An idea to simplify how migratory populations of ducks, geese and cranes are counted first hatched around a campfire. The project took flight in Bosque Del Apache earlier this month for its first-ever waterfowl survey.
Every winter, wildlife managers are challenged to count the migratory waterfowl that fly down into refuges. Creating the counts is difficult and often involves scaring birds into the air to be counted by making loud sounds or soaring past them in low-flying airplanes. Researchers at The University of New Mexico, in collaboration with the U.S. Fish and Wildlife Service in the Department of the Interior, are working to develop a machine-learning model prototype that can count the birds using images taken by drones in a project titled Drones for Ducks.
Christopher Lippitt, associate dean in the College of Arts and Sciences and professor in the Department of Geography and Environmental Studies, and Rowan Converse, a Ph.D. candidate in the Department of Geography & Environmental Studies, have spent the past three years working to develop an AI model that can distinguish and count the birds in photos taken by drones. This winter, they will test the technology in a complete waterfowl survey for the first time and compare the results to what is collected by wildlife managers using traditional methods. Fieldwork is planned in Bosque Del Apache and Texas Chenier Plain Refuge Complex.
"Ideally it makes collecting the data more quick and efficient, it helps the biologists get the information they need more quickly by running it through the model, and they get an output that they can interpret in a timely manner," Converse said. "We also hope that this is less disruptive for wildlife. A lot of the species seem to not react much to the drone going over, whereas some of the traditional methods involve flushing them out so they can be counted."
The model is still in development but the results have been promising. Early indications are that it can detect birds with 95% accuracy. The team hopes that the new system will make data collection more accurate and give wildlife managers faster access to information they use to make intervention decisions. Accurate counts of these migratory populations of birds are important because they help measure how populations have been impacted by climate change, if wildlife managers need to put out additional food for the birds, and other information.
Other universities are working on models that can count birds in snow or on water, but UNM's research is unique in its exploration of detecting birds in more diverse landscapes like farm fields, flooded wetlands and river sandbars. Training a computer to judge such diverse imagery takes time – and a lot of volunteers.
In order to test and train the AI model's accuracy, researchers have employed the help of volunteers on Zooniverse, a website where anyone can sign up and participate in research as a citizen scientist. Citizen scientists look at the images taken by drones, mark what objects appear to be birds, and differentiate if each bird is a goose, duck, or crane. The task is more difficult than it may sound due to the diversity of the landscape and the unique bird's-eye point of view. The information submitted by the volunteers is then reviewed and filtered. If a majority of volunteers have agreed on something, it is generally accepted as true.
Want to get involved with the research?
The new flock of images collected this winter means the Drones for Ducks team will need volunteers to help review images. People interested in participating can do so on Zooniverse in February 2024. Check out the team's other wildlife project to help in the meantime.
The idea for the research first came about during a camping trip organized by the state's chapter of the American Society for Photogrammetry and Remote Sensing where Converse, Lippitt, and people from other fields were able to discuss different challenges they face in their work.
Steven Sesnie, a spatial ecologist from the U.S. Fish and Wildlife Service, brought up the challenges of counting the migratory bird populations. Lippitt, who specializes in remote sensing and spatial modeling, proposed researching a way to use AI for the counts. The project, with its combination of wildlife management and machine learning, was perfect research for Converse to explore in her Ph.D. It was campfire science at work. Drones for Ducks, which is funded by the U.S. Fish and Wildlife Service, is now on its third research award.
"We all work as a team— the folks at the Fish and Wildlife Service and us— so we've gone from the early days of, 'let's see if this is crazy,' with an initial round of funding, to now starting to get down to surveying refuges to test out the potential for large-scale operations," Lippitt said.
The project speaks to the importance of scientists collaborating with outside entities on research.
"That's the kind of thing that happens when we work closely with our different stakeholders and collaborators," Lippitt said. "The more that UNM students and faculty have access to the problems that are faced by our state and federal agencies, the more we are able to partner with them to help solve them and that is mutually beneficial in every possible respect."