Machine learning-based geographical origin traceability of Gastrodia elata via integrated elemental-functional fingerprints and environmental response analysis

The tuber of Gastrodia elata (GR), a traditional food and medicinal homologous substance, exhibits significant variation in quality and market value across geographical origins, highlighting the need for reliable authentication methods. This study established a high-precision classification model for the origin of the mainstream cultivated variety Gastrodia elata f. elata (Hongtianma) by integrating multi-element fingerprints and functional chemicals, while also examining the response of these factors to bioclimatic factors. A total of 270 batches of GR samples from 23 counties and cities across four major producing regions were analyzed for inorganic elements and functional compounds. Nine machine learning algorithms were systematically compared through repeated stratified sampling combined with nested cross-validation, and key discriminatory variables were further correlated with bioclimatic factors using Mantel tests and Redundancy Analysis. Results revealed distinct regional chemical patterns of GR, with the Support Vector Machine (SVM) model demonstrating the best overall performance, achieving a test set accuracy of 92.53% alongside superior generalization capability (Kappa = 0.88, F1 = 87.16%). Key discriminant variables included elements such as Cd and Ca, as well as compounds such as p-hydroxybenzyl alcohol and parishin C. Environmental factors, particularly temperature regimes (e.g., MTCO, AMT) and precipitation (AP), significantly correlated with chemical characteristics and contributed to the geographical differentiation of GR quality. This integrated data strategy provides a reliable tool for GR origin traceability, elucidates environmental driving mechanisms, and offers scientific support for its quality standardization, market regulation, and sustainable cultivation.