Estimation of Nitrogen Concentration in Winter Wheat Leaves Based on a Novel Spectral Index and Machine Learning...

Assessing crop nitrogen status is crucial for optimizing fertilization strategies and promoting sustainable production. Although hyperspectral data offer significant advantages for monitoring subtle physiological changes in crops, accurately determining nitrogen status based on spectral information remains challenging. In this study, field experiments were conducted during the jointing stage of winter wheat on the Loess Plateau from 2018 to 2020. Concurrent measurements of leaf nitrogen concentration (LNC) and hyperspectral reflectance were collected to derive three types of spectral parameters: traditional vegetation indices, two-dimensional optimal spectral indices, and three-dimensional optimal spectral indices. Spectral parameters exhibiting a significant correlation with LNC (p < 0.05) were selected and combined as inputs for three machine learning models-extreme learning machine (ELM), back-propagation neural network (BPNN), and random forest (RF)-to develop LNC estimation models. The results demonstrated that, among the traditional indices, the Double Difference Index (DDn) showed the strongest correlation with LNC (r = 0.674). Within the multidimensional optimal indices, the differential three-dimensional scattering index (DTSI) exhibited the highest sensitivity to LNC (r = 0.721) at wavelength combinations of 833 nm, 755 nm, and 802 nm. Moreover, Model Input Combination 5 (comprising empirical indices plus three-dimensional optimal indices) further enhanced estimation accuracy. The RF model using Combination 5 achieved the best performance on the validation set (R² = 0.827, RMSE = 2.803 mg g^-1, MRE = 7.664%), significantly outperforming other model-input combinations. This study confirms the feasibility and high accuracy of winter wheat LNC inversion using novel multidimensional spectral indices and provides a new approach for real-time, non-destructive monitoring of nitrogen status in winter wheat.