Hyperspectral inversion models for soil organic matter content in the Yellow River Delta

【Objectives】Soil organic matter (SOM) can improve soil structure, promote formation of granular structure, increase soil loose, improve soil air permeability and water permeability, and promote plant growth and development. Although the traditional method is acurate in determination of SOM, it cannot provide real-time measurement. The sensitive wavelengthshigh correlated with the SOM content are screened to build a quick statistic model to estimate the SOM content in Yellow River delta.【Methods】Sixty soil samples were collected in the Yellow River Delta. Mathematical statistics was used to build the inversion model by studying the relationship between the spectral reflectance and the SOM content. The SOM contents of soil samples were analyzed by a chemical method, and their hyperspectral reflectences were measured in an indoor dark room environment by ASD Fieldspec3 spectrometer. Sensitive wavelengths were screened according to their correction with the SOM content, after the handling method of continuum-removal. The principal component regression analysis, multiple linear regression analysis, quadratic polynomial stepwise regression analysis and the support vector machine (SVM) regression analysis methods were used to establish models for predicting the organic content.【Results】 From the soil spectral reflectance curve, around 1400 nm, 1900 nm and 2200 nm bands have obvious moisture absorption valley. It is obviously improved the correlation between the SOM content and spectral reflectance after processing method of continuum-removal by comparing correlation. The seven higher correlated wavelengths, 1278 nm, 1307 nm, 1314 nm, 1322 nm, 1328 nm, 1334 nm and 1343 nm, are considered as the sensitive wavelengths to estimate the SOM content.The quadratic polynomial stepwise regression model is prevent to be the best inversion model for SOM content prediction in the Yellow River delta with the determination coefficientof 0.865 and the the largest determination coefficient of the validation set, which reaches 0.837.【Conclusions】 The processing method of continuum-removal could improve the correlation between the SOM and spectral reflectance. The near infraredspectral of 1278 nm, 1307 nm, 1314 nm, 1314 nm, 1328 nm, 1334 nm and 1343 nm are the sensitive wavelengths for estimating the SOM content. The quadratic polynomial stepwise regression model has the best fitting effect, is appropriate for estimating the SOM content in Yellow River Delta.