Discriminating flint and chert raw materials with LA-ICP-MS/MS, compositional data analysis, and supervised machine learning for archaeological applications

Abstract

Identifying lithic production and distribution networks in flint and chert extraction areas is a key issue for better understanding of Neolithic societies. Determining artefact provenance by petrographic characterisation of flint and chert facies is generally efficient, but this method may sometimes be limited by factors such as the opaque white patina that covers many artefacts. This study presents a protocol for geochemical analysis using laser ablation–inductively coupled plasma–tandem mass spectrometry (LA-ICP-MS/MS), compositional data analysis (CoDA), and supervised machine learning (Linear Discriminant Analysis, LDA, Random Forest Leave-One-Out, RFLOO) to complement the petrographic determination of flint and chert sources. The protocol sought to discriminate siliceous raw material exploited in the Saint-Gond Marshes region (SGM; Marne, France) from the Campanian (Ca; Late Cretaceous) and the Bartonian (Ba; Eocene), and flint raw material exploited in the Pays d’Othe region (PO; Aube, France) from the Coniacian (Co; Late Cretaceous). We tested the discrimination potential of geochemical signatures at different spatial (inter- and intra-regional) and stratigraphic scales (Late Cretaceous vs Eocene; Coniacian vs Campanian), formed in different environments (marine vs lacustrine, or within the same sedimentary basin) on a corpus of 52 samples from 9 sites. Our results demonstrate that all three groups are clearly discriminated, as are all but one of the 9 sampling sites. The CoDA procedure transforms raw data into log-ratios discriminated by LDA and RFLOO: the results obtained are more robust than those obtained from the unprocessed raw data. Discrimination is linked to the geochemical and diagenetic processes involved in the formation of flint and chert nodules. Flints and cherts formed in different palaeo-environments will possess distinctive geochemical signatures. These results are very encouraging for sourcing archaeological artefacts, particularly when their provenance cannot be determined petrographically.