Hyperspectral extraction of soil organic matter content based on principal component regression

Abstract

Soil organic matter (SOM) content is an important index reflecting soil fertility. It provides important information for digital soil management and resource evaluation if SOM can be estimated using hyperspectral technology. In this experiment, the NIR-visible spectral reflectance of soil samples was measured using an ASD2500 hyperspectrometer. Correlation analysis indicated that SOM has stronger correlation with the first derivative than with the original spectral reflectance. The sensitive bands were sub-divided according to bandwidth and the position of the most sensitive wavelengths. Then, the treatments including the significant variables were selected, collinearity diagnosis and data transformation were done through stepwise regression and principal component regression (PCR). Finally, averages of the first derivative in the bands 540–570, 880–930, 1240–1300, 1570–1600 and 1600–1630 nm were retained in the model, which was established for the black soil of the north-east of China. Furthermore, an F-test validated that the model is effective and practicable with a correlation coefficient of 0.78. © 2007 Taylor & Francis Group, LLC.