Abstract: To improve the prediction accuracy of soil organic matter content in typical black soil region via spectral processing and optimization of model methods, partial least squares regression (PLSR) model, back propagation neural network (BPNN) model and support vector machine (SVM) model were established based on the ground measured data and the first-order, second-order differential and principal component analysis data of image reflectance, and the best characteristic band selection and regression prediction of soil organic matter content in the study area were carried out. The results showed that different mathematical transformation of image band could enlarge some fine absorption characteristics of image data and highlight sensitive spectral information. The contribution rate of the spectral data after treatment was quantified by using the PLSR standardized model, and the best characteristic band was screened out along with the correlation coefficient. Among all the established models, the decision coefficient, root mean square error, and relative percent difference of the SVR model on the test set were 0.89, 2.81 g·kg^-1 and 2.14, respectively, which exhibited the best performance. The present study could provide a new method for the selection of the best characteristic band in the inversion modeling of soil organic matter content in black soil region, and provide reference for selection of the best soil organic matter content inversion model. Meanwhile, the established SVR model could be used for rapid monitoring of soil organic matter content in typical black soil region, and could provide digital support and theoretical basis for the future effective development of cultivated land.

Key words: soil organic matter, contribution rate analysis, fractional differential, composite model, support vector machine