91TV

How did the idea for the paper come about?

The Antarctic fur seal has been extensively studied over the past four decades on Bird Island, South Georgia. Through traditional field monitoring, we have substantial understanding of the species’ phenology and the threats that Antarctic fur seal pups face during their early development. Historically, the principal danger arose from adult males, who during territorial brawl can accidentally trample and thus fatally injure pups, with a smaller proportion of pups succumbing to predation. Given that it is a less common interaction, little research has focused on the predator-prey interactions within the breeding colonies. However, shortly before we installed the remote observatory camera, a predator-driven Allee effect was uncovered in the focal study population sparking curiosity about the predator-prey dynamics.

Please can you introduce your study and tell us what it’s about?

Our investigated how population density influences predator-prey interactions within a breeding colony of Antarctic fur seals. Density is a major determinant of population dynamics. In our system, higher densities increase the risk of pups getting trampled by territorial males or pups getting separated from their mothers, while at low densities they face a greater risk of predation. Thus, we wanted to understand how fine-scale differences in density shape predator-prey interactions.

Bird Island is an ideal place to address this question. Here, two Antarctic fur seal breeding colonies are located just a few hundred metres apart, yet they differ about fourfold in density. Pups born at the lower-density colony suffer higher predation from three associated bird species: giant petrels, brown skuas and snowy sheathbills. These three species represent a predator to scavenger gradient, with giant petrels being the main predator of the pups, whereas snowy sheathbills are commonly considered opportunistic scavengers. This is the colony we chose to study.

We used an autonomous time-lapse camera system overlooking the colony and recorded an image every minute across an entire breeding season, and combined this with a custom-trained neural network to automatically detect and classify every adult male, adult female and pup, along with the three birds (6 classes in total) — around 4.1 million detections in total.

What are the findings and what makes them significant?

We first confirmed that the camera-and-network setup could recover the known seasonal abundance patterns of Antarctic fur seals: males arriving first to establish territories, then females coming ashore to give birth and afterwards nurse their pups. Next, we mapped the area where each class spent time within the colony. Here, we found a strong spatial overlap between brown skuas, adult females and pups and a strong overlap between snowy sheathbills and adult males. Meanwhile, only giant petrels occupied the shoreline, reflecting one of their main hunting techniques (drowning). These patterns suggest that predator and scavenger species partition the colony according to the distribution of the different seal categories and their foraging strategies.

Then, we found that pups were located closer to adult females, other pups, and snowy sheathbills than expected by chance, whereas the opposite was true for adult males and giant petrels. This finding provides insight into the spatial behaviour of the pups - showing that they seek out areas which could provide maternal care and social interactions while avoiding areas associated with risk.

Finally, we used the distance between birds and pups as a proxy for predation risk, alongside the distance a pup kept from its neighbouring seals. The key result: a pup being close to an adult seal — of either sex — reduced its predation risk, whereas being close to other pups did not. Thus, proximity to adult seals is more important for reducing predation risk than aggregation with other pups. Taken together, these results demonstrate that predation risk within dense breeding colonies is shaped by colony density but also fine-scale relationships among individuals. Understanding such interactions is essential for predicting how changes in colony density and composition may affect pup survival.

What is the relevance of the article to the wider scientific community?

The broader contribution is methodological. By combining a low-cost, automated camera system with a neural network, we were able to monitor a colonial species continuously and non-invasively, generating millions of observations with minimal disturbance to the animals.

The key advantage of this approach is statistical power. Traditional field counts and even systematic observations provide only intermittent snapshots of colony dynamics, limiting our ability to detect the fine-scale spatial and temporal patterns we observed here. Millions of automated detections, by contrast, give you the sample sizes to see them clearly. Although automated detections are not as precise as careful manual annotations on an image-by-image basis, the vastly greater volume of data more than compensates for this limitation.

What makes this exciting beyond our own study is how accessible these tools have become. Off-the-shelf detectors like YOLO are now genuinely usable even without extensive machine-learning experience, with large language models lowering the threshold further by helping write data-preparation scripts, set up training and run inference. As a result, automated monitoring is no longer restricted to large, highly specialised studies. We hope our study encourages the wider use of these approaches to uncover patterns and processes that are difficult to detect through conventional methodologies alone.

What is the future for research in this area?

We see remote observation as a powerful complement to classical fieldwork rather than a replacement for it. Its real promise lies in bringing together different scales of observation, data sources and disciplines. Colonial breeders are particularly well suited to this, because they aggregate predictably in space and time, making them tractable for automated monitoring.

Future work could combine continuous imaging with the long-term demographic, individual-based and genetic records already available for many study populations. Such integration would make it possible to ask new questions about how behaviour and fine-scale relationships influences survival and reproductive success, how social and spatial interactions are structured within colonies, and whether these patterns have a genetic basis.

About the authors:

This work was carried out by researchers based at Bielefeld University and Friedrich-Alexander University Erlangen-Nürnberg in collaboration with the British Antarctic Survey, led by Ane Liv Berthelsen and Johannes Bartl.

Ane Liv is a molecular ecologist at the Evolutionary Population Genetics group at Bielefeld University, Germany. Her work aims at disentangling phenotypic trait variation in Antarctic fur seals and exploring the effects of density on Antarctic fur seal pups using image-based, physiological and transcriptomic data.

Johannes is a physicist at the Biophysics Group at Friedrich-Alexander University Erlangen-Nürnberg, Germany. His research applies machine learning to large image-based datasets of polar species, such as Antarctic fur seals and emperor penguins, to monitor their populations and behaviour remotely.

Read the to find out more.  is an open access journal that welcomes the submission of all high-quality science. More information about the submission process can be found on our .

Images:

Image 1: Antarctic fur seal pups and giant petrel © Svenja Stöhr.

Image 2: The Antarctic fur seal colony on Bird Island captured by the remote camera system. Bounding boxes outline all detected animals by the neural network.