The influence of soil particle sizes on hyperspectral prediction of soil organic matter content

The purpose of this study is to find the influence of soil particle sizes on hyperspectral prediction of soil organic matter content in order to get the appropriate soil sample particle size and reduce unnecessary work. Soil samples were sieved by 10, 20, 60 and 100 meshes and then were measured in the laboratory with ASD Fieldspec Pro Spectroradiometer. The soil spectral reflectance (R) was mathematical transformed into first derivatives of reflectance (R') and the logarithm of the inverse of the reflectance (Log (1/R)). Methods of Partial Least Squares Regression (PLSR), Support Vector Machine (SVM) and PLSR-SVM were used to build SOM hyperspectral prediction models. Results show that the soil particle size has an obvious effect on the soil spectral reflectance. The smaller the soil particles, the higher the soil spectral reflectance. When the soil particle size is less than 0.25mm, there are no more helpful for improving SOM prediction accuracy. SOM estimation accuracy by using PLSR-SVM method is higher than only using PLSR or SVM. This paper is helpful for soil sample pretreatments when using hyperspectral method to predict SOM content.