Extraction Of Spectral Characteristic Information And Quantitative Analysis For SOM By Using Near-infrared Spectroscopy

Soil organic matter (SOM) which is always used as a vital indicator to reflect the soil fertility is one of the important soil components. Timely monitoring the content of soil organic matter is the foundation of field management. The traditional test method to obtain the soil organic matter has been difficult to meet the requirements of modern agricultural production. Hyperspectral technology has become an important approach to obtain soil attributes information due to the characteristics of rapid, simple, no pollution and no damage, etc.The 0-20cm topsoil samples were collected from the winter wheat field based on the nitrogen application experiment in Shanxi Agricultural University'agricultural. The SOM and its soil spectra were measured by using the 2 mm soil. Three method styles of conventional processing (smoothing methods), transformation processing and correction processing were applied on preprocessing the soil spectra. The correlation analysis between the SOM and these preprocessed soil spectra and the quantitative models of SOM were established based on the full spectrum by using the method of partial least squares regression (PLSR). The MLR models were also constructed based on the sensitive wavelengths extracted with the method of successive projections algorithm (SPA). Comprising the performance of PLSR model and SPA-MLR model, the optimal model for predicting the SOM in the winter wheat field was determined. The conclusions as follows:1) The preprocessed spectrum could improve the correlation coefficient between SOM and the spectra compared with the correlation with the origin spectra. Among these spectral preprocessed methods, second order differential, standard normal variable transformation, continuum removal method had a better performance.2) The sensitive wavebands extracted with the SPA method were mainly located in the center of 1400 nm and the region of 2125-2450 nm, and these wavebands proved to be directly or indirectly correlative with SOM.3) The PLSR and MLR model based on the continuum removed reflectance performed best than other preprocessed methods with high predictive accuracy (R~2=0.648, RMSE=0.405, RPD=1.46) and (R~2= 0.612, RMSE= 0.432, RPD= 1.256), respectively. It indicated that the continuum removed method was superior to other preprocessed methods. For the multivariate methods, the PLSR model with full spectrum performed more accurate and robust than the SP A-MLR models with the extracted sensitive wavebands.