The successful conservation of any species is predicated on our ability to understand its abundance and distribution (Stander 1998, Hussain 2003, Gusset and Burgener 2005, Houser et al. 2009, Trolliet et al. 2014). While some taxa (e.g. birds) have the benefit of being relatively well studied, many species such as the larger mammalian carnivores are notoriously difficult to directly monitor in the field (Stephens et al. 2006). Traditionally, indirect methods such as locating den sites, and scat or spoor (tracks, pugmarks) surveys have been utilised as a highly cost effective method (Gusset and Burgener 2005) to determine the presence or absence, abundance or population density for species such as the leopard Panthera pardus, snow leopard Uncia uncia, cougar Felis concolor, lion Panthera leo, caracal Caracal caracal, tiger Panthera tigris and pine marten Martes martes (Beier and Cunningham 1996, Zalewski 1999, Hussain 2003, Melville and Bothma 2006, Sharma et al. 2005, Houser et al. 2009, Sanei et al. 2011, Sheehy et al. 2014).

It has been suggested that the sex of tracked Panthera spp. and cougar Felis concolor can be judged from size differ enees in spoor (Bothma 1984, Stander et al. 1997, Stander 1998, Sharma et al. 2003, Sanei et al. 2011, Gu et al. 2014); indeed, if enough features are recorded, leopard, snow leopard, tiger, white rhino Ceratotherium simum and black rhino Diceros bicornis individuals may be identified from spoor alone (Stander et al. 1997, Riordan 1998, Jewell et al. 2001, Sharma et al. 2005, Alibhai et al. 2008, Balme et al. 2009). However, using spoor alone to estimate tiger densities in the field has been criticised for giving unreliable results, as simple measurements or analyses fail to provide the discrimination required (Karanth et al. 2003). Established methods were to take plaster casts or tracings of spoor on acetate from which measurements were taken (Lewison et al. 2001). Sharma et al. (2005) highlighted that data gathered in this way were of poor quality and inconsistently collected.

Most feline spoor studies have been investigated in snow (Hayward et al. 2002), tropical (Sanei et al. 2011) or clay substrates (Garcia et al. 2010), sandy loam (Lewison et al. 2001), or thick soil (Sharma et al. 2005), which are often more than 5 mm in depth. Lewison et al. (2001) found linear measurements from tracings to be more accurate in thick sandy loam. However it is not always possible to have these optimum substrate depths in more arid environments which often contain sandy substrates overlaying hard ground (but see Jewell et al. 2001). Such habitats are typical of much of the range of species such as leopard in parts of eastern and southern Africa.

Sharma et al. (2005) also found that a substrate depth of 0.5–1 cm was most effective for detecting tiger spoor and gathering data in the form of tracings and photographs, which were scanned and measured utilising computer software; a technique which has later been used for black rhino, white rhino and puma with a good level of accuracy (Jewell et al. 2001, Alibhai et al. 2008, Garcia et al. 2010). Therefore, while detailed spoor measurements can allow individual identification of some species, simple spoor measurements are considered unlikely to be reliable in determining individual identification. However, it is not clear if such measurements would allow the tracking of individuals in a single tracking period where we are more concerned with our ability to follow single individuals. If this were so, then a set of spoor that is broken due to unsuitable terrain or crossed by another individual could be reliably followed, allowing further information to be gleaned by relatively untrained fieldworkers on the movement patterns and habitat utilisation of the species of interest.

More recently, the widespread availability of camera trap technology has greatly changed the approach taken to monitoring larger terrestrial mammals of conservation concern across the world (e.g. tigers, Karanth and Nichols 1998; leopard, Trolle and Kery 2005; snow leopard, Jackson et al. 2006; Rowcliffe and Carbone 2008; ocelot Leopardus pardalis, Trolliet et al. 2014). Camera traps offer the benefit of allowing species identification with relative ease, and in identifying individuals for species which show variation in pelage markings or natural features (Karanth and Nichols 1998, Trolle and Kery 2005, Jackson et al. 2006, Negroes et al. 2012, Maputla et al. 2013, Pirie et al. 2014). They can have a significant advantage over indirect means of recording individuals, often allowing the population size of cryptic species (notably the large cats) to be estimated using capturerecapture models (Karanth 1995, Karanth and Nichols 1998, Trolle and Kery 2005, Jackson et al. 2006, Balme et al. 2009, Royle et al. 2009, Chapman and Balme 2010, Negroes et al. 2012, Maputla et al. 2013, Tobler and Powell 2013).

Although abundance can be estimated relatively accurately when using camera traps (Chapman and Balme 2010) providing that the probability of detection is high, cameras are positioned appropriately and camera avoidance is low (Maputla et al. 2013), there remains a lack of a standardised method of camera trap-based mark—recapture (Kelly 2008) and issues such as camera performance and efficacy have still not been satisfactorily addressed (Maputla et al. 2013, Urlus et al. 2014). However, it is detectability which presents the greatest challenge in effectively sampling the species abundance in a surveyed area (Royle and Nichols 2003) as little is known about how species vary in their likelihood of being recorded by camera traps (Balme et al. 2009, Ballard et al. 2014). In most studies the proportion of individuals which enter the camera trap range and fail to trigger the unit is unknown, resulting in under-estimation of distribution and/ or abundance.

One approach is to compare camera trap results with spoor. Lyra-Jorge et al. (2008) found that compared to spoor, film cameras with a trigger delay of one second collectively under recorded by 1.65 times puma, Puma concolor; maned wolf, Chrysocyon brachyurus; mazama, Mazama sp; striped hog-nosed skunk, Conepatus semistriatus; armadillo, Dasypus sp. and forest rabbit, Sylvilagus brasiliensis. In such studies, spoor is usually recorded in sand traps, which maximises the likelihood of successfully recording a passing individual. In field work such perfect recording opportunities are rare and conditions often affect ability to record spoor (Lyra-Jorge et al. 2008, Balme et al. 2009).

In this paper we report the results of a study of large to medium mammalian predators of the African savannah, where we explore 1) the effectiveness of camera traps compared to spoor recorded in natural settings by trained amateur trackers for a variety of predators ranging in size from leopard to Meilers mongoose Rhynchogale melleri and 2) if observer accuracy in recording measurements and substrate depth may limit the utility of simple measurements of spoor in arid environments to track individual animals. We studied the latter using two approaches; first using a cast of a leopard print, we investigated inter-observer variation in spoor measurements. Second, using one observer following three leopards in two depths of fine sand substrate types we measured intra-observer variation in spoor measurements.

Methods

Study location

The study took place in July and August 2013 and was conducted at Thaba Tholo Wilderness Reserve, Mpumalanga, South Africa (24°57′40.4″S, 30°2l′10.5″E, Fig. 1). The reserve was established in 2002 as a 1500-ha privately owned game reserve, rehabilitating land previously used for cattle with small areas used for cultivation, evidence of which still remains. In 2009 the boundary was increased to 5400 ha and it is now run as a commercial reserve with South African giraffe Giraffa camelopardalis giraffe, plains, bush and mountain antelope and a number of carnivores, the largest being the brown hyena Hyaena brunnea and leopard.

Figure 1.

Map of Thaba Tholo Wilderness Reserve, South Africa showing the four camera sites which were surveyed and their relation to base camp. (Google Maps 2014 and QGIS 2.6.1).

The area is situated between the Steenkampsberg and Mauchsberg mountain ranges and lies on the cusp of two major biomes formally classified as savannah in the valleys and northern section of the reserve and grassland on top of the mountains in the southern section of the reserve. Altitudes range from 1100–2000 m and it has an average annual summer rainfall of 700–900 mm falling mainly October— February. Rock types include granite, gneiss and sandstone in the mountains with mudstone, sandstone, quartz, shale and gneiss in the valleys. The vegetation is mixed veld, predominantly Vachellia spp., Combretum spp. and Commiphora spp, Themeda triandra, Hyperthelia spp. and Cymbopogan spp.

Spoor surveys

Spoor surveys were carried out for 38 consecutive days on foot during the early morning, by a minimum of two people who had received spoor recognition training for a week prior to the survey. This is the best time of day to locate and view tracks as the angle of the sun creates shadows in the spoor (Liebenberg 2005). In addition, recording spoor early in the morning reduced the possibility that the movement of diurnal species and reserve vehicles would damage the spoor of nocturnal animals.

Spoor of leopard, brown hyena, African civet, Meiler's mongoose, genet spp. and jackal spp. were recorded. Genet species cannot be distinguished by spoor so were grouped and there are difficulties in separating black-backed jackal Canis mesomelas and side-striped jackal Canis adustus so these were also grouped (Liebenberg 2005, Gutteridge and Liebenberg 2013). Carnivores were studied owing to the relative ease of species spoor recognition compared to that of herbivores. In addition, their generally solitary nature reduced confusion between spoor of different individuals. In order to check for camera shyness, 25 m either side of each camera trap were surveyed for spoor.

Once recording was complete, the whole survey area was swept clean of all prints by walking from one end of the area to the other, using a feather duster to sweep the whole road; ensuring spoor found the following day was fresh. The substrate was otherwise left in a natural state, open to all elements such as wind and dew, which can affect substrate conditions (Alibhai et al. 2008, Lyra-Jorge et al. 2008). Spoor which was situated within the range of the cameras took priority and was marked as being at the camera. Spoor outside this area was marked as not at the camera.

Figure 2.

Representative image of carnivore spoor taken for confirmation of species identification with measurement zeroed at the back of the metatarsal or metacarpal.

A photograph was taken with a Nikon D3100 SLR for each carnivore spoor located at all sites, with a standard mm ruler zeroed from the back point of the metatarsal or metacarpal pad in the frame for size reference (Fig. 2). The thickness of substrate was recorded as being below 2 mm or over 2 mm (rarely more than 5 mm). Only complete and clear spoor were recorded. Partial prints were not recorded to avoid misidentification. Species identification from spoor recorded in the field were checked by observers using Liebenberg (2005) and re-checked from the image by an experienced tracker. For both camera trap and spoor surveys, presence of a given species was logged once per night per location, irrespective of the number of spoor and/or images recorded.

Statistical analysis

All analyses were conducted using Mini tab 16 and R (< www.r-project.org>). To ensure that cameras were> optimally placed and that we had no evidence of animals avoiding camera trap locations we compared the number of spoor outside the camera trapping area with numbers located inside using a G-test. The effectiveness of the cameras in detecting each species compared to spoor was analysed using a paired t-test. In order to meet assumptions of normality, + 1 was added to the data then Log₁₀ transformed prior to analysis. Detection rate was calculated by dividing number of total images or spoor by number of hours of exposure (840 h; following Lyra-Jorge et al. 2008).

Figure 3.

Spoor size was recorded in four dimensions. Full spoor length was taken as the tip of the longest toe to the furthest point of the tarsal pad (A), widest part of spoor (B), length of hind pad (C), width of hind pad (D).

We investigated a possible relationship between the size and length of the study species and under-recording by camera trap compared to spoor located using the percentage difference between the two methods with a one-tailed Pearsons correlation.

A Friedman test was used to analyse the spoor measurements taken by four different observers from the same track and MANOVA was used to analyse the measurements taken by a single observer from three different animal trails in two substrate thicknesses in the field. Values for full spoor width were analysed following Box—Cox transformation and hind pad width following Johnson transformation to ensure data met assumptions of normality.

Table 1.

Number of spoor recorded Inside and outside of the camera trap area for each species during the study.

Table 2.

Number of recordings per species during the study period as camera trap images alone, spoor alone, or both images and spoor.

Results

Discussion

The ability to document the presence (and potentially abundance) of species of conservation concern is of great interest to field ecologists (Karanth and Nichols 1998, Trolle and Kery 2005, Jackson et al. 2006). Understanding the limitations of the main means of recording these species is therefore of considerable importance. In this study of mammalian carnivores inhabiting savannah habitat in South Africa, we found that camera traps significantly under-recorded the number of animals passing a trapping area when compared with those identified using spoor. This under-recording ranged from 22% for brown hyena to 64% for civet and there is a suggestion that animal size affected the likelihood of being recorded. This study illustrates that spoor can provide us with an opportunity to calibrate camera traps. However, the ability to detect and/or identify spoor is affected by the tracker's expertise as well as ground characteristics (Alibhai et al. 2008, Lyra-Jorge et al. 2008) which do need to be taken into consideration, but could be controlled where necessary.

Figure 4.

Sum of images of each study species recorded inside the camera trap area by spoor or camera trap images.

Figure 5.

Relationship between study species body length (cm) and likelihood of being under-recorded by camera trap (considering the percentage difference between spoor and image records) in this study.

We found that inter-observer variation in the simple measurements recorded in our study meant that spoor size data collected by amateur volunteers are unlikely to provide useful data beyond the presence/absence of leopards. However, experienced researchers were more consistent in measuring spoor, although spoor size was influenced by substrate, as well as other factors such as terrain and speed of the animal's movement (Riordan 1998, Liebenberg 2005, Sanei et al. 2011). We found that mean leopard spoor size was significantly larger in substrate with a depth of > 2 mm when compared with those recorded in < 2 mm, which may result from the greater displacement of the toes from the metatarsal pad as the foot pushes further into the substrate. Full spoor length was significantly different between animals, which was more apparent in substrates > 2 mm depth, although we advise that measurements taken directly in field are best done in < 2 mm sandy substrate as this reduces variance. This is in contrast to Lewison et al. (2001) and Sharma et al. (2005), who found thicker soil to be more beneficial for tracings and photographs, especially for digital measurements. Therefore, while simple spoor measurements collected directly in the field in sandy soil may not reliably allow the identification of individuals, simple spoor measurements of the leopard of interest can be taken, so that where the trail is broken or crossed by another individual, size measurements provide evidence that the focal leopard's trail has been relocated.

Table 3.

Inter-quartile and median artificial spoor dimensions (mm; Fig. 3) as recorded by four observers. Values for Friedman analyses are provided below each spoor dimension.

Table 4.

MANOVA values for each spoor measurement (mm; Fig. 3) for three leopard hind left spoor sets taken in the field. None of the interaction terms was significant.

Images can be recorded of the spoor for computer analysis as described by Sharma et al. (2005). The lack of optimum recording substrate could be a limitation in more arid habitats, although research conducted on white rhino spoor has been undertaken in similar environments (Alibhai et al. 2008). Nevertheless under the right circumstances, spoor may still provide additional useful information on individual leopards in terms of movement and behaviour which will prove useful in developing conservation strategies.

In contrast, camera traps readily allow identification of species and individuals when variable pelage markings are evident, and often their general size or sex can be determined. Increasing the likelihood of detecting focal species is extremely important to avoid under-recording (Maputla et al. 2013). Our work suggests that this could be a more significant factor to consider when utilising camera traps than perhaps previously thought.

Trolle and Kery (2005) found dirt roads to be more effective than game trails in capturing carnivores on camera, so we are likely to have maximised our capture rate, but evidently many individuals were not recorded by the camera traps. While the results for leopard are based on a small sample, results from our long term study (Pirie, Thomas and Fellowes unpubl.) also suggest camera traps record the presence of this species approximately twice as often as spoor. All four sites were picked partly because of the ideal substrate conditions, although these can alter between days due to changing temperatures, humidity, rainfall and wind strength (Alibhai et al. 2008, Lyra-Jorge et al. 2008). On two occasions, leopard spoor was registered outside of the camera trap zone and the leopard was photographed by the camera trap, but no complete leopard spore was detected within the camera trap zone. This is likely due to slight variations in the substrate within the camera trap zone. Partial spoor was not recorded to avoid misidentification, so was not recorded as being inside the trap zone.

Table 5.

Mean and range (mm) of spoor measurements (Fig. 3) for three leopards hind left spoor, recorded under two substrate conditions In the field.

Overall, under-recording appears to occur more frequently with smaller species. Previous studies (Kelly 2008, Lyra-Jorge et al. 2008, but see Negroes et al. 2012) support this view. In contrast, Urlus et al. (2014) found that larger Australian mammals were more likely to be under-recorded by camera traps. It is likely that this is as a result of the much greater size difference in their study, with smaller mammals having a more restricted home range around the traps, and their use of bait stations to attract smaller species (Urlus et al. 2014). Trigger time, the interval between sensors detecting movement and an image being taken, may also play a role in under recording given that small animals may not trigger the sensors or move more rapidly out of the capture zone once the sensors have been activated. Negroes et al. (2012) found having two cameras per station was on average 29% more effective than a single camera based on photographic rate. In addition, camera traps can vary greatly in their likelihood of capturing different species (Urlus et al. 2014, Swan et al. 2014) and understanding how equipment varies is of considerable importance.

Assessing predator abundance and distribution in savannah remains a challenge. We argue that the traditional approach of using spoor and the skills associated with tracking in the field still have a place in monitoring the larger mammalian predators of savannah habitat in two key ways. First, spoor allows us to evaluate how effective camera traps are at recording the study species; our study species were 37% more likely to be recorded by spoor than by our camera traps. This study also supports the view that camera trap studies should be calibrated against other methods (Gompper et al. 2006, Balme et al. 2009) in order to gain more accurate data to determine actual density, relative abundance and movements of species. Camera traps are beneficial but their reliability in capturing each animal that triggers the sensors is uncertain and depends on the camera model, habitat and as we show, species. Second, while it is unlikely that simple direct spoor measurements can be used to reliably identify unknown leopards due to interobserver variation, such spoor measurements enable known individuals to be followed by experienced field trackers with some certainty, allowing researchers to link records at static camera trap locations with the wider movements of an individual. We therefore suggest that while camera traps are an exceptionally helpful addition to the ecologist's armoury, we should take care to ensure that the benefits of good field craft are not neglected.