Biologists trying to count endangered Antillean manatees in Costa Rica and Panama face a major challenge: the animals live in murky waters, making them virtually impossible to see. “I rowed back and forth [along Panama's] San San River every day for two years, and all I got to see were some noses,” biologist and computer scientist Mario Rivera-Chavarria says. “I could hear them, but I never saw them.”
In 2013 Rivera-Chavarria, then at the University of Costa Rica (UCR), and his colleagues at the Smithsonian Tropical Research Institute launched a manatee census in Panama's San San Pond Sak wetlands, an area that borders Costa Rica and includes the San San River. Using a boat equipped with side-scan sonar, which produces images by bouncing sound waves off submerged animals and their environment, the team estimated the manatee population in an 18-kilometer stretch of the San San was as low as two individuals in some months and as high as 33 in others.
But sonar can be disruptive to these animals, so Rivera-Chavarria wanted to prove that the census could be carried out using a less invasive technique. Manatee vocalizations have specific characteristics that allow a trained ear—or computer—to distinguish one individual from another. Rivera-Chavarria recorded the animals' songs with underwater microphones suspended from his kayak as he paddled down the San San. His colleague Jorge Castro, a computer scientist now at Costa Rica's National Center for High Technology, developed an algorithm to automatically count manatees based on the recordings. Castro showed, using a sample of 54 calls that belonged to four different manatees, that his algorithm was 100 percent accurate.
The algorithm breaks the process into four main steps: chopping the recordings into short chunks, canceling out noise, labeling the manatees' calls and clustering the calls by individual. The noise cancellation step takes the longest to process; to speed things up, Castro and his colleague Esteban Meneses used a supercomputer. They translated the algorithm into a programming language that would allow them to carry out tasks in parallel, making the process 120 times faster, the team reported last July at the IEEE International Work Conference on Bioinspired Intelligence.
Next, Castro and his team plan to adjust the algorithm to identify the calls of the clay-colored thrush, Costa Rica's national bird. Roberto Vargas-Masís, an expert in bioacoustics at the National Distance Education University of Costa Rica, who was not involved in the manatee study but aims to participate in the bird research, says: “This technology will allow us to gather and analyze large quantities of data and very quickly determine if the species is present in a specific region.”