Characterize and Explore the Flavor Profiles of Green Teas from Different Tenderness Levels of Camellia sinensis cv. Fudingdabai by E-Nose, E-Tongue and HS-GC-IMS Combined with Machine Learning

Understanding how leaf tenderness influences flavor attributes in green tea is crucial for optimizing harvest timing and processing strategies. This study comprehensively characterized the flavor profiles of Fudingdabai green teas at three tenderness levels—single bud (FDQSG), one bud + one leaf (FDMJ1G), and one bud + two leaves (FDTC2G)—using a multimodal approach integrating electronic nose, electronic tongue, HS-GC-IMS, relative odor activity value (rOAV) evaluation, and machine learning algorithms. FDQSG was dominated by fresh, grassy, and herbal volatiles such as (Z)-3-hexen-1-ol and nonanal, suggesting active LOX pathways. FDMJ1G showed a metabolic surge in floral compounds, especially linalool (rOAV > 7400), along with minty and fruity notes. FDTC2G featured roasted and cocoa-like aromas due to enhanced Maillard reactions and phenylalanine metabolism. KEGG analysis revealed significant enrichment in butanoate metabolism and monoterpenoid biosynthesis. Random forest–SHAP analysis identified 20 key flavor markers, mostly VOCs, that effectively discriminated samples by tenderness grade. ROC–AUC validation further confirmed their diagnostic performance (accuracy > 0.8). These findings provide a scientific basis for flavor-driven harvest management and the quality-oriented grading of Fudingdaibai green tea.