Camera Trapping for Wildlife Research

‘We need to recognize at the outset that any attempt to define behavior is bound to be invalidated in time. The more we know about behavior, the more our definition changes. In a real sense, the whole aim of a science of behavior is to define behavior.’ (Baum 2013)

In this chapter we introduce a range of different approaches and technologies used to study animal behaviour and discuss their advantages and disadvantages compared to camera trapping. We then present different applications of camera trapping in behavioural studies and provide insights into what needs to be considered when choosing camera trap sites for a variety of study aims. Finally, we present three examples as case studies: the first on marking behaviour in the Eurasian lynx (Lynx lynx), Box 8.2; the second on tree rubbing behaviour in the brown bear (Ursus arctos), Box 8.3; and the third is an example of temporal interactions between species, such as competing species, or predators and preys – specifically, we look at activity pattern overlap between the Eurasian lynx and its two main preys, i.e. roe deer (Capreolus capreolus) and chamois (Rupicapra rupicapra) in Switzerland (see section 8.6.2).

Direct observations of animals in the field has been the predominant technique in ethological studies prior to the arrival of bio-logging and remote sensors such as camera traps, and despite many limitations it is still a common approach. The irreplaceable strengths of direct observational studies are detailed and complex data collection that would be difficult to achieve when researchers are not physically present with the animal. There are also positive impacts associated with the presence of researchers, e.g. incentives from ecotourism (Knight 2009), or a decrease in wildlife poaching around highly used research stations and field sites (Campbell et al. 2011). However, direct observation carries substantial weaknesses that can lead to spurious conclusions, such as behavioural or activity changes due to presence of the observer, unequal observability of different categories of individuals (e.g. Boyer-Ontl and Pruetz 2014), observer-dependent bias in the data, and a limited number of focal individuals from which inferences can be drawn.

Bio-logging technology seeks to overcome these problems by enabling the remote measurement of continuous data streams on earth and in the air for free-ranging, undisturbed subjects without requiring continuous human support (Cooke et al. 2004; Ropert-Coudert and Wilson 2005). These technologies provide the opportunity to focus on animal behaviour and physiology on a variety of scales (e.g. temporal, spatial or biological organisation - from organ system, to individual to community interactions; Cook et al. 2004) and enable the linking of the behaviour and physiology of free-ranging animals in their own environment by collecting a wealth of data in the field including location information from telemetry, motion patterns from accelerometers, temperature and depth for aquatic animals, estimates of proximity to other animals, still images and video, and physiological data (e.g. metabolic rate) from a range of sensors (Moll et al. 2007). As all animals are handled when transmitters are attached, most individual specific parameters, such as individual identity, sex, age and reproductive status, are also gathered. However, as with all techniques, a suite of challenges and limitations exists that must be balanced against the positive aspects. Inferences based on bio-logging have inherent limitations due to biases associated with trapping and handling, variation among individuals, and potentially high costs which in some cases could lead to small sample sizes. Battery size and longevity continue to limit research on small organisms and to restrict long-term monitoring. Because data can be collected in real time there is the danger of collecting large data sets that are difficult to manage. Interpreting patterns of large data sets can be extremely difficult and challenging and needs to be coupled with other techniques, including detailed visual (Löttker et al. 2009) and/or video observation or by combining multiple biosensors. Data interpretation is scale dependent, for example in GPS data the ability to statistically discriminate different movement modes will depend on the sampling interval (Hebblewhite and Haydon 2010), and the monitoring frequency can also impact estimates of environmental effects on site selection (Carvalho et al. 2015). Time series, the repeated sampling of data of the same individual, are non-independent and could require complex statistical techniques but the current approaches are poorly suited to these types of data (Cook et al. 2004).

Recently, animal-borne video and environmental data collection systems (AVEDs), enabling investigators to ‘see’ what the animals themselves see in the field, offer new research potential in the field of bio-logging especially when fine-scale and continuous data for individuals are needed (Moll et al. 2007). These new technologies are most appropriate for elusive species living in inaccessible environments (e.g. deep-diving marine species) and for fine-scale assessments of animal behaviour such as food selection, reproduction, social behaviour, species interactions and disease transmission, and might also help to mitigate human–wildlife conflicts (e.g. reducing animal–vehicle collisions through studying road-crossing behaviour) and provide insight into habitat use if sampling schemes are well designed (Moll et al. 2007). AVEDs have the greatest potential for explaining ecological mechanisms when video is integrated with other animal-borne sensors, because data can subsequently be interpreted within the context of animal activity (Moll et al. 2007). However, unless the interval between images is small, it can be difficult to interpret animal behaviour in these snapshots of activity because important details are missing. Like direct observational studies and other bio-logging technologies, still imaging and AVEDs suffer from small sample sizes and thus may not permit the population-wide inferences that researchers have come to expect from camera traps (Nichols et al. 2011). An additional shortcoming of AVEDs is a greater potential to affect the natural behaviour of animals than other bio-logging technologies (e.g. telemetry) because these devices are larger, heavier and cannot be implanted subcutaneously. Furthermore these systems are prone to damage and lens obstruction, especially in terrestrial habitats.

Camera trapping is a valid alternative to bio-logging as it essentially allows the placing of remote sensors in the environment rather than on animals. Camera trapping has indeed a long history in behaviour studies (see Bridges and Noss 2011 for a review). It combines many of the advantages of the techniques described above while offering a number of improvements. In the first place it is a highly non-invasive tool (see also Chapter 1), and although the camera trap itself, the flash and the noise in the infra-ultrasonic range produced may disturb the animal (Meek et al. 2014a), and hence potentially modify its behaviour, this disturbance is likely to be of much lower impact than that produced by observers. In addition, camera trapping can be applied over relatively large areas, including secluded and inaccessible areas, and provide data for a larger number of individuals, and a wider range of species relative to the earlier-mentioned techniques that require trapping and tagging. Importantly, moreover, newly developed camera traps with video mode can, unlike single photographs, record behaviour. Finally, photographs and/or videos from camera traps, similar to bio-logging data, can be archived and shared with the scientific community, allowing for a common interpretation of the observational data and future reanalysis as new scientific knowledge is acquired.

In recognition of the non-overlapping advantages of the different techniques, recently researchers have started to combine camera trapping with telemetry and sometimes also with direct visual observations. For example, Leuchtenberger et al. (2014) recorded the activity pattern of giant otters (Pteronura brasiliensis) using telemetry combined with visual observation during the day. The animals were radiotracked from boats and whenever visual observation was possible the predominant behaviour of the majority of the group was observed and recorded by focal group sampling. Camera traps positioned at entrances of active dens and on latrines provided information about the patterns of den use and scent marking over 24 h periods. Suselbeek et al. (2014) used an unique combination of an automated radiotelemetry system (ARTS; Kays et al. 2011), manual radiotelemetry and camera trapping to test whether activity at high-risk times declined with food availability as predicted by risk allocation hypothesis in a neotropical forest rodent, the Central American agouti (Dasyprocta punctate) in relation to the temporal pattern of predation risk by its principal predator, the ocelot (Leopardus pardalis). The measure of daily activity pattern for agoutis and ocelot at the population level was inferred from camera trapping. To determine which periods had elevated predation risk for agoutis the ratio of ocelot to agouti activity over time was quantified. ARTS, which enabled activity to be tracked continuously and recorded the exact time of death for radiocollared agouti, was used to determine how much of agouti mortality was due to ocelots and fell during the daily period of high predation risk. The ARTS included a wireless network making these data, including mortality events, available in real time to the investigator via a web-accessible database. As a mortality event was detected, the carcass was located to check for bite marks and a camera was deployed nearby to record any predators returning to their kills. Activity patterns between agoutis that lived in areas of contrasting food abundances were compared to determine how food abundance affected individual agouti activity patterns.

Camera trapping has seen a vast field of applications in behavioural studies, reviewed by Bridges and Noss (2011). These include studies of activity patterns (e.g. Di Bitetti et al. 2006; Story et al. 2014); niche partitioning (spatial and temporal separation or avoidance; e.g. Ridout and Linkie 2009; Di Bitetti et al. 2010; Monterroso et al. 2014; Rowcliffe et al. 2014); habitat and corridor usage (e.g. Gužvica et al. 2014); refugia (e.g. study on badger setts; Mori et al. 2015) and reproduction (e.g. quantify behaviour of nesting birds; Enderson et al. 1972); feeding ecology (e.g. chickadees, tits and owls at nest boxes; Royama 1970; Juillard 1987); foraging (e.g. DeVault et al. 2004; Bauer et al. 2005; Vernes et al. 2014; Zhi-Pang Huang et al. 2014); predation events mainly focusing on predation of bird nests (Cutler and Swann 1999); species ranging and grouping behaviours (e.g. by using camera traps Boyer-Ontl and Pruetz (2014) were able to observe chimpanzees grooming, playing, resting, feeding and performing agonistic displays); behavioural interactions among and between species, as well as between predator and prey (e.g. Díaz et al. 2005; Foster et al. 2013; Kuijper et al. 2014); social structure (Séquin et al. 2003; Vogt et al. 2014) and group compositions (e.g. Boyer-Ontl and Pruetz 2014); and marking behaviour (e.g. Vogt et al. 2014; Tattoni et al. 2015).

Behavioural studies have capitalised on the ability of camera traps to monitor fixed (i.e. predetermined) locations where a specific behaviour or resource use, as well as interactions among and between species, occur. A non-exhaustive list of fixed locations include: birds’ nests to study predation rates and to identify predators of eggs and fledglings (e.g. Major and Gowing 1994); entry or exit points of animal dens, setts or burrows (e.g. Mori et al. 2015); foraging areas including mast-fruiting tree resources (e.g. Miura et al. 1997); truffle patches (e.g. Vernes et al. 2014) to study seed dispersal, mineral licking or geophagy (i.e. the deliberate consumption of earth material; Galvis et al. 2014); kill sites or carcasses to identify carnivores feeding at the kill or stock raiders (e.g. Zimmermann et al. 2011) or the set of scavengers consuming the carcass (e.g. Bauer et al. 2005; Zhi-Pang Huang et al. 2014); and water resources such as water holes (e.g. Bleich et al. 1997).

Camera traps have also been used to monitor specific features such as: wood piles to study the marking behaviour of the Eurasian lynx (Vogt et al. 2014; see Box 8.2) or trees and poles to study the rubbing behaviour of brown bears (e.g. Tattoni et al. 2015; see Box 8.3); and green bridges or any other natural or artificial structure that facilitates the movement of animals (e.g. Gužvica et al. 2014; Taylor and Goldingay 2014).

The sampling design should always be related to the research question and focal species. Despite an accurate description of the sampling design being central to the interpretation of the results, there are still studies that do not report the methodological details in sufficient detail (guidelines for reporting on camera trapping studies can be found in Meek et al. 2014b). Furthermore, to our knowledge, no research has directly tested the influence of sampling design on behavioural studies. Therefore, we can only provide here some basic recommendations. When studying specific behaviours at fixed, predefined sites, investigators should try to sample a representative number of individuals of the studied population at a representative number of sites, to avoid individuals showing non-typical behaviours being over-represented in the sample. If the study focuses on behaviour occurring at different categories of sites, a representative number of sites of each category should be sampled. Researchers should bear in mind that specific sampling designs (i.e. camera traps placed only at predefined locations) are adopted to meet specific objectives and as such they allow conclusions to be drawn only on the behaviour of the focal species at those sites. Because species do not use fixed locations such as clustered resources to the same extent across the daily cycle, this sampling design would certainly need some adjustment for activity pattern and temporal species interaction studies (see section 8.6.2 below) as it could lead to bias in activity level estimation. The central assumption regarding sampling design for activity pattern measurements is that camera trap sites are placed randomly with respect to diel patterns of movement (Rowcliffe et al. 2014). In practice, this means that activity pattern estimation requires that animals are not attracted to the camera trap site by means of lures or baits (as this is not generally recommended for occupancy and abundance estimation: see Chapters 6 and 7), as this could introduce non-random diel patterns of space use that would prevent accurate estimation of activity level. Most camera trapping studies place camera traps on trails to maximise trap rate. This strategy would be valid for activity level estimation as long as animals use trails to the same extent across the daily cycle (they could, for example, go off trail selectively during peak foraging hours). This potential bias can be tested by comparing trap rate patterns on and off trail. According to Rowcliffe et al. (2014), it will probably be safer to use a random camera trap site placement strategy in most cases, although this strategy may still not work in cases where some important habitat used by the target species is not suitable for camera trapping. For example, for semi-arboreal species, such as the pine marten Martes martes, it will not be practical to camera trap representatively in the tree canopy as well as on ground, which will lead to bias in activity level estimation.

Finally, behavioural studies should adopt sampling designs and analytical frameworks that take into account imperfect detection (i.e. detectability; see Chapters 6 and 7) instead of using detection rates as the metric of choice. Methodologies incorporating imperfect detection into modelling allow, for example, investigation of species co-occurrence patterns and can address questions about the importance of interspecific interactions such as competition and predator–prey relationships as potential determinants of community structure (MacKenzie et al. 2004).

Data from camera traps that record the time of the day at which photographs are taken are used widely to study daily activity patterns (e.g. Bridges et al. 2004; Di Bitteti et al. 2006; Gerber et al. 2012; Kamler et al. 2012; Leuchtenberger et al. 2014). Camera trap pictures are commonly grouped into regular (e.g. 1 h, 2 h) discrete time categories (e.g. Jácomo et al. 2004). To construct frequency histograms, two parameters are required, the bin width (e.g. 1 h) and origin (e.g. 00:00 h). Both parameters can have dramatic effects on the shape of the resulting histogram of the observed frequency. Another disadvantage is that the histogram estimators are usually not smooth, displaying bumps that may have been observed only due to noise. Therefore it is preferable to present kernel density estimates which require only one parameter, the window width, and will therefore always be a more robust estimate of the underlying probability density function (see Wand and Jones 2005). The independent detection records for each target species are regarded as a random sample from the underlying circular continuous temporal distribution that describes the probability of a photograph being taken within any particular interval of the day (Ridout and Linkie 2009). The circular probability density function of this distribution is estimated non-parametrically using kernel density (Ridout and Linkie 2009). More details about statistical developments, and in particular the kernel density estimators, can be found in Ridout and Linkie (2009) and Rowcliffe et al. (2014).

The coexistence of similar species is difficult to explain if two species share very similar ecological niches, as competitive exclusion principles predict the extinction of the inferior competitor (Hutchinson 1978; Soberon 2007). Alternatively, competition can drive niche differentiation by which competing species pursue dissimilar patterns of resource use. Therefore coexistence is acquired through the segregation of ecological niches (Hutchinson 1978). How species use time and distribute their activity within the diel cycle is an important niche dimension, although it is often being regarded as the least important of the three main niche axes (i.e. spatial, temporal and resource exploitation). Indeed species may reduce intraguild competition and predation risk, and thus increase niche segregation, by minimising temporal overlap with similar species or predators. However, the activity pattern of a species along the diel cycle is not only regulated by competition and predation risk. It is also internally regulated by each species’ endogenous clock (Kronfeld-Schor and Dayan 2003) and by external abiotic factors. Temporal niche segregation by ecologically similar carnivores has been demonstrated in diverse systems (Chen et al. 2009; Di Bitteti et al. 2009; Hayward and Slotow 2009; Lucherini et al. 2009; Monterroso et al. 2014). More contrasting temporal patterns were observed between predators and their main preys. While some studies have revealed marked temporal niche overlap (e.g. Núñez et al. 2000; Linkie and Ridout 2011; Ramesh et al. 2012; Foster et al. 2013; Ross et al. 2013), others have provided evidence of prey animals concentrating their activity at times of relatively low predation risk (e.g. Diaz et al. 2005; Ross et al. 2013; Suslebeek et al. 2014). Predator–prey avoidance may be enhanced in areas with higher food abundance for the prey (Sulsebeek et al. 2014).

Here we review the definition and additional assumptions, besides the sampling design considerations already outlined in section 8.4, presented in Rowcliffe et al. (2014), which are the fundamentals for all studies of the activity patterns and temporal interactions of species (see section 8.6.2 for a case study). Camera traps typically detect animals only during the animals’ routine daily movements, i.e. when they are outside their refuges (e.g. shelters, nests or sleeping sites). Following Rowcliffe et al. (2014), animals are defined as active whenever they move out of these locations, which cannot be observed by camera traps. While this definition covers the fundamental characteristic of activity, i.e. a more costly behavioural state than the rest, it differs from the finer categories of behaviour commonly used by ethologists (i.e. foraging, vigilant, sleeping or grooming) as these could potentially take place either within or outside refuges. By assuming that activity level is the only determinant of the rate at which animals are detected by camera traps, the trap rate at a given time of the day will be proportional to the level of activity in the population at that time, and the total amount of activity will be proportional to the area under the trap rate curve. Another assumption central to the method is that all animals are active when the camera trap rate reaches its maximum in the daily cycle (see Rowcliffe et al. 2014 for the detailed reasoning behind this). Given the paucity of evidence currently available, Rowcliffe et al. (2014) caution a stronger research focus on such synchrony to demonstrate the validity of this assumption. As activity peaks are not always synchronised, the method clearly needs to be applied cautiously with this in mind. However, besides the proportion of the population being active, several additional factors could potentially affect trap rate. These include animal speed while active, camera detection zone size and animal density (Rowcliffe et al. 2008). Constant density and population closure over the daily cycle can reasonably be assumed if camera trap sites are placed randomly with respect to diel patterns of movement (see section 8.4). While Rowcliffe et al. (2014) did not find evidence for significant diel variation in animal speed among 12 Panamanian forest species, the camera detection radius was 21% higher during the day than during the night. Based on these findings, they developed a method that can correct for bias due to factors other than activity influencing the diel variation in trap rate (R package activity; Rowcliffe et al. 2014).

Many different measures of overlap have been suggested for quantifying the affinity of overlap of two probability density functions (see Ridout and Linkie 2009 for a review). Ridout and Linkie (2009) used the coefficient of overlapping proposed by Weitzman (1970) for pairwise comparison of activity patterns. The coefficient of overlap Δ ranges from 0 (no overlap) to 1 (complete overlap) and is obtained taking the minimum of the density functions of the two cycles being compared at each time point. Non-parametric estimation of Δ was studied in more detail by Schmid and Schmidt (2006), who noted several equivalent mathematical expressions for Δ which lead to five different estimators. However, for circular data, the first two are equivalent (

₁,

₂) and the third (

₃) was excluded because it is not invariant to the choice of origin (Ridout and Linkie 2009). Therefore three distinct estimators need to be considered:

₁,

₄ and

₅ (see section 8.6.2). A smoothed bootstrap should be used to estimate the precision of the coefficient of overlapping. At least 1000 resamples (preferably 10,000) should be done. The evaluation of Δ values and consequent definition of high or low overlap between two distinct activity patterns is largely subjective. Some authors (e.g. Monterroso et al. 2014) ranked the activity overlap values resulting from overall pairwise comparisons into low, moderate or high based on percentiles (e.g. low = Δ ≤ 50th percentile of the samples, moderate = 50th percentile < Δ ≤ 75th percentile and high = Δ > 75th percentile). The significance of pairwise comparisons, either between relative activity levels at different times of the day or between overall activity levels, can be estimated by means of a Wald test (Wald and Wolfowitz 1940).

Box 8.1 Studies at high latitudes and covering long periods

Suppose we find a nocturnal species that emerges immediately after sunset and a diurnal species which goes to roost just before sunset. Their activity patterns clearly do not overlap. However, in higher latitudes, as the time of sunset changes seasonally, there will be an apparent overlap if the study lasts over longer periods, which is an artefact of pooling the data. Caution is therefore required when the time of sunrise and sunset vary considerably. While this problem is negligible in the tropics and in short studies, variations can be dramatic over longer periods at higher latitudes. Peaks in activity are usually tuned to sunrise and sunset, and the progression of these times therefore flattens peaks and overestimates activity level (Aschoff 1966), and consequently the coefficient of overlap. Care is needed when comparing activity patterns or coefficients of overlap among study areas or periods with significant seasonal differences, and data should not be pooled across such study areas and/or periods unless specific steps are taken to account for this issue (see below for ways of dealing with this). In these specific situations, probability density functions should be fitted to solar time (the deviation of clock time from sunrise and sunset). The function (SunTIME) written by Nouvellet et al. (2012) offers equations and R code to transform clock time data to deviation from sunrise (sunset), based on the location and date at which photographs were taken. This approach works well for simple activity patterns with unimodal distributions but has limitations when the distributions get more complicated. Activity patterns are usually concentrated around dusk and dawn (Aschoff 1966) and thus have a bimodal distribution, though dealing with such distributions is complex. Therefore, we recommend that instead of trying to account for variable day length retrospectively in the analyses, researchers should rather take this aspect into account when planning the study. For example, at higher latitudes where day length changes quickly from one day to the next, it would be better to set more camera traps over a short duration than a few camera traps over a long duration. In these regions it is highly recommended to compare the activity patterns between different study areas solely on surveys conducted roughly at the same latitude and around the same time of year.

Scent-marking with faeces, urine or glandular secretion is widespread among mammals (see Gostling and Roberts 2001a,b for a review). This behaviour has also been observed in different felid (Macdonald 1985) and bear species (in the form of rubbing against trees; see Box 8.3). Although the behaviour is well described, little is known about its function in wild felid and bear populations.

Box 8.2 Scent-marking behaviour of wild Eurasian lynx

We review the study by Vogt et al. (2014) which investigated the pattern of scent-marking and its role in intra- and intersexual communication among resident and non-resident Eurasian lynx (Lynx lynx) (Figure 8.1) by monitoring marking sites by means of infrared camera traps. The study was conducted in a 1,424 km² study area located in the northwestern Swiss Alps, on a population whose spatial and social structure is well known from several previous radiotelemetry studies (Haller and Breitenmoser 1986; Breitenmoser-Würsten et al. 2001) and repeated camera trapping sessions (e.g. Pesenti and Zimmermann 2013; Zimmermann et al. 2013). The sampled marking sites were located along trails and forest roads frequently used by lynx and were previously identified either while setting and checking sites during camera trapping sessions, or during radio- or snow-tracking. Scent-marks are usually placed on visually conspicuous objects (i.e. wood piles, tree trunks, rocks, small spruce trees and the corner of a wooden shed), where lynx hair can be found and urine marks can be smelled, even by humans. Infrared (RC55) and no-glow (PC90) LED Reconyx camera traps, set to take 10 pictures at each trigger with no delay between triggers, were used in this study. Unlike the infrared video trap models available at the time, these camera traps had a fast trigger speed and offered a high enough image quality so that at least one of the multiple pictures per trigger was sufficient to recognise individuals by their fur pattern. Between 4 and 14 camera traps were operating at the same time. Each camera trap was adjusted to observe one marking site (i.e. one repeatedly marked object). From December 2009 to July 2012 a total of 22 marking sites were observed by means of camera trapping. Observational periods for different marking sites ranged from 4 months to 2.5 years. For the analyses of seasonal variation in marking activity data from two periods of comparable length (15 December–14 July 2010/11 and 2011/12) were used. During a total of 7,598 realised trap nights, 338 lynx visits of 40 individuals (19 males, 10 females, 11 unknown) were observed over the course of the study. The results showed that lynx marking activity was highest during the mating season and was lowest during the time when females gave birth and lactated. Both sexes scent-marked, but males visited marking sites more often than females and marked relatively more during visits. Most lynx at marking sites were residents but non-resident lynx also scent-marked occasionally. Juveniles were never observed marking. The presence of another individual’s scent-mark triggered over-marking (i.e. scent-marking on the same area of an object, including scent-marking on top, touching, as well as adjacent to, the previous mark) in male lynx. Males responded similarly to the presence of another individual’s scent-mark irrespective of whether it was the top scent-mark or the underlying scent-mark in a mixture of scent-marks they encountered - over-marking was provoked in both cases. These results suggest that lynx when over-marking do not cover completely the scent-marks left by others so that only the latest scent-mark would be smelt. Marking sites could therefore serve as ‘chemical bulletin boards’, where males advertise their presence and gain information on ownership relationships in a given area. Female lynx over-marked mostly only the scent-marks left by the resident males, but further studies are needed to understand the function of over-marking in females.

Figure 8.1 Typical behavioural sequence of Eurasian lynx scent-marking beginning with sniffing (a) the object, in this case a wood pile, to be marked followed by rubbing of cheeks and neck (b–d), and concluding with urine spraying (e). (KORA)

Box 8.3 Rubbing trees and brown bears

The behaviour displayed by brown bears (Ursus arctos) of rubbing against trees (Figure 8.2) is known to occur across the range of the species and is considered primarily a means of intraspecific communication (e.g. Clapham et al. 2014). Studies in North America have shown that rubbing is performed more by adult male bears and more during the mating season, indicating that the prime communication function is associated with reproductive strategies (Clapham et al. 2014). We here review the study by Tattoni et al. (2015), which showed how camera trapping was an essential tool for studying the use of rubbing trees by brown bears in northeast Italy (Trentino province, Adamello-Brenta Natural Park).

A large (>150) sample of rubbing trees had been previously identified by the local Wildlife Service, which uses hairs left in these trees by rubbing bears as samples to genetically monitor the population. Tattoni et al. (2015) set camera traps in front of a pool of 20 randomly selected rubbing trees to systematically detect passing bears over a period of 3 years (2012–2014), sampling each year the entire season for which bears were active (April–November). The study used UV565HD digital camera traps equipped with an infrared LED flash and set to record video; the cameras recorded continuous, i.e. with no delays between triggers, 20 s videos when triggered. Camera traps were fixed to a tree facing the rubbing tree, at a height of about 2 m and at an average distance of 4 m. The camera traps were checked every 3 weeks for data download and battery replacement.

The study realised 9,302 camera days overall and individual video sequences of bears were concatenated to derive 546 complete events of passing bears, which were then screened for sex/age identification and for classifying the rubbing behaviour. Four categories of rubbing behaviour were identified: (1) ‘indifferent’: the bear passed without considering the tree; (2) ‘investigate’: the animal sniffed or stopped to inspect the tree; (3) ‘rub’: the animal scratched its back or other body parts on the tree; (4) ‘investigate and rub’: when rubbing followed an obvious investigation. Sex could be determined only for adult bears, which comprised events for 37 females and 215 males, while the remaining 294 events were of undetermined sex. Among these, 14 were cubs and 55 sub-adults, leaving the remaining 255 bear events undetermined for age/sex. Table 8.1 reports the observed behaviours according to sex/age classification. Bears which rubbed the trees were predominantly males. The behaviours of ‘rubbing’ and of ‘investigate and rubbing’ were significantly (Pearson’s chi-squared test P < 0.01) more frequent during the breeding season (May–July), while investigation of the trees occurred throughout the period of activity and by both sexes. Females passing rubbing trees were mainly indifferent, though they did, occasionally, investigate; however, only three females were observed to rub, only after investigation and only during the non-breeding season.

Figure 8.2 Adult male brown bear Ursus arctos rubbing against a tree in the Adamello-Brenta Natural Park, northeast Italy. (Francesco Rovero

Table 8.1 Events of rubbing behaviour by age/sex classes, performed by the brown bear (Ursus arctos) in the eastern Alps as detected by camera trapping. From Tattoni et al. (2015).

This study provided new insights into the behaviour of brown bears at rubbing trees and showed that camera trapping was instrumental to revealing aspects that could not have been revealed through other methods, such as hair sampling for genetic analysis. Indeed, camera trapping allowed the sampling of bears belonging to different age/sex classes, hence overcoming the known male-bias of hair sampling. The authors found that adult males were the main performers of rubbing, confirming that rubbing has a primary role of intraspecific communication related to reproductive strategies. Ongoing analysis (C. Tattoni unpublished data) is extending the investigation to the triggering effect that a bear using a rubbing tree has on other individuals. Preliminary results indicate that the rubbing sequences during a span of 90 days at any rubbing tree are triggered by a male bear in 82% of cases, and after that, 43% of the time the second bear was another male (and 28% of the time a bear of undetermined age/sex). This further supports the inter-male communication role of bear rubbing.

We here provide an example of quantitative estimation of the overlap in activity pattern between a predator, the Eurasian lynx, and its two main prey species, roe deer and chamois, and used the data collected during the same survey presented in Chapter 7, where details on study area, season and sampling design are found. The complete data set file can be downloaded as Appendix 8.1.

In order to remove consecutive photographs of the same animal, which may bias the activity estimations towards some overrepresented moments of the day, some data preparation is needed. The raw data were subsampled to delete all consecutive detections of a given species on the same site occurring within less than 30 min. Different analytical routines allow this data preparation, and for further details and a practical example we refer to Chapter 5. Users should be aware that modifying and saving data files with Microsoft Excel may change the format of the data, especially the date and time format. Therefore it is strongly suggested to perform any data preparation directly in R or with a text editor such as Notepad.

8.6.2.2Running the analysis in R (adapted from Meredith and Ridout 2014)

To set the working directory of the R console, type the following command line. This will open an explorer window for easy browsing:

In order to get the current working directory and verify that it has been correctly set:

Choose the file containing the data, in this case ‘true_events_NWA2013_14.csv’.

In order to verify that the file has been correctly imported, you can display the header of the table. Furthermore, this will serve to remind you of the exact column names in your data set, which is crucial for successfully running the functions presented hereinafter.

Before proceeding with the analyses, it is always good to explore the data set. For instance, in order to see from how many different study areas the data come from, type:

If you wish to learn how many species and events per species are available, type:

The data set contains the time of photographic capture from one study area for 15 species and several animals that could not be identified to the species level (e.g. Aves, Ungulata and Martes). The time unit is the day, so values range from 0 to 1. The package overlap works entirely in radian units, but the conversion is straightforward:

Once we have checked the integrity of our data set and collected the most important information, we can start fitting the kernel density. After having loaded the package overlap (Meredith and Ridout 2014), we can extract the data for the Eurasian lynx and plot a kernel density curve:

Figure 8.3 shows the activity pattern. Periods of 3 h before and after midnight (the area in grey) were repeated as a reminder that the activity patterns are circular. The original records are shown at the foot of Figure 8.3 as a ‘rug’.

The degree of smoothing of the density estimation is controlled by the argument adjust of the densityPlot function (values >1, the default value, give a flatter curve, values <1 give a more ‘spiky’ curve) as shown in Figure 8.4. It goes without saying that the choice of adjust affects the estimate of species’ activity overlap.

The area under a density curve is by definition unity. The package overlap uses the coefficient of overlapping proposed by Weitzman (1970). As shown in Figure 8.5, the coefficient of overlapping Δ is the area lying under both of the density curves. Five non-parametric estimators of the coefficient of overlapping were proposed by Schmid and Schmidt (2006). For circular distributions, the first two are equivalent and the third is unworkable (Ridout and Linkie 2009) and thus three were retained in the package overlap:

₁,

₄ and

₅. On the basis of simulations, Ridout and Linkie (2009) recommend using adjust = 0.8 to estimate

₁, adjust = 1 for

₄, and adjust = 4 for

₅. These are the default values for overlap functions. According to the simulations conducted by Ridout and Linkie (2009) and Meredith and Ridout (2014), the best estimator depends on the size of the smaller of the two samples: when the smaller has fewer than 50 records,

₁ performed best, while

₄ is better when the sample size is greater than 75. The coefficient of overlapping

₅ was found not to be useful as it turned out to be unstable – small, incremental changes in the data produce discontinuous changes in the estimate – and can give estimates >1.

We provide a practical example using the northwestern Alps data set. We will first extract the data for the Eurasian lynx and its prey, the roe deer, using the same procedure described above:

Both of the samples have more than 75 observations, so the

₄ estimate, Dhat4 in R, is the most appropriate, giving an estimate of overlap of 0.51.

The best way to estimate the confidence interval of our coefficient of overlapping is to perform a bootstrap analysis. The usual bootstrap method assumes that the existing sample is fully representative of the population and generates a large number of new samples by randomly resampling observations with replacements from the original sample. However, this may not work very well when estimating activity patterns. For example, bootstrapping a nocturnal species with an observed activity range between 20:00 h and 06:00 h will never yield an observation outside that window, whereas a genuine sample from nature may do so. An alternative to strict bootstrapping is smoothed bootstrap. In this case a kernel density curve is fitted to the original data and then random simulated observations are drawn from this distribution. Most simulated observations would fall in the same range as the observed ones, but a few will fall outside. In the overlap package, bootstraps are generated with resample and a smoothing argument can be specified: if smooth = TRUE (the default), smoothed bootstraps are generated.

For this example we will choose 10,000 smoothed bootstrap samples for Eurasian lynx and roe deer:

This produces matrices with a column for each bootstrap sample. The bootstrap sample size is the same as the original sample size. To generate estimates of the overlap from each pair of samples, these two matrices are passed to the function bootEst(). Since the size of the smaller of the two samples is greater than 75 only

₄ should be considered; consequently, the estimation of the others can be suppressed by setting adjust = c(NA, 1, NA), which considerably reduces the computation time.

The function bootEst() takes a while to generate estimates of the overlap from each pair of samples (about 1 min). The values resulting from the simulations may differ slightly; this is due to the random component of the bootstrapping process.

Note that the bootstrap mean,

, differs from

: 0.54 vs. 0.51. The difference,

–

, is the bootstrap bias and needs to be taken into account when calculating the confidence interval. Although the bootstrap bias would have been a good estimator of the original sampling bias, a better estimator of Δ would be

= 2

–

. Simulations show that

results in higher root mean square deviation (RMSE) than the original

, so it is not recommended to apply this correction.

The following estimates of the confidence interval for the Eurasian lynx–roe deer data are obtained after having extracted the right column from the bootstrap matrix:

perc corresponds to the 2.5% and 97.5% percentiles for a 95% confidence interval. As seen above, the bootstrap values differ from the estimates because of the bootstrap bias. Therefore the raw percentiles produced by perc need to be adjusted to account for this bias. The appropriate confidence interval is perc – (

–

) which corresponds to basic0 in the bootCI output. An alternative approach is to consider the standard deviation of the bootstrap results, s_BS, as an estimate of the spread of the sampling distribution and then to calculate the confidence interval as

± z_α/2s_BS. Using z_0.025 = 1.96 gives the usual 95% confidence interval. This corresponds to norm0 in the bootCI output. This procedure assumes that the sampling distribution is normal. If this is the case, norm0 will be close to basic0; however, if the distribution is skewed, as will be the case if

is close to 0 or 1, basic0 is the better estimator. bootCI produces two further estimators: basic and norm. These are analogous to basic0 and norm0 but are intended for use with the bias-corrected estimator,

. The coefficient of overlapping takes a value in the interval [0,1]. All the confidence interval estimators except perc involve additive correction which might result in values outside of this range. bootCIlogit can avoid this problem by carrying out the corrections on a logistic scale and back-transforming.

Based on simulations, Meredith and Ridout (2014) recommend using the basic0 output from bootCI with smoothed bootstraps as the confidence interval. Users should, however, be aware that it will be too narrow for small sample sizes and Δ close to 1.

The overlaps for Eurasian lynx and chamois, including the estimation of the confidence interval, were estimated following the same procedure:

chamois<-timeRad[events$area==1 & events$species=='Rupicapra rupicapra']

min(length(lynx), length(chamois))

[1] 92

lynxchamoisest<-overlapEst(lynx,chamois)

lynxchamoisest

Dhat1	Dhat4	Dhat5
0.4668991	0.4608886	0.4552277

overlapPlot(lynx,chamois, rug=T)

legend('topright', c("Eurasian lynx", "Chamois"), lty=c(1,2), col=c(1,4), bty='n')

chamoisboot<-resample(chamois, 10000)

lynxchamois<-bootEst(lynxboot, chamoisboot, adjust=c(NA,1,NA))

BSMean<-colMeans(lynxchamois)

BSMean

Dhat1	Dhat4	Dhat5
NA	0.4837872	NA

tmp<-lynxchamois[,2]

bootCI(lynxchamoisest[2],tmp)

	lower	upper
norm	0.3400009	0.5359790
norm0	0.3628995	0.5588776
basic	0.3408138	0.5361775
basic0	0.3627010	0.5580647
perc	0.3855996	0.5809633

bootCIlogit(lynxchamoisest[2],tmp)

	lower	upper
norm	0.3442636	0.5370180
norm0	0.3651424	0.5596105
basic	0.3451875	0.5380054
basic0	0.3642222	0.5586023
perc	0.3855996	0.5809633

The coefficient of overlap is purely descriptive and thus does not provide a threshold value below which two activity patterns might be significantly different. The function compareCkern() in the package activity (Rowcliffe 2015) provides a test probability that two sets of circular observations come from the same distribution. The computation takes about 6 min when reps is set to 10,000 (the progress is indicated via a progress bar).

The results of both comparisons therefore support what can be seen in Figure 8.5, i.e. the activity pattern of the Eurasian lynx is significantly different from that of its two main prey species. However, the activity patterns of the two prey species do not differ significantly.

The significance of pairwise comparisons, either between relative activity levels at different times of day or between overall activity levels, can be estimated by a Wald test using the functions compareTimes(fit, times) and compareAct(fits) provided in the package activity (Rowclife 2015), respectively. A circular kernel density on the original data set is fitted by means of the fitact() function. Users should be aware of their sample size, because the coverage of the confidence interval seems to be better estimated with sample = "model" when the sample size is greater than 100–200, whereas smaller sample sizes should be investigated using sample = "data" (see the manual of activity written by Marcus Rowclife). In order to account for the previously mentioned issue of unrepresentative bootstrapping with regards to the true activity of the target species, we will choose the bootstrapping on a model basis for lynx and roe deer given their sample size (>100), which should represent an approach closer to the ‘smooth’ bootstrapping described above (M. Rowclife, personal communication 2015). Furthermore, it is important to notice that the parameter adj of the fitact() function is not the same as the parameter adjust of the densityPlot() and bootEst() functions in the package overlap written by Ridout and Linkie (2009). However, they are tightly linked since they are reciprocal to each other, e.g. if the argument adjust of fitact() equals 2, it will be ½ = 0.5 in the densityPlot() and bootEst() functions (M. Rowclife, personal communication 2015), allowing for straightforward compatibility between the two packages. Since the recommended value of adjust in the densityPlot() function is 1 for

₄ (default value in the function), adj in the fitact() function was set to 1, the reciprocal.

The function fitact() takes a while to fit the circular kernel density to radian time-of-day data and to bootstrap the distribution (about 10 min). The progress of the computation is indicated via a progress bar.

Then the bootstrapped activity patterns of lynx and roe deer are compared by means of a Wald statistic on a chi-square distribution with one degree of freedom, in order to test for significant differences at the 5% level:

The Wald test indicates that the lynx and the roe deer in the western Swiss Alps do not show the same overall activity level. We perform the same analysis for the Chamois:

As the sample size for chamois is <100, the argument sample of the function was set to data (see above).

On the other hand, the overall activity level of both prey species does not differ significantly.

Another interesting investigation would be to compare the activity level of both the predator and the prey for different periods of the day (e.g. day, night, dawn and dusk). This could be performed by means of the function compareTimes (fit, times) of the package activity. It should, however, be stressed that cameras must be randomly distributed over the study area in order to obtain reliable inferences about the true activity pattern of a given species (see Chapters 5 and 6 for details on robust sampling designs). If the camera sites are chosen exclusively on forest roads and hiking trails, as in the present study, the resulting activity refers only to animal movements along roads and trails. Therefore overlap indicates the extent to which two species move on roads and trails at the same period of the day. According to this definition, a browsing roe deer and the lynx stalking it are probably both inactive. As a consequence, conclusions about temporal species interactions need to be drawn with care.

This chapter addressed the use of camera trapping in behavioural studies. We first discussed the advantages and disadvantages of camera trapping compared to other technologies to study animal behaviour (e.g. direct observations of animals in the field, bio-logging, animal-borne video and AVEDs). We highlighted that camera trapping is a valid alternative to other technologies as it combines many of their advantages while offering a number of improvements. Compared to other technologies, it is a highly noninvasive tool that can be applied over relatively large areas, and that potentially provides data for a larger number of individuals and a wider range of species. In recognition of the non-overlapping advantages of these technologies, researchers have started to combine camera trapping with telemetry and direct visual observations. However, camera trapping alone has been applied to a range of behaviour studies such as reproduction, grouping behaviour, social structure, marking behaviour and activity patterns.

We then discussed the importance of choosing the site in relation to a variety of study aims and showed that behavioural studies have capitalised on the ability of camera traps to monitor fixed (i.e. predetermined) locations, where specific behaviours occur. As no study has directly tested the influence of sampling design on behavioural studies, we provided basic recommendations. We suggest that more research is needed to test the effect of sampling design on behavioural studies. Through case studies of the scent-marking behaviour of Eurasian lynx and the tree-rubbing behaviour of brown bears, we showed how camera trapping has the potential to reveal aspects not revealed through other technologies. The last example was chosen to highlight how camera trapping and appropriate analytical methods enable the study of temporal interactions between species. Finally, we are confident that in the near future newly developed video camera traps and the integration of camera trapping with other technologies will allow for enhanced use of this technology in behavioural studies.

Appendix 8.1 Case study comparison of activity patterns: ‘true_events_NWA2013_14.csv’

We are grateful to Clara Tattoni for her analysis of rubbing behaviour by brown bears featured as one of the case studies in this chapter and Marcus Rowcliffe for his help on the use of the R package ‘activity’.

Aschoff, J. (1966) Circadian activity patterns with two peaks. Ecology 47: 657–662.

Baum, W.M. (2013) What counts as behavior? The molar multiscale view. Behavior Analyst 36: 283−293.

Bauer, J.W., Logan, K.A., Sweanor, L.L. and Boyce, W.M. (2005) Scavenging behavior in Puma. Southwestern Naturalist 50: 466–471.

Bleich, V.C., Bowyer, R.T. and Wehausen, J.D. (1997) Sexual segregation in mountain sheep: resources or predation? Wildlife Monographs 134: 1–50.

Boyer-Ontl, K.M. and Pruetz, J.D. (2014) Giving the forest eyes: the benefits of using camera traps to study unhabituated chimpanzees (Pan troglodytes verus) in southeastern Senegal. International Journal of Primatology 35: 881–894.

Breitenmoser-Würsten, C., Zimmermann, F., Ryser, A., Capt, S., Laass, J., Siegenthaler, A. and Breitenmoser, U. (2001) Untersuchungen zur Luchspopulation in den Nordwestalpen der Schweiz 1997–2000. KORA-Report 9. 88 pp. (in German; summary, tables and figures in English and French).

Bridges, A. and Noss, A. (2011) Behavior and activity patterns. In: A.F. O’Connell, J.D. Nichols and K.U. Karanth (eds), Camera Traps in Animal Ecology Methods and Analyses. New York: Springer. pp. 57–69.

Bridges, A.S. Vaughan, M.R. and Klenzendorf, S. (2004) Seasonal variation in American black bear Ursus americanus activity patterns: quantification via photography. Wildlife biology 10: 277−284.

Campbell, G., Kuehl, H., Diarrassouba, A., N’Goran, P.K. and Boesch, C. (2011) Long-term research sites as refugia for threatened and over-harvested species. Biology Letters 7: 723–726.

Carvalho, F., Carvalho, R., Galantinho, A., Mira, A. and Beja, P. (2015) Monitoring frequency influences the analysis if resting behaviour in a forest carnivore. Ecological Research 30: 537–546.

Chen, M.-T., Tewes, M.E., Pei, J.J. and Grassman, L.I. (2009) Activity patterns and habitat use of sympatric small carnivores in southern Taiwan. Mammalia 73: 20−26.

Clapham, M., Nevin, O.T., Ramsay, A.D. and Rosell, F. (2014) Scent-marking investment and motor patterns are affected by the age and sex of wild brown bears. Animal Behaviour 94: 107–116.

Cook, S.J., Hinch, S.G., Wikelski, M., Andrews, R.D., Kuchel, L.J., Wolcott, T.G. and Butler, P.J. (2004) Biotelemetry: a mechanistic approach to ecology. Trends in Ecology and Evolution 19: 334−343.

DeVault, T.L., Brisbin, I.L. and Rhodes, O.E. (2004) Factors influencing the acquisition of rodent carrion by vertebrate scavengers and decomposers. Canadian Journal of Zoology 82: 502–509.

Díaz, M., Torre, I., Peris, A. and Tena, L. (2005) Foraging behavior of wood mice as related to presence and activity of genets. Journal of Mammalogy 86: 1178−1185.

Di Bitetti, M.S., Paviolo, A. and De Angelo, C. (2006) Density, habitat use and activity patterns of ocelots (Leopardus pardalis) in the Atlantic Forest of Misiones, Argentina. Journal of Zoology 270: 153−163.

Di Bitteti, M.S., Di Blanco, Y.E., Pereira, J.A., Paviolo, A. and Pérez I.J. (2009) Time partitioning favors the coexistence of sympatric crab-eating foxes (Cerdocyon thous) and Pampas foxes (Lycalopex gymnocercus). Journal of Mammalogy 90: 479–490.

Di Bitteti, M.S., De Angelo, C.D., Di Blanco, Y.E. and Paviolo, A. (2010) Niche partitioning and species coexistence in a Neotropical felid assemblage. Acta Oecologia 36: 403–412.

Enderson, J.H., Temple, S.A. and Swartz, L.G. (1972) Time-lapse photographic records of nesting peregrine falcons. Living Bird 11: 113−128.

Foster, V.C., Sarmento, P., Sollmann, R., Tôrres, N., Jácomo, A.T.A., Negrões, N., Fonseca, C. and Silveira, L. (2013) Jaguar and puma activity patterns and predator-prey interactions in four Brazilian biomes. Biotropica 45: 373−379.

Galvis, N., Link, A. and Di Fiore, A. (2014) A novel use of camera traps to study the demography and life history in wild animals: a case study of spider monkeys (Ateles belzebuth). International Journal of Primatology 35: 908–918.

Gerber, B.D., Karpanty, S.M. and Randrianantenaina, J. (2012) Activity pattern of carnivores in the rain forests of Madagascar: implications for species coexistence. Journal of Mammalogy 93: 667–676.

Gosling, L.M. and Roberts, S.C. (2001a) Scent-marking by male mammals: cheat-proof signals to competitors and mates. Advances in the Study of Behavior 30: 169–217.

Gosling, L.M. and Roberts, S.C. (2001b) Testing ideas about the function of scent marks in territories from spatial patterns. Animal Behaviour 62: F7–F10.

Gužvica, G., Bošnjak, I., Bielen, A., Babić, D., Radanović-Gužvica, B. and Šver, L. (2014) Comparative analysis of three different methods for monitoring the use of green bridges by wildlife. PLoS ONE 9: e106194.

Haller, H. and Breitenmoser, U. (1986) Zur Raumorganisation der in den Schweizer Alpen wiederangesiedelten Population des Luchses Lynx lynx. Zeitschrift für Säugetierkunde 51: 289–311 (in German).

Hayward, M.W. and Slotow, R. (2009) Temporal partitioning of activity in large African carnivores: tests of multiple hypotheses. South African Journal of Wildlife Research 39: 109−125.

Hebblewhite, M. and Haydon, D.T. (2010) Distinguishing technology from biology: a critical review of the use of GPS telemetry data in ecology. Philosophical Transactions of the Royal Society B – Biological Sciences 365: 2303−2312.

Huang, Z.-P., Qi, X.-G., Garber, P.A., Jin, T., Guo, S.-T., Li, S. and Li, B.-G. (2014) The use of camera traps to identify the set of scavengers preying on the carcass of a golden snub-nosed monkey (Rhinopithecus roxellana). PLoS ONE 9: e87318.

Hutchinson, G.E. (1978) An Introduction to Population Ecology. New Haven, CT: Yale University Press.

Jácomo, A.T.A., Silveira, L. and Diniz-Filho, J.A.F. (2004) Niche separation between the maned wolf (Chrysocyon brachyurus), the crab-eating fox (Dusicyon thous) and the hoary fox (Dusicyon vetulus) in central Brazil. Journal of Zoology 262: 99−106.

Juillard, M. (1987) La photographie sur pellicule infrarouge, une méthode pour l'étude du régime alimentaire des oiseaux cavicoles. Terre et Vie 84: 223−287.

Kamler, J.F., Johnson, A., Vongkhamheng, C. and Bousa, A. (2012) The diet, prey selection, and activity of dholes (Cuon alpinus) in northern Laos. Journal of Mammalogy 93: 627–633.

Kays, R., Tilak, S., Crofoot, M., Fountain, T., Obando, D., Ortega, A., Kuemmeth, F., Mandel, J., Swenson, G., Lambert, T., Hirsch, B. and Wikelski, M. (2011) Tracking animal location and activity with an automated radio telemetry system in a tropical rainforest. Computer Journal 54: 1931−1948.

Knight, J. (2009) Making wildlife viewable: habituation and attraction. Society and Animals 17: 167−184.

Kronfeld-Schor, N. and Dayan, T. (2003) Partitioning of time as an ecological resource. Annual Review of Ecology Evolution and Systematics 34: 153–181.

Kuijper, D.P.J., Verwijmeren, M., Churski, M., Zbyryt, A., Schmidt, K., Jedrzejewska, B. and Smit, K. (2014) What cues do ungulates use to assess predation risk in dense temperate forests? PLoS ONE 9: e84607.

Leuchtenberger, C., Zucco, C.A., Ribas, C., Magnusson, W. and Mourão, G. (2014) Activity patterns of giant otters recorded by telemetry and camera traps. Ethology Ecolgy & Evolution 26: 19–28.

Linkie, M. and Ridout, M.S. (2011) Assessing tiger–prey interactions in Sumatran rainforests. Journal of Zoology 284: 224–229.

Löttker, P., Rummel, A., Traube, M., Stache, A., Šustr, P., Müller, J. and Heurich, M. (2009) New possibilities of observing animal behaviour from a distance using activity sensors in GPS-collars: an attempt to calibrate remotely collected activity data with direct behavioural observations in red deer Cerphus elaphus. Wildlife Biology 15: 425–434.

Lucherini, M., Reppucci, J., Walker, R., Villalba, M., Wurstten, A., Gallardo, G., Iriarte, A., Villalobos, R. and Perovic, P. (2009) Activity pattern segregation of carnivores in the high Andes. Journal of Mammalogy 90: 1404–1409.

Macdonald, D.W. (1985) The carnivores: order carnivora. In: R.E Brown and D.W. Macdonald (eds), Social Odours in Mammals. Oxford: Clarendon Press. pp. 619–722.

MacKenzie, D.I., Bailey, L.L. and Nichols, J.D. (2004) Investigating species co-occurrence patterns when species are detected imperfectly. Journal of Animal Ecology 73: 546–555

Major, R.E. and Gowing, G. (1994) An inexpensive photographic technique for identifying nest predators at active nests of birds. Wildlife Research 21: 657–666.

Meek, P.D., Ballard, G.-A., Fleming, P.J.S., Schaefer, M., Williams, W. and Falzon, G. (2014a) Camera traps can be heard and seen by animals. PLoS ONE 9: e110832.

Meek, P.D., Ballard, G., Claridge, A., Kays, R., Moseby, K., O’Brien, T., O’Connell, A., Sanderson, J., Swann, D.E., Tobler, M. and Townsend, S. (2014b) Recommended guiding principles for reporting on camera trapping research. Biodiversity and Conservation 23: 2321–2343.

Moll, R.G., Millspaugh, J.J., Beringer, J., Startwell, J. and He, Z. (2007) A new ‘view’ of ecology and conservation through animal-borne video systems. Trends in Ecology and Evolution 22: 660–668.

Monterroso, P., Alves, P.C. and Ferreras, P. (2014) Plasticity in circadian activity patterns of mesocarnivores in Southwestern Europe: implications for species coexistence. Behavioral Ecology and Sociobiology 68: 1403−1417.

Mori, E., Menchetti, M. and Balestrieri, A. (2015) Interspecific den sharing: a study on European badger setts using camera traps. Acta Ethologica 18: 121–126.

Miura, S., Yasuda, M. and Ratnam, L.C. (1997) Who steals the fruits? Monitoring frugivory of mammals in a tropical rain-forest. Malayan Nature Journal 50: 183–193.

Nichols, J.D., O’Connell, A.F. and Karanth, K.U. (2011) Camera traps in animal ecology and conservation: what’s next? In: A.F. O’Connell, J.D. Nichols and K.U. Karanth (eds), Camera Traps in Animal Ecology Methods and Analyses. New York: Springer. pp. 253–263.

Nouvellet, P., Rasmussen, G.S.A., Macdonald, D.W. and Courchamp, F. (2012) Noisy clocks and silent sunrises: measuremnt methods of daily activity pattern. Journal of Zoology 286: 179–184.

Núñez, R., Miller, B. and Lindzey, F. (2000) Food habits of jaguars and pumas in Jalisco, Mexico. Journal of Zoology 252: 373–379.

Pesenti, E. and Zimmermann, F. (2013) Density estimation of Eurasian lynx Lynx lynx in the Swiss Alps. Journal of Mammalogy 94: 73–81.

Ramesh, T., Kalle, R., Sankar, K. and Qureshi, Q. (2012) Spatio-temporal partitioning among large carnivores in relation to major prey species in Western Ghats. Journal of Zoology 287: 269–275.

Ridout, M.S. and Linkie, M. (2009) Estimating overlap of daily activity patterns from camera trap data. Journal of Agricultural, Biological, and Environmental Statistics 14: 322–337.

Ropert-Coudert, Y. and Wilson, R.P. (2005) Trends and perspectives in animal-attached remote sensing. Frontiers in Ecology and the Environment 3: 437–444.

Ross, J., Hearn, A.J., Johnson, P.J. and Macdonald, D.W. (2013) Activity patterns and temporal avoidance by prey in response to Sunda clouded leopard predation risk. Journal of Zoology 290: 96–106.

Rowcliffe, J.M., Field, J., Turvey, S.T. and Carbone, C. (2008) Estimating animal density using camera traps without the need for individual recognition. Journal of Applied Ecology 45: 1228–1236.

Rowcliffe, J.M., Kays, R., Kranstauber, B., Carbone, C. and Jansen, P.A. (2014) Quantifying levels of animal activity using camera trap data. Methods in Ecology and Evolution 5: 1170–1179.

Royama, T. (1970) Factors governing the hunting behaviour and selection of food by great tit (Parus major L.). Journal of Animal Ecology 39: 110–111.

Schmid, F. and Schmidt, A. (2006) Nonparametric estimation of the coefficient of overlapping – theory and empirical application. Computational Statistics and Data Analysis 50: 1583–1596.

Story, G., Driscoll, D. and Banks, S. (2014) What can camera traps tell us about the diurnal activity of the nocturnal bare-nosed wombat (Vombatus ursinus)? In: P.D. Meek, A.G. Ballard, P.B. Banks, A.W. Claridge, P.J.S. Fleming, J.G. Sanderson and D.E. Swann (eds), Camera Trapping Wildlife Management and Research. Collingwood, Australia: CSIRO Publishing. pp. 35–43.

Séquin, E.S., Jaeger, M.M., Brussard, P.F. and Barrett, R.H. (2003) Wariness of coyotes to camera traps relative to social status and territory boundaries. Canadian Journal of Zoology 81: 2015–2025.

Soberon, J. (2007) Grinnellian and Eltonian niches and geographic distributions of species. Ecological Letters 10: 1115–1123.

Suselbeek, L., Emsens, W.-J., Hirsch, B.T., Kays, R., Rowcliffe, J.M., Zamora-Gutierrez, V. and Jansen, P.A. (2014) Food acquisition and predator avoidance in a Neotropical rodent. Animal Behaviour 88: 41–48.

Tattoni, C., Bragalanti, N., Groff, C. and Rovero, F. (2015) Patterns in the use of rub trees by the Eurasian brown bear. Hystrix: the Italian Journal of Mammalogy doi:10.4404/hystrix-26.2-11414.

Taylor, B. and Goldingay, R. (2014) Using camera traps to monitor use of roadside glide poles and rope canopy-bridges by Australian gliding mammals. In: P.D. Meek, A.G. Ballard, P.B. Banks, A.W. Claridge, P.J.S. Fleming, J.G. Sanderson and D.E. Swann (eds), Camera Trapping Wildlife Management and Research. Collingwood, Australia: CSIRO Publishing. pp. 245–252.

Vernes, K., Smith, M. and Jarman, P.J. (2014) A novel camera-based approach to understanding the foraging behaviour of mycophagous mammals. In: P.D. Meek, A.G. Ballard, P.B. Banks, A.W. Claridge, P.J.S. Fleming, J.G. Sanderson and D.E. Swann (eds), Camera Trapping Wildlife Management and Research. Collingwood, Australia: CSIRO Publishing. pp. 215–224.

Vogt, K., Zimmermann, F., Kölliker, M. and Breitenmoser, U. (2014) Scent-marking behaviour and social dynamics in a wild population of Eurasian lynx Lynx lynx. Behavioural Processes 106: 98−106.

Wald, A. and Wolfowitz, J. (1940) On a test whether two samples are from the same population. Annals of Mathematical Statistics 11: 147−162.

Weitzman, M.S. (1970) Measure of the Overlap of Income Distribution of White and Negro Families in the United States. Technical Report No. 22, US Department of Commerce, Bureau of Census, Washington, DC

Zimmermann, F., Molinari-Jobin, A., Ryser, A., Breitenmoser-Würsten, Ch., Pesenti, E. and Breitenmoser, U. (2011) Status and distribution of the lynx (Lynx lynx) in the Swiss Alps 2005–2009. Acta Biologica Slovenica 54: 74–84.

Zimmermann, F., Breitenmoser-Würsten, C., Molinari-Jobin, A. and Breitenmoser, U. (2013) Optimizing the size of the area surveyed for monitoring a Eurasian lynx (Lynx lynx) population in the Swiss Alps by means of photographic capture-recapture. Integrative Zoology 8: 232–243.

	lower	upper
norm	0.3961900	0.5834619
norm0	0.4207476	0.6080195
basic	0.3966518	0.5825330
basic0	0.4216765	0.6075577
perc	0.4462341	0.6321153

	lower	upper
norm	0.3958140	0.5837067
norm0	0.4199636	0.6077879
basic	0.3950336	0.5820023
basic0	0.4216721	0.6085658
perc	0.4462341	0.6321153

library(activity)
compareCkern(lynx, roe, reps = 10000)
Overlap	p
0.5143836	0.0000000
compareCkern(lynx, chamois, reps = 10000)
Overlap	p
0.4608886	0.0000000
compareCkern(roe, chamois, reps = 10000)
Overlap	p
0.8618328	0.5773000