Abstract:

Aims Our objectives were to determine the feasibility of estimating nitrogen content in fresh and dry wheat leaves using near-infrared (NIR) spectroscopy and chemometrics and to establish the near-infrared model for estimating nitrogen content in wheat leaves in order to lay a foundation for wheat nitrogen management.

Methods We conducted three field experiments with different years, wheat varieties and nitrogen rates and determined time-course near-infrared absorbance spectroscopy and total nitrogen content from fresh and dry wheat leaves. The methods of partial least squares (PLS), back-propagation neural network (BPNN) and wavelet neural network (WNN) were used to establish the calibration models, and a dataset selected at random was used to evaluate the established models.

Important findings Near infrared calibration models based on PLS, BPNN and WNN could be used to estimate nitrogen content in wheat leaves with high precision and stable performance, especially WNN. The validation results showed that the root mean square errors of prediction (RMSEP) for the power model are 0.147, 0.101 and 0.094, respectively, while those for the fresh leaves model are 0.216, 0.175 and 0.169, respectively. The correlation coefficients (R²) for all models are >0.84. Therefore, near-infrared spectrometry can be an efficient method to estimate the nitrogen nutrition of crops.

Key words: leaf, near-infrared spectroscopy, neural network, partial least squares, total nitrogen content, wheat