Knowledge, Innovation and Technology in Scientific Research Vol. 2

Closed-loop precision irrigation couples continuous measurement of soil, plant or atmospheric state to automatic actuation of water delivery, and the migration of inference from remote servers to battery-powered field devices has been presented as the development that makes such control practical at farm scale. This review critically evaluates the evidence supporting that proposition across three interdependent layers: the sensing and network hardware that constrains what can be measured and transmitted, the algorithms that convert measurements into irrigation decisions under severe memory and energy budgets, and the field deployments that test whether the resulting systems conserve water without agronomic penalty. Literature published between January 2015 and the final search date was identified through open scholarly databases and indexes, supplemented by targeted verification of bibliographic records and by searches of institutional sources, with foundational earlier work retained where conceptually necessary. The synthesis indicates that the strongest and most reproducible evidence concerns component-level performance. Quantised neural networks now execute on microcontroller-class hardware at sub-millijoule energy per inference, low-power wide-area radios support kilometre-scale field coverage, and soil-specific calibration reduces low-cost dielectric sensor error to values approaching those of research-grade instruments. Evidence for the closed loop as a whole is considerably weaker. Reported water savings, which cluster between roughly fifteen and forty-five per cent, derive predominantly from single-season, spatially unreplicated comparisons against loosely specified conventional practice, and the most striking algorithmic gains originate in crop-model simulation rather than in commanded irrigation. Energy autonomy is frequently asserted from datasheet arithmetic that field measurements contradict by an order of magnitude. Persistent gaps include the absence of shared benchmarks and open field datasets, the near-total lack of multi-season controlled trials with hydraulic verification of delivered water, and a geographical concentration of deployments that limits transferability to the smallholder systems most often invoked as beneficiaries. Confidence in current water-saving estimates should therefore remain provisional.