A Systematic Review of IoT, Edge Computing, Time-Series Modelling and Control Systems in Precision Irrigation

Abstract

This review critically appraises the literature on the convergence of Internet of Things (IoT) sensing, edge computing and time-series machine learning for precision irrigation, checking predictive soil-moisture forecasting, edge-deployed neural controllers and LoRaWAN telemetry directly against independently published, peer-reviewed sources rather than taking them at face value. A structured search across four bibliographic databases identified 214 candidate records, of which 33 met the review’s evidentiary criteria and were carried forward into systematic comparisons spanning forecasting architectures, edge-hardware deployment and the broader artificial-intelligence and machine-learning literature on soil-moisture prediction. The underlying sensing, forecasting and edge-deployment techniques are found to be technically sound and increasingly well validated, but a recurring weakness emerges in how they are communicated: single-site accuracies, prototype hardware benchmarks and unreferenced market figures are frequently presented with the same confidence as multi-season, independently reproduced results, without distinguishing one evidentiary tier from another. Building on this pattern, the review develops a cross-cutting synthesis, illustrated with original figures and nineteen governing equations spanning hydrology, neural-network forecasting, sensor physics, and radio-link and power budgeting, showing that a single calibration-generalisation caution runs through statistical forecasting, edge-hardware and connectivity claims alike: a predictive irrigation model is only ever as trustworthy as the independent, multi-season validation behind it. A dedicated section generalises this synthesis across crops, formalising bang-bang, proportional-integral-derivative and model-predictive irrigation control, deriving governing equations and stability conditions from first principles, connecting them through Pontryagin’s Minimum Principle, and situating them against control practice reported for major world crops. A further section applies FAIR data principles to the underlying datasets, finding that published results are citable while the sensor logs, firmware and quantisation parameters behind them are rarely archived for independent verification. The review closes with recommendations for cross-site and cross-season validation, transparent hardware and power-budget reporting, FAIR-by-design data management, and explicit evidentiary tiering in future work.