Estimation of soil organic matter content using selected spectral subset of hyperspectral data

Soil organic matter (SOM) content plays an important role in the global carbon cycle and agricultural activities. Reflectance spectroscopy has been recognized as a promising method to rapidly estimate SOM content. However, the existing estimation methods mainly apply partial least squares regression (PLSR) to the entire spectral region of hyperspectral data. Here we proposed a method to extract the informative spectral subset based on spectral characteristics of soil constituents, which was then used to estimate SOM content with PLSR. Genetic algorithm (GA) and variable importance in the projection (VIP) score of PLSR were adopted to further select spectral bands separately. Both laboratory spectra of soil samples collected from an agricultural area and a hyperspectral satellite image were used to evaluate the performance of the method. For the estimations of SOM content using laboratory spectra, compared with the estimation using the entire spectral region of 400–2400 nm, the model accuracy was improved by using the spectral bands associated with clay minerals and the combined spectral bands of organic matter and clay minerals. For the estimations using soil spectra from hyperspectral remote sensing image, the RMSE and R2 values were improved from 0.91% and 0.34 to 0.55% and 0.76 by using the spectral bands associated with organic matter in comparison with the entire spectral region of 390–1029 nm. The estimation model developed with GA-PLSR using soil spectra from the hyperspectral satellite image was applied to map SOM content. Results suggest that estimating SOM content using informative spectral subset is promising and can be transferred to the hyperspectral satellite image to map SOM content.