Geochemical background in polluted river sediments: How to separate the effects of sediment provenance and grain size with statistical rigour?

• Toxic metal concentrations are strongly influenced by grain size in fluvial sediments.
• Ti is a better normalizing element than Al and Rb, but empirical testing is needed.
• Robust LTS regression of Pb, Zn, Cu, Cr vs. Ti removes the grain-size effect.
• Suitable subcompositions are needed for log-ratio approach to background definition.

Defining anomalies from geochemical background is a principal step in anthropogenic contamination studies and environmental risk assessment, and yet the complexity of processes affecting the composition of sediments makes this task extremely difficult. We address grain size and sediment provenance as two parameters influencing the geochemistry of contaminated fluvial sediments. We analysed 576 sediment samples from channels, floodplains, oxbow lakes, and dam reservoirs of the Morava River, Czech Republic. The geochemical data are compared with grain size and statistically processed using primary observations and log-ratios. An effective way to remove the grain size and provenance effect is to use enrichment factors calculated from least trimmed squares regression of Pb, Zn, Cu and Cr data vs. normalizing Ti, which is sensitive to grain size. Based on empirical testing, titanium proved more suitable for normalizing than did Al and Rb. Isometric log-ratio coordinates of the Pb, Zn, Cu and Cr data show a strong misfit with those defined by the enrichment factors. A log-ratio approach considering just subcompositions of elements with similar geochemical behaviour provides much better results.