Hyperspectral Response to Leaf Nitrogen in Sugarcane: Dynamic Effects of Cultivar, Growth Stage, and Leaf Position with Model Inversion

Rapid and non-destructive monitoring of leaf nitrogen (N) content (LNC) is essential for precision N management in sugarcane ( Saccharum officinarum L .). However, the accuracy of hyperspectral estimation is challenged by the dynamic interactions among cultivar, growth stage, and leaf position. This study systematically investigated the effects of these three factors on LNC and leaf hyperspectral reflectance (400–1000 nm) across six main sugarcane varieties. We identified sensitive spectral bands and developed LNC inversion models using Partial Least Squares Regression (PLSR) and Random Forest (RF). The results revealed highly significant interactive effects (P < 0.01) of cultivar, growth stage, and leaf position on both LNC and spectral responses. The 2nd leaf was identified as the most stable and optimal leaf for spectral monitoring, thus is recommended as the standard sampling target for field diagnosis. Correlation analysis showed a consistent negative correlation between LNC and reflectance in the visible region (520–570 nm), a positive correlation in the near-infrared (930–999 nm), and cultivar-specific responses in the red-edge region (700–730 nm). The first derivative (FD) transformation most effectively enhanced spectral features and correlations. The RF algorithm significantly outperformed PLSR, with the "FD + RF" combination yielding the best prediction performance for most varieties (e.g., test set R² = 0.50 for ROC22). This study elucidates the hyperspectral response mechanisms of sugarcane LNC under multi-factor interactions and provides a robust framework for developing cultivar-specific models, paving the way for precision N diagnosis in sugarcane production. Context: Although previous studies has resulted in substantial knowledge on crop N monitoring via spectroscopy, systematic investigations into the "leaf N content–spectral characteristics" response mechanisms in sugarcane at the leaf level remain limited. Aims: This study was designed to address these critical research gaps. The specific objectives were to: (1) quantify the independent and interactive effects of cultivar, growth stage, and leaf position on sugarcane LNC and spectral reflectance; (2) identify the most sensitive spectral bands responsive to LNC changes across different varieties; and (3) construct and evaluate robust estimation models for sugarcane LNC using advanced machine learning algorithms. Our findings are expected to provide a solid theoretical foundation and technical support for developing remote sensing technologies tailored for precision N management in sugarcane fields.