An Intelligent Fog–Cloud Smart Irrigation System with Hybrid Machine Learning and Evapotranspiration-Based Control

Abstract

Sustainable agricultural water management is increasingly important in arid and semi-arid regions, where limited freshwater resources, climatic variability, and inefficient irrigation practices can constrain long-term food production. This study proposes an intelligent fog–cloud smart irrigation framework that integrates Internet of Things sensing, LoRa/LoRaWAN communication, evapotranspiration-based irrigation calculation, machine-learning prediction, and cloud-assisted historical search within a hierarchical decision architecture. The system prioritizes local fog-layer processing for routine irrigation decisions, while cloud resources are reserved for long-term storage and fallback support when sensor information is incomplete or local predictions are insufficiently reliable. A K-nearest-neighbor classifier was evaluated using 104 site-day observations with stratified five-fold cross-validation and achieved 71.2% accuracy, outperforming the evaluated Random Forest, Logistic Regression, Decision Tree, and SVM-RBF models in several performance measures. Prototype testing verified the end-to-end operation of sensing, irrigation actuation, water-flow monitoring, and remote visualization. Using the recorded 52-day sensing dataset, the adaptive irrigation strategy reduced combined water use by 58.7% relative to a simulated daily fixed-schedule baseline and by 17.3% relative to an every-two-day schedule. Simulation-based system evaluation further indicated a 95.2% reduction in mean decision latency for fog processing compared with cloud-centered processing, while modeled communication overhead decreased by up to 87.5% in message count and 75.0% in transmitted data volume under complete sensor availability. Missing-sensor experiments demonstrated that the proposed fallback mechanism maintains decision continuity by progressively shifting uncertain cases from evapotranspiration calculation to K-NN prediction and cloud-assisted retrieval. The proposed framework demonstrates the potential of combining adaptive irrigation, distributed intelligence, and reliability-aware decision making to support more resource-efficient and sustainable agricultural water management.   Keywords: sustainable agriculture, Smart irrigation, Water conservation, Fog computing, Internet of Things, Machine learning, Evapotranspiration, LoRaWAN