Using computer vision to understand the global biogeography of ant color
Using computer vision to understand the global biogeography of ant color
复制标题
DOI：
10.1111/ecog.06279
复制
发表时间：
2023-01-24
期刊：
影响因子：
5.9
通讯作者：
Fisher, Brian L. L.
中科院分区：
文献类型：
--
作者：
Idec, Jacob H. H.;Bishop, Tom R. R.;Fisher, Brian L. L.
生物体利用颜色来提供各种生物功能，包括伪装，配偶吸引和体温调节。颜色的潜在适应作用通常通过检查地理，栖息地和生活史梯度的变化模式来研究。然而，这种方法提出了一个数据收集的权衡，即研究人员必须最大限度地提高种内的细节或分类和地理覆盖范围。这限制了我们在全球范围内充分了解整个分类群体的颜色变化的能力。我们提供了一个解决方案，从超过44 000个蚂蚁个体标本中提取颜色数据，代表超过14 000个物种和形态种，使用计算机视觉算法对蚂蚁头部图像。我们对这个数据集的分析表明，蚂蚁是占主导地位的变化，在黑暗中，苍白的颜色光谱，这种变化的大部分是在物种内举行，而且，总体而言，一套流行的生态地理学假说是无法解释内和种间变异蚂蚁颜色。这与以前在蚂蚁和其他无脊椎动物组合水平上的工作形成对比，这些工作表明温度和整个组合的平均颜色等变量之间存在明确而强烈的联系。我们的工作应用了一种新的计算方法来研究大规模的性状多样性。通过这样做，我们揭示了以前未知的种内变异水平。类似的方法可能会释放大量的数据驻留在博物馆和标本数据库，并建立一个数字平台的数据收集革命的功能地理学。
Organisms use color to serve a variety of biological functions, including camouflage, mate attraction and thermoregulation. The potential adaptive role of color is often investigated by examining patterns of variation across geographic, habitat and life-history gradients. This approach, however, presents a data collection trade-off whereby researchers must either maximize intraspecific detail or taxonomic and geographic coverage. This limits our ability to fully understand color variation across entire taxonomic groups at global scales. We provide a solution by extracting color data from more than 44 000 individual specimens of ants, representing over 14 000 species and morphospecies, using a computer vision algorithm on ant head images. Our analyses on this dataset reveal that ants are dominated by variation in the dark-pale color spectrum, that much of this variation is held within species, and that, overall, a suite of popular ecogeographic hypotheses are unable to explain intra- and interspecific variation in ant color. This is in contrast to previous work at the assemblage level in ants and other invertebrates demonstrating clear and strong links between variables such as temperature and the average color of entire assemblages. Our work applies a novel computational approach to the study of large-scale trait diversity. By doing so, we reveal previously unknown levels of intraspecific variation. Similar approaches may unlock a vast amount of data residing in museum and specimen databases and establish a digital platform for a data collection revolution in functional biogeography.