Surviving in steep terrain: a lab-to-field assessment of locomotor costs for wild mountain lions (Puma concolor)

Identifying animal behaviours from accelerometers: Improving predictive accuracy of machine learning by refining the variables selected, data frequency, and sample duration

Observing animals in the wild often poses extreme challenges, but animal-borne accelerometers are increasingly revealing unobservable behaviours. Automated machine learning streamlines behaviour identification from the substantial datasets generated during multi-animal, long-term studies; however, the accuracy of such models depends on the qualities of the training data. We examined how data processing influenced the predictive accuracy of random forest (RF) models, leveraging the easily observed domestic cat (Felis catus) as a model organism for terrestrial mammalian behaviours. Nine indoor domestic cats were equipped with collar-mounted tri-axial accelerometers, and behaviours were recorded alongside video footage. From this calibrated data, eight datasets were derived with (i) additional descriptive variables, (ii) altered frequencies of acceleration data (40 Hz vs. a mean over 1 s) and (iii) standardised durations of different behaviours. These training datasets were used to generate RF models that were validated against calibrated cat behaviours before identifying the behaviours of five free-ranging tag-equipped cats. These predictions were compared to those identified manually to validate the accuracy of the RF models for free-ranging animal behaviours. RF models accurately predicted the behaviours of indoor domestic cats (F-measure up to 0.96) with discernible improvements observed with post-data-collection processing. Additional variables, standardised durations of behaviours and higher recording frequencies improved model accuracy. However, prediction accuracy varied with different behaviours, where high-frequency models excelled in identifying fast-paced behaviours (e.g. locomotion), whereas lower-frequency models (1 Hz) more accurately identified slower, aperiodic behaviours such as grooming and feeding, particularly when examining free-ranging cat behaviours. While RF modelling offered a robust means of behaviour identification from accelerometer data, field validations were important to validate model accuracy for free-ranging individuals. Future studies may benefit from employing similar data processing methods that enhance RF behaviour identification accuracy, with extensive advantages for investigations into ecology, welfare and management of wild animals.

Anthropogenic activities and age class mediate carnivore habitat selection in a human-dominated landscape

Human activities increasingly challenge wild animal populations by disrupting ecological connectivity and population persistence. Yet, human-modified habitats can provide resources, resulting in selection of disturbed areas by generalist species. To investigate spatial and temporal responses of a generalist carnivore to human disturbance, we investigated habitat selection and diel activity patterns in caracals (Caracal caracal). We GPS-collared 25 adults and subadults in urban and wildland-dominated subregions in Cape Town, South Africa. Selection responses for landscape variables were dependent on subregion, animal age class, and diel period. Contrary to expectations, caracals did not become more nocturnal in urban areas. Caracals increased their selection for proximity to urban areas as the proportion of urban area increased. Differences in habitat selection between urban and wildland caracals suggest that individuals of this generalist species exhibit high behavioral flexibility in response to anthropogenic disturbances that emerge as a function of habitat context.

Assessing microhabitat, landscape features and intraguild relationships in the occupancy of the enigmatic and threatened Andean tiger cat (Leopardus tigrinus pardinoides) in the cloud forests of northwestern Colombia

Mesocarnivores play a key role in ecosystem dynamics through the regulation of prey populations and are sensitive to environmental changes; thus, they are often considered good model organisms for conservation planning. However, data regarding the factors that influence the habitat use of threatened small wild felids such as the Andean tiger cat (Leopardus tigrinus pardinoides) are scarce. We conducted a two-year survey with 58 camera trap stations to evaluate the determinants of Andean tiger cat habitat use in three protected areas in the Middle Cauca, Colombia. We developed site occupancy models and found that Andean tiger cat habitat use increased with leaf litter depth at intermediate elevations and far from human settlements. Through conditional cooccurrence models, we found that Andean tiger cat habitat use was invariant to the presence of prey or potential intraguild competitors and killers/predators, but its detectability increased when they were present and detected. This suggests that Andean tiger cats may be more likely to be detected in sites with high prey availability. We found that Andean tiger cats preferred sites with deep leaf litter, which is a particular feature of cloud forests that provides suitable conditions for ambush hunting and hiding from intraguild enemies. Our results showed that Andean tiger cats avoided human settlements, which may minimize potential mortality risks in those areas. Moreover, the restricted use of middle elevations by Andean tiger cats suggested that they could be used as a sentinel species to track the effects of climate change since their suitable habitat is likely to be projected upward in elevation. Future conservation actions must be focused on identifying and mitigating human-related threats close to the Andean tiger cat habitat while preserving microhabitat conditions and the existing networks of protected areas.

Combining accelerometry with allometry for estimating daily energy expenditure in joules when in-lab calibration is unavailable

BACKGROUND

All behaviour requires energy, and measuring energy expenditure in standard units (joules) is key to linking behaviour to ecological processes. Animal-borne accelerometers are commonly used to infer proxies of energy expenditure, termed 'dynamic body acceleration' (DBA). However, converting acceleration proxies (m/s2) to standard units (watts) involves costly in-lab respirometry measurements, and there is a lack of viable substitutes for empirical calibration relationships when these are unavailable.

METHODS

We used past allometric work quantifying energy expenditure during resting and locomotion as a function of body mass to calibrate DBA. We used the resulting 'power calibration equation' to estimate daily energy expenditure (DEE) using two models: (1) locomotion data-based linear calibration applied to the waking period, and Kleiber's law applied to the sleeping period (ACTIWAKE), and (2) locomotion and resting data-based linear calibration applied to the 24-h period (ACTIREST24). Since both models require locomotion speed information, we developed an algorithm to estimate speed from accelerometer, gyroscope, and behavioural annotation data. We applied these methods to estimate DEE in free-ranging meerkats (Suricata suricatta), and compared model estimates with published DEE measurements made using doubly labelled water (DLW) on the same meerkat population.

RESULTS

ACTIWAKE's DEE estimates did not differ significantly from DLW (t(19) = - 1.25; P = 0.22), while ACTIREST24's estimates did (t(19) = - 2.38; P = 0.028). Both models underestimated DEE compared to DLW: ACTIWAKE by 14% and ACTIREST by 26%. The inter-individual spread in model estimates of DEE (s.d. 1-2% of mean) was lower than that in DLW (s.d. 33% of mean).

CONCLUSIONS

We found that linear locomotion-based calibration applied to the waking period, and a 'flat' resting metabolic rate applied to the sleeping period can provide realistic joule estimates of DEE in terrestrial mammals. The underestimation and lower spread in model estimates compared to DLW likely arise because the accelerometer only captures movement-related energy expenditure, whereas DLW is an integrated measure. Our study offers new tools to incorporate body mass (through allometry), and changes in behavioural time budgets and intra-behaviour changes in intensity (through DBA) in acceleration-based field assessments of daily energy expenditure.

The role of individual variability on the predictive performance of machine learning applied to large bio-logging datasets

Animal-borne tagging (bio-logging) generates large and complex datasets. In particular, accelerometer tags, which provide information on behaviour and energy expenditure of wild animals, produce high-resolution multi-dimensional data, and can be challenging to analyse. We tested the performance of commonly used artificial intelligence tools on datasets of increasing volume and dimensionality. By collecting bio-logging data across several sampling seasons, datasets are inherently characterized by inter-individual variability. Such information should be considered when predicting behaviour. We integrated both unsupervised and supervised machine learning approaches to predict behaviours in two penguin species. The classified behaviours obtained from the unsupervised approach Expectation Maximisation were used to train the supervised approach Random Forest. We assessed agreement between the approaches, the performance of Random Forest on unknown data and the implications for the calculation of energy expenditure. Consideration of behavioural variability resulted in high agreement (> 80%) in behavioural classifications and minimal differences in energy expenditure estimates. However, some outliers with < 70% of agreement, highlighted how behaviours characterized by signal similarity are confused. We advise the broad bio-logging community, approaching these large datasets, to be cautious when upscaling predictions, as this might lead to less accurate estimates of behaviour and energy expenditure.

Striped hyaena den site selection in Nilgiri Biosphere Reserve, India

The den is a multi-purpose critical space of carnivores and provides a growth conducive refuge which ensures both substances and protection from interspecific predation and harsh climate. Selection of optimal den sites determined by various site-specific factors potentially reduces aversive interspecific interactions and provides cost-effective access to food sources. In this study, we have assessed the factors determining the den site selection by a small population of striped hyaena, Hyaena hyaena in a shared landscape dominated by large carnivores. We assessed den site selection as a function of vegetation patch characteristics, site-specific anthropogenic threats/activities and topographical variables using Bayesian algorithm through field collected binomial data on den use by the species. Our model suggested that hyaenas select rocky refugia surrounded by trees and tall grasses, situated on mountain slopes proximate to a water body. Our study consolidated the importance of undulating terrain in the species ecology and postulated the slope as an ‘energy-expensive’ terrain that refrains frequent movement of other carnivores, in turn providing more affordable denning space for the striped hyaena. This study provides critical information on denning ecology of last remaining major breeding population striped hyaena of southern India.

Asymmetry-induced order in multilayer networks

Symmetries naturally occur in real-world networks and can significantly influence the observed dynamics. For instance, many synchronization patterns result from the underlying network symmetries, and high symmetries are known to increase the stability of synchronization. Yet here we find that general macroscopic features of network solutions such as regularity can be induced by breaking their symmetry of interactions. We demonstrate this effect in an ecological multilayer network where the topological asymmetries occur naturally. These asymmetries rescue the system from chaotic oscillations by establishing stable periodic orbits and equilibria. We call this phenomenon asymmetry-induced order and uncover its mechanism by analyzing both analytically and numerically the absence of dynamics on the system's synchronization manifold. Moreover, the bifurcation scenario describing the route from chaos to order is also disclosed. We demonstrate that this result also holds for generic node dynamics by analyzing coupled paradigmatic Rössler and Lorenz systems.

Do all roads lead to resistance? State road density is the main impediment to gene flow in a flagship species inhabiting a severely fragmented anthropogenic landscape

AIM

Connectivity conservation is ideally based on empirical information on how landscape heterogeneity influences species-specific movement and gene flow. Here, we present the first large-scale evaluation of landscape impacts on genetic connectivity in the European wildcat (Felis silvestris), a flagship and umbrella species for connectivity conservation across Europe.

LOCATION

The study was carried out in the core area of the distributional range of wildcats in Germany, covering about 186,000 km2 of a densely populated and highly fragmented landscape.

METHODS

We used data of 975 wildcats genotyped at 14 microsatellites and an individual-based landscape genetic framework to assess the importance of twelve landscape variables for explaining observed genetic connectivity. For this, we optimized landscape resistance surfaces for all variables and compared their relative impacts using multiple regression on distance matrices and commonality analysis.

RESULTS

Genetic connectivity was best explained by a synergistic combination of six landscape variables and isolation by distance. Of these variables, road density had by far the strongest individual impact followed by synergistic effects of agricultural lands and settlements. Subsequent analyses involving different road types revealed that the strong effect of road density was largely due to state roads, while highways and federal roads had a much smaller, and county roads only a negligible impact.

MAIN CONCLUSIONS

Our results highlight that landscape-wide genetic connectivity in wildcats across Germany is strongly shaped by the density of roads and in particular state roads, with higher densities providing larger resistance to successful dispersal. These findings have important implications for conservation planning, as measures to mitigate fragmentation effects of roads (e.g., over- or underpasses) often focus on large, federally managed transportation infrastructures. While these major roads exert local barrier effects, other road types can be more influential on overall connectivity, as they are more abundant and more widespread across the landscape.

An evaluation of machine learning classifiers for next-generation, continuous-ethogram smart trackers

BACKGROUND

Our understanding of movement patterns and behaviours of wildlife has advanced greatly through the use of improved tracking technologies, including application of accelerometry (ACC) across a wide range of taxa. However, most ACC studies either use intermittent sampling that hinders continuity or continuous data logging relying on tracker retrieval for data downloading which is not applicable for long term study. To allow long-term, fine-scale behavioural research, we evaluated a range of machine learning methods for their suitability for continuous on-board classification of ACC data into behaviour categories prior to data transmission.

METHODS

We tested six supervised machine learning methods, including linear discriminant analysis (LDA), decision tree (DT), support vector machine (SVM), artificial neural network (ANN), random forest (RF) and extreme gradient boosting (XGBoost) to classify behaviour using ACC data from three bird species (white stork Ciconia ciconia, griffon vulture Gyps fulvus and common crane Grus grus) and two mammals (dairy cow Bos taurus and roe deer Capreolus capreolus).

RESULTS

Using a range of quality criteria, SVM, ANN, RF and XGBoost performed well in determining behaviour from ACC data and their good performance appeared little affected when greatly reducing the number of input features for model training. On-board runtime and storage-requirement tests showed that notably ANN, RF and XGBoost would make suitable on-board classifiers.

CONCLUSIONS

Our identification of using feature reduction in combination with ANN, RF and XGBoost as suitable methods for on-board behavioural classification of continuous ACC data has considerable potential to benefit movement ecology and behavioural research, wildlife conservation and livestock husbandry.

Quantifying energetic costs and defining energy landscapes experienced by grizzly bears

Animal movements are major determinants of energy expenditure and ultimately the cost–benefit of landscape use. Thus, we sought to understand those costs and how grizzly bears (Ursus arctos) move in mountainous landscapes. We trained captive grizzly bears to walk on a horizontal treadmill and up and down 10% and 20% slopes. The cost of moving upslope increased linearly with speed and slope angle, and this was more costly than moving horizontally. The cost of downslope travel at slower speeds was greater than the cost of traveling horizontally but appeared to decrease at higher speeds. The most efficient walking speed that minimized cost per unit distance was 1.19±0.11 m s−1. However, grizzly bears fitted with GPS collars in the Greater Yellowstone Ecosystem moved at an average velocity of 0.61±0.28 m s−1 and preferred to travel on near-horizontal slopes at twice their occurrence. When traveling uphill or downhill, grizzly bears chose paths across all slopes that were ∼54% less steep and costly than the maximum available slope. The net costs (J kg−1 m−1) of moving horizontally and uphill were the same for grizzly bears, humans and digitigrade carnivores, but those costs were 46% higher than movement costs for ungulates. These movement costs and characteristics of landscape use determined using captive and wild grizzly bears were used to understand the strategies that grizzly bears use for preying on large ungulates and the similarities in travel between people and grizzly bears that might affect the risk of encountering each other on shared landscapes.

Energetics and fear of humans constrain the spatial ecology of pumas

Energetic demands and fear of predators are considered primary factors shaping animal behavior, and both are likely drivers of movement decisions that ultimately determine the spatial ecology of wildlife. Yet energetic constraints on movement imposed by the physical landscape have only been considered separately from those imposed by risk avoidance, limiting our understanding of how short-term movement decisions scale up to affect long-term space use. Here, we integrate the costs of both physical terrain and predation risk into a common currency, energy, and then quantify their effects on the short-term movement and long-term spatial ecology of a large carnivore living in a human-dominated landscape. Using high-resolution GPS and accelerometer data from collared pumas (Puma concolor), we calculated the short-term (i.e., 5-min) energetic costs of navigating both rugged physical terrain and a landscape of risk from humans (major sources of both mortality and fear for our study population). Both the physical and risk landscapes affected puma short-term movement costs, with risk having a relatively greater impact by inducing high-energy but low-efficiency movement behavior. The cumulative effects of short-term movement costs led to reductions of 29% to 68% in daily travel distances and total home range area. For male pumas, long-term patterns of space use were predominantly driven by the energetic costs of human-induced risk. This work demonstrates that, along with physical terrain, predation risk plays a primary role in shaping an animal's "energy landscape" and suggests that fear of humans may be a major factor affecting wildlife movements worldwide.