Generalized nonlinearity in animal ecology: Research, review, and recommendations

Generalized linear models (GLMs) are an integral tool in ecology. Like general linear models, GLMs assume linearity, which entails a linear relationship between independent and dependent variables. However, because this assumption acts on the link rather than the natural scale in GLMs, it is more easily overlooked. We reviewed recent ecological literature to quantify the use of linearity. We then used two case studies to confront the linearity assumption via two GLMs fit to empirical data. In the first case study we compared GLMs to generalized additive models (GAMs) fit to mammal relative abundance data. In the second case study we tested for linearity in occupancy models using passerine point-count data. We reviewed 162 studies published in the last 5 years in five leading ecology journals and found less than 15% reported testing for linearity. These studies used transformations and GAMs more often than they reported a linearity test. In the first case study, GAMs strongly out-performed GLMs as measured by AIC in modeling relative abundance, and GAMs helped uncover nonlinear responses of carnivore species to landscape development. In the second case study, 14% of species-specific models failed a formal statistical test for linearity. We also found that differences between linear and nonlinear (i.e., those with a transformed independent variable) model predictions were similar for some species but not for others, with implications for inference and conservation decision-making. Our review suggests that reporting tests for linearity are rare in recent studies employing GLMs. Our case studies show how formally comparing models that allow for nonlinear relationships between the dependent and independent variables has the potential to impact inference, generate new hypotheses, and alter conservation implications. We conclude by suggesting that ecological studies report tests for linearity and use formal methods to address linearity assumption violations in GLMs.

Spatial patterns of reproduction suggest marginal habitat limits continued range expansion of black bears at a forest-desert ecotone

Investigating spatial patterns of animal occupancy and reproduction in peripheral populations can provide insight into factors that form species range boundaries. Following historical extirpation, American black bears (Ursus americanus) recolonized the western Great Basin in Nevada from the Sierra Nevada during the late 1900s. This range expansion, however, has not continued further into the Great Basin despite the presence of additional habitat. We aimed to quantify whether reduced reproduction toward the range edge contributes to this range boundary. We analyzed black bear detections from 100 camera traps deployed across black bear distribution in western Nevada using a multistate occupancy model that quantified the probability of occupancy and reproduction (i.e., female bears with cubs occupancy) in relation to changes in habitat type and habitat amount toward the range boundary. We detected a strong effect of habitat amount and habitat type on the probability of black bear occupancy and reproduction. At similar levels of landscape-scale habitat amount (e.g., 50%), estimated probability of occupancy for adult bears in piñon-juniper woodlands near the range boundary was 0.39, compared to ~1.0 in Sierra Nevada mixed-conifer forest (i.e., core habitat). Furthermore, estimated probability of cub occupancy, conditional on adult bear occupancy, in landscapes with 50% habitat was 0.32 in Great Basin piñon-juniper woodlands, compared to 0.92 in Sierra Nevada mixed-conifer forest. Black bear range in the western Great Basin conforms to the center-periphery hypothesis, with piñon-juniper woodland at the range edge supporting ecologically marginal habitat for the species compared to habitat in the Sierra Nevada. Further geographic expansion of black bears in the Great Basin may be limited by lower occupancy of reproducing females in piñon-juniper woodland. Center-periphery range dynamics may be common in large carnivore species, as their dispersal ability allows them to colonize low-quality habitat near range edges.

Estimating the density of small mammals using the selfie trap is an effective camera trapping method

Camera trapping to study wildlife allows for data collection, without the need to capture animals. Traditionally, camera traps have been used to target larger terrestrial mammal species, though recently novel methods and adjustments in procedures have meant camera traps can be used to study small mammals. The selfie trap (a camera trapping method) may present robust sampling and ecological study of small mammals. This study aimed to evaluate the selfie trap method in terms of its ability to detect species and estimate population density. To address this aim, standard small mammal live trapping was undertaken, immediately followed by camera trapping using the selfie trap. Both methods were set to target the arboreal sugar glider (Petaurus breviceps) and semi-arboreal brown antechinus (Antechinus stuartii). The more ground-dwelling bush rat (Rattus fuscipes) was also live trapped and recorded on camera. Across four survey areas, the probability of detection for each of the three species was higher for selfie traps than for live trapping. Spatially explicit capture-recapture models showed that selfie traps were superior at estimating density for brown antechinus and sugar gliders, when compared to simulated live trapping data. Hit rates (number of videos per various time intervals) were correlated with abundance. When correlating various hit rate intervals with abundance, the use of 10-min hit rate was best for predicting sugar glider abundance (R2 = 0.94). The abundance of brown antechinus was estimated from selfie traps using a 24-h hit rate as a predictor (R2 = 0.85). For sugar gliders, the selfie trap can replace live trapping as individuals can be identified through their unique facial stripes and natural ear scars, and thus used in capture-recapture analysis. This method may be useful for monitoring the abundance of other small mammal species that can also be individually recognized from photographs.

The Integral Nature of Encounter Rate in Predicting Livestock Depredation Risk

Carnivore depredation of livestock is one of the primary drivers of human-carnivore conflict globally, threatening the well-being of livestock owners, and fueling large carnivore population declines. Interventions designed to reduce carnivore depredation typically center around predictions of depredation risk. However, these spatial risk models tend to be informed by data depicting the number of livestock attacked by carnivores. Importantly, such models omit key stages in the predation sequence which are required to predict predation risk, or in this case depredation risk. Applying the classic predation risk model defined by Lima and Dill demonstrates that depredation risk is dependent upon quantifying the rates at which carnivores encounter livestock before attacking. However, encounter rate is challenging to estimate, necessitating novel data collection systems. We developed and applied such a system to quantify carnivore-livestock encounters at livestock corrals (i.e., bomas) across a 9-month period in Central Kenya. Concurrently, we monitored the number of livestock attacked by carnivores at these bomas. We calculated carnivore-livestock encounter rates, attack rates, and depredation risk at the boma. We detected 1,383 instances in which carnivores encountered livestock at the bomas. However, we only recorded seven attacks. We found that the encounter rate and attack rate for spotted hyenas were almost six and three times higher than that for any other species, respectively. Consequently, spotted hyenas posed the greatest depredation risk for livestock at the boma. We argue that better understanding of carnivore-livestock encounter rates is necessary for effective prediction and mitigation of carnivore depredation of livestock.

Phylogeny is a stronger predictor of activity than allometry in an African mammal community

In promoting coexistence, sympatric species often partition shared resources along spatio-temporal domains. Similarly sized and phylogenetically close species, for instance, partition the times of day in which they are active to limit interference competition. Given that variation in species body mass has evolutionary underpinnings, species activity levels (time spent active in a 24-h daily cycle) within animal communities might be structured by phylogeny. However, few studies have tested this hypothesis across animal communities, and none among medium-sized to large mammals. We quantified the relative contributions of phylogeny and body mass in predicting activity levels in a community of 22 sympatric mammal species in Murchison Falls National Park, Uganda. We show that phylogeny is a stronger predictor of species activity levels than body mass. Our findings provide empirical evidence for the phylogenetic structuring of mammal activity in diverse communities. More broadly, our results suggest that evolutionary relationships mask allometry in predicting species traits in diverse animal communities.

Hunter-Gatherers in context: Mammal community composition in a northern Tanzania landscape used by Hadza foragers and Datoga pastoralists

In many regions of sub Saharan Africa large mammals occur in human-dominated areas, yet their community composition and abundance have rarely been described in areas occupied by traditional hunter-gatherers and pastoralists. Surveys of mammal populations in such areas provide important measures of biodiversity and provide ecological context for understanding hunting practices. Using a sampling grid centered on a Hadza hunter-gatherer camp and covering 36 km² of semi-arid savannah in northern Tanzania, we assessed mammals using camera traps (n = 19 stations) for c. 5 months (2,182 trap nights). In the study area (Tli’ika in the Hadza language), we recorded 36 wild mammal species. Rarefaction curves suggest that sampling effort was sufficient to capture mammal species richness, yet some species known to occur at low densities in the wider area (e.g. African lions, wildebeest) were not detected. Relative abundance indices of wildlife species varied by c. three orders of magnitude, from a mean of 0.04 (African wild dog) to 20.34 capture events per 100 trap-nights (Kirk’s dik dik). To contextualize the relative abundance of wildlife in the study area, we compared our study’s data to comparable camera trap data collected in a fully protected area of northern Tanzania with similar rainfall (Lake Manyara National Park). Raw data and negative binomial regression analyses show that wild herbivores and wild carnivores were generally detected in the national park at higher rates than in the Hadza-occupied region. Livestock were notably absent from the national park, but were detected at high levels in Tli’ika, and cattle was the second most frequently detected species in the Hadza-used area. We discuss how these data inform current conservation efforts, studies of Hadza hunting, and models of hunter-gatherer foraging ecology and diet.

An apex carnivore's life history mediates a predator cascade

Apex predators can shape communities via cascading top-down effects, but the degree to which such effects depend on predator life history traits is largely unknown. Within carnivore guilds, complex hierarchies of dominance facilitate coexistence, whereby subordinate species avoid dominant counterparts by partitioning space, time, or both. We investigated whether a major life history trait (hibernation) in an apex carnivore (black bears Ursus americanus) mediated its top-down effects on the spatio-temporal dynamics of three sympatric mesocarnivore species (coyotes Canis latrans, bobcats Lynx rufus, and gray foxes Urocyon cinereoargenteus) across a 15,000 km² landscape in the western USA. We compared top-down, bottom-up, and environmental effects on these mesocarnivores using an integrated modeling approach. Black bears exerted top-down effects that varied as a function of hibernation and were stronger than bottom-up or environmental impacts. High black bear activity in summer and fall appeared to buffer the most subordinate mesocarnivore (gray foxes) from competition with dominant mesocarnivores (coyotes and bobcats), which were in turn released by black bear hibernation in winter and early spring. The mesocarnivore responses occurred in space (i.e., altered occupancy and site visitation intensity) rather than time (i.e., diel activity patterns unaffected). These results suggest that the spatio-temporal dynamics of mesocarnivores in this system were principally shaped by a spatial predator cascade of interference competition mediated by black bear hibernation. Thus, certain life history traits of apex predators might facilitate coexistence among competing species over broad time scales, with complex implications for lower trophic levels.

What are we missing? How the delay-period setting on camera traps affects mesopredator detection

Trigger-delays are often set on camera traps to save labour costs involved with servicing camera traps and reviewing images. However, the effects that delays of various length could have on data quantity and quality are unmeasured. Here, we aimed to assess how varying trigger-delays (5, 10, 30 and 60min) or using an ‘independent event’ classification (i.e. a series of images taken less than 5min apart on the same camera trap) affects detection rates and the number of individuals per trigger for feral cats and European red foxes. Using real camera trap images, we simulated trigger-delays of between 5min and 60min and compared with the independent events. Trigger-delays caused inaccuracies in detection frequencies of 3.6–22.0% for feral cats and 3.1–24.0% for foxes. Further, 68% of independent events in which two individual foxes were present were reduced to a single fox trigger when implementing a 5-min trigger-delay (n=65). Using trigger-delays likely reduces accuracy and reliability of data generated for wildlife monitoring programs and could affect the types of observations and analyses able to be made from imagery so obtained.

The effect of camera-trap viewshed obstruction on wildlife detection: implications for inference

Abstract ContextCamera traps are one of the most popular tools used to study wildlife worldwide. Numerous recent studies have evaluated the efficiency and effectiveness of camera traps as a research tool. Nonetheless, important aspects of camera-trap methodology remain in need of critical investigation. One such issue relates to camera-trap viewshed visibility, which is often compromised in the field by physical obstructions (e.g. trees) or topography (e.g. steep slopes). The loss of visibility due to these obstructions could affect wildlife detection rates, with associated implications for study inference and management application. AimsWe aimed to determine the effect of camera-trap viewshed obstruction on wildlife detection rates for a suite of eight North American species that vary in terms of ecology, commonness and body size. MethodsWe deployed camera traps at 204 sites throughout an extensive semi-urban park system in Cleveland, Ohio, USA, from June to September 2016. At each site, we quantified camera-trap viewshed obstruction by using a cover-board design. We then modelled the effects of obstruction on wildlife detection rates for the eight focal species. Key resultsWe found that detection rates significantly decreased with an increasing viewshed obstruction for five of the eight species, including both larger and smaller mammal species (white-tailed deer, Odocoileus virginianus, and squirrels, Sciurus sp., respectively). The number of detections per week per camera decreased two- to three-fold as visibility at a camera site decreased from completely free of obstruction to mostly obstructed. ConclusionsThese results imply that wildlife detection rates are influenced by site-level viewshed obstruction for a variety of species, and sometimes considerably so. ImplicationsResearchers using camera traps should address the potential for this effect to ensure robust inference from wildlife image data. Accounting for viewshed obstruction is critical when interpreting detection rates as indices of abundance or habitat use because variation in detection rate could be an artefact of site-level viewshed obstruction rather than due to underlying ecological processes.