Smart Agriculture and Environmental Monitoring: An Edge-AI, IoT, and Federated Learning Framework for Climate-Resilient Farming

This study presents a system architecture, algorithm, and evaluation protocol for smart agriculture and environmental monitoring. It is a method/architecture contribution and does not report field trial results. World food production is still under stress from climate variability, the availability of freshwater resources, and the cost of farm inputs, offering an incentive to transition to data-driven and sensor-based farm management practices. Internet of Things (IoT)-based agricultural monitoring is well developed, but most systems deployed focus on a single parameter (e.g., soil moisture) or the parameter is inferred solely in the cloud, adding latency and fragility to the deployment of systems in rural areas. In this study, we propose an integrated architecture for low-power multiparameter environmental sensing, LoRaWAN/NB-IoT connectivity, edge-based lightweight inference, and privacy-preserving federated learning with cross-farm model aggregation, which is further complemented by a digital twin layer for scenario simulation. The decision layer of the framework can combine data from ground sensors with UAV and satellite images for the multi-scale monitoring of crop health and environmental quality (air, water, and GHG proxies). We contribute the full system design, an algorithmic pipeline, a sensor and model configuration specification, and a pre-registered evaluation protocol intended to guide future field deployment; no field data have yet been collected and clearly labelled non-empirical templates showing the expected reporting format only, rather than measured outcomes. The architecture and protocol are instead grounded in, and cross-checked against, published state-of-the-art results, which report an accuracy range of 82–97% for federated crop-yield prediction and 20–30% water savings from sensor-controlled irrigation.