# Uncovering Hidden Predators: Thermal Drone Detection of Antarctic Fur Seals in Tussac Grass at South Georgia

## TL;DR

Thermal drones help detect Antarctic fur seals hidden in grass, improving population monitoring at South Georgia.

## Contribution

Fixed-wing drones with thermal and RGB sensors provide a scalable method for detecting cryptic fur seals in tussac grass.

## Key findings

- Thermal sensors outperformed RGB in detecting fur seals in tussac grass.
- Combining thermal and RGB data improved population estimates across varied terrain.
- The method is scalable for monitoring declining fur seal populations in complex habitats.

## Abstract

Antarctic fur seals are an important predator in the Southern Ocean, with > 95% of the population breeding at South Georgia. Female seals generally pup on open beaches, but many move into long tussac grass to suckle offspring, where their presence can be concealed by vegetation. This makes it difficult to assess population changes, introducing considerable uncertainty. Broad‐scale, time‐efficient monitoring capable of detecting fur seals in tussac is therefore required to better understand population trends throughout the island, especially given recent reports of declines associated with reduced food availability, as well as important negative impacts from HPAI. This study utilises a fixed‐wing drone to provide both red/green/blue (RGB) imagery and thermal imagery for detecting fur seals in tussac grass as well as along beaches for assessing populations. Thermal sensors proved highly effective at detecting fur seals in tussac relative to RGB, with a much more efficient processing workflow. However, a combination of both is necessary to accurately identify seals across the range of coastal terrain in which they are found.

We demonstrate that fixed‐wing drones equipped with dual thermal and RGB sensors can effectively detect Antarctic fur seals, including those concealed in tussac grass at South Georgia. Thermal imagery significantly improved detection rates compared with RGB alone, while a combined workflow provided the most accurate population estimates. This approach offers a scalable, time‐efficient method for monitoring a declining Southern Ocean predator and has broader applications for surveying cryptic wildlife in complex habitats.