Abstract: For the prediction problem in data analysis on near infrared diffuse reflection spectroscopy of soil nutrient, in this paper, the principal component regression and partial least squares regression were used to establish the mathematical models of the near infrared spectra of soil samples with different total nitrogen contents, and the prediction accuracy of the models were compared. The results show that the RMSEP is 0.040 by principal component regression and 0.034 by partial least squares regression respectively, with determination coefficient R²=0.873 1 by principal component regression and R²=0.903 5 by partial least squares regression through correlation analysis between the predicted value and the actual value of the total nitrogen content by means of model validation, which indicates the prediction accuracy of modeling by partial least squares regression is superior to that by principal component regression. The research results provides the basis for improving the detection accuracy of soil nutrients by near-infrared spectroscopy.

Key words: near infrared spectroscopy, soil, principal component regression, partial least squares, total nitrogen content