Multi-scale geographical origin discrimination of geographical indication rice based on multi-source data fusion...

Growing fraud incidences have made it more difficult to authenticate the provenance of geographical indication (GI) rice. Taking Panjin rice (a GI product recognized by both China and the EU) as a case, this study developed a multi-scale traceability model integrating stable isotope analysis, near-infrared spectroscopy (NIRS), mineral element profiling, and metabolomics with machine learning. At the inter-provincial scale, NIRS-based models showed superior discrimination performance, with linear discriminant analysis (LDA) achieving the highest accuracy (88.5%) after appropriate spectral preprocessing. Within Liaoning Province at the intra-provincial scale, both stable isotope and NIRS-based models performed excellently, enabling LDA-based models to achieve 100.0% classification accuracy. However, discrimination at the municipal-scale within Panjin City proved challenging when relying solely on either stable isotope analysis or NIRS. By incorporating mineral element and metabolomic profiles through mid-level data fusion, robust traceability at the municipal-scale was achieved, with multiple classifiers attaining 100.0% accuracy. These results underscore the necessity of a flexible, multi-technique, and data-fusion-driven strategy for high-precision geographical authentication across varying spatial scales.