Instrument Reading Matches Tasters on Bai Mudan Grade

A chemical fingerprint read by machine learning assigned the correct official grade to 23 of 24 unseen bai mudan white tea samples, a study reported.

The work was published online on 28 June 2026 in Rapid Communications in Mass Spectrometry by researchers at Pusan National University in South Korea and Fuzhou University in China.

The pluck and cultivar fix bai mudan as a style. Within that style, China's standard further sorts the leaf into four official sub-grades, Special, Grade I, Grade II, and Grade III, and that finer sort is where a trained sensory panel, working by smell and taste, still does the work. The study tested whether the same sub-grade distinctions can be measured on an instrument instead.

The researchers took 120 samples, 30 per grade. Aroma compounds in the air above the leaf were measured by proton-transfer-reaction time-of-flight mass spectrometry, which sorts the compounds by mass as they come off. Non-volatile compounds, among them amino acids and soluble sugars, were measured separately by liquid chromatography and by assay. A two-stream model was trained on 96 samples and tested on the remaining 24, where it reached 95.8 percent accuracy. The single error fell between two adjacent high grades.

The higher grades were associated with floral and sweet aroma compounds and with higher amino acid and soluble sugar content. The lower grades were associated with greener and woodier compounds and with kaempferol-related markers.

The study is a demonstration on 120 samples from one dataset, not a grading system in commercial use. The grades themselves are unchanged.

Sources: Rapid Communications in Mass Spectrometry, Multimodal Transformer Fusion of PTR-ToF-MS Volatiles and Targeted Non-Volatile Metabolites Enables Data-Driven Grading of Baimudan White Tea; Chromatography Online (LCGC International), HPLC, PTR-ToF-MS Enable AI Grading of White Tea.