Abstract: Soil organic matter (SOM) is a key indicator for soil fertility. To explore the estimation ability of hyperspectral information for soils with low SOM content, this study focused on the surface soil (0-20 cm) of Hongsibu and Xidatan in Ningxia, and used field hyperspectral reflectance and SOM content as data sources. Four mathematical transformations were performed on the original spectral reflectance: reciprocal transformation first-order differential (RTFD), logarithm transformation first-order differential (LTFD), logarithmic reciprocal first-order differential (LRFD), and first-order differential (FD). Correlation analysis (PCC) and stepwise regression (SR) were used to screen for the sensitive bands. A SOM estimation model was established based on principal component regression (PCR), partial least squares regression (PLSR), support vector machine (SVM), and geographic weighted regression (GWR). The correlation between the original spectral reflectance and SOM was significantly enhanced after mathematical transformation, with the largest increase in FD transformation. The combination of PCC and SR was used to screen sensitive bands, which greatly reduced the dimensionality of the data, reduced the computational difficulty of the model, and improved its estimation ability. Based on the LTFD-SVM model, the estimation performance was optimal, with modeling determination coefficients (R_c²) and validation determination coefficients (R_p²) of 0.7593 and 0.9321, respectively. The root mean square error (RMSE) and relative analysis error (RPD) of the validation set were 0.62 and 3.47, respectively. The results can provide a reference basis for quantitative monitoring of low-content SOM in the study region and similar regions.

Key words: hyperspectrum, soil organic matter, estimation model, geographically weighted regression, support vector machine, inverse distance weighting