Non-destructive quick detection based on Vis-NIR spectra for content of soil available nitrogen

Abstract. In this paper, visible and near infrared (Vis-NIR) spectral technique and chemometrics were investigated for the fast and non-destructive determination of soil available nitrogen content. The Vis-NIR diffuse reflectance spectra of 2 mm dried soil surface was collected by USB2000+ spectrometer, and the available nitrogen of soil was detected by alkaline hydrolysis diffusion method. Three different pretreatment methods were used to process the raw spectral data, including Savitzky-Golay (SG) smoothing, multiplicative scatter correction (MSC), standardized normal variable (SNV), and then the preprocessed spectral data was combined with partial least squares regression (PLSR) method to establish the prediction model. In addition, the precision of prediction model was evaluated by the correlation coefficient and the standard deviation of the model. The results showed that the optimal pretreatment method was Savitzky-Golay (SG), the highest precision model was established by Savitzky-Golay (SG) smoothing pretreatment method combined with PLSR, and the correlation coefficient of the validation set (Rv) was 0.8389, the validation standard deviation (SEP) was 9.6564. Therefore, it is feasible to apply Vis-NIR spectral technique in fast and non-destructive determination of soil available nitrogen content, which provides a theoretical basis for further development in real-time and in-situ detection equipment for soil available nitrogen content.