PREDICTION OF FARMLAND SOIL ORGANIC MATTER CONTENT BASED ON DIFFERENT MODELING METHODS

Soil Organic Matter (SOM) is an important source of crop growth. Its content can reflect the status of soil fertility, has a significant impact on the growth and development of crops, and is one of the indicators of land quality. In this study, 190 soil samples were collected in the Weihe Plain as the research area. ASD Fieldspec 4 was used to obtain the spectral data of the soil sample in the 350-2500nm band, and the relationship between the SOM content and the soil reflectance spectrum was analyzed, and the effect of FD, SD, Log, FDL, SDI, CR combined with CARS in the characteristic band extraction was compared and analyzed. Cubist is used to establish a SOM estimation model to provide a basis for hyperspectral estimation of SOM content. The results show that the number of feature bands extracted based on CR-CARS is the least, and the accuracy is the best. Compare six kinds of transformations, CR has the least number of characteristic bands, which are 428 nm, 454 nm, 466 nm, 468 nm, 477 nm, 535 nm, 1406 nm, 1500 nm, 1609 nm, 1686 nm, 2172 nm, 2398 nm, 2399 nm, 2438 nm, 2449 nm. In the Cubist model, the coefficient of determination (Rv(2)) in the validation ranges from 0.60 to 0.97, and the ranges for RPD is 1.19 to 3.02. Comparing the model calibration set Rc(2), RMSEC and the validation set Rv(2), RMSEP and scatter plot distribution, the CR-Cubist model has the highest prediction accuracy, the calibration set Rc 2 is 0.96, the RMSE is 0.49, the validation set Rv(2) is 0.97, the RMSEP is 0.51, and RPD=3.02, which is stable, and the estimated model has better potential.