It is a feasible method to detect soil total nitrogen (TN) content with remote sensing based on NIR spectroscopy.
However, the accuracy of soil TN model was affected seriously by soil particle size. The spectral scanning results
showed that at the same soil TN content level, with the decrease of the soil particle size, the reflectance of soil samples
was reduced and the trend was not linear relationship. At the short wavelength (760-1100 nm) wave bands, there were a
little of differences; while at the long wavelength (1100-2500 nm) wave bands, there were great differences. Two
methods were adopted to eliminate the effect of soil particle size. The first method was to establish TN model by the first
order differential preconditioning method of the spectral data. The second method was to establish TN model with mixed
calibration set of different particle size soil samples after data preprocessing. Through the combination of the two
methods, The RC, RV, RMSEC, RMSEP and RPD of the model improved from 0.85, 0.31, 0.046, 0.132, 0.866 to 0.92,
0.86, 0.018, 0.091, 2.700 respectively. The results showed that the effect of soil particle size on prediction of soil TN can
be eliminated effectively.