Abstract: In recent years, tea production in China has faced challenges of labor shortages, driving urgent industrial demands for mechanized harvesting technologies. The prevalent issues of high leaf fragmentation and low harvest quality in current mechanical systems are closely related to tea plant phenotypic traits. This study investigated seven widely cultivated tea cultivars mainly being suitable for black/oolong tea production in Guangdong Province. Phenotypic data from 100 quadrat sample plots of seven tea cultivars was systematically collected. Multiple analytical approaches, combined with mechanical harvesting quality predictive models, were employed to identify critical determining traits. The correlation analysis demonstrates that leaf length, leaf length-to-width ratio, length of one bud with two leaves, and length of one bud with three leaves were positively correlated with mechanical harvesting quality, whereas the total bud density showed a negative correlation. Grey relational analysis reveals significant associations between the mechanical harvesting quality and the leaf length-to-width ratio, leaf width, length of one bud with one leaf, and length of one bud with two leaves. Random forest analysis identifies the leaf length-to-width ratio, third internode length, and fourth internode length as critical traits influencing the superior harvest rate, intact rate, and fragmentation rate in mechanical harvesting, respectively. Further validation combined four phenotypic trait groups (full trait set, correlation-significant traits, grey relational>0.9, random forest top 6) with four machine learning algorithms to predict three core mechanical harvesting traits. The results demonstrate the grey relational traits achieved the optimal prediction for superior harvest rate (R²=0.65, RMSE=0.04), and random forest-selected traits were suitable for intact rate prediction (R²=0.86, RMSE=0.02), and complete trait group shows the highest fragmentation rate accuracy (R²=0.92, RMSE=0.01). Integrated modeling analysis identifies the length of one bud with two leaves and third internode length as key mechanical harvestable traits influencing black and oolong tea production. These findings established theoretical foundations and provided critical parameters for breeding tea cultivars optimized for mechanical harvesting.

Key words: tea plant, mechanical harvestable traits, phenotypic screening, machine learning