Abstract: Spectroscopic real-time measurement plays an important role on fine fertilizer in soil, but the difference of soil particle size has brought great influences on the results of the measurement. It is studied the correlation between the reflectance spectrum of different particle sizes of soil and the prediction results of total nitrogen contents of soil by means of partial least squares, which is based on Fourier transform near infrared reflectance spectroscopy detection technology. Results show that the prediction models, established by the soil samples of same total nitrogen contents with different particle sizes（20, 40, 60, 80, 100 mesh）, exist significant differences, with RMSEP up to 3.99×10^-4, 3.04×10^-4, 1.78×10^-4, 2.23×10^-4 and 1.73×10^-4 g/kg, respectively. It is indicated that the smaller the particle size was, the higher the prediction precision was. And it is verified that the particle size has an obvious influence on soil nutrient detection with near infrared reflectance spectroscopy, which has practical values for further study on correction to particle size.

Key words: soil particle size, total nitrogen content, near infrared reflectance spectroscopy, prediction accuracy