How to treat mixed behavior segments in supervised machine learning of behavioural modes from inertial measurement data.

The application of supervised machine learning methods to identify behavioural modes from inertial measurements of bio-loggers has become a standard tool in behavioural ecology. Several design choices can affect the accuracy of identifying the behavioural modes. One such choice is the inclusion or exclusion of segments consisting of more than a single behaviour (mixed segments) in the machine learning model training data. Currently, the common practice is to ignore such segments during model training. In this paper we tested the hypothesis that including mixed segments in model training will improve accuracy, as the model would perform better in identifying them in the test data. We test this hypothesis using a series of data simulations on four datasets of accelerometer data coupled with behaviour observations, obtained from four study species (Damaraland mole-rats, meerkats, olive baboons, polar bears). Results show that when a substantial proportion of the test data are mixed behaviour segments (above ~ 10%), including mixed segments in machine learning model training improves the accuracy of classification. These results were consistent across the four study species, and robust to changes in segment length, sample size, and degree of mixture within the mixed segments. However, we also find that in some cases (particularly in baboons) models trained with mixed segments show reduced accuracy in classifying test data containing only single behaviour (pure) segments, compared to models trained without mixed segments. Based on these results, we recommend that when the classification model is expected to deal with a substantial proportion of mixed behaviour segments (> 10%), it is beneficial to include them in model training, otherwise, it is unnecessary but also not harmful. The exception is when there is a basis to assume that the training data contains a higher rate of mixed segments than the actual (unobserved) data to be classified-such a situation may occur particularly when training data are collected in captivity and used to classify data from the wild. In this case, excess inclusion of mixed segments in training data should probably be avoided.

Revealing principles of autonomous thermal soaring in windy conditions using vulture-inspired deep reinforcement-learning.

Thermal soaring, a technique used by birds and gliders to utilize updrafts of hot air, is an appealing model-problem for studying motion control and how it is learned by animals and engineered autonomous systems. Thermal soaring has rich dynamics and nontrivial constraints, yet it uses few control parameters and is becoming experimentally accessible. Following recent developments in applying reinforcement learning methods for training deep neural-network (deep-RL) models to soar autonomously both in simulation and real gliders, here we develop a simulation-based deep-RL system to study the learning process of thermal soaring. We find that this process has learning bottlenecks, we define a new efficiency metric and use it to characterize learning robustness, we compare the learned policy to data from soaring vultures, and find that the neurons of the trained network divide into function clusters that evolve during learning. These results pose thermal soaring as a rich yet tractable model-problem for the learning of motion control.

Estimating individual exposure to predation risk in group-living baboons, Papio anubis.

In environments with multiple predators, vulnerabilities associated with the spatial positions of group-living prey are non-uniform and depend on the hunting styles of the predators. Theoretically, coursing predators follow their prey over long distances and attack open areas, exposing individuals at the edge of the group to predation risk more than those at the center (marginal predation). In contrast, ambush predators lurk unnoticed by their prey and appear randomly anywhere in the group; therefore, isolated individuals in the group would be more vulnerable to predators. These positions of vulnerability to predation are expected to be taken by larger-bodied males. Moreover, dominant males presumably occupy the center of the safe group. However, identifying individuals at higher predation risk requires both simultaneous recording of predator location and direct observation of predation events; empirical observations leave ambiguity as to who is at risk. Instead, several theoretical methods (predation risk proxies) have been proposed to assess predation risk: (1) the size of the individual 'unlimited domain of danger' based on Voronoi tessellation, (2) the size of the 'limited domain of danger' based on predator detection distance, (3) peripheral/center position in the group (minimum convex polygon), (4) the number and direction of others in the vicinity (surroundedness), and (5) dyadic distances. We explored the age-sex distribution of individuals in at-risk positions within a wild baboon group facing predation risk from leopards, lions, and hyenas, using Global Positioning System collars. Our analysis of the location data from 26 baboons revealed that adult males were consistently isolated at the edge of the group in all predation risk proxies. Empirical evidence from previous studies indicates that adult male baboons are the most frequently preyed upon, and our results highlights the importance of spatial positioning in this.

Ecological and social pressures interfere with homeostatic sleep regulation in the wild.

Sleep is fundamental to the health and fitness of all animals. The physiological importance of sleep is underscored by the central role of homeostasis in determining sleep investment - following periods of sleep deprivation, individuals experience longer and more intense sleep bouts. Yet, most sleep research has been conducted in highly controlled settings, removed from evolutionarily relevant contexts that may hinder the maintenance of sleep homeostasis. Using triaxial accelerometry and GPS to track the sleep patterns of a group of wild baboons (Papio anubis), we found that ecological and social pressures indeed interfere with homeostatic sleep regulation. Baboons sacrificed time spent sleeping when in less familiar locations and when sleeping in proximity to more group-mates, regardless of how long they had slept the prior night or how much they had physically exerted themselves the preceding day. Further, they did not appear to compensate for lost sleep via more intense sleep bouts. We found that the collective dynamics characteristic of social animal groups persist into the sleep period, as baboons exhibited synchronized patterns of waking throughout the night, particularly with nearby group-mates. Thus, for animals whose fitness depends critically on avoiding predation and developing social relationships, maintaining sleep homeostasis may be only secondary to remaining vigilant when sleeping in risky habitats and interacting with group-mates during the night. Our results highlight the importance of studying sleep in ecologically relevant contexts, where the adaptive function of sleep patterns directly reflects the complex trade-offs that have guided its evolution.

Big-data approaches lead to an increased understanding of the ecology of animal movement.

Understanding animal movement is essential to elucidate how animals interact, survive, and thrive in a changing world. Recent technological advances in data collection and management have transformed our understanding of animal "movement ecology" (the integrated study of organismal movement), creating a big-data discipline that benefits from rapid, cost-effective generation of large amounts of data on movements of animals in the wild. These high-throughput wildlife tracking systems now allow more thorough investigation of variation among individuals and species across space and time, the nature of biological interactions, and behavioral responses to the environment. Movement ecology is rapidly expanding scientific frontiers through large interdisciplinary and collaborative frameworks, providing improved opportunities for conservation and insights into the movements of wild animals, and their causes and consequences.

Animal lifestyle affects acceptable mass limits for attached tags.

Animal-attached devices have transformed our understanding of vertebrate ecology. To minimize any associated harm, researchers have long advocated that tag masses should not exceed 3% of carrier body mass. However, this ignores tag forces resulting from animal movement. Using data from collar-attached accelerometers on 10 diverse free-ranging terrestrial species from koalas to cheetahs, we detail a tag-based acceleration method to clarify acceptable tag mass limits. We quantify animal athleticism in terms of fractions of animal movement time devoted to different collar-recorded accelerations and convert those accelerations to forces (acceleration × tag mass) to allow derivation of any defined force limits for specified fractions of any animal's active time. Specifying that tags should exert forces that are less than 3% of the gravitational force exerted on the animal's body for 95% of the time led to corrected tag masses that should constitute between 1.6% and 2.98% of carrier mass, depending on athleticism. Strikingly, in four carnivore species encompassing two orders of magnitude in mass (ca 2-200 kg), forces exerted by '3%' tags were equivalent to 4-19% of carrier body mass during moving, with a maximum of 54% in a hunting cheetah. This fundamentally changes how acceptable tag mass limits should be determined by ethics bodies, irrespective of the force and time limits specified.

Locomotor compromises maintain group cohesion in baboon troops on the move.

When members of a group differ in locomotor capacity, coordinating collective movement poses a challenge: some individuals may have to move faster (or slower) than their preferred speed to remain together. Such compromises have energetic repercussions, yet research in collective behaviour has largely neglected locomotor consensus costs. Here, we integrate high-resolution tracking of wild baboon locomotion and movement with simulations to demonstrate that size-based variation in locomotor capacity poses an obstacle to the collective movement. While all baboons modulate their gait and move-pause dynamics during collective movement, the costs of maintaining cohesion are disproportionately borne by smaller group members. Although consensus costs are not distributed equally, all group-mates do make locomotor compromises, suggesting a shared decision-making process drives the pace of collective movement in this highly despotic species. These results highlight the importance of considering how social dynamics and locomotor capacity interact to shape the movement ecology of group-living species.

Social foraging and individual consistency in following behaviour: testing the information centre hypothesis in free-ranging vultures.

Uncertainties regarding food location and quality are among the greatest challenges faced by foragers and communal roosting may facilitate success through social foraging. The information centre hypothesis (ICH) suggests that uninformed individuals at shared roosts benefit from following informed individuals to previously visited resources. We tested several key prerequisites of the ICH in a social obligate scavenger, the Eurasian griffon vulture (Gyps fulvus), by tracking movements and behaviour of sympatric individuals over extended periods and across relatively large spatial scales, thereby precluding alternative explanations such as local enhancement. In agreement with the ICH, we found that 'informed' individuals returning to previously visited carcasses were followed by 'uninformed' vultures that consequently got access to these resources. When a dyad (two individuals that depart from the same roost within 2 min of each other) included an informed individual, they spent a higher proportion of the flight time close to each other at a shorter distance between them than otherwise. Although all individuals occasionally profited from following others, they differed in their tendencies to be informed or uninformed. This study provides evidence for 'following behaviour' in natural conditions and demonstrates differential roles and information states among foragers within a population. Moreover, demonstrating the possible reliance of vultures on following behaviour emphasizes that individuals in declining populations may suffer from reduced foraging efficiency.

Decision-making by a soaring bird: time, energy and risk considerations at different spatio-temporal scales.

Natural selection theory suggests that mobile animals trade off time, energy and risk costs with food, safety and other pay-offs obtained by movement. We examined how birds make movement decisions by integrating aspects of flight biomechanics, movement ecology and behaviour in a hierarchical framework investigating flight track variation across several spatio-temporal scales. Using extensive global positioning system and accelerometer data from Eurasian griffon vultures (Gyps fulvus) in Israel and France, we examined soaring-gliding decision-making by comparing inbound versus outbound flights (to or from a central roost, respectively), and these (and other) home-range foraging movements (up to 300 km) versus long-range movements (longer than 300 km). We found that long-range movements and inbound flights have similar features compared with their counterparts: individuals reduced journey time by performing more efficient soaring-gliding flight, reduced energy expenditure by flapping less and were more risk-prone by gliding more steeply between thermals. Age, breeding status, wind conditions and flight altitude (but not sex) affected time and energy prioritization during flights. We therefore suggest that individuals facing time, energy and risk trade-offs during movements make similar decisions across a broad range of ecological contexts and spatial scales, presumably owing to similarity in the uncertainty about movement outcomes.This article is part of the themed issue 'Moving in a moving medium: new perspectives on flight'.

Adult vultures outperform juveniles in challenging thermal soaring conditions.

Due to the potentially detrimental consequences of low performance in basic functional tasks, individuals are expected to improve performance with age and show the most marked changes during early stages of life. Soaring-gliding birds use rising-air columns (thermals) to reduce energy expenditure allocated to flight. We offer a framework to evaluate thermal soaring performance, and use GPS-tracking to study movements of Eurasian griffon vultures (Gyps fulvus). Because the location and intensity of thermals are variable, we hypothesized that soaring performance would improve with experience and predicted that the performance of inexperienced individuals (<2 months) would be inferior to that of experienced ones (>5 years). No differences were found in body characteristics, climb rates under low wind shear, and thermal selection, presumably due to vultures' tendency to forage in mixed-age groups. Adults, however, outperformed juveniles in their ability to adjust fine-scale movements under challenging conditions, as juveniles had lower climb rates under intermediate wind shear, particularly on the lee-side of thermal columns. Juveniles were also less efficient along the route both in terms of time and energy. The consequences of these handicaps are probably exacerbated if juveniles lag behind adults in finding and approaching food.

AcceleRater: a web application for supervised learning of behavioral modes from acceleration measurements.

BACKGROUND: The study of animal movement is experiencing rapid progress in recent years, forcefully driven by technological advancement. Biologgers with Acceleration (ACC) recordings are becoming increasingly popular in the fields of animal behavior and movement ecology, for estimating energy expenditure and identifying behavior, with prospects for other potential uses as well. Supervised learning of behavioral modes from acceleration data has shown promising results in many species, and for a diverse range of behaviors. However, broad implementation of this technique in movement ecology research has been limited due to technical difficulties and complicated analysis, deterring many practitioners from applying this approach. This highlights the need to develop a broadly applicable tool for classifying behavior from acceleration data. DESCRIPTION: Here we present a free-access python-based web application called AcceleRater, for rapidly training, visualizing and using models for supervised learning of behavioral modes from ACC measurements. We introduce AcceleRater, and illustrate its successful application for classifying vulture behavioral modes from acceleration data obtained from free-ranging vultures. The seven models offered in the AcceleRater application achieved overall accuracy of between 77.68% (Decision Tree) and 84.84% (Artificial Neural Network), with a mean overall accuracy of 81.51% and standard deviation of 3.95%. Notably, variation in performance was larger between behavioral modes than between models. CONCLUSIONS: AcceleRater provides the means to identify animal behavior, offering a user-friendly tool for ACC-based behavioral annotation, which will be dynamically upgraded and maintained.

Mixed strategies of griffon vultures' (Gyps fulvus) response to food deprivation lead to a hump-shaped movement pattern.

BACKGROUND: The need to obtain food is a critical proximate driver of an organism's movement that shapes the foraging and survival of individual animals. Consequently, the relationship between hunger and foraging has received considerable attention, leading to the common conception that hunger primarily enhances a "food-intake maximization" (FIMax) strategy and intensive search. A complementary explanation, however, suggests a trade-off with precautions taken to reduce the risk of physiological collapse from starvation, under a strategy we denote as "energy-expenditure minimization" (EEMin). The FImax-EEmin trade-off may interact with the forager's hunger level to shape a complex (non-monotonic) response pattern to increasing hunger. Yet, this important trade-off has rarely been investigated, particularly in free-ranging wild animals. We explored how hunger affects the movements of adult griffon vultures (Gyps fulvus) in southern Israel. Transmitters combining GPS and accelerometers provided high-resolution data on vultures' movements and behavior, enabling the identification of feeding events and the estimation of food deprivation periods (FDPs, measured in days), which is used as a proxy for hunger. RESULTS: Data from 47 vultures, tracked for 339 ± 36 days, reveal high variability in FDPs. While flight speed, flight straightness and the proportion of active flights were invariant in relation to food deprivation, a clear hump-shaped response was found for daily flight distances, maximal displacements and flight elevation. These movement characteristics increased during the first five days of the FDP sequence and decreased during the following five days. These characteristics also differed between short FDPs of up to four days, and the first four days of longer FDP sequences. These results suggest a switch from FIMax to EEMin strategies along the FDP sequence. They also indicate that vultures' response to hunger affected the eventual duration of the FDP. During winter (the vultures' incubation period characterized by unfavorable soaring meteorological conditions), the vultures' FIMax response was less intensive and resulted in longer starvation periods, while, in summer, more intensive FIMax response to hunger resulted in shorter FDPs. CONCLUSIONS: Our results show a flexible, non-monotonic response of free-ranging wild animals to increasing hunger levels, reflecting a trade-off between increasing motivation to find food and the risk of starvation. The proposed trade-off offers a unifying perspective to apparently contradictory or case-specific empirical findings.

Using tri-axial acceleration data to identify behavioral modes of free-ranging animals: general concepts and tools illustrated for griffon vultures.

Integrating biomechanics, behavior and ecology requires a mechanistic understanding of the processes producing the movement of animals. This calls for contemporaneous biomechanical, behavioral and environmental data along movement pathways. A recently formulated unifying movement ecology paradigm facilitates the integration of existing biomechanics, optimality, cognitive and random paradigms for studying movement. We focus on the use of tri-axial acceleration (ACC) data to identify behavioral modes of GPS-tracked free-ranging wild animals and demonstrate its application to study the movements of griffon vultures (Gyps fulvus, Hablizl 1783). In particular, we explore a selection of nonlinear and decision tree methods that include support vector machines, classification and regression trees, random forest methods and artificial neural networks and compare them with linear discriminant analysis (LDA) as a baseline for classifying behavioral modes. Using a dataset of 1035 ground-truthed ACC segments, we found that all methods can accurately classify behavior (80-90%) and, as expected, all nonlinear methods outperformed LDA. We also illustrate how ACC-identified behavioral modes provide the means to examine how vulture flight is affected by environmental factors, hence facilitating the integration of behavioral, biomechanical and ecological data. Our analysis of just over three-quarters of a million GPS and ACC measurements obtained from 43 free-ranging vultures across 9783 vulture-days suggests that their annual breeding schedule might be selected primarily in response to seasonal conditions favoring rising-air columns (thermals) and that rare long-range forays of up to 1750 km from the home range are performed despite potentially heavy energetic costs and a low rate of food intake, presumably to explore new breeding, social and long-term resource location opportunities.