Calibration of XRF core scanners for quantitative geochemical logging of sediment cores: Theory and application

On-line analysis of split sediment cores by XRF core scanners has become increasingly popular in the past decade, because it allows nondestructive extraction of near-continuous records of element intensities from sediment cores with a minimum of analytical effort. A disadvantage of XRF core scanning relative to conventional geochemical analysis is the problematic conversion of core-scanner output to element concentrations. The main reason for this long-standing problem is the poorly constrained measurement geometry, attributable to inhomogeneity of the specimens (e.g. variable water content and grain-size distribution), irregularities of the split core surface, and in some setups, spatial variations in thickness of an adhesive pore-water film which forms directly below a protective foil covering the core surface. We propose a log-ratio calibration model for XRF core scanners, derived from a combination of XRF-spectrometry theory, principles of compositional data analysis, and empirical evidence. The log-ratio calibration model provides accurate and precise predictions of sediment composition (element concentrations) from XRF core-scanner output with a limited number of parameters, namely 2(D-1), where D equals the number of chemical elements whose concentrations are to be estimated. The model can accommodate the inherent non-linearity of the relation between (relative) intensities and concentrations. which is apparent from the fact that it provides unbiased estimates. An immediate corollary of our results is that log-ratios of element intensities, which are related to log-ratios of element abundances by a simple linear transformation, provide the most easily interpretable signals of relative changes in chemical composition. Consistent use of log-ratios of element intensities or concentrations should minimise the risk of drawing erroneous conclusions from geochemical proxies. The relative standard deviation (precision) of predicted element concentrations in core GeoB7920 is less than 2%. Stochastic simulations indicate that this level of precision can be attained with 40 randomly selected calibration specimens. Improved control over input errors and development of robust goodness-of-fit statistics allows XRF core scanning to be developed into a rigorous quantitative Measurement technology. The log-ratio calibration equation derived in this study may be adapted to inter-laboratory and inter-instrument calibration as well. (C) 2008 Elsevier B.V. All rights reserved.