Prediction of salinity ion content in different soil layers based on hyperspectral data

Abstract: Soil salinization is a worldwide environmental problem with severe economic and social consequences. It is necessary to develop a soil-salinity-estimation model to project the spatial distribution of soil salinity. In this study, the spectra characteristics and salinization parameters of the soils in the different layers in northern Ningxia Yinchuan were measured. Based on the soil science and geostatistics methods, the sensitive wavelengths and the best transformation forms of spectral reflectance to salinity parameters (soil pH value, electric conductivity(EC) and salt ions) in 0-5 cm and 0-20 cm were selected respectively, and then the soil salinity parameters monitoring model was established. The results showed that: 1) The variation trend of soil spectral characteristic curves of five different types and different salinization degrees was similar. Saline soil had the highest spectral reflectance, and slightly SO42- type soil had the lowest reflectance. Salinized soils had good spectral response characteristics in visible and near infrared spectra region. The reflectance had the most closely related to the content of SO42- in all salinity parameters (coefficient of correlation was 0.910 4) of 0-5cm layer. There were non-significant relationships between reflectance and the contents of CO32-, HCO3- and Cl-. The average coefficient of correlation of reflectance and SO42- in 0-20 cm layer was decreased 0.232 2 than in 0-5 cm. However, the average coefficient of correlations of reflectance and Cl-, K+, HCO3-, EC were increased 0.433 1, -0.343 3, 0.303 2, 0.296 2, and got significant level. 2) After the spectral reflectance were transformed in different methods, the correlation between the most sensitive wavelengths and each salinity parameters were enhanced to some extent, especially after the (R)′ (first order differential conversion) and (CR)′ (the first order differential after continuous removal). In 0-5 cm layer, (R)′ was the optimal transformation forms of reflectance for pH value, SO42-, K+, Mg2+, and the (CR)′ was best for EC, CO32-, HCO3-, Cl-, Na+ and Ca2+. In 0-20 cm layer, (CR)′ was the optimal index for soil pH value, lg(1/R)′ was the optimal index for EC, HCO3-, SO42-, Na+, Ca2+ were, and (R)′ were the best one for CO32-, K+, Mg2+. In addition, there are different sensitive wavelengths in different soil layers about the same salinity parameters. 3) In the models of PLSR (Partial least squares regression), the average determination coefficient (R2) between sensitive wavelengths of 10 salinity parameters were 0.820 8 and 0.890 7 in 0-5cm and 0-20 cm soil layer, respectively. The determination coefficient between sensitive reflectance and SO42- was 0.967 6 in 0-5 cm layer, and it was higher 0.077 6 than in 0-20 cm layer. The numbers of sensitive wavelengths reduced and R2 decreased that used PLSR method to established prediction model than used the SR (step-wised regression) method, but the R2 of the SR method also got the significant level. The results conformed that the prediction accuracy of models for SO42-, CO32- in 0-20 cm were lower than in 0-5 cm. However, the prediction ability of models for other salinity parameters in 0-20 cm was stronger than in 0-5 cm. The study provide some beneficial references for regional soil salinity prediction and configuration of plant structure.