Color pattern is an easily observable biological trait, and with bold patterns, bright colors, and variability in species-specific phenotypes, much attention is given to ‘showy’ species by professional scientists and amateur naturalists alike. Within a single species, color pattern variation (color polymorphism) can aid evolutionary maintenance of an organism as distinct phenotypes experience different selective pressures. This work presents a workflow and machine learning approach for classifying color patterns of animals from digital photographs. To illustrate the value of this computer vision model, I focus on a use-case of a striped/unstriped color pattern polymorphism in the geographically widespread and abundant Eastern Red-backed Salamander (Plethodon cinereus). The ecological and evolutionary mechanisms influencing the geographic patterns of coloration in P. cinereus color morphs remains unclear, and no studies have examined range-wide patterns of the polymorphism. For this work, we extracted 15,777 images of P. cinereus from the online citizen science platform, iNaturalist. Using the desktop software program ImageAnt, a small group of volunteers manually annotated 4,000 salamander images. These 4,000 images served as training data for developing an automated image analysis pipeline, based on convolutional neural networks and classical computer vision techniques, which we then used to analyze the remaining 11,777 images. Using ecological niche modeling we test whether the two color morphs partition available niche space, thereby contributing to the maintenance of this polymorphism. Overall, this work provides a means of reliably scoring color pattern variation from large-scale image datasets to test outstanding biological questions.
Dr. Maggie Hantak is broadly interested in the ecological and evolutionary processes that maintain phenotypic variation within and among populations. Much of her research has been focused on the striped/unstriped color pattern polymorphism in the Eastern Red-backed Salamander (Plethodon cinereus) and how these morphs are maintained over space and time. She is interested in using machine learning to take advantage of the wealth of underutilized image data from community science platforms, which will greatly contribute to our understanding of the relationship between polymorphism and diversity.