关键词: 蜜蜂, 中华蜜蜂, 工蜂, 形态, 协同演化, 机器学习

Abstract: 【Aim】The morphology of Apis cerana cerana is influenced by geographical and ecological differences. This study aims to explore the morphological differences of A. c. cerana in different regions of Guizhou Province, Southwest China, so as to provide a reference for understanding its morphological differentiation across regions. 【Methods】A total of 7 524 worker samples of the A. c. cerana Yunnan-Guizhou Plateau ecotype were collected from 33 regions across Guizhou Province. The morphological parameters related to proboscis, legs, tergites, sternites and forewings were measured post-dissection. The random forest (RF), neural network (NN) and support vector machine (SVM) methods were employed for parameter tuning to identify the optimal morphological analysis model and screen the importance of variables. The K-means unsupervised cluster analysis was conducted based on the importance of variables. The Pearson correlation analysis was performed between important variables and temperature, precipitation and vegetation values. 【Results】 The machine learning results indicated that the optimal parameter tuning was achieved with the mtry set at 5 and random tree at 550. Among the three models, SVM outperformed RF and NN in terms of precision, recall and F1 score. The importance values for the wax mirror interval on sternite Ⅲ, wax mirror length on sternite Ⅲ, wax mirror width on sternite Ⅲ, cubital2 and forewing length were notably high. The cluster analysis results revealed that the worker samples of A. c. cerana from 12 regions such as Zhijin, Zhenfeng and Chishui clustered into one group, those from 8 regions such as Nayong, Hezhang and Ceheng clustered into another, while those from 13 regions such as Xifeng, Pingtang and Luodian were distributed across both groups. The correlation analysis result showed a highly significant positive correlation between the wax mirror length on sternite Ⅲ and the wax mirror width on sternite Ⅲ and forewing width, and the wax mirror width on sternite Ⅲ was also highly positively correlated with the forewing width. The wax mirror interval on sternite Ⅲ was significantly negatively correlated with the cumulative spring-winter temperature difference over three years, and cubital2 exhibited a significant negative correlation with the cumulative spring-winter temperature difference over three years. 【Conclusion】 The machine learning results indicate morphological differentiation of A. c. cerana across different regions in Guizhou Province. This SVM method offers new insights into bee classification and aids in understanding the evolutionary trends of A. c. cerana morphology within Guizhou Province. It also has implications for the conservation of local A. c. cerana germplasm resources in Guizhou Province.

Key words: Honey bee; Apis cerana cerana, worker bee, morphology, coevolution, machine learning