Año Nuevo Island Animal Count: analyzing citizen science pinniped counts from drone imagery

• Main Author: Wood, Sarah
• Format: Dataset
• Language: unknown
• Published: 2020
• Subjects: drone imagery; Marine mammals; crowd-sourced data; Elephant Seals
• Online Access: https://zenodo.org/record/4279795; https://doi.org/10.7291/D1J66X

Fluctuations in marine mammal abundance can reveal changes in local ecosystem health and inform conservation strategies. Unmanned aircraft systems (UAS) such as drones are increasingly being used to photograph and count marine mammals in remote locations; however, counting animals in images is a laborious task. Crowd-sourced science has the potential to considerably reduce the time required to conduct these censuses but must first be validated against expert counts to confirm accuracy. Our objectives were to examine the citizen science counts for accuracy, identify costs and benefits of drone imagery and citizen science for pinniped censuses, and make recommendations for future uses of the data. We obtained and uploaded drone imagery of Año Nuevo Island in California to a custom citizen science website (sealcount.com) that instructed volunteers to count seals and sea lions. Across 212 days, over 1,500 volunteers counted northern elephant seals, harbor seals, California sea lions, and Steller sea lions in 90,000 photographs. We created five simple algorithms to extract one count per photograph from the crowd-sourced data and then analyzed each algorithm for accuracy by comparing to expert counts. We found that the median was the most accurate metric for extracting counts of seals but not sea lions. Volunteers consistently underestimated sea lions, so removing minimum values was the best strategy for extracting accurate counts of sea lions. We also found that while citizen scientists were able to accurately count adult seals, their accuracy was lower during pupping season, when small pups were present but difficult to detect. With proper precautions, citizen science saves money, labor, and time, while producing large amounts of accurate data that can be used to analyze a suite of biological patterns. Future applications include analyses of geo-spatial patterns within and between species, quantifying interspecific niche partitioning, and life history phenology.