Butterfly Wing Diversification & Evolution with Citizen Science

Butterflies are a species of conservation importance and are one of the main species representing insects of the wider countryside in the United Kingdom. Butterflies act as an effective indicator of biodiversity and ecosystem function, so understanding which species exist and their population trends is paramount to supporting rich and functional ecosystems in the face of such extreme change; be that land use or climate. They typically have short lifecycles, and therefore have the potential to evolve rapidly in response to these changes. However, many species are subject to local or even global extinction and therefore often become extinct before they are even formally identified. This relates to the concept of the taxonomic impediment: the notion by which the tools, skills and knowledge required to identify and fill gaps in our knowledge cannot keep pace with the species requiring identification. Without addressing this, species are lost to us before we even know they existed.

Image: Species such as the Peacock butterfly (Aglais io) have highly distinctive wing markings that resemble eye spots to ward off predators, a form of patterning know as aposematic colouration. © Willow Neal.

To help resolve this, detailed and rapid identification is required to understand patterns in butterfly evolution and species response to change. The vision of our project is to build an AI algorithm to identify butterfly species that can be used for multiple different applications. Our proposed system goes beyond a simple black-box AI neural network, but instead, envisions a system that can understand wing details from an image of a butterfly’s wing that an expert taxonomist might use. These experts are invaluable, and arguably underappreciated for their ‘behind-the-scenes’ contributions to butterfly taxonomy but are constrained by time and resources. Therefore, a supportive tool that is built with this in mind is one pathway to help address these issues. 

Identification apps exist that can recognise a species and report back to the user what they have found, perhaps with some accompanying information about the species to go along with it. However, our proposed ‘interpretable AI’ system will have a focus on wing morphology, including unique markings, patterns, wing venation, and shape, to not only allow users to recognise a species from an image, but to identify it, by indicating to them what characteristic patterns were the reason for this butterfly to be designated as that species. A sophisticated, detailed algorithm which can catalogue and report back which features it has used to determine a species allows users to imbue this knowledge for future use the next time they are out in nature and engage more closely with butterfly taxonomy. Building familiarity with the appearance patterns of common butterflies may even allow them to potentially identify, perhaps to family level, a species which they have never seen before. 

We will work with butterfly taxonomists to ensure that the interpretations provided by the AI model are scientifically valid. On the other hand, from a butterfly researcher’s perspective, these fine details can be used to compare individual specimens in a collection to each other, and contribute to the process of describing new species, which are often determined through sometimes minor variations in morphology.

Image: Many predators aim for the head, so some eyespots are used as a distracting feature to confuse them into targeting a safer part of the wing. This Gatekeeper (Pyronia tithonus) could still fly! © Willow Neal.

A critical element of conservation is engaging communities, and helping people connect with the natural world. This is good for our health, good for the environment, and is the cornerstone of much of our understanding of ecology, from Victorian records and well-regarded work such as ‘Butterflies of the British Isles’ by F.W. Frohawk  in 1914, to citizen science recorders that contribute to the UK Butterfly Monitoring scheme today. By using the citizen science platform Zooinverse (https://www.zooniverse.org/), we plan to present volunteer users with an image of a butterfly’s wing and ask them to highlight the fine details they see on the image, to provide the dataset that will teach the AI model which features to search for.

Image: Diagnostic features of Papilio genus butterflies include some highly distinctive wing structures (H6) (Owens, et al. 2020)

By engaging communities online, supporting scientific research on butterflies, and using AI to understand the impacts of environmental change, our project seeks to understand what even the finest wing details can mean on a species level. We feel strongly that engaged research is an important element of any project, particularly one which supports passion and education of those interested in nature. We are committed to keeping our research, progress and results open access to allow our findings to cross boundaries, be that social, geographic or otherwise. 

You can read more on the Challenge page.

Other team members: James Pearson, Stephen Serjeant, Patrick Wong, Yoseph Araya and Lida Shahmiri.