Prediction of soil organic matter by Kubelka-Munk based airborne hyperspectral moisture removal model

Obtaining high-precision soil organic matter (SOM) spatial distribution information is of great significance for applications such as precision agriculture. But in the current hyperspectral SOM inversion work, soil moisture greatly influences the representation of the sensitive information of SOM on the spectrum. Therefore, a Kubelka-Munk theory based spectral correction model for soil moisture removal is proposed to improve the spectral sensitivity of SOM. Firstly, the soil moisture content was obtained by the use of a Kubelka-Munk based physical soil moisture model and an unmixing method. Then, the spectral correction model for soil moisture removal was implemented based on the quantitative description of the Beer-Lambert law. The results show that the proposed spectral correction model for soil moisture removal can significantly enhance the expression of the sensitive spectral features of SOM, especially for the short-wave infrared range. After moisture removal, the imaging spectral data were used for inversion, using the sensitive band at 0.69 μm and a support vector machine regression (SVR) modeling method. The Kubelka-Munk moisture removal model for soil moisture removal can improve the accuracy of SOM inversion by at least 22% comparing with the 0.69 μm original spectral inversion model, with R2 of 0.42. Moreover, the proposed model can also improve the accuracy of SOM inversion by 19% for the SVR statistical regression method, with R2 of 0.69. Finally, the SOM distribution maps based on sensitive band model and SVR regression method were analyzed. Findings show that the two methods have high consistency, but the statistical method obtains better details of the SOM spatial distribution, due to its higher accuracy.