ABSTRACT
Explaining colour variation among animals at broad geographic scales remains challenging. Here we demonstrate how deep learning-a form of artificial intelligence-can reveal subtle but robust patterns of colour feature variation along an ecological gradient, as well as help identify the underlying mechanisms generating this biogeographic pattern. Using over 20,000 images with precise GPS locality information belonging to nearly 2,000 moth species from Taiwan, our deep learning model generates a 2048-dimension feature vector that accurately predicts each species' mean elevation based on colour and shape features. Using this multidimensional feature vector, we find that within-assemblage image feature variation is smaller in high elevation assemblages. Structural equation modeling suggests that this reduced image feature diversity is likely the result of colder environments selecting for darker colouration, which limits the colour diversity of assemblages at high elevations. Ultimately, with the help of deep learning, we will be able to explore the endless forms of natural morphological variation at unpreceded depths.
