Estimation of leaf C/N in crops based on hyperspectral measurements and machine learning methods

Ratio of carbon to nitrogen (C/N) from crop leaves, defined as the ratio of LCC (leaf carbon concentration) to LNC (leaf nitrogen concentration), is one vital index for evaluating the balance of carbon and nitrogen, nutrient status and growth vigor in crop plants. Therefore, it is of great importance for efficient assessment of crop growth status in field to monitor and estimate leaf $\mathbf{C / N}$ quickly and accurately. In terms of close relationships between chlorophyll, nitrogen (N) and $\mathrm{C} / \mathrm{N}$, some typical indices aimed at N and chlorophyll estimation were tested to estimate $\mathrm{C} / \mathrm{N}$ in radish and grape leaves. The multi-temporal hyperspectral data from the two growth stages of grape and four stages of radish in 2023 were collected to extract the selected spectral indices for evaluating $\mathbf{C / N}$ in radish and grape. The results showed that two tested indices such as VI $\mathrm{I}_{\text {opt }}$ and RVI2 had the better performance of estimating C/N for the two crops. Machine Learning (ML) methods, LASSO regression and OWC (Optimal Weight Combination) algorithm were adopted along with spectral indices to improve the $\mathbf{C / N}$ estimates for the two crops, and both of the two methods acquired the better results with $R^{2}$ over 0.6. It indicates that monitoring of leaf C/N in radish and grape based on hyperspectral reflectance measurements with ML appears very potential.