Integration of Biosensors, the Internet of Things and Artificial Intelligence for Sustainable Agriculture: A Critical Narrative Review of Analytical Capability, Systems Integration and Outcome Evidence

Résumé fourni par la source

Continuous, molecule-level measurement of crops, soils and livestock, transmitted through networked instrumentation and interpreted by learning algorithms, is widely presented as a route to agricultural systems that consume fewer inputs while sustaining output. The three constituent technology families, namely biosensors, the Internet of Things and artificial intelligence, have matured along largely separate research trajectories, and their combination is more often described than evaluated. This review asks whether the accessible evidence supports the sustainability claims attached to integrated biosensor, Internet of Things and artificial intelligence systems, and identifies where the integration fails in practice. Literature was identified through structured searching of open scholarly indexes and through citation tracking, appraised for methodological adequacy, and synthesised around four problems: the translation of analytical performance into agronomic decision value; the architectural constraints imposed by rural connectivity, energy budgets and interoperability; the generalisation behaviour of predictive models trained on curated data; and the quality of outcome evidence linking these systems to resource use and environmental effect. The strongest evidence concerns component-level performance. Plant-integrated and nanomaterial-based biosensors demonstrate sensitive and rapid detection of pesticide residues, stress signalling molecules and pathogen-associated volatiles, while supervised learning models achieve high classification accuracy on curated image collections. Evidence weakens at each subsequent stage. Field evaluations of sensor durability, calibration stability and sampling representativeness remain scarce; model accuracy degrades under conditions unlike those of the training data; and demonstrations of end-to-end closed-loop operation are uncommon. Outcome evidence is weakest of all, since input savings are frequently reported while economic returns remain inconsistent and system-level environmental accounting that would capture manufacturing burdens and behavioural rebound is largely absent. Adoption research indicates that realised benefit depends at least as much on farm structure, advisory capacity and data governance as on device performance, and that research effort is concentrated in regions containing a minority of the world's farms. Priorities are proposed for field-realistic validation, outcome-anchored trial design and evaluation within smallholder systems.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.