Remote Sensing Inversion for Simulation of Soil Salinization Based on Hyperspectral Data and Ground Analysis in Yinchuan, China

The Pingluo area, as an experimental study area in Yinchuan, has been subjected to major environmental degradation due to soil salinization problems. Soil salinization is one of the main problems of land degradation in arid and semiarid regions. In the present study, remote sensing was integrated with mathematical modeling to evaluate soil salinization adequately. To detect soil salinization, soil water content and electrical conductivity of soil samples were analyzed. The reflectance of soil samples was measured using a spectrometer (SR-3500) with 1024 bands. Indices of soil salinity, vegetation and drought were analyzed using Landsat images over the study area. Based on Landsat images, physicochemical analysis, reflectance of sensitive bands for soil salinization and environmental indices, canopy response salinity index (CRSI), perpendicular drought index (PDI) and enhanced normalized difference vegetation index (ENDVI), a new model was established for simulation and prediction of soil salinization in the study area. Correlation analyses and multiple regression methods were used to construct an accurate model. The results showed that green, blue and near-infrared light was significantly correlated with soil salinity and that the spectral parameters improved this correlation significantly. Therefore, the model was more effective when combining spectral parameters with sensitive bands with modeling. After mathematical transformation of soil reflectance, the correlations of bands sensitive to soil salinization were 0.739 and 0.7 for electrical conductivity and water content, respectively. After transformation of vegetation reflectance, the correlation coefficient of soil salinity became 0.577. After inversion of the model based on soil hyperspectral and water content, the significance became 0.871 and 0.726, respectively, which can be used to predict soil salinity and water content. The spectral soil salinity model had a coefficient of 0.739 for soil salinity prediction. Among the salinity indices, the CRSI was selected as the most significant, with R2 of 0.571, whereas the R2 for PDI reached only 0.484. Among the vegetation indices, the ENDVI had the highest response to soil salinity, with R2 of 0.577. After scale conversion, the correlation percentages between CRSI and measured soil salinity and between ENDVI and measured soil salinity increased to 16.2% and 8.5%, respectively. Following the correlation between PDI and soil water content, the percentage of correlation increased to 11.6%. The integration of hyperspectral remote sensing, ground methods and an inversion method for salinity is a very important and effective technique for rapid and nondestructive monitoring of soil salinization.