Tea quality estimation based on multi-source information from leaf and soil using machine learning algorithm

Mineral nutrients play a significant role in influencing the quality of tea. In order to detect the quantitative relationships between tea quality and mineral elements from the soil and tea plant, samples of soil and tea leaves from 'Baiyeyihao' and 'Huangjinya' cultivars were collected from 160 tea plantations, and these were used to determine 16 types of soil mineral elements, 16 leaf nutrient elements, and 10 key tea quality compositions. Three predictive models including linear regression, multiple linear regression (MLR) and random forest (RF) were applied to predict the main constituents of tea quality. The usage of mineral elements in the soil and tea leaves improved the estimation accuracy of tea quality compositions, the RF performed best for EGCG (R² = 0.67-0.77), amino acid (R² = 0.61-0.88), tea polyphenols (R² = 0.68-0.77) and caffeine (R² = 0.59-0.68), while the MLR performed well for predicting the soluble sugars (R² = 0.54-0.84). The multi-source information demonstrated a superior accuracy in predicting the biochemical components of tea when compared to individual mineral elements.