Horticultural Salinity-Stress Phenotyping and Tolerance Inference: A Critical Evidence Map and Validation Framework for AI-Related Claims

Abstract

Salinity stress constrains horticultural production in protected cultivation, hydroponics, coastal agriculture and reclaimed-water irrigation. This critical narrative review and evidence map synthesizes a DOI-verified core corpus of 160 peer-reviewed journal articles to ask how artificial intelligence (AI) can support, rather than overstate, salinity-stress inference in horticultural crops. The evidence base is uneven: 22 retained records were AI-, sensing- or phenotyping-relevant, six treated ML, deep learning, edge intelligence or agentic AI as a central method, and four directly tested salinity- or water-stress AI/sensor phenotyping in a crop-relevant system. Among the six explicit-AI records, none externally validated a salinity-specific AI model; one distinguished salinity from drought, and none reported a prospective AI-guided intervention trial. Accordingly, this article is framed as a validation and reporting framework, not as a quantitative meta-analysis of model or intervention efficacy. Across the broader corpus, evidence for salinity tolerance centres on osmotic limitation, Na+ and Cl− toxicity, K+ retention, ROS regulation, photosynthetic protection, hormonal signalling, root hydraulics, rhizosphere processes and metabolic reprogramming. The review links these mechanisms to measurable traits and to claim-specific AI validation requirements. We propose a mechanism-to-AI map, a validation ladder, an intervention maturity framework and a reporting checklist. The conclusion is deliberately conservative: AI can improve salinity research when it is constrained by rigorous metadata, physiological grounding, external validation, stress-confusion testing and explicit uncertainty, but current evidence is insufficient to support autonomous, economically validated or field-ready AI-based salinity management.

Plant phenotyping relevance

植物の塩ストレス表現型推定に関するAI・センシング研究を体系的に整理し、検証枠組み、検証段階、報告チェックリストを提案する方法論的レビューであり、表現型計測・推定が中心です。

Code and data availability

This is a narrative review/evidence map. It references Supplementary Information S1 (search logs and audit tables) but provides no public URL, deposit, or availability statement for any dataset, code, model, or phenotyping data. No paper-specific public asset is identified in the supplied blocks.

No evidence-backed public reproduction asset is currently recorded.