UAV-Based Nitrogen Assessment in Wheat: A Systematic Review of Target Traits, Validation Rigor, Growth Stage Evidence, and Machine Learning Approaches

Excess or deficiency of nitrogen affects wheat yield significantly. Several destructive and non-destructive methods are used for nitrogen diagnosis to support precision fertilizer management in wheat. Recently, UAV-based remote sensing combined with machine learning has emerged as a promising approach for wheat nitrogen assessment. A systematic review was conducted to identify strengths and gaps in the methodologically diverse literature. The PRISMA approach was used to identify relevant literature from Scopus and Web of Science databases. The extracted data were used for comparative quantitative analysis to evaluate whether studies targeted direct nitrogen variables or proxy traits, how validation rigor influenced reported performance, and which growth stages were most commonly associated with nitrogen diagnosis. Across studies with comparable reported performance, direct nitrogen studies showed a median selected R2 of 0.855, while close-proxy and indirect-proxy studies showed median selected R2 values of 0.868 and 0.841, respectively. Validation design also differed markedly across the literature. Most studies relied on internal-only validation, and these studies showed a higher median selected R2 (0.860) than studies using independent-like validation (0.825), suggesting that reported performance may often be optimistic under less rigorous validation frameworks. Growth-stage analysis showed that nitrogen diagnosis was most commonly investigated from jointing to grain filling with most studies focusing on multiple growth stages rather than on a single stage. This indicates the use of a broader diagnostic window rather than identifying single stages of practical importance. In conclusion, the reviewed literature represents a mixture of direct nitrogen and proxy or indirect studies with stronger within-study predictive capacity than in providing robust and transferable performance for practical nitrogen management. Future research should focus on direct nitrogen diagnosis and adopt independent validation designs to link diagnosis outputs to actionable precision nutrient management.