Artificial intelligence, community-driven science, and seadragons
SeadragonSearch is a community-based research initiative with a mission to collect data about wild seadragon populations across Australia through underwater photography. Our project has partnered with the non-profit software developer Wild Me, using machine learning tools to identify seadragons in photos and track individual dragons over time.
Seadragons are beautiful, unique fish living exclusively along the southern coast of Australia. Researchers have compiled important data about seadragons over the years through electronic tagging, photo identification projects, and tissue sampling for genetic analysis. A new species (the ruby seadragon) was even discovered in 2015. However, we have yet to deeply understand aspects of seadragons’ lives that require long-term and widespread data collection. How long do seadragons live? How far are they capable of moving? How are their populations being impacted by changing ocean conditions? Seadragons’ coastal habitats are subject to pollution from land, habitat destruction from commercial fishing, and warming ocean temperatures, to name just a few human impacts. How resilient might seadragon populations be to the aggregate effects of such stressors?
Leafy and common seadragons depend on kelp beds and seagrass meadows for hunting and camouflage, and documented declines in these types of habitats across Australian waters in recent years have raised concerns about how seadragons will persist. Both common and leafy seadragons are slow swimmers and likely don’t move far within a small home range. Due to the lack of long-term, comprehensive data on seadragon populations, there is uncertainty around the extent to which they should be considered threatened, and how they should be strategically protected across their range.
SeadragonSearch shares the conservation goals of other seadragon monitoring projects, which have included electronic tagging and various versions of photo identification and matching. Because seadragons are so visually spectacular, scuba divers love to observe them and they are popular subjects with underwater photographers. Over the years, seadragon monitoring programs have collected photos from divers and used them to match individual seadragons from different encounter dates based on visual characteristics, such as face and body patterns or appendage shapes. The goal of our new initiative is to use machine learning to automate this type of matching, so that we can track more seadragons through their lifespans and across their distributions, efficiently gathering large volumes of data from all Australian states where seadragons occur. Automating the matching process with machine learning aims to increase the accuracy with which we can monitor seadragon populations in real time. To that end, our new initiative builds upon the Wildbook platform developed by the non-profit Wild Me, combining computer vision, machine learning, and community science to gather data on individual wild seadragons.
We plan to run SeadragonSearch for at least a decade, and in that time continue to work in partnership with Wild Me to determine how ongoing innovations in artificial intelligence tools can be applied to seadragon research, including future machine and deep learning applications. SeadragonSearch is a collaborative effort between marine biologists, software engineers, and underwater photographers, and we’re excited to see what we can discover in the coming years about seadragon populations and how to best protect both the dragons and their incredibly biodiverse habitats.
If you’d like to learn more about how artificial intelligence tools are being developed and used for wildlife population monitoring, pay a visit to Wild Me.
If you’d like to learn more about sharing your seadragon photography with SeadragonSearch, check out these videos.