Effect of wetting on the determination of soil organic matter content using visible and near-infrared spectrometer

The assessment of soil organic matter (SOM) content by proximal sensing using Visible and Near-Infrared (VNIR) reflectance spectroscopy of field soil samples is complicated by interactions with various soil constituents and moisture content. This study examined a total of 486 archived agricultural soil samples, covering a wide range of soil reflectance characteristics and soil textures. Spectral reflectance was measured with a spectrometer for samples with wetness ranges from air-dry to near saturation. Prediction models for soil water and SOM content were then developed using partial least square regressions (PLSR) combined with various reflectance spectrum pre-processing methods (standard normal variate transformation and detrend, as well as spectral derivatives). Our results indicate that the PLSR model based on spectra measured on sufficiently wet soil samples had slightly better accuracy for SOM predictions than models based on air-dried samples. The mechanisms of wetting to increase prediction accuracy were also explored. Important reflectance wavelengths associated with organic functional groups such as aromatics, aliphatics, and amides were identified through analysis of their variable importance in projections (VIP). Robustness of the developed models was tested against other two independent datasets (comprised of sample numbers 126 and 99 each), achieving prediction accuracies of mean bias difference (MBD) = 0.02%, root mean square difference (RMSD) = 0.99%, Ratio of Performance to Inter-Quartile (RPIQ) = 2.90 and MBD = −0.23%, RMSD = 1.35%, RPIQ = 1.44, respectively. These findings suggest that developing SOM prediction PLSR models based on sufficiently wetted soil samples may be a viable approach, particularly when developing models for operational use in the field.