Estimation of the ratio of leaf carbon to nitrogen in winter wheat based on hyperspectral data and machine learning method

Plant C/N ratio, the ratio of carbon to nitrogen, is an important factor in agriculture and ecology, especially for the comprehensive diagnosis of C/N balance, growth and nutritional status of wheat plants. Rapid and accurate monitoring of spectral characteristics of crop canopy by remote sensing technology, fast and non-destructive, is of great significance for the dynamic regulation of crop growth in the field. In this study, partial least squares regression (PLSR) and random forest regression (RF) models were constructed to estimate C/N of winter wheat leaves by optimizing spectral variables. Results showed that (1) successive projections algorithm (SPA) algorithm combined with correlation analysis selected spectral variables with good sensitivity and weak multicollinearity; (2) Compared with the two inversion models of C/N in winter wheat leaves, the accuracy of PLSR model $\left(R^{2}=0.65, R M S E=1.97\right)$ was higher than that of $R F$ model $\left(R^{2}=0.63\right.$, RMSE $\left.=2.02\right)$; (3) The PLSR model has well accuracy and stability through the cross-validation of PLSR and RF models by random combination of multi-year samples. In this study, hyperspectral remote sensing accurately estimated the leaf C/N of winter wheat, which is helpful to further study the carbon and nitrogen metabolism status of winter wheat at different growth stages, and has greater exploration potential.