Statistical Understanding on the Pre-processing of Vnir Spectra Data from Soil Samples with Different Preparations

Statistical analyses of the relationship between visible and near-infrared (VNIR) spectra and soil parameters usually require the normal distribution of wavelength variables. Most previous studies examined only the distribution of soil parameters, while the distribution of each VNIR spectra wavelength was ignored. Moreover, how the sample preparation process and spectral pre-processing procedure influence the distribution of each spectral wavelength has seldom been reported with data collected at a regional scale. With 71 soil samples collected from the Le'an River floodplain, China, our study analyzed the distribution of spectral wavelength variables in the VNIR region (350–2500 nm). More specifically, by means of the Lilliefors normality test plot and the kurtosis and skewness curve plot proposed by this study, we aimed to (1) test the normality of each wavelength using the Lilliefors normality test; (2) compare the kurtosis and skewness of wavelength variables between soil samples, with and without preparations. Results show that most wavelength variables from differently transformed spectra are not normally distributed. Soil sample preparation processes such as air-drying, grinding, and 0.2 mm sieving, can significantly change the distribution of wavelength variables. Thus, statistical methods with normality assumption in the wavelength variables would not be appropriate for data of this kind.