We’re Getting Better at Surveying Seabirds

Anna Vallery, Coastal Program Manager for Audubon Washington, has long been interested in improving methods used for monitoring seabird and waterbird colonies. She is excited to keep pushing this field forward through a partnership with Audubon Texas, developing a suite of aerial imagery monitoring tools called SeeBird.  

Working in a waterbird colony is an experience that overwhelms all the senses in the best way possible. The smell is probably the most notable, with guano hitting the nostrils long before you physically arrive at the island. Then the noise hits you – a cacophony of waterbird calls and growls. Many colonies are also a visual spectacle. Birds are coming and going, parents on nests are fussing at their neighbors or handing off duties to their partners, and the density of birds can be hard to comprehend. Though these places are incredible to experience, they can be extremely challenging to monitor. This is especially true for the waterbird colonies of the Texas coast, sites that support hundreds of thousands of nesting pelicans, terns, herons, egrets, and skimmers every single year.  

These sites were the focus of my master’s research and sparked my interest in ensuring seabird and waterbird colonies everywhere are protected and monitored as effectively – and with as little disturbance – as possible. For decades, waterbirds have been monitored at colonies by either walking transects across the colony or by circumnavigating the colony in a boat. Both methods, however, can have unintended negative consequences to the birds. Surveying a waterbird colony on foot can cause nesting adults to flush, exposing eggs and chicks to the elements or predators (sometimes the other birds nearby). Boat-based surveys solve the disturbance problem but introduce another challenge – a low vantage point that makes it easy to miss birds obscured by vegetation or packed tightly together. Crewed aerial surveys are an excellent option as they offer a better view, but these surveys come with significant costs and logistical constraints that put them out of reach for many monitoring programs. 

In 2018 I was fresh out of my master’s degree, where I had looked into the feasibility of using drones to monitor waterbirds along the Texas coast. I had just started as a Conservation Specialist with Houston Audubon, a chapter responsible for stewarding several of the nesting colonies in Galveston Bay. Houston Audubon had recently acquired a drone, and our team was eager to see how drone counts compared to our boat-based estimates.  

Our initial findings were striking. The drone counted birds at one of the Galveston Bay colonies far more accurately than we could from a boat. But going through the imagery took weeks. So Richard Gibbons, our Conservation Director, and I decided to get some help automating the process and started a collaboration with computer scientists at Rice University’s Data to Knowledge Lab. Krish Kabra joined as the lead student on the project in 2021. 

The tool we’ve since developed pairs drone surveys with a machine learning model trained to detect and identify waterbirds in aerial imagery. It was built and tested at Chester Island, a major nesting colony in Matagorda Bay managed by Audubon Texas and one of the most productive colonies along the Texas coast. 

From the beginning, our goal was to make the tools we developed open source so other participants in the Texas Colonial Waterbird Survey, and anyone else doing this kind of monitoring, could use it. We published our first version in 2022 and have been improving it since. The current tool can now distinguish between visually similar species, including Royal and Sandwich Tern, two species that nest side by side and are notoriously difficult to tell apart even for experienced observers. 

To put it to the test, we compared four monitoring approaches at Chester Island: traditional ground counts, manual drone annotation, computer-assisted review using the model with human verification, and fully automated detection. The computer-assisted workflow detected 89% of the birds identified by a biologist going through every single image while requiring only about 19% of the time. Fully automating the location and identification of the birds in images took under an hour. We’re really excited that hose results were just submitted for publication and the tool is available for use. 

We’re not done yet. Our four-way comparison demonstrated that bird detection generalized well across species, even those that were somewhat rare in our images. The team is now working to train a generalized bird detector, one that can locate individuals in drone imagery regardless of species. We’re calling this growing suite of tools SeeBird. A first round of training the generalized detector is complete, with imagery from colonies across the globe already in hand, including some from colonies right here in Washington and Oregon. Our hope is that SeeBird will become a useful tool for biologists and managers across the Pacific Northwest and beyond. 

Washington’s marine waters, from the Salish Sea to the outer coast, support significant colonies of ground-nesting seabirds, including alcids, cormorants, and gulls, on islands managed by partners like the U.S. Fish and Wildlife Service and the Washington Department of Fish and Wildlife. A tool that cuts image review time by 80% doesn’t just save hours. It makes it feasible to survey more sites, more often, and to get data into managers’ hands sooner. 

Want to learn more about how A.I. is helping support avian monitoring and conservation? Check out “A.I. in the Wild” in Audubon Magazine’s Summer 2026 issue.