Reflectance spectroscopy for the assessment of soil salt content in soils of the Yellow River Delta of China

There has been growing interest in the use of reflectance spectroscopy as a rapid and inexpensive tool for soil characterization. In this study, we collected 95 soil samples from the Yellow River Delta of China to investigate the level of soil salinity in relation to soil spectra. Sample plots were selected based on a field investigation and the corresponding soil salinity classification map to maximize variations of saline characteristics in the soil. Spectral reflectances of air-dried soil samples were measured using an Analytical Spectral Device (ASD) spectrometer (350-2500nm) with an artificial light source. In the Yellow River Delta, the dominant chemical in the saline soil was NaCl and MgCl2. Soil spectra were analysed using two-thirds of the available samples, with the remaining one-third withheld for validation purposes. The analysis indicated that with some preprocessing, the reflectance at 1931-2123nm and 2153-2254nm was highly correlated with soil salt content (S-SC). In the spectral region of 1931-2123nm, the correlation R ranged from -0.80 to -0.87. In the region of 2153-2254nm, the S-SC was positively correlated with preprocessed reflectance (0.79-0.88). The preprocessing was done by fitting a convex hull to the reflectance curve and dividing the spectral reflectance by the value of the corresponding convex hull band by band. This process is called continuum removal, and the resulting ratio is called continuum removed reflectance (CR reflectance). However, the S-SC did not have a high correlation with the unprocessed reflectance, and the correlation was always negative in the entire spectrum (350-2500nm) with the strongest negative correlation at 1981nm (R=-0.63). Moreover, we found a strong correlation (R=0.91) between a soil salinity index (S-SI: constructed using CR reflectance at 2052nm and 2203nm) and S-SC. We estimated S-SC as a function of S-SI and S-SI' (S-SI': constructed using unprocessed reflectance at 2052nm and 2203nm) using univariate regression. Validation of the estimation of S-SC was conducted by comparing the estimated S-SC with the holdout sample points. The comparison produced an estimated root mean squared error (RMSE) of 0.986 (S-SC ranging from 0.06 to 12.30gkg(-1)) and R-2 of 0.873 for S-SC with S-SI as independent variable and RMSE of 1.248 and R-2 of 0.8 for S-SC with S-SI' as independent variable. This study showed that a soil salinity index developed for CR reflectance at 2052nm and 2203nm on the basis of spectral absorption features of saline soil can be used as a quick and inexpensive method for soil salt-content estimation.