Experimental Study on Soil Electrical Conductivity Inversion by Ground Spectral Measurement and SAR Data

Abstract

In this paper, the ground-measured spectral reflectance was combined with C-band microwave radar quadrupolarized backscattering data, and the characteristic bands were selected using partial least squares and correlation coefficient methods, and a model was developed to evaluate the degree of soil salinization. Using the spectral reflectance and its logarithmic, first-order and second-order derivatives of the four spectral data, correlation analysis was performed and found that the first and second order derivatives of the spectra were better correlated compared to the first two. The correlations of soil EC values in the four bands of 1584-1588 nm, 1802-1806 nm, 2201-2205 nm, and 2344-2348 nm transformed by second-order derivatives were 0.27, 0.34, 0.33, and 0.35, respectively, and there existed two bands of 1802-1806 nm and 2344-2348 nm that both had better soil EC correlation. The bands selected by the partial least squares method are more backward than those selected by the correlation coefficient method, and there are extremely sensitive bands, and the fit of the second-order derivative transformation model is better compared with that of the correlation coefficient method. By combining the second-order derivatives of reflectance, surface roughness and radar backscatter coefficients, the neural network model with the second-order derivatives of reflectance and radar backscatter characteristics was the best prediction model, and its R2 for soil EC was 0.8666.