Comparison of Four Chemometric Techniques for Estimating Leaf Nitrogen Concentrations in Winter Wheat (Triticum Aestivum) Based on Hyperspectral Features

Four chemometric techniques for estimating LNC in winter wheat were compared by spectral features. The predictive power and impact of sample size were evaluated. Key results include: (1) partial least squares regression (PLSR) and support vector machines regression (SVR) performed better than the other two methods, with coefficient of determination (r (2)) values in the calibration set of 0.82 and 0.81 and the normalized root mean square error (NRMSE) values in the validation set of 5.48 and 5.94%, respectively; (2) the lowest accuracy was achieved using stepwise multiple linear regression (SMLR), with r (2) and NRMSE values of 0.78 and 6.52%, respectively; (3) the predictive power of the back propagation neural network (BPN) was enhanced as sample size increased. Sample size less than 80 is not recommended when using BPN. These results suggest that PLSR and SVR are preferred choices to estimate LNC in winter wheat, and BPN is recommended when a sufficient sample size is available.