Smith Colleen, Bachelder Nick, Russell Avery L, Morales Vanessa, Mosher Abilene R, Seltmann Katja C
Cheadle Center for Biodiversity and Ecological Restoration, University of California, Harder South Building 578, Santa Barbara, CA, 93106, USA.
Department of Biology, Missouri State University, 910 S John Q Hammons Parkway, Temple Hall, Springfield, MO, 65897, USA.
Oecologia. 2024 Dec 18;207(1):13. doi: 10.1007/s00442-024-05653-5.
An animal's diet breadth is a central aspect of its life history, yet the factors determining why some species have narrow dietary breadths (specialists) and others have broad dietary breadths (generalists) remain poorly understood. This challenge is pronounced in herbivorous insects due to incomplete host plant data across many taxa and regions. Here, we develop and validate machine learning models to predict pollen diet breadth in bees, using a bee phylogeny and occurrence data for 682 bee species native to the United States, aiming to better understand key drivers. We found that pollen specialist bees made an average of 72.9% of their visits to host plants and could be predicted with high accuracy (mean 94%). Our models predicted generalist bee species, which made up a minority of the species in our dataset, with lower accuracy (mean 70%). The models tested on spatially and phylogenetically blocked data revealed that the most informative predictors of diet breadth are plant phylogenetic diversity, bee species' geographic range, and regional abundance. Our findings also confirm that range size is predictive of diet breadth and that both male and female specialist bees mostly visit their host plants. Overall, our results suggest we can use visitation data to predict specialist bee species in regions and for taxonomic groups where diet breadth is unknown, though predicting generalists may be more challenging. These methods can thus enhance our understanding of plant-pollinator interactions, leading to improved conservation outcomes and a better understanding of the pollination services bees provide.
动物的食性广度是其生活史的核心方面，然而，决定为何有些物种食性广度狭窄（特化物种）而另一些物种食性广度宽泛（泛化物种）的因素仍知之甚少。由于许多分类群和地区的寄主植物数据不完整，这一挑战在食草昆虫中尤为突出。在此，我们开发并验证了机器学习模型，以预测蜜蜂的花粉食性广度，利用蜜蜂系统发育关系和682种原产于美国的蜜蜂物种的出现数据，旨在更好地理解关键驱动因素。我们发现，花粉特化蜜蜂平均72.9%的访花行为是针对寄主植物的，并且能够以高精度（平均94%）进行预测。我们的模型对泛化蜜蜂物种（在我们的数据集中占少数）的预测准确率较低（平均70%）。在空间和系统发育上进行了数据划分的测试表明，食性广度最具信息量的预测因子是植物系统发育多样性、蜜蜂物种的地理分布范围和区域丰度。我们的研究结果还证实，分布范围大小可预测食性广度，并且雄性和雌性特化蜜蜂大多访花其寄主植物。总体而言，我们的结果表明，尽管预测泛化物种可能更具挑战性，但我们可以利用访花数据来预测食性广度未知的地区和分类群中的特化蜜蜂物种。因此，这些方法可以增强我们对植物 - 传粉者相互作用的理解，从而改善保护成果，并更好地理解蜜蜂提供的授粉服务。
