Quantitative Prediction of Soil Salinity Content with Visible-Near Infrared Hyper-Spectra in Northeast China

Studying the spectral property of salinized soil is an important work, for it is the base of monitoring soil salinization by remote sense. To investigate the spectral property of salinized soil and the relationship between the soil salinity and the hyper-spectral data, the field soil samples were collected in the region of Northeast China and then reflectance spectra were measured. The partial least squares regression (PLSR) model was established based on the statistical analysis of the soil salinity content and the reflectance of hyperspectra. The feasibility of soil salinity prediction by hyperspectra was decided by analyzed calibration model and independent validation. Models accuracy was also analyzed, which was established in the conditions of different treatment methods and different re-sampling intervals. The results showed that it was feasible to predict soil salinity content based on measured reflectance spectrum. The results also revealed that it was necessary to smooth measured hyperspectra for spectral prediction accuracy to be improved significantly after smoothing. The best model was established based on smoothed and log(1/x) transformed hyperspectra with high determination coefficients (R-2) of 0.667 7 and RPD=1.61, which showed that this math transformation could eliminate noise effectively and so as to improve the prediction accuracy. The largest re-sampling interval is 8 nm that could meet the accuracy of the soil salinity prediction. Therefore, it provided scientific reference of monitoring soil salinization by remote sensing from satellite platform.