Aroma evolution of Anhua Qianliang tea across different storage years: Machine learning-assisted discrimination of storage stages and screening of marker volatile compounds.

Anhua Qianliang tea (QLT) is a representative compressed dark tea whose aroma quality is progressively remodeled during long-term storage, yet the volatile basis underlying this transition remains poorly understood. In this study, quantitative descriptive analysis (QDA), GC × GC-QTOFMS, multivariate statistics, and machine learning were integrated to resolve the sensory evolution, volatile remodeling, and stage-related aroma markers of QLT across different storage years. Aging shifted the aroma profile of QLT from green and fresh notes in the early stage to woody, stale, and herbal characteristics in the late stage. Volatile profiling identified 281 compounds, and storage-stage differentiation was mainly associated with coordinated changes in lipid oxidation-derived volatiles, carotenoid-derived compounds, oxygenated terpenoids, and methoxybenzene derivatives. Correlation analysis further linked lipid-derived volatiles to early-stage aroma expression, whereas carotenoid-derived compounds, oxygenated terpenoids, and methoxybenzene derivatives were more closely associated with aged aroma. Differential screening identified linalool, β-ionone, dihydroactinidiolide, safranal and other compounds as potential stage-related markers. Machine learning further showed that CatBoost achieved the best classification performance, while SHAP analysis highlighted octanal, hexanal, (Z)-4-heptenal, among others, as major contributors to storage-stage discrimination. Together, these results support a stage-dependent volatile remodeling framework for QLT aging and provide a chemical basis for storage-stage identification and aroma quality evaluation.