Soil organic carbon determination for Louisiana soils via portable x-ray fluorescence spectroscopy

Abstract

As scientists work to investigate the mechanisms underlying the depletion of carbon from terrestrial ecosystems, there has been an urgent call for the development of test methods that offer reduced analysis times for soil organic carbon (SOC) determinations. Traditional laboratory techniques can be time-consuming and costly, making high-volume sample analyses problematic. Portable x-ray fluorescence spectroscopy (PXRF) provides quantitative, multi-elemental data for soil samples in as little as 60 seconds; and, unlike other spectroscopic methods, differences in elemental concentrations between field-moist and oven-dry samples are considered to be negligible when soil moisture contents are less than 20%, by weight. This study aims to evaluate the performance of various SOC prediction models, constructed from PXRF elemental data from 300 soil samples collected from alluvium and loess parent materials found in Louisiana, USA. Elemental data, in addition to pH and depth measurements, were used in the construction of prediction models using multiple linear regression (MLR) analysis and principal components analysis (PCA) statistical techniques. Previous research indicates that the use of a stability index may enhance SOC prediction modeling capabilities. Therefore, models utilized relative elemental abundances on the basis on Zr and Ti concentrations, and performances were compared to those resulting from models developed from ‘Raw’ PXRF data. Results show that models constructed using field-moist PXRF elemental data provide excellent SOC prediction capabilities (R > 0.90) for both alluvium and loess datasets. Optimal performances resulted from the use of Ti as a stability index for field-moist datasets, producing accurate SOC predictions for both wet and dry validation sub-datasets. Findings indicate that PXRF elemental analysis, conducted under field conditions, provide for accurate SOC content determinations via MLR modeling of Louisiana alluvium and loess soil types examined in this study.