Decadal increase in vessel interactions by a scavenging pelagic seabird across the North Atlantic.

Fisheries waste is used by many seabirds as a supplementary source of food,1 but interacting with fishing vessels to obtain this resource puts birds at risk of entanglement in fishing gear and mortality.2 As a result, bycatch is one of the leading contributors to seabird decline worldwide,3 and this risk may increase over time as birds increasingly associate fishing vessels with food. Light-level geolocators mounted on seabirds can detect light emitted from vessels at night year-round.4 We used a 16-year time series of geolocator data from 296 northern fulmars (Fulmarus glacialis) breeding at temperate and arctic colonies to investigate trends of nocturnal vessel interactions in this scavenging pelagic seabird. Vessel attendance has progressively increased over the study period despite no corresponding increase in the number of vessels or availability of discards over the same time frame. Fulmars are highly mobile generalist surface feeders,5 so this may signal a reduction in available prey biomass in the upper water column, leading to increased reliance on anthropogenic food subsidies6 and increased risk of bycatch mortality in already threatened seabird populations. Individuals were consistent in the extent to which they interacted with vessels, as shown in other species,7 suggesting that population-level increases may be due to a higher proportion of fulmars following vessels rather than changes at an individual level. Higher encounter rates were correlated with lower time spent foraging and a geographically restricted overwintering distribution, suggesting an energetic advantage for these scavenging strategists compared with foraging for natural prey.

A general deep learning model for bird detection in high-resolution airborne imagery.

Advances in artificial intelligence for computer vision hold great promise for increasing the scales at which ecological systems can be studied. The distribution and behavior of individuals is central to ecology, and computer vision using deep neural networks can learn to detect individual objects in imagery. However, developing supervised models for ecological monitoring is challenging because it requires large amounts of human-labeled training data, requires advanced technical expertise and computational infrastructure, and is prone to overfitting. This limits application across space and time. One solution is developing generalized models that can be applied across species and ecosystems. Using over 250,000 annotations from 13 projects from around the world, we develop a general bird detection model that achieves over 65% recall and 50% precision on novel aerial data without any local training despite differences in species, habitat, and imaging methodology. Fine-tuning this model with only 1000 local annotations increases these values to an average of 84% recall and 69% precision by building on the general features learned from other data sources. Retraining from the general model improves local predictions even when moderately large annotation sets are available and makes model training faster and more stable. Our results demonstrate that general models for detecting broad classes of organisms using airborne imagery are achievable. These models can reduce the effort, expertise, and computational resources necessary for automating the detection of individual organisms across large scales, helping to transform the scale of data collection in ecology and the questions that can be addressed.

Sampling strategies for species with high breeding-site fidelity: A case study in burrow-nesting seabirds.

Sampling approaches used to census and monitor populations of flora and fauna are diverse, ranging from simple random sampling to complex hierarchal stratified designs. Usually the approach taken is determined by the spatial and temporal distribution of the study population, along with other characteristics of the focal species. Long-term monitoring programs used to assess seabird population trends are facilitated by their high site fidelity, but are often hampered by large and difficult to access colonies, with highly variable densities that require intensive survey. We aimed to determine the sampling effort required to (a) estimate population size with a high degree of confidence, and (b) detect different scenarios of population change in a regionally important species in the Atlantic, the Manx shearwater (Puffinus puffinus). Analyses were carried out using data collected from tape-playback surveys on four islands in the North Atlantic. To explore how sampling effort influenced confidence around abundance estimates, we used the heuristic approach of imagining the areas sampled represented the total population, and bootstrapped varying proportions of subsamples. This revealed that abundance estimates vary dramatically when less than half of all plots (n dependent on the size of the site) is randomly subsampled, leading to an unacceptable lack of confidence in population estimates. Confidence is substantially improved using a multi-stage stratified approach based on previous information on distribution in the colonies. In reality, this could lead to reducing the number of plots required by up to 80%. Furthermore, power analyses suggested that random selection of monitoring plots using a matched pairs approach generates little power to detect overall population changes of 10%, and density-dependent changes as large as 50%, because variation in density between plots is so high. Current monitoring programs have a high probability of failing to detect population-level changes due to inappropriate sampling efforts. Focusing sampling in areas of high density with low plot to plot variance dramatically increases the power to detect year to year population change, albeit at the risk of not detecting increases in low density areas, which may be an unavoidable strategy when resources are limited. We discuss how challenging populations with similar features to seabirds might be censused and monitored most effectively.

Seabird movement reveals the ecological footprint of fishing vessels.

Exploitation of the seas is currently unsustainable, with increasing demand for marine resources placing intense pressure on the Earth's largest ecosystem [1]. The scale of anthropogenic effects varies from local to entire ocean basins [1-3]. For example, discards of commercial capture fisheries can have both positive and negative impacts on scavengers at the population and community-level [2-6], although this is driven by individual foraging behaviour [3,7]. Currently, we have little understanding of the scale at which individual animals initiate such behaviours. We use the known interaction between fisheries and a wide-ranging seabird, the Northern gannet Morus bassanus[3], to investigate how fishing vessels affect individual birds' behaviours in near real-time. We document the footprint of fishing vessels' (≥15 m length) influence on foraging decisions (≤11 km), and a potential underlying behavioural mechanism, by revealing how birds respond differently to vessels depending on gear type and activity. Such influences have important implications for fisheries, including the proposed discard ban [8]), and wider marine management.

Space partitioning without territoriality in gannets.

Colonial breeding is widespread among animals. Some, such as eusocial insects, may use agonistic behavior to partition available foraging habitat into mutually exclusive territories; others, such as breeding seabirds, do not. We found that northern gannets, satellite-tracked from 12 neighboring colonies, nonetheless forage in largely mutually exclusive areas and that these colony-specific home ranges are determined by density-dependent competition. This segregation may be enhanced by individual-level public information transfer, leading to cultural evolution and divergence among colonies.