibited a 15% improvement in water use efficiency and a 10% increase in yield simulation compared to a baseline cloud-dependent system under intermittent connectivity conditions. The framework also showed robustness in adapting to dynamic environmental changes, offering a viable solution for precision agriculture in challenging connectivity scenarios.