Vis-Near IR reflectance spectroscopy for soil organic carbon content measurement in the Canadian prairies

Abstract

The objective of this study was to measure soil organic carbon (SOC) content for Canadian prairies soils using Vis-near IR reflectance (VisNIR) spectroscopy. Specifically, the effects of the spectral data pretreatment method and number of latent variables on SOC prediction were determined. In addition, the capability of VisNIR spectroscopy to capture SOC variability was evaluated. For these purposes, 491 soil samples of 0-30cm from two perpendicular 1600m long sampling transects (NS and WE transects with 239 and 252 samples, respectively) in the Canadian prairies were scanned by VisNIR spectroscopy. SOC content at one transect was predicted by models calibrated at the other transect. The potential of VisNIR spectroscopy in predicting SOC was verified in this area. Smoothing using cubic smoothing spline outperformed other pretreatment methods. The performance of SOC prediction improved and then worsened with the increase in the number of latent variables in the models. Wavelet transform indicated a similar pattern of high variances in the scale-location domains between the predicted and measured SOC; the predicted SOC showed a decreased structured variability at the NS transect and increased nugget effect at the WE transect. At both transects, weaker, but with the same levels, spatial dependency and greater correlation length were observed for the predicted SOC compared with the measured SOC. The obtained results can be applicable for SOC measurements with VisNIR spectroscopy.