European Journal of Sustainable Agroecosystems

Background: Background: Nitrogen (N) is the single most yield-limiting and most frequently mismanaged nutrient in wheat (Triticum aestivum L.) production, and conventional destructive tissue analysis cannot support the timely, spatially explicit fertilization decisions required by modern precision agriculture. Hyperspectral reflectance sensing, which resolves canopy and leaf optical signals across hundreds of narrow contiguous bands spanning the visible, near-infrared, and shortwave-infrared regions, has emerged as a non-destructive alternative capable of capturing subtle biochemical and structural signatures associated with plant N status.
Objective: This review synthesizes peer-reviewed literature published mainly between 2015 and 2026 on hyperspectral reflectance-based estimation of wheat N status, with the aim of critically comparing methodological families, quantifying reported predictive performance, and identifying unresolved research gaps.
Sources of literature: A structured search of Scopus, Web of Science, PubMed, and Google Scholar identified 612 initial records; after de-duplication, title/abstract screening, and full-text assessment following a PRISMA-style workflow, 40 studies met the inclusion criteria for detailed qualitative and comparative synthesis.
Major findings: Four methodological families dominate the literature: empirical vegetation-index regression, chemometric/partial least squares regression (PLSR) full-spectrum models, machine-learning and deep-learning approaches, and physically based or hybrid radiative-transfer models. Red-edge-centred indices and their derivatives consistently show the strongest and most transferable correlations with leaf N concentration (LNC) and leaf N accumulation (LNA), with coefficients of determination typically between 0.60 and 0.90 across independent validation datasets. Machine-learning and deep-learning models frequently outperform classical indices in accuracy but at the cost of transferability and interpretability. Unmanned aerial vehicle (UAV) and airborne platforms have substantially expanded the operational scale of hyperspectral N diagnosis, while spaceborne imaging spectroscopy missions remain in an early operational phase.
Research gaps: Persistent gaps include limited cross-cultivar and cross-site model transferability, confounding of N signals with water stress and disease, inconsistent validation protocols, and a scarcity of studies translating spectral N diagnostics into field-deployable, cost-effective decision support tools.
Conclusion: Hyperspectral reflectance sensing offers a scientifically mature but operationally incomplete pathway toward precision N management in wheat. Future work should prioritize standardized multi-site benchmark datasets, hybrid physical-machine-learning frameworks, and integration with emerging satellite imaging spectroscopy missions to bridge the gap between research prototypes and on-farm application.