Integrated multi-elemental and 87Sr/86Sr isotopic fingerprinting coupled with explainable artificial intelligence for geographical authentication of wheat.

This study presents an integrated framework combining multi-elemental profiling and 87Sr/86Sr isotopic fingerprinting with machine learning and explainable artificial intelligence (XAI) for the geographical authentication of Italian wheat. A total of 122 samples collected from Northern, Central and Southern Italy over two harvest years (2023-2024) were analysed by ICP-MS and MC-ICP-MS. Elemental composition exhibited pronounced interannual variability, whereas the 87Sr/86Sr ratio showed greater temporal stability and a consistent link to geological background. Random Forest models achieved three-class classification accuracies of 0.75 ± 0.08 for 2023 data and 0.81 ± 0.06 for 2024. SHAP analysis identified the isotopic ratio, together with Zn, Ni, Mn and Cu, as the main contributors to classification. Results demonstrate that wheat geographical origin is reliably characterised by an integrated elemental-isotopic signature interpreted through explainable machine learning, supporting provenance assessment across the major Italian macro-areas over different harvest years.