Exploring the joint compositional variability of major components and trace elements in the TELLUS soil geochemistry campaign (Northern Ireland)

Geochemical datasets are growing in both their number of samples and components, including elements which typical values cover the ranges of magnitude from % to ppm or even ppb. Moreover, the sources of variability increase as well. It is becoming thus increasingly necessary to have appropriate tools to explore this geochemical variability. An example of such framework is provided by any modern regional geochemistry campaign, typically having thousands of samples analysed for several tens of elements covering diverse geological units in non-homogeneous climate-landscape environments. Using the TELLUS soil geochemical survey of Northern Ireland, this paper primarily aims to present one such exploratory case study.

The typical tools to use in this context are all based on principal component analysis (PCA) and variants. Several different alternatives are considered: raw data, standardized data and log-ratio transformed data; classical PCA vs robust PCA; only major components vs. a mixture of major and trace elements. Using the 12 resulting analyses, insights into the nature of the dominant factors of geochemical variability are extracted, and the advantages of a log-ratio approach highlighted.

From a methodological point of view, we show: (i) that raw data should seldom be considered; (ii) standardised data do not show any difference relevant between robust and classical PCA; and (iii) that standardised data and log-ratio data deliver a roughly similar picture, but (iv) logratio data offer much more insights, especially when comparing robust and classical PCA using all variables available.

From the perspective of the case study presented, we conclude that the first two sources of soil geochemical variation are related to the background geology and the presence/absence of blanket peat (a major feature of the landscape of Northern Ireland). Which of the two factors is the dominant one depends on whether robust or classical PCA was used. Besides this (rather expected) result, the analysis showed in addition striking positive correlations between the logratios Se/LOI, MnO/CaO and I/Br; between Br/Ni, Cl/Cr, Sb/TiO2 and Cd/MnO; or between Cs/I, Rb/Zn and Cr/Ni. Also null correlations are detected between the ratios of the second and third groups. Taking into account information on the major lithologies present in Northern Ireland, the third group of ratios appears to describe lithological changes, while the second group of ratios is strongly related. to peat building processes, which not only enrich in LOI (as expected) but in Br, Cl, Cd and Sb as well.