Inversion, Based On Laboratory Spectra Of Soil Nutrient Elements

In this study, using the laboratory spectra of Tianjin area, we analyzed the relationships between the nutrient elements (N, P and K) and VNIR (visible - near infrared) reflectance. This paper introduced the soil sampling and spectral pre-processing, spectral feature analysis, and the regression analysis between the nutrient elements and spectral feature parameters.Main content and conclusion as follows:A parabola correction was performed to the original spectra to eliminate the broken point between the VIS range and the NIR range. Then a Hamming window with length 9 was used as the filter to smooth the spectra.Several parameters of spectra were calculated, including first derivative reflectance spectra (FDR), inverse-log spectra (log (1/R)), Band-Depth, Band-Depth normalized by Depth (BND) and Band-Depth normalized by Area (BNA)) to establish the statistical equations for nutrient elements by SMLR and PLSR method, and the regression models were validated using the validation dataset.The results indicate that the concentrations of soil nutrient elements have significant correlations with spectral response. Partial least squares regression (PLSR) has unique advantages, which is also approved by the results of our study. In summary, the precisions of the PLSR models are higher than the SMLR models, but the differences are not very significant and the regression coefficients of SMLR models are far fewer than PLSR models. Thus these two regression methods both have its advantages and disadvantages. In SMLR method, Reflectance and log (1/R) get better results. While in PLSR method, Depth and FDR get better results and higher convergence speeds. Either in SMLR analysis or in PLSR analysis, Reflectance and log (1/R) have almost same results. From this study, we find that log (1/R) could hardly improve the predicted precisions.