Characterizing typical farmland soils in China using Raman spectroscopy

Raman spectroscopy has been rarely applied in soil characterization due to the interference of fluorescence resulting from soil organic matter (SOM). However, valuable structural information is likely stored in the soil Raman spectra. This study was undertaken to investigate the potential of Raman spectroscopy in soil identification as an alternative tool to traditional methods, and the partial least squares (PLS) model was developed from Raman spectra to make a quantitative prediction of SOM. Diverse soil samples (n = 200) representing four typical farmlands in China were scanned with a portable Raman spectrometer (i-Raman® Plus, USA) from the spectra range of 180 to 3200 cm− 1. The Raman shift range of 1000–2000 cm− 1 was selected to establish an identification model. Probabilistic neural network (PNN) combined with principal component analysis (PCA) of the spectra data was employed to identify 200 soils and an acceptable result was obtained with an accuracy of 96% in validation. The PLS model with cross-validation was constructed to predict the content of SOM, and the best prediction model was calculated using spectra with the linear exclusion in selected range (RPDv = 1.92, Rv2 = 0.74, and RMSEP = 8.16 g kg− 1). This study illustrates that soil Raman spectra contain information of soil constituents even in the presence of fluorescence interference; moreover, it demonstrates that this contained information can play a vital role in the characterization of symmetric structures of SOM, which can provide complementary knowledge to understand molecular structure of SOM that infrared spectroscopy cannot offer.