Vis/NIR Spectral Prediction Of Soil Carbon And Nitrogen With Different Particle Size

Vis/NIR spectral technology is a quick soil quantitative analysis without loss, which is a widely used and effective method. Numerous studies show that there is a strong linear correlation between the soil OC, water, N, other properties and spectroscopy. However the field measurements are affected by various environmental factors and weaken this linear relationship to some extent. Generally the spectral measurements are on dried and ground samples in the lab to remove the impacts of moisture, soil type, and texture and so on. The current researches are given only qualitative results about the accuracy of the prediction model affected by different particle sizes.But the reason why led to differences and the quantitative interpretation of the differences are not found yet.In this study, the relationships among the soil TN, OC, and Vis-NIR reflectance of soil sample of different particle sizes were analyzed using PLSR.32 samples were obtained from Hou Lake Farm in Qianjiang, Hubei province, of the particle sizes varied from 0.85～0.25mm,0.25～0.18mm,0.18-0.15mm, and<0.15mm which were separated by nylon sieves.The spectra data processed by intercepting, re-sampling, SG smoothing and 1st derivative were used to establish the PLSR models of TN and OC. The importance of each wavelength in the calibration models was also determined by VIP Graph.(1)Smaller the particle sizes and brighter the samples always present higher the spectra. The continuum removed method shows that particles<0.15mm had the least absorption depth than others which have no linearity correlation with their reflectance.(2) The concentrations of TN and OC have significant correlations with spectral response. An accurate TN prediction was obtained when the particle sizes varied from 0.18～0.15mm with Rval2 of 0.806. And an accurate OC prediction was obtained when the particle sizes varied from 0.85-0.25mm with Rval2 of 0.803.(3)The reason is that soil spectroscopy in different particle sizes have different VIP, which lead the detecting bands in different regions. So exactitude prediction expects proper samples processing.