Estimating corn leaf nitrogen content using airborne VNIR-SWIR hyperspectral imagery

Abstract Purpose Applying nitrogen at the right timing and rate is critical to improving nitrogen use efficiency and maximizing yield. This study aimed to evaluate an airborne hyperspectral imaging system covering the visible–near-infrared (VNIR) and shortwave infrared (SWIR) spectrum (400–2500 nm) for nondestructive quantification of corn leaf nitrogen content (LNC). Methods A field experiment was arranged as a randomized complete block design with six nitrogen rates (0–336 kg ha⁻¹) and two application timings (at-planting or split applied), each replicated four times. Hyperspectral imagery was acquired during the critical vegetative stage, and leaves from five consecutive plants per plot were sampled for laboratory LNC analysis. Images were processed using the first derivative (FD), Savitzky–Golay smoothing, and Gaussian filtering, then followed by six machine learning algorithms to predict LNC. Results Partial least squares regression (PLSR) achieved the highest accuracy (R² = 0.82; RMSE = 0.23%, g per 100 g dry weight), and FD outperformed the other two pre-processing methods. A three-band vegetation index, (R 588 − R 1667 )/(R 588 + R 1667 − R 1205 ), was strongly correlated with LNC ( r = 0.93). LNC at V8–V9 was positively correlated with final grain yield ( r = 0.62 and 0.83 for single and split applications, respectively). Conclusion The results demonstrate that combining VNIR-SWIR hyperspectral imaging with FD-based processing provides a rapid, nondestructive approach for precision nitrogen management in corn.