Deciphering dilution, grain size, and provenance in sediment geochemistry

A considerable number of geochemical and granulometric datasets from various sediment sequences was gathered during the recent decades in the context of palaeoenvironmental and palaeoclimate reconstructions, assessment of human impacts on earth surface processes, and provenance tracing in fluvial environments. Although the large importance of grain-size control on sediment geochemistry has been known for many years and was explicitly declared in some review papers on geochemical provenance tracing, it still forms a challenge for current research. The problem is that provenance, grain-size, and weathering (post-depositional alterations) jointly control the resulting chemical composition of paleosol-loess sequences or floodplain deposits, and need hence to be distinguished from each other. However, in several recent studies data processing was simplified and the results were presented in an unequivocal manner, although interpretation of sediment composition is always rather equivocal. This was especially the case when geochemical datasets were subjected to automated data processing by software routines, instead of an expert-based examination of the individual datasets and a correct qualitative distinguishing of the individual controlling factors.

Data assessment should always start from understanding the major geochemical and sedimentological factors and processes behind data variability. This phase cannot be automated, and should mandatorily precede the selection of appropriate data processing routines. On the one hand geochemical compositions may be mainly controlled by varying percentages of ‘diluting’ components such as quartz (usually sand) or (detritic or autochthonous) carbonate, that can be corrected for by rationally chosen element ratios. Numerous complex mathematical approaches have been designed to address that issue, however, they do not always produce interpretable results and therefore need empirical (expert-based) verification. One the other hand, ‘dilution’ effects can interfere with grain-size control, that can be revealed by scatterplots of element ratios or the visualisation of element ratios and grain size classes. Furthermore, the recently established Bayes space methodology for modelling and analysing continuous distributive data can visualise the grain size control of element ratios for entire granulometric curves. Combined with regression modelling this allows statistically sound conclusions about grain size effects on the element ratios desired for interpretation. For example, varying grain-size preferences of feldspars or zircons can point to distinct source rocks and thus qualitatively reveal provenance changes. Provenance changes can only be quantified after deciphering and considering ‘dilution’ and grain-size effects, and only if the sediment sources have really distinct geochemistry. The provenance tracing cannot be automated or based only on the formal performance of statistical tools such as low values of RMSE.

Concluding, provenance tracing should be based on geochemically interpretable element concentration ratios after cross-checking for ‘dilution’ and grain-size control, best done ‘manually’ by assessing a series of (old-fashioned) scatterplots, preferably with the granulometry information implemented using the Bayes space methodology. To obtain sound conclusions it is also essential to phrase clear and testable research questions before any research, acquire suitable data really representing variability in studied sediment sequences and potential provenance areas, and use statistical methods respecting real data complexity.