Prediction of Soil Organic Carbon Contents in Tibet Using a Visible Near-Infrared Spectral Library

Accurate soil organic carbon (SOC) data are very important for management of agricultural production and climate change mitigation. Visible near-infrared diffuse reflectance spectroscopy is an inexpensive, non-destructive, efficient, and reliable technique for monitoring soil properties. Soil spectral libraries can contain large sets of diverse soil data for empirical calibration. In this study, we focused on improving the prediction accuracy of the SOC content at the local field scale in Tibet using field-wet, intact spectra and different spectral libraries. The direct standardization algorithm and piecewise direct standardization algorithm were used to remove the influence of environmental factors from the in situ vis-NIR spectra. These algorithms effectively removed the influence of environment factors from the field-wet, intact spectra. The ratio of performance to deviation values for prediction of the SOC content using the field and laboratory spectra with the local spectral library were 1.57 and 1.98, respectively. The local spectral library models outperformed spiked national spectral library models and had higher ratio of performance to deviation values for shrub meadows, forests, and the total dataset.