Errors induced by spectral measurement positions and instrument noise in soil organic carbon prediction using vis-NIR on intact soil

Soil spectroscopy potentially would become a routine method for the measurement of soil properties in the field on soil cores. However, many factors could lead to errors in the measurement of spectra from intact soil cores. So far, errors induced by spectral measurement positions and instrument noise have rarely been quantified. The present study evaluated these errors based on 160 intact ring core samples collected from 20 profiles in a forest, southwest China. Each ring sample was scanned using a visible near-infrared (vis-NIR) spectrometer at nine evenly distributed positions at both ends of the ring to characterize within-sample variability. In addition, 10 scans were made at each position to characterize instrument error. A position spectrum was then calculated by averaging the 10 scans at each position, while a sample spectrum was calculated by averaging all 180 scans of each sample. The samples were then tested for soil organic carbon (SOC) content using the wet oxidation method. Based on the sample spectra and SOC content, a partial least squares regression (PLSR) model was calibrated. The modeling error was empirically calculated. Each scan, position spectrum, and sample spectrum was then fed into the calibrated PLSR model to predict SOC content, and the prediction results were evaluated for errors induced by instrument noise and different spectral measurement positions. Results showed that the error induced by different spectral measurement positions was about 75% of the modeling error, while the error induced by instrument noise was small and negligible. These three errors in terms of standard error of prediction (SEP) accounted for 18.6%, 23.8% and 1.46% of the average measured SOC content of the samples, respectively. Besides, the error due to different soil core sampling positions was also considerable. In terms of SEP, this error accounted for 12.5% of the average measured SOC content. These quantified error budgets would help to account for uncertainty in the measurement of soil properties using spectroscopy in the field.