Estimation of SO ion in saline soil using VIS-NIR spectroscopy under different human activity stress.

SO ion is an important indicator of soil salinization degree, but there are few researches on quantitative inversion of SO content based on hyperspectral and fractional-order derivative (FOD). This study aimed to improve the prediction accuracy of SO content in arid regions using visible and near-infrared (VIS-NIR) spectroscopy. The study area was divided into three regions according to different human activity stress, namely, lightly affected region (Region A), moderately affected region (Region B) and severely affected region (Region C). The combination estimation method of spectral transformations (R, R, 1/R, lgR, 1/lgR), FOD, significance test band (STB), and partial least squares regression (PLSR) were been constructed, and four models (FULL-PLSR, FOD-FULL-PLSR, IOD-STB-PLSR, FOD-STB-PLSR) were also used to compare and analyze the estimation accuracy. Simulation results show that the optimal prediction model of three regions is FOD-STB-PLSR, its spectral transformation is established by R, 1/R and R in Region A, B, and C, respectively. Its RPD is 2.4701, 3.4679 and 1.9781, and its optimal FOD derivative is located at 1.8-, 1.1- and 1.1-order, respectively. It means that FOD can fully extract VIS-NIR spectroscopy details, the higher-order FOD is more capable of extracting characteristic data than low-order FOD, and the predictive ability of the best estimation model is very good, extremely strong and relatively good in Region A, B and C, respectively. Compared with the best IOD-STB-PLSR of each region, the RPD of the optimal FOD-STB-PLSR model has increased more than 38%, 32%, and 19%, respectively. This study shows that the proposed FOD-STB-PLSR model is suitable for estimating the SO ion content of saline soil under different human activity stresses, and the study can provide a certain technical reference value for the monitoring of saline soil in arid areas.