Search

Conservation of breeding seabirds typically requires detailed data on where they feed at sea. Ecological niche models (ENMs) can fill data gaps, but rarely perform well when transferred to new regions. Alternatively, the foraging radius approach simply encircles the sea surrounding a breeding seabird colony (a foraging circle), but overestimates foraging habitat. Here, we investigate whether ENMs can transfer (predict) foraging niches of breeding tropical seabirds between global colonies, and whether ENMs can refine foraging circles. We collate a large global dataset of tropical seabird tracks (12000 trips, 16 species, 60 colonies) to build a comprehensive summary of tropical seabird foraging ranges and to train ENMs. We interrogate ENM transferability and assess the confidence with which unsuitable habitat predicted by ENMs can be excluded from within foraging circles. We apply this refinement framework to the Great Barrier Reef (GBR), Australia to identify a network of candidate marine protected areas (MPAs) for seabirds. We found little ability to generalise and transfer breeding tropical seabird foraging niches across all colonies for any species (mean AUC: 0.56, range 0.4-0.82). Low global transferability was partially explained by colony clusters that predicted well internally but other colony clusters poorly. After refinement with ENMs, foraging circles still contained 89% of known foraging areas from tracking data, providing confidence that important foraging habitat was not erroneously excluded by greater refinement from high transferability ENMs nor minor refinement from low transferability ENMs. Foraging radii estimated the total foraging area of the GBR breeding seabird community as 2,941,000 km2, which was refined by excluding between 197,000 km2 and 1,826,000 km2 of unsuitable foraging habitat. ENMs trained on local GBR tracking achieved superior refinement over globally trained models, demonstrating the value of local tracking. Our framework demonstra

Plastic pollution is distributed patchily around the world’s oceans. Likewise, marine organisms that are vulnerable to plastic ingestion or entanglement have uneven distributions. Understanding where wildlife encounters plastic is crucial for targeting research and mitigation. Oceanic seabirds, particularly petrels, frequently ingest plastic, are highly threatened, and cover vast distances during foraging and migration. However, the spatial overlap between petrels and plastics is poorly understood. Here we combine marine plastic density estimates with individual movement data for 7137 birds of 77 petrel species to estimate relative exposure risk. We identify high exposure risk areas in the Mediterranean and Black seas, and the northeast Pacific, northwest Pacific, South Atlantic and southwest Indian oceans. Plastic exposure risk varies greatly among species and populations, and between breeding and non-breeding seasons. Exposure risk is disproportionately high for Threatened species. Outside the Mediterranean and Black seas, exposure risk is highest in the high seas and Exclusive Economic Zones (EEZs) of the USA, Japan, and the UK. Birds generally had higher plastic exposure risk outside the EEZ of the country where they breed. We identify conservation and research priorities, and highlight that international collaboration is key to addressing the impacts of marine plastic on wide-ranging species.

Plastic pollution is distributed patchily around the world’s oceans. Likewise, marine organisms that are vulnerable to plastic ingestion or entanglement have uneven distributions. Understanding where wildlife encounters plastic is crucial for targeting research and mitigation. Oceanic seabirds, particularly petrels, frequently ingest plastic, are highly threatened, and cover vast distances during foraging and migration. However, the spatial overlap between petrels and plastics is poorly understood. Here we combine marine plastic density estimates with individual movement data for 7137 birds of 77 petrel species to estimate relative exposure risk. We identify high exposure risk areas in the Mediterranean and Black seas, and the northeast Pacific, northwest Pacific, South Atlantic and southwest Indian oceans. Plastic exposure risk varies greatly among species and populations, and between breeding and non-breeding seasons. Exposure risk is disproportionately high for Threatened species. Outside the Mediterranean and Black seas, exposure risk is highest in the high seas and Exclusive Economic Zones (EEZs) of the USA, Japan, and the UK. Birds generally had higher plastic exposure risk outside the EEZ of the country where they breed. We identify conservation and research priorities, and highlight that international collaboration is key to addressing the impacts of marine plastic on wide-ranging species.

Conservation of breeding seabirds typically requires detailed data on where they feed at sea. Ecological niche models (ENMs) can fill data gaps, but rarely perform well when transferred to new regions. Alternatively, the foraging radius approach simply encircles the sea surrounding a breeding seabird colony (a foraging circle), but overestimates foraging habitat. Here, we investigate whether ENMs can transfer (predict) foraging niches of breeding tropical seabirds between global colonies, and whether ENMs can refine foraging circles. We collate a large global dataset of tropical seabird tracks (12000 trips, 16 species, 60 colonies) to build a comprehensive summary of tropical seabird foraging ranges and to train ENMs. We interrogate ENM transferability and assess the confidence with which unsuitable habitat predicted by ENMs can be excluded from within foraging circles. We apply this refinement framework to the Great Barrier Reef (GBR), Australia to identify a network of candidate marine protected areas (MPAs) for seabirds. We found little ability to generalise and transfer breeding tropical seabird foraging niches across all colonies for any species (mean AUC: 0.56, range 0.4-0.82). Low global transferability was partially explained by colony clusters that predicted well internally but other colony clusters poorly. After refinement with ENMs, foraging circles still contained 89% of known foraging areas from tracking data, providing confidence that important foraging habitat was not erroneously excluded by greater refinement from high transferability ENMs nor minor refinement from low transferability ENMs. Foraging radii estimated the total foraging area of the GBR breeding seabird community as 2,941,000 km2, which was refined by excluding between 197,000 km2 and 1,826,000 km2 of unsuitable foraging habitat. ENMs trained on local GBR tracking achieved superior refinement over globally trained models, demonstrating the value of local tracking. Our framework demonstra

Animal tracking has afforded insights into patterns of space use in numerous species and thereby informed area-based conservation planning. A crucial consideration when estimating spatial distributions from tracking data is whether the sample of tracked animals is representative of the wider population. However, it may also be important to track animals in multiple years to capture changes in distribution in response to varying environmental conditions. Using GPS-tracking data from 23 seabird species, we assessed the importance of multi-year sampling for identifying important sites for conservation during the chick-rearing period, when seabirds are most spatially constrained. We found a high degree of spatial overlap among distributions from different years in most species. Multi-year sampling often captured a significantly higher portion of reference distributions (based on all data for a population) than sampling in a single year. However, we estimated that data from a single year would on average miss only 5 % less of the full distribution of a population compared to equal-sized samples collected across three years (min: −0.3 %, max: 17.7 %, n = 23). Our results suggest a key consideration for identifying important sites from tracking data is whether enough individuals were tracked to provide a representative estimate of the population distribution during the sampling period, rather than that tracking necessarily take place in multiple years. By providing an unprecedented multi-species perspective on annual spatial consistency, this work has relevance for the application of tracking data to informing the conservation of seabirds.

Animal-borne sensors ('bio-loggers') can record a suite of kinematic and environmental data, which can elucidate animal ecophysiology and improve conservation efforts. Machine learning techniques are used for interpreting the large amounts of data recorded by bio-loggers, but there exists no common framework for comparing the different machine learning techniques in this domain. To address this, we present the Bio-logger Ethogram Benchmark (BEBE), a collection of datasets with behavioral annotations, as well as a modeling task and evaluation metrics. BEBE is to date the largest, most taxonomically diverse, publicly available benchmark of this type, and includes 1654 hours of data collected from 149 individuals across nine taxa. In addition, using BEBE, we test a novel self-supervised learning approach to identifying animal behaviors based on bio-logger data, using a deep neural network pre-trained with self-supervision on data collected from human wrist-worn accelerometers. We show that this approach out-performs common alternatives, especially in a setting with a low amount of training data. Datasets, models, and evaluation code are made publicly available at https://github.com/earthspecies/BEBE, to enable community use of BEBE as a point of comparison in methods development., Comment: For associated code repositories, see https://github.com/earthspecies/BEBE/ and https://github.com/earthspecies/BEBE-datasets/ . For data repository, see https://zenodo.org/record/7947104

Plastic pollution is distributed patchily around the world's oceans. Likewise, marine organisms that are vulnerable to plastic ingestion or entanglement have uneven distributions. Understanding where wildlife encounters plastic is crucial for targeting research and mitigation. Oceanic seabirds, particularly petrels, frequently ingest plastic, are highly threatened, and cover vast distances during foraging and migration. However, the spatial overlap between petrels and plastics is poorly understood. Here we combine marine plastic density estimates with individual movement data for 7137 birds of 77 petrel species to estimate relative exposure risk. We identify high exposure risk areas in the Mediterranean and Black seas, and the northeast Pacific, northwest Pacific, South Atlantic and southwest Indian oceans. Plastic exposure risk varies greatly among species and populations, and between breeding and non-breeding seasons. Exposure risk is disproportionately high for Threatened species. Outside the Mediterranean and Black seas, exposure risk is highest in the high seas and Exclusive Economic Zones (EEZs) of the USA, Japan, and the UK. Birds generally had higher plastic exposure risk outside the EEZ of the country where they breed. We identify conservation and research priorities, and highlight that international collaboration is key to addressing the impacts of marine plastic on wide-ranging species.

Arboreal herbivores require large digestive tracts for leaf fermentation and detoxification; however, they must also have a low body mass that allows them to reach the foliage. The three-toed sloth, Bradypus tridactylus, experiences this trade-off, as leaves comprise 97.2% of its diet. Their calorie intake is extremely low owing to the low available caloric density of leaves and slow digestive processes related to leaf fibre fermentation and secondary compound detoxification. Sloths may require a high body temperature to assist fermentation; however, thermogenesis is energy-consuming. To investigate how sloths accomplish thermoregulation using marginal energy, we attached heart rate (HR) and temperature loggers to wild B. tridactylus individuals inhabiting the Amazon rainforest and recorded their HR and body surface temperature (T[skin]). T[skin] changed with ambient temperature (Ta) but was higher than Ta in 99.2% of cases. Increases in T[skin] and HR did not coincide, suggesting that the increases were not caused by thermogenesis. Instead, they may passively increase T[skin] by selecting warmer microhabitats and sunbathing. Consequently, 90.5% of T[skin] were within 27.6–36.0 °C while the Ta fluctuated between 21.5 and 42.9 °C. This low-cost thermoregulation results in a low HR. In this study, the mean HR during observation was approximately 38.4% of the expected value based on the mammalian allometric relationship between body mass and HR. Thus, these properties may contribute to the low metabolic rates of sloths, alleviating their restricted energy intake.

Body-mounted accelerometers provide a new prospect for estimating power use in flying birds, as the signal varies with the two major kinematic determinants of aerodynamic power: wingbeat frequency and amplitude. Yet wingbeat frequency is sometimes used as a proxy for power output in isolation. There is, therefore, a need to understand which kinematic parameter birds vary and whether this is predicted by flight mode (e.g. accelerating, ascending/descending flight), speed or morphology. We investigate this using high-frequency acceleration data from (i) 14 species flying in the wild, (ii) two species flying in controlled conditions in a wind tunnel and (iii) a review of experimental and field studies. While wingbeat frequency and amplitude were positively correlated, R2 values were generally low, supporting the idea that parameters can vary independently. Indeed, birds were more likely to modulate wingbeat amplitude for more energy-demanding flight modes, including climbing and take-off. Nonetheless, the striking variability, even within species and flight types, highlights the complexity of describing the kinematic relationships, which appear sensitive to both the biological and physical context. Notwithstanding this, acceleration metrics that incorporate both kinematic parameters should be more robust proxies for power than wingbeat frequency alone.

Body-mounted accelerometers provide a new prospect for estimating power use in flying birds, as the signal varies with the two major kinematic determinants of aerodynamic power: wingbeat frequency and amplitude. Yet wingbeat frequency is sometimes used as a proxy for power output in isolation. There is, therefore, a need to understand which kinematic parameter birds vary and whether this is predicted by flight mode (e.g. accelerating, ascending/descending flight), speed or morphology. We investigate this using high-frequency acceleration data from (i) 14 species flying in the wild, (ii) two species flying in controlled conditions in a wind tunnel and (iii) a review of experimental and field studies. While wingbeat frequency and amplitude were positively correlated, R2 values were generally low, supporting the idea that parameters can vary independently. Indeed, birds were more likely to modulate wingbeat amplitude for more energy-demanding flight modes, including climbing and take-off. Nonetheless, the striking variability, even within species and flight types, highlights the complexity of describing the kinematic relationships, which appear sensitive to both the biological and physical context. Notwithstanding this, acceleration metrics that incorporate both kinematic parameters should be more robust proxies for power than wingbeat frequency alone.

Anthropogenic marine debris is a threat to marine organisms. Understanding how this debris spatially distributes at sea and may become associated with marine wildlife are key steps to tackle this current issue. Using bird-borne GPS- and video-loggers on 13 black-footed albatrosses Phoebastria nigripes breeding in Torishima, Japan, we examined the distribution of large floating debris in the Kuroshio Current area, western North Pacific. A total of 16 floating debris, including styrofoam (n=4), plastic pieces (n=3), plastic sheet (n=1), fishery-related items (rope or netting, n=4), and unidentified debris (n=4), were recorded across the 9003 km covered by nine birds. The debris was concentrated in the southern area of the Kuroshio Current, where the surface current was weak, and the albatrosses were foraging. The albatrosses displayed changes in flight direction towards the debris when at a mean distance of 4.9 km, similarly to when approaching prey, and one bird was observed pecking at a plastic sheet; indicating that albatrosses actively interacted with the debris. This paper shows the usefulness of studying wide-ranging marine predators through the use of combined biologging tools, and highlights areas with increased risk of debris exposure and behavioral responses to debris items.

Anthropogenic marine debris is a threat to marine organisms. Understanding how this debris spatially distributes at sea and may become associated with marine wildlife are key steps to tackle this current issue. Using bird-borne GPS- and video-loggers on 13 black-footed albatrosses Phoebastria nigripes breeding in Torishima, Japan, we examined the distribution of large floating debris in the Kuroshio Current area, western North Pacific. A total of 16 floating debris, including styrofoam (n=4), plastic pieces (n=3), plastic sheet (n=1), fishery-related items (rope or netting, n=4), and unidentified debris (n=4), were recorded across the 9003 km covered by nine birds. The debris was concentrated in the southern area of the Kuroshio Current, where the surface current was weak, and the albatrosses were foraging. The albatrosses displayed changes in flight direction towards the debris when at a mean distance of 4.9 km, similarly to when approaching prey, and one bird was observed pecking at a plastic sheet; indicating that albatrosses actively interacted with the debris. This paper shows the usefulness of studying wide-ranging marine predators through the use of combined biologging tools, and highlights areas with increased risk of debris exposure and behavioral responses to debris items.

Anthropogenic marine debris is a threat to marine organisms. Understanding how this debris spatially distributes at sea and may become associated with marine wildlife are key steps to tackle this current issue. Using bird-borne GPS- and video-loggers on 13 black-footed albatrosses Phoebastria nigripes breeding in Torishima, Japan, we examined the distribution of large floating debris in the Kuroshio Current area, western North Pacific. A total of 16 floating debris, including styrofoam (n=4), plastic pieces (n=3), plastic sheet (n=1), fishery-related items (rope or netting, n=4), and unidentified debris (n=4), were recorded across the 9003 km covered by nine birds. The debris was concentrated in the southern area of the Kuroshio Current, where the surface current was weak, and the albatrosses were foraging. The albatrosses displayed changes in flight direction towards the debris when at a mean distance of 4.9 km, similarly to when approaching prey, and one bird was observed pecking at a plastic sheet; indicating that albatrosses actively interacted with the debris. This paper shows the usefulness of studying wide-ranging marine predators through the use of combined biologging tools, and highlights areas with increased risk of debris exposure and behavioral responses to debris items.

Extracting the interaction rules of biological agents from movement sequences pose challenges in various domains. Granger causality is a practical framework for analyzing the interactions from observed time-series data; however, this framework ignores the structures and assumptions of the generative process in animal behaviors, which may lead to interpretational problems and sometimes erroneous assessments of causality. In this paper, we propose a new framework for learning Granger causality from multi-animal trajectories via augmented theory-based behavioral models with interpretable data-driven models. We adopt an approach for augmenting incomplete multi-agent behavioral models described by time-varying dynamical systems with neural networks. For efficient and interpretable learning, our model leverages theory-based architectures separating navigation and motion processes, and the theory-guided regularization for reliable behavioral modeling. This can provide interpretable signs of Granger-causal effects over time, i.e., when specific others cause the approach or separation. In experiments using synthetic datasets, our method achieved better performance than various baselines. We then analyzed multi-animal datasets of mice, flies, birds, and bats, which verified our method and obtained novel biological insights., Comment: 24 pages, 5 figures, to appear in NeurIPS 2021

An efficient inverse reinforcement learning for generating trajectories is proposed based of 2D and 3D activity forecasting. We modify reward function with $L_p$ norm and propose convolution into value iteration steps, which is called convolutional value iteration. Experimental results with seabird trajectories (43 for training and 10 for test), our method is best in terms of MHD error and performs fastest. Generated trajectories for interpolating missing parts of trajectories look much similar to real seabird trajectories than those by the previous works., Comment: The 2019 International Workshop on Frontiers of Computer Vision (IW-FCV2019)

How surface-feeding albatrosses feed on deep-sea squids has long been a mystery. We investigated foraging behavior during daylight hours of 20 Laysan albatrosses Phoebastria immutabilis breeding in Hawaii using GPS- and camera-loggers. The birds traveled to the North Pacific Transition Zone up to 600 km north of their breeding site. The camera images showed that Laysan albatrosses fed on large (～ 1 m body length), intact floating dead squids (6 events) and floating fragmented squids (10 events) over deep oceanic water (>2000 m) while they flew in a straight path without sinuous searching. Feeding events on squids were not observed during trips when fishing vessels were photographed and seemed to be distributed randomly and sparsely. Thus, this study suggests that Laysan albatrosses found large, presumably post-spawning, squids opportunistically while they were traveling during daylight hours. Although we did not find cetaceans in our surface pictures, we could not rule out the possibility that birds fed on squids, especially fragmented specimens, in the regurgitates of cetaceans at depth. This study demonstrates the usefulness of combining animal-borne GPS- and camera-loggers on wide-ranging top predators for studying the distribution of little known deep-sea squids and their importance in the diet of marine top predators.

How surface-feeding albatrosses feed on deep-sea squids has long been a mystery. We investigated foraging behavior during daylight hours of 20 Laysan albatrosses Phoebastria immutabilis breeding in Hawaii using GPS- and camera-loggers. The birds traveled to the North Pacific Transition Zone up to 600 km north of their breeding site. The camera images showed that Laysan albatrosses fed on large (～ 1 m body length), intact floating dead squids (6 events) and floating fragmented squids (10 events) over deep oceanic water (>2000 m) while they flew in a straight path without sinuous searching. Feeding events on squids were not observed during trips when fishing vessels were photographed and seemed to be distributed randomly and sparsely. Thus, this study suggests that Laysan albatrosses found large, presumably post-spawning, squids opportunistically while they were traveling during daylight hours. Although we did not find cetaceans in our surface pictures, we could not rule out the possibility that birds fed on squids, especially fragmented specimens, in the regurgitates of cetaceans at depth. This study demonstrates the usefulness of combining animal-borne GPS- and camera-loggers on wide-ranging top predators for studying the distribution of little known deep-sea squids and their importance in the diet of marine top predators.

How surface-feeding albatrosses feed on deep-sea squids has long been a mystery. We investigated foraging behavior during daylight hours of 20 Laysan albatrosses Phoebastria immutabilis breeding in Hawaii using GPS- and camera-loggers. The birds traveled to the North Pacific Transition Zone up to 600 km north of their breeding site. The camera images showed that Laysan albatrosses fed on large (～ 1 m body length), intact floating dead squids (6 events) and floating fragmented squids (10 events) over deep oceanic water (>2000 m) while they flew in a straight path without sinuous searching. Feeding events on squids were not observed during trips when fishing vessels were photographed and seemed to be distributed randomly and sparsely. Thus, this study suggests that Laysan albatrosses found large, presumably post-spawning, squids opportunistically while they were traveling during daylight hours. Although we did not find cetaceans in our surface pictures, we could not rule out the possibility that birds fed on squids, especially fragmented specimens, in the regurgitates of cetaceans at depth. This study demonstrates the usefulness of combining animal-borne GPS- and camera-loggers on wide-ranging top predators for studying the distribution of little known deep-sea squids and their importance in the diet of marine top predators.

The energetics of adult Streaked Shearwaters Calonectris leucomelas during the chick-rearing period were examined on Awa Island, Japan, in 2008 and 2009. Basal metabolic rates (BMR) were quantified using an open-flow respirometry system and field metabolic rates (FMR) were quantified using a doubly labelled water (DLW) method. In addition, we used activity loggers to estimate time allocations for different activities at sea. BMR was 0.0124 kJ g(-1) h(-1) (+/- 0.0153, N=4) on average and corresponded to 54% of the value predicted from allometric equations. FMR was 0.0634 kJ g(-1) h(-1) (+/- 0.0331, N=3) and was equivalent to 5.1 times BMR, which was higher than values reported for albatrosses (2-4 times BMR). Shearwaters made 50.3 landings a day (+/- 9.8, N=12) and spent 44.8% (+/- 8.0, N=12) of their time sitting on the water. They landed on water approximately twice as often as albatrosses (which have been well-studied using DLW), but they both spent similar proportions of their time on water. Frequent landings at sea, and frequent takeoffs, may generate incremental energetic expenses because of the use of flapping flight; therefore, the Streaked Shearwater's relatively high FMR may be related to its high number of landings.

The energetics of adult Streaked Shearwaters Calonectris leucomelas during the chick-rearing period were examined on Awa Island, Japan, in 2008 and 2009. Basal metabolic rates (BMR) were quantified using an open-flow respirometry system and field metabolic rates (FMR) were quantified using a doubly labelled water (DLW) method. In addition, we used activity loggers to estimate time allocations for different activities at sea. BMR was 0.0124 kJ g(-1) h(-1) (+/- 0.0153, N=4) on average and corresponded to 54% of the value predicted from allometric equations. FMR was 0.0634 kJ g(-1) h(-1) (+/- 0.0331, N=3) and was equivalent to 5.1 times BMR, which was higher than values reported for albatrosses (2-4 times BMR). Shearwaters made 50.3 landings a day (+/- 9.8, N=12) and spent 44.8% (+/- 8.0, N=12) of their time sitting on the water. They landed on water approximately twice as often as albatrosses (which have been well-studied using DLW), but they both spent similar proportions of their time on water. Frequent landings at sea, and frequent takeoffs, may generate incremental energetic expenses because of the use of flapping flight; therefore, the Streaked Shearwater's relatively high FMR may be related to its high number of landings.

Otsuka R., Yoshimura N., Tanigaki K., et al. Exploring deep learning techniques for wild animal behaviour classification using animal-borne accelerometers. Methods in Ecology and Evolution 15, 716 (2024); https://doi.org/10.1111/2041-210X.14294., Machine learning-based behaviour classification using acceleration data is a powerful tool in bio-logging research. Deep learning architectures such as convolutional neural networks (CNN), long short-term memory (LSTM) and self-attention mechanism as well as related training techniques have been extensively studied in human activity recognition. However, they have rarely been used in wild animal studies. The main challenges of acceleration-based wild animal behaviour classification include data shortages, class imbalance problems, various types of noise in data due to differences in individual behaviour and where the loggers were attached and complexity in data due to complex animal-specific behaviours, which may have limited the application of deep learning techniques in this area. To overcome these challenges, we explored the effectiveness of techniques for efficient model training: data augmentation, manifold mixup and pre-training of deep learning models with unlabelled data, using datasets from two species of wild seabirds and state-of-the-art deep learning model architectures. Data augmentation improved the overall model performance when one of the various techniques (none, scaling, jittering, permutation, time-warping and rotation) was randomly applied to each data during mini-batch training. Manifold mixup also improved model performance, but not as much as random data augmentation. Pre-training with unlabelled data did not improve model performance. The state-of-the-art deep learning models, including a model consisting of four CNN layers, an LSTM layer and a multi-head attention layer, as well as its modified version with shortcut connection, showed better performance among other comparative models. Using only raw acceleration data as inputs, these models outperformed classic machine learning approaches that used 119 handcrafted features. Our experiments showed that deep learning techniques are promising for acceleration-based behaviour classification of w

It has been proposed that predators searching for prey acquire food according to a probabilistic framework, where success is based on ‘luck’ and the odds of success vary with prey abundance. If true, this has major ramifications for variation in the rates of energy acquisition within animal populations, which is particularly pertinent in offspring provisioning and breeding success, because smaller animals (the young) cannot starve for as long as the adults. However, despite much general speculation about rates of food acquisition, no study has measured whether food encounter is probabilistic in wild animals. We used animal-mounted cameras to document all prey captures by wild imperial shags Leucocarbo atriceps as they hunted underwater and show that, although they mostly do not have inter-prey acquisition time distributions that accord with a ‘luck-based’ framework assuming a constant probability of finding prey over time, there is no difference in the predicted amount of food captured between models that use the empirical data or theoretical Poisson-based fits of the data. We also noted considerable inter-individual differences in foraging success that far exceeded any differences between empirical and theoretical inter-prey acquisition time distributions. The data were used in a probabilistic foraging model that made explicit the mechanistic link between random prey encounters and food-dependent breeding success, indicating that ‘less lucky’ individuals could not provision their broods at rates commensurate with normal growth while the ‘lucky’ birds could do so easily. Given the nature of food encounter in these birds, coupled with substantial inter-individual variation in foraging success, we suggest that more successful individuals are particularly choosey about when, how and where to forage, which results in them operating with higher odds of success.

It has been proposed that predators searching for prey acquire food according to a probabilistic framework, where success is based on ‘luck’ and the odds of success vary with prey abundance. If true, this has major ramifications for variation in the rates of energy acquisition within animal populations, which is particularly pertinent in offspring provisioning and breeding success, because smaller animals (the young) cannot starve for as long as the adults. However, despite much general speculation about rates of food acquisition, no study has measured whether food encounter is probabilistic in wild animals. We used animal-mounted cameras to document all prey captures by wild imperial shags Leucocarbo atriceps as they hunted underwater and show that, although they mostly do not have inter-prey acquisition time distributions that accord with a ‘luck-based’ framework assuming a constant probability of finding prey over time, there is no difference in the predicted amount of food captured between models that use the empirical data or theoretical Poisson-based fits of the data. We also noted considerable inter-individual differences in foraging success that far exceeded any differences between empirical and theoretical inter-prey acquisition time distributions. The data were used in a probabilistic foraging model that made explicit the mechanistic link between random prey encounters and food-dependent breeding success, indicating that ‘less lucky’ individuals could not provision their broods at rates commensurate with normal growth while the ‘lucky’ birds could do so easily. Given the nature of food encounter in these birds, coupled with substantial inter-individual variation in foraging success, we suggest that more successful individuals are particularly choosey about when, how and where to forage, which results in them operating with higher odds of success.

Otsuka R., Yoshimura N., Tanigaki K., et al. Exploring deep learning techniques for wild animal behaviour classification using animal-borne accelerometers. Methods in Ecology and Evolution 15, 716 (2024); https://doi.org/10.1111/2041-210X.14294., Machine learning-based behaviour classification using acceleration data is a powerful tool in bio-logging research. Deep learning architectures such as convolutional neural networks (CNN), long short-term memory (LSTM) and self-attention mechanism as well as related training techniques have been extensively studied in human activity recognition. However, they have rarely been used in wild animal studies. The main challenges of acceleration-based wild animal behaviour classification include data shortages, class imbalance problems, various types of noise in data due to differences in individual behaviour and where the loggers were attached and complexity in data due to complex animal-specific behaviours, which may have limited the application of deep learning techniques in this area. To overcome these challenges, we explored the effectiveness of techniques for efficient model training: data augmentation, manifold mixup and pre-training of deep learning models with unlabelled data, using datasets from two species of wild seabirds and state-of-the-art deep learning model architectures. Data augmentation improved the overall model performance when one of the various techniques (none, scaling, jittering, permutation, time-warping and rotation) was randomly applied to each data during mini-batch training. Manifold mixup also improved model performance, but not as much as random data augmentation. Pre-training with unlabelled data did not improve model performance. The state-of-the-art deep learning models, including a model consisting of four CNN layers, an LSTM layer and a multi-head attention layer, as well as its modified version with shortcut connection, showed better performance among other comparative models. Using only raw acceleration data as inputs, these models outperformed classic machine learning approaches that used 119 handcrafted features. Our experiments showed that deep learning techniques are promising for acceleration-based behaviour classification of w