Biology subjects

Farine, D.

Publications and source records attributed to Farine, D..

11 recordsLinked to original sources

bioRxiv2025

Resource scarcity increases foraging activity despite thermal risk in an arid-adapted bird

Behavioural flexibility is considered key for species to cope with the effects of climate change. Periods of resource scarcity, such as during climate change-driven droughts, may force animals to increase their foraging activity to meet energetic demands. However, doing so may increase the thermal risk inherent to the trade-off between foraging and exposure to higher temperatures. We use biologging data (internal body temperature and activity budgets classified via accelerometers) from a population of wild vulturine guineafowl (Acryllium vulturinum) to quantify this trade-off across varying resource availability (quantified through NDVI) over 2.5 years, including a period of severe drought. We start by linking internal body temperature measurements to external operative temperature to identify a thermal threshold--where body temperature reaches a maximum--at an external temperature of 33.8{degrees}C. We interpret this threshold as the point where birds must prioritise heat dissipation (physiologically and/or behaviourally) to avoid hyperthermia. Relevantly, we find that a larger part of the day exceeds this external thermal threshold as resources decline. We then show that guineafowl feeding activity positively correlates with resource scarcity, suggesting the birds compensate for diminishing resources by increasing foraging time investment, despite thermal risks. However, the compensatory foraging does not appear to buffer the birds against food shortfalls, as birds still lose weight as resource scarcity increases. Sublethal effects (e.g., body mass loss) of compounding environmental stressors have fitness consequences (e.g., postponement of breeding, as was noted in our population). Our results could point to an evolutionary mismatch; the projected intensification of resource scarcity and higher temperatures under climate change may alter trade-offs in such a way that existing physiological and behavioural responses to cyclical dry seasons are no longer adaptive, in vulturine guineafowl, and in other arid-system organisms more broadly.

ecology↗

bioRxiv2025

Human activities drive novel behaviours and transitions in dolphins

Intensifying human activities are reshaping coastal ecosystems, yet their impact on wildlife behaviour and survival remain poorly understood. We conducted drone-based focal-group follows of endangered Indian Ocean humpback dolphin groups in Goa, India, to quantify how tourism and fisheries jointly impact behavioural states and transitions. By integrating machine learning and georeferencing techniques, we found that both human activities triggered distinct behavioural transitions absent in undisturbed groups. Dolphins expressed escape behaviours exclusively near tourist boats and were more likely to transition into foraging near fishing nets (mostly purse-seines). Smaller groups reacted more strongly to tourist boats and interacted more frequently with fishing nets, especially in seasons when fish landing data showed a decline in their prey. These findings offer new insights into the behavioural mechanisms underlying the co-occurrence of coastal dolphins with humans and provide broader guidelines for managing tourism and fisheries to reduce anthropogenic pressure on threatened delphinids.

ecology↗

bioRxiv2025

Space use fidelity of non-territorial vulturine guineafowl groups is shaped by both environmental and social processes

Animals often use consistent areas. Some are territorial, restricting their space use within territorial boundaries, whereas others at not territorial animals but still restrict their space use despite not being constrained by surrounding conspecifics. Staying within a familiar area can provide a range of benefits, such as using previous knowledge (i.e. memory) to efficiently exploit resources or because they can consistently return to key locations (such as a nest or sleeping site). In group-living animals, consistent space use could reduce the complexity of decision-making time (e.g. by choosing among known foraging sites), facilitating group cohesion. However, to date, little research has explicitly asked what factors determine whether groups use consistent areas. Here we used repeated movements by groups of vulturine guineafowl (Acryllium vulturinum)--leaving and returning back to the same areas in response to seasonal conditions--to examine and disentangle social processes from spatial and ecological factors that might shape the distribution of animals over space. Specifically, we quantified (i) how groups distribute themselves over the landscape, (ii) if their space use is consistent across seasons with similar environmental conditions, (iii) how different social and spatial factors shape the consistency of space use by groups over time, and (iv) how social and spatial factors affect home range overlap between groups. We found that groups were highly consistent in their space use over time and that home ranges were distinct across groups. Fidelity to the core home range area was higher when group composition was more stable, while overall home range fidelity was higher when groups recently experienced milder ecological conditions. Overlap in core areas and the overall home ranges among groups were greater among groups that shared roosts and groups that were fused in the previous season. Home range overlap was also lowest during long intermediate seasons (i.e. a sampling period that immediately follows intermediate season conditions, as opposed to sampling periods that followed dry or wet conditions), suggesting that extended intermediate conditions allow groups to increasingly partition their overall space use. These results provide insights into how the movement decisions by groups, the distribution of animals, and group-level space use emerge, and the role of social and ecological conditions as potential precursors to territoriality.

ecology↗

bioRxiv2025

Integrated deep learning and geo-referencing for drone-based animal tracking with flexible camera angles

O_LIThe ability of drones to provide detailed information on animals and their surroundings makes them ideal for studying animal behaviour at fine scales. While drones can provide high-resolution images of what animals are doing, they should also, in theory, be able to provide data on where they are. However, reconstructing geo-referenced tracks from drone videos that follow animals is challenging, particularly because current methods require specific drone flight patterns and large computational power. C_LIO_LIHere, we combine deep learning and object tracking methods with a novel geo-referencing algorithm which allows us to track individuals across video frames and reconstruct their geo-referenced trajectories. We used a Region-based Convoluted Neural Network to detect animals and a Hungarian tracking algorithm to link detections across video frames, and then geo-referenced each detection in every frame to reconstruct individual trajectories. C_LIO_LIWe tested our geo-refencing algorithm through multiple drone flights with varying flight parameters over known Ground Control Points. The median (95% CI) geo-referencing error was 2.81 (0.74 - 23.23) meters, which reduced by 50% when the drone camera was positioned between -90{degrees} and -40{degrees}. Error increased with drone height and camera angle (-90{degrees} refers to the camera pointing towards the ground) but was not impacted by drone orientation. C_LIO_LIWe then demonstrate the utility of our framework with empirical examples using consumer-level drones. First, we tracked a volunteer carrying a high-resolution GPS unit and overlayed their GPS tracks on our estimated tracks to quantify tracking error. Next, we used drone videos of two delphinid species (Tursiops truncates gephyreus, and Sousa plumbea) representing varying environmental and flight conditions. We were able to successfully infer individual tracks across all conditions except when individuals formed tight clusters, in which case tracks were assigned a group identifier. C_LIO_LIOur framework demonstrates an easy and robust approach to translate drone videos of moving animals into geo-referenced animal tracks which is applicable in many research contexts. A major advance over previous methods is that our algorithm is robust to different camera angles, and provides tracks with accuracy on-par with, or even exceeding, the accuracy from GPS tracking. C_LI HeadlineAutomated drone-based animal tracking

ecology↗

bioRxiv2024

Moving towards more holistic validation of machine learning-based approaches in ecology and evolution

Machine-learning (ML) is revolutionizing field and laboratory studies of animals. However, a challenge when deploying ML for classification tasks is ensuring the models are reliable. Currently, we evaluate models using performance metrics (e.g., precision, recall, F1), but these can overlook the ultimate aim, which is not the outputs themselves (e.g. detected species or individual identities, or behaviour) but their incorporation for hypothesis testing. As improving performance metrics has diminishing returns, particularly when data are inherently noisy (as human-labelled, animal-based data often are), researchers are faced with the conundrum of investing more time in maximising metrics versus doing the actual research. This raises the question: how much noise can we accept in ML models? Here, we start by describing an under-reported factor that can cause metrics to underestimate model performance. Specifically, ambiguity between categories or mistakes in labelling validation data produces hard ceilings that limit performance metrics. This likely widespread issue means that many models could be performing better than their metrics suggest. Next, we argue and show that imperfect models (e.g. low F1 scores) can still be useable. Using a case study on ML-identified behaviour from vulturine guineafowl accelerometer data, we first propose a simulation framework to evaluate robustness of hypothesis testing using models that make classification errors. Second, we show how to determine the utility of a model by supplementing existing performance metrics with biological validations. This involves applying ML models to unlabelled data and using the models outputs to test hypotheses for which we can anticipate the outcome. Together, we show that effects sizes and expected biological patterns can be detected even when performance metrics are relatively low (e.g., F1: 60-70%). In doing so, we provide a roadmap for validation approaches of ML classification models tailored to research in animal behaviour, and other fields with noisy, biological data. HighlightsEvaluating machine learning (ML) models must go beyond performance metrics Mislabels in validation data leads to underestimation of models performance Underestimated metrics can cause research delays despite models being useful We propose simulations and biological validations to evaluate model performance Models with low standard metrics can still be powerful for hypothesis testing

ecology↗

bioRxiv2021

Machine learning reveals cryptic dialects that guide mate choice in a songbird

Culturally transmitted communication signals - such as human language or bird song - can change over time through a process of cultural drift, and may consequently enhance the separation of populations, potentially leading to reproductive isolation1-4. Local song dialects have been identified in bird species with relatively simple songs where individuals show high cultural conformity5-10. In contrast, the emergence of cultural dialects has been regarded as unlikely11-13 for species with more variable song, such as the zebra finch (Taeniopygia guttata). Instead, it has been proposed that selection for individual recognition and distinctiveness may lead to a complete spread across the space of acoustic and syntactical possibilities11-15. However, another possibility is that analytical limitations have meant that subtle but possibly salient group differences have not yet been discovered in such species. Here we show that machine learning can distinguish the songs from multiple captive zebra finch populations with remarkable precision, and that these cryptic song dialects drive strong assortative mating in this species. We studied mating patterns across three consecutive generations using captive populations that have evolved in isolation for about 100 generations. Cross-fostering eggs within and between these populations and quantifying social interactions of the resulting offspring later in life revealed that mate choice primarily targets cultural traits that are transmitted during a short developmental time window. Detailed social networks showed that females preferentially approached males whose song resembled that of their adolescent peers. Our study shows that birds can be surprisingly sensitive to cultural traits for mating that have hitherto remained cryptic, even in this well-studied species that is used as a model for song-learning13,14,16-28.

animal behavior and cognition↗

bioRxiv2019

Season-specific carry-over of early-life associations in a monogamous bird species

Social relationships can have important fitness consequences. Although there is increasing evidence that social relationships carry over across contexts, few studies have investigated whether relationships formed early in life are carried over to adulthood. For example, juveniles of monogamous species go through a major life-history stage transition--pair formation--during which the pair bond becomes a central unit of the social organization. At present, it remains unclear if pair members retain their early-life relationships after pair formation. We investigated whether same-sex associations formed early in life carry over into adulthood and whether carry-over was dependent on season, in a monogamous species. Moreover, we investigated the role of familiarity, genetic relatedness and aggression on the perseverance of social associations. We studied the social structure before and after pair formation in captive barnacle geese (Branta leucopsis), a highly social, long-lived, monogamous species. We constructed association networks of groups of geese before pair formation, during the subsequent breeding season, and in the following wintering season. Next, we studied how these associations carried over during seasonal changes. We found that early-life associations in females were lost during the breeding season, but resurfaced during the subsequent wintering season. In males, the early-life associations persisted across both seasons. Association persistence was not mediated by genetic relatedness or familiarity. The high level of aggressiveness of males, but not females, in the breeding season suggests that males may have played a key role in shaping both their own social environment and that of their partners. We show that early-life social relationships can be maintained well into later life. Such relationships can be sustained even if they are temporarily disrupted, for example due to reproductive behaviour. Our findings therefore highlight that the early-life social environment can have life-long consequences on individuals social environment.

animal behavior and cognition↗

bioRxiv2019

Spurious inference when comparing networks

Comparing networks is challenging. Social networks can be shaped by many factors. Failing to adequately consider non-social processes, including sampling artefacts, can lead to spurious conclusions about differences in social networks among groups. Here we demonstrate that incorrect application of statistical testing methods when comparing networks can generate very high rates of false positives. We then show that null models, specifically pre-network permutation tests, can control for non-social differences in networks and substantially reduce rates of false positives.

animal behavior and cognition↗

bioRxiv2018

Nepotism masks evidence for reciprocity in cooperation networks

Nepotism and reciprocity are not mutually exclusive explanations for cooperation, because helping decisions can depend on both kinship cues and past reciprocal help. The importance of these two factors can therefore be difficult to disentangle using observational data. We developed a resampling procedure for inferring the statistical power to detect observational evidence of nepotism and reciprocity. We first applied this procedure to simulated datasets resulting from perfect reciprocity, where the probability and duration of helping events from individual A to B equaled that from B to A. We then assessed how the probability of detecting correlational evidence of reciprocity was influenced by (1) the number of helping observations and (2) varying degrees of simultaneous nepotism. Last, we applied the same analysis to empirical data on food sharing in vampire bats and allogrooming in mandrills and Japanese macaques. We show that at smaller sample sizes, the effect of kinship was easier to detect and the relative role of kinship was overestimated compared to the effect of reciprocal help in both simulated and empirical data, even with data simulating perfect reciprocity and imperfect nepotism. We explain the causes and consequences of this difference in power for detecting the roles of kinship versus reciprocal help. To compare the relative importance of genetic and social relationships, we therefore suggest that researchers measure the relative reliability of both coefficients in the model by plotting these coefficients and their detection probability as a function of sampling effort. We provide R scripts to allow others to do this power analysis with their own datasets.

animal behavior and cognition↗

bioRxiv2017

An automated barcode tracking system for behavioural studies in birds

O_LIRecent advances in technology allow researchers to automate the measurement of animal behaviour. These methods have multiple advantages over direct observations and manual data input as they reduce bias related to human perception and fatigue, and deliver more extensive and complete data sets that enhance statistical power. One major challenge that automation can overcome is the observation of many individuals at once, enabling whole-group or whole-population tracking.\nC_LIO_LIWe provide a detailed description for implementing an automated system for tracking birds. Our system uses printed, machine-readable codes mounted on backpacks. This simple, yet robust, tagging system can be used simultaneously on multiple individuals to provide data on bird identity, position and directionality. Further, because our codes and backpacks are printed on paper, they are very lightweight.\nC_LIO_LIWe describe the implementation of this automated system on two flocks of zebra finches. We test different camera options, and describe their advantages and disadvantages. We show that our method is reliable, relatively easy to implement and monitor, and with proper handling, has proved to be safe for the birds over long periods of time. Further, we highlight how using single-board computers to control the frequency and duration of image capture makes this system affordable, flexible, and adaptable to a range of study systems.\nC_LIO_LIThe ability to automate the measurement of individual positions has the potential to significantly increase the power of both observational and experimental studies. The system can capture both detailed interactions (using video recordings) and repeated observations (e.g. once per second for the entire day) of individuals over long timescales (months or potentially years). This approach opens the door to tracking life-long relationships among individuals, while also capturing fine-scale differences in behaviour.\nC_LI

animal behavior and cognition↗

bioRxiv2017

Association Indices For Quantifying Social Relationships: How To Deal With Missing Observations Of Individuals Or Groups

Social network analysis has provided important insight into many population processes in wild animals. Constructing social networks requires quantifying the relationship between each pair of individuals in the population. Researchers often use association indices to convert observations into a measure of propensity for individuals to be seen together. At its simplest, this measure is just the probability of observing both individuals together given that one has been seen (the simple ratio index). However, this probability becomes more challenging to calculate if the detection rate for individuals is imperfect. We first evaluate the performance of existing association indices at estimating true association rates under scenarios where (i) only a proportion of all groups are observed (group location errors), (ii) not all individuals are observed despite being present (individual location errors), and (iii) a combination of the two. Commonly-used methods aimed at dealing with incomplete observations perform poorly because they are based on arbitrary observation probabilities. We then derive complete indices that can be calibrated for the different types of observation probabilities to generate accurate estimates of association rates. These are provided in an R package that readily interfaces with existing routines. We conclude that using calibration data is an important step when constructing animal social networks, and that in their absence, researchers should use a simple estimator and explicitly consider the impact of this on their findings.

animal behavior and cognition↗