Hyperspectral Modeling of Soil Organic Matter Based on Characteristic Wavelength in East China

Soil organic matter (SOM) is a key index of soil fertility. Visible and near-infrared (VNIR, 350–2500 nm) reflectance spectroscopy an effective method for modeling SOM content. Characteristic wavelength screening spectral transformation may improve the performance prediction. This study aimed to explore optimal combination characteristic selection hyperspectral SOM. A total 219 topsoil (0–20 cm) samples were collected from two types in East China. VNIR spectra measured laboratory. Firstly, after (inverse-log (LR), continuum removal (CR) first-order derivative (FDR)) spectra, wavelengths selected by competitive adaptive reweighted sampling (CARS) uninformative variables elimination (UVE) algorithms. Secondly, prediction models constructed based on partial least squares regression (PLSR), random forest (RF) support vector (SVR) methods using full wavelengths, respectively. Finally, types. The results as follows: (1) CARS algorithm screened 40–125 spectra. UVE 105–884 wavelengths. (2) For improved precision PLSR SVR methods. coefficient determination (R2) value validation CARS-PLSR (PLSR model combined with CARS) CARS-SVR (SVR ranged 0.69 0.95, relative percent deviation (RPD) 1.74 4.31. Lin’s concordance correlation (LCCC) values 0.83 0.97. UVE-PLSR UVE-SVR showed moderate precision. (3) accuracies Paddy better than those Shajiang black soil. RF performed worse both types, R2 ranging 0.22 0.68 RPD 1.01 1.60. (4) soil, (highest RPD, lowest root mean square error (RMSE)) CR-CARS-PLSR (R2 RMSE: 0.97 1.21 g/kg calibration sets, 0.95 1.72 RPD: 4.31) CR-CARS-SVR 0.98 1.04 0.91 2.24 3.37). LR-CARS-PLSR 0.93 0.86 1.44 2.62) FDR-CARS-SVR 0.99 0.45 1.58 2.38). suggested that CR FDR can significantly accuracy