Artificial Intelligence in Conservation: Counting Guadalupe Fur Seal Populations

The Challenge: Counting Newborn Fur Seals

Currently, the Guadalupe fur seal population is threatened in the U.S. and needs consistent record counts to ensure its protection. Thought to be extinct in the early 1900s, the fur seal is making a cautious comeback. But, unfortunately, the only known breeding site for these seals is Guadalupe Island, a remote, volcanic island off the coast of Northern Baja California. Still, scientists regularly venture to the island to count fur seals during breeding season.

Tenaya Norris, a California-based Marine Mammal Biologist, has spent countless hours tracking these seals with a group of scientists in Mexico lead by Fernando Elorriaga-Verplancken. The activity is laborious, intrusive, and sometimes dangerous. Seals often hide behind rocks and crevices, and counting these animals often involves flushing or moving them, including moms and their pups. “Sometimes, these pups are only a day old, so that disturbance can be significant,” explained Tenaya. In addition, Tenaya noted that the activity of ground counting is also potentially dangerous to scientists. 

That’s why Tenaya and Fernando partnered with Oceans Unmanned, a DroneDeploy.org partner, to empower researchers to utilize drone technology when collecting marine spatial data. Together, the team uses DroneDeploy software to map Punta Sur, the southern tip of Guadalupe Island.

While collecting data with DroneDeploy is much more streamlined than ground counts, it can still be very time-consuming back in the office. Without an AI tool, scientists must study raw or stitched-together orthomosaic images and count each seal manually. The lack of automated counting tools is a significant hurdle to putting drone data to use. “There are researchers across the globe with years of collected data that has yet to be processed due to the number of hours required for analysis,” said Brian Taggart, co-founder of Oceans Unmanned.

The Solution: a DroneDeploy Hackathon Team

This challenge sparked interest from DroneDeploy’s internal engineering team, who took on the project during our most recent Hackathon:a three-day event where teams of engineers across DroneDeploy work on technical problems and test ideas. These quarterly Hackathons not only drive community and team-building across our organization but also generate innovative designs and product features. Some of our most popular product offerings, including Live Map, began as Hackathon ideas, and this June, our engineering team partnered with scientists to identify a way to more efficiently count seals.

To count seals through aerial imagery, the team used DroneDeploy’s count tool in conjunction with Picterra’s technology. Picterra is a geospatial intelligence platform for machine learning and includes many training and processing libraries for extracting information from imagery, including specific tools for counting wildlife. The team integrated DroneDeploy data into the Picterra platform for training and processing. Using this method, they were able to identify and count both Guadalupe fur seals and their pups with a trained detector.

Next, the engineers facilitated the integration of the trained seal detector into the DroneDeploy platform. Using the detector, scientists can now highlight areas on their maps and count both Guadalupe fur seal pups and adults in imagery. This tool may make Guadalupe fur seal population monitoring more streamlined and efficient.

Broader Implications for World Conservation

The automation of seal counts is a small step forward in a much larger push for AI in conservation. The World Wildlife Fund estimated that the world had lost two-thirds of its wildlife in the past 50 years, and monitoring habitats and landscapes is a crucial tool in unlocking this future for conservation. Faster and easier tools in wildlife monitoring can help overcome some of the hurdles of traditional methods, such as how often surveys can occur and how many people are needed to conduct these surveys.