Comparing centred log ratio and conventional standardisation for principal component analysis of portable X ray fluorescence soil geochemistry in a peri urban retention basin

Soil geochemical data generated by pXRF measurements are usually treated as ordinary multivariate measurements, yet the analytical values are compositional. Each element is reported relative to a constrained total. This paper evaluates how that constraint affects the multivariate interpretation of geochemical sources in a peri-urban dry retention basin in Slovenia, central Europe. Sixty-one georeferenced soil profiles were sampled in 2 layers and analysed for 17 elements using an Olympus Delta Premium 50 pXRF. Conventional z-score PCA was compared with centred log-ratio (CLR) PCA, and the sensitivity of the first ordination axis was evaluated with MCD-based robust PCA. The contrast was large. In the upper layer, z-score PCA on the same 17-element matrix explained 26.1% and 21.1% of variance on PC1 and PC2, whereas CLR-PCA explained 44.8% and 33.7%. In the lower layer, the corresponding values were 31.5% and 21.7% for z-score PCA and 49.7% and 36.0% for CLR-PCA. CLR transformation sharpened the two leading axes into a clearer carbonate–phosphate versus silicate contrast and reduced the tendency of high-abundance elements to dominate ordination geometry; correlating these axes with geoaccumulation indices showed a moderate association with Cu and Zn enrichment ( r ≈ 0.4–0.5) and a weak association with Pb and As ( r < 0.15), so the leading CLR axes are best read as compositional contrasts that are only partly aligned with anthropogenic metal enrichment. However metal enrichment was better explained in the following PC’s. Robust PCA gave a moderate upper-layer correlation with classical PC1 scores, r = 0.787, but near-identity in the lower layer, r = 0.994, indicating that the upper layer is more sensitive to localised organic- and metal-enriched samples. The results show that preprocessing choice can change the apparent hierarchy of geochemical gradients in peri-urban soil monitoring, with consequences for exploratory zonation and map-based interpretation, although regulatory exceedance itself depends on individual element concentrations rather than on ordination choice.CLR-PCA offered a more defensible ordination space than z-score PCA, while robust PCA provided a useful diagnostic for outlier leverage; broader validation across sites, element suites and independent source evidence is needed before CLR-PCA can be generalised as a universal default.